The bacterial strains and plasmids used in this work are listed in Table 1. K. pneumoniae, K. aerogenes and E. coli DH5α strains were routinely grown at 37°C with shaking (180 rpm) in Luria-Bertani broth (10 g/L triptone, 5 g/L yeast extract, 5 g/L NaCl) until late-exponential phase of growth. K. pneumoniae ATCC 700603 and ATCC BAA-1705 were kindly provided by Dr. Roberto Vidal from Facultad de Medicina, Universidad de Chile. For fluorescence labeling, bacterial strains were transformed by electroporation with pFCcGi, allowing the constitutive expression of the mCherry protein (Figueira et al., 2013; Addgene Plasmid #59324), or with pFCcGi-TMP. This last plasmid is a pFCcGi-derivative constructed by our group, where a dihydrofolate reductase gene expression cassette (conferring resistance to trimethoprim) was inserted in the PvuI site, interrupting the beta-lactamase gene.

D. discoideum strains AX4 (DBS0302402) and AX2 vatM-GFP (DBS0235537) were obtained from Dicty Stock Center (Kreppel et al., 2004; Basu et al., 2013; Fey et al., 2013), and cultured according to standard protocols (Fey et al., 2007). Briefly, D. discoideum strains were maintained at 23°C in SM medium (10 g/L glucose, 10 g/L peptone, 1 g/L yeast extract, 1 g/L MgSO₄ × 7H₂O, 1.9 g/L KH₂PO₄, 0.6 g/L K₂HPO₄, 20 g/L agar), growing on a confluent lawn of Klebsiella aerogenes DBS0305928. Before the assays, amoebae were grown at 23°C with agitation (180 rpm) in liquid HL5 medium (14 g/L tryptone, 7 g/L yeast extract, 0,35 g/L Na₂HPO₄, 1,2 g/L KH₂PO₄, 14 g/L glucose, pH 6,3) in the absence of bacteria (axenic cultures). When required, media were supplemented with streptomycin (300 mg/L), ampicillin (100 mg/L) or G418 (geneticin, 10 mg/L). Amoebae were harvested in early exponential phase (1–2 × 10⁶ cells/mL) and centrifuged at 500 × g for 5 min. Prior to the infection assays, the supernatant was discarded and the pellet was washed three times using Soerensen buffer (2 g/L KH₂PO₄, 0.36 g/L Na₂HPO₄ × 2H₂O, pH 6.0) or adjusted to 5 × 10⁷ cells/mL in HL5 medium for the social development assays. Determination of viable amoeba cells was performed by Trypan blue exclusion and counting in a Neubauer chamber.

For social development assays, overnight cultures of each bacterial strain (30 μL) were homogeneously included per well of a 24-well plate containing N agar (1 g peptone, 1 g glucose, 20 g agar in 1 L of 17 mM Soerensen phosphate buffer) (Sillo et al., 2011) and grown overnight at 23°C. A drop of a cellular suspension corresponding to 5 × 10⁵ D. discoideum AX4 cells in HL5 was spotted in the middle of each well and the plates further incubated at 23°C for 5 days. The social development of amoebae was monitored daily during 5 days and the phase reached was scored, being classified as “aggregation,” “elevation” and “culmination.” A score of “1” was assigned when amoebae aggregated forming a phagocytosis plate, “2” when across all the well surface elevated structures such as worms or fingers were observed, and “3” when fruiting bodies were formed across all the well surface. Transitions among any of these three phases were scored with half of the value corresponding to the closest next stage. Additionally, images of social development were obtained at days 1, 2, and 3 by using an Olympus MVX10 stereomicroscope with a total magnification of 20X.

D. discoideum AX2 vatM-GFP grown axenically (~1 × 10⁷ cells) was co-incubated with each bacterial strain at 23°C with agitation (180 rpm) in 5 mL of Soerensen buffer using a multiplicity of infection (MOI) of 50 bacteria/amoeba. To evaluate the intracellular survival, amoebae were washed three times with Soerensen buffer to remove extracellular bacteria after 1 h of co-incubation. Then, infected cells were suspended in 5 mL of Soerensen buffer (t = 0) and further incubated at 23°C with agitation. Aliquots were obtained at 0, 1.5, 3, 6, and 24 h post-infection. Viable amoebae were determined at each time point. In parallel, infected amoebae recovered at each time point were treated for 10 min with 20 μg/mL gentamycin to remove extracellular bacteria, washed with Soerensen buffer, and lysed with 0.2% Triton X-100. Titers of intracellular bacteria were determined by serial dilutions and plating on LB agar. Statistical significance was determined using a one-way ANOVA and two-way ANOVA with a Fisher's LSD post-test, as described previously (Riquelme et al., 2016). Also, aliquots of the mixed cultures were mounted for laser-scanning confocal microscopy analysis.

To quantify extracellular bacteria, D. discoideum AX2 vatM-GFP (1 × 10⁷ cells) and mCherry-labeled bacteria were co-incubated at 23°C for 24 h in 5 mL of Soerensen buffer using a MOI of 100 bacteria/amoeba. Aliquots were obtained at 0, 3, and 24 h post-infection. Amoebae were removed by centrifugation at 500 × g for 5 min and extracellular bacteria were determined at each point by serial dilutions and plating on LB agar. Statistical significance was determined using a two-way ANOVA with Fisher's LSD post-test. In parallel, aliquots of the mixed cultures were extracted and mounted for laser-scanning confocal microscopy analysis.

Images of infected amoeba cells from phagocytosis assays were acquired using a Zeiss LSM 700 laser scanning confocal microscope equipped with an Alpha Plan Fluar 100X/1.45 Oil M27 optic setup. Prior to observation, cells were mounted on a thin layer of 1.5% agarose in PBS buffer and deposited on a glass slide. To visualize GFP-associated fluorescence (amoebae), the sample was excited at 488 nm with an argon laser and emission was detected using a filter in the 493–549 nm range. To visualize mCherry-associated fluorescence (bacteria), the sample was excited at 543 nm with a HeNe laser and emission was detected using a filter in the 548–679 nm range. Images were acquired using the ZEN 2010 software (Zeiss), and analyzed using Fiji and ImageJ softwares (Schindelin et al., 2012; Schneider et al., 2012), as described previously (Riquelme et al., 2016). Microphotographs of the phagocytosis assays are presented with a total magnification of 550X. Images of capsule-stained K. pneumoniae were obtained using a Zeiss LSM 700 laser scanning confocal microscope equipped with an Alpha Plan Fluar 100X/1.45 Oil M27 optic setup, and a laser-mediated bright field mode. Microphotographs of stained K. pneumoniae cells are presented with a total magnification of 3000X (Figure 6) or of 2000X (Supplementary Figure 4).

Zebrafish (Danio rerio) embryos were obtained by natural spawning of Tab5 (wild type), Tg(BACmpo:gfp) (Renshaw et al., 2006) and Tg(fli1:egfp) lines (Lawson and Weinstein, 2002). Fertilized eggs were raised in Petri dishes containing E3 medium (5 mM NaCl, 0.17 mM KCl, 0.33 mM CaCl₂, 0.3 mM MgSO4) and 0.1% methylene blue until 3 days post-fertilization (dpf). All procedures complied with national guidelines of the Animal Use Ethics Committee of the University of Chile and the Bioethics Advisory Committee of FONDECYT-CONICYT (the national funding agency for this work).

Bacterial injection experiments were adapted from previously described methods to evaluate S. Typhimurium virulence (Díaz-Pascual et al., 2017; Varas et al., 2017a,b). For the injection assays, the bacterial cultures were washed and subsequently suspended in phosphate-buffered saline (PBS) and concentrated 5-fold (K. pneumoniae strains) or 7.5-fold (E. coli DH5α). Zebrafish larvae (3-dpf) were anesthetized with 4.2% tricaine and mounted on 1% low melting point agarose. The zebrafish were kept under anesthesia during the whole injection procedure. 1–5 nL (15,000–18,000 CFU) were injected in the circulatory system via the caudal artery (to evaluate survival), or in the otic vesicle (to evaluate neutrophil recruitment and bacterial burden). To determine the CFU injected, the droplets were also diluted in sterile PBS and plated on LB-agar. Infected larvae were transferred into 6-well plates containing fresh E3 medium and kept at 28°C for 72 h. Death events were quantified every 24 h, and survival curves were constructed and analyzed by means of the Kaplan-Meier method using the GraphPad Prism 6.0 software. Survival experiments were performed in triplicate for a total of 57 larvae per condition, while neutrophils recruitment experiments were done in triplicate for a total of 21 larvae per condition.

Recruitment of neutrophils to the otic vesicle was observed using the Tg(BACmpo:gfp) zebrafish line. The larvae were observed using an Olympus MVX10 fluorescence stereomicroscope at 24 h post-injection (hpi). To measure neutrophil recruitment and the bacterial burden, injected zebrafish larvae were anesthetized with 0.1% tricaine and mounted on 1% low melting point agarose to photograph the injected area. Subsequently, each image was processed by quantifying the fluorescence intensity from the red channel (mCherry-labeled bacteria) and green channel (EGFP-labeled neutrophils) (ImageJ software). The region used to quantify the fluorescence was defined as the total area corresponding to the otic vesicle, which can be easily distinguished from bright field images of the larvae. In our conditions, this corresponded to a circular area of 550 × 550 pixels, which was the same for all the individuals evaluated. Microphotographs of the otic vesicle are presented with a total magnification of 40X. Statistical differences were assessed by means of a Kruskal-Wallis analysis followed by Dunn's multiple comparisons test using the GraphPad Prism 6.0 software.

Immersion assays for K. pneumoniae were adapted from similar methods performed previously to study S. Typhimurium virulence (Díaz-Pascual et al., 2017; Varas et al., 2017a,b). Bacterial cultures were grown until exponential phase (OD₆₀₀ = 0.4–0.6) and the cells were harvested by centrifugation, washed, and subsequently suspended in sterilized E3 medium. To determine the viable bacteria (CFU/mL), the suspension was serially diluted and plated on LB-agar. Zebrafish larvae (3-dpf) were washed with E3 medium and groups of 10 larvae were placed per well in a sterile 6-well plate. The wells were filled up to 8 mL with bacteria (5–8 × 10⁸ CFU/mL) and zebrafish larvae were incubated in this suspension at 28°C for 48 h. After the co-incubation, the larvae were washed with E3 medium and placed in a sterile 6-well plate filled with fresh E3 medium. The progression of the infection was monitored until 96 h post-exposition (hpe). Immersion experiments were performed in a total of 6 independent replicates, using a total of 75 larvae per bacterial strain. The progression of infection and bacterial colonization was monitored through live-cell imaging using an Olympus MVX10 fluorescence stereomicroscope at 48, 72, and 96 hpe. Photographs of the larvae are presented with a total magnification of 12X. To evaluate intestinal colonization, the gastrointestinal tract of the larvae was divided arbitrarily into three segments: anterior, middle and posterior. Four qualitative degrees of intestinal colonization were established based on the fluorescent signal observed under stereomicroscope: (0) no colonization, (1) low colonization, (2) intermediate colonization, and (3) high colonization. The total intestinal colonization score of each individual was calculated as the sum of the scores observed in each segment of the intestine, in a scale ranging from 0 to 9.

Capsule staining was performed as described previously (Breakwell et al., 2009). For this, cultures of each bacterial strain were grown overnight at 37°C without shaking. An aliquot of the culture was mixed with a drop of India ink (Mars® matic, Staedtler) and smeared over a glass slide, letting it to air-dry. Then, the smear was covered with a drop of crystal violet during 1 min, washed carefully with distilled water, and let to air-dry. The stained samples were observed by laser-scanning confocal microscopy using a 543 nm HeNe laser-mediated bright field mode.

Biofilm formation in abiotic surfaces was measured as described previously for K. pneumoniae (Bruchmann et al., 2015). Briefly, bacterial cultures were grown overnight at 37°C with shaking (180 rpm), and the cells were washed and suspended in fresh LB medium to an OD₆₀₀ of 0.2. Wells of a sterile 96-well microtiter plate were inoculated with 100 μL of the bacterial suspensions, and incubated at 37°C without shaking. After 24 h, the wells were washed three times with 200 μL of sterile distilled water and stained with 150 μL of a 0.1% crystal violet aqueous solution for 30 min. Then, wells were washed three times with 200 μL of distilled water, crystal violet was extracted adding 200 μL of 95% ethanol and incubating for 30 min. After incubation, the absorbance at 590 nm was measured.

The whole-genome sequences of the strains K. pneumoniae RYC492, ATCC 700603, and ATCC BAA-1705 were retrieved from the NCBI database, under the accession numbers APGM01000001.1, CP014696.2, and GCA_000349265.2, respectively. Genomic islands identification was performed through sequence analysis and multiple alignment using the platforms Artemis, ProgressiveMauve, and NCBI BLAST, as described previously (Marcoleta et al., 2016). The identification of capsule synthesis loci was performed using Kaptive (Wyres et al., 2016) and Kleborate (Lam et al., in review). The identification of genes related to yersiniabactin, colibactin, salmochelin and other siderophore production systems was done used the tool Kleborate. For the identification of fimbriae-related genetic clusters, all the known K. pneumoniae proteins participating in fimbriae biogenesis were retrieved from the UniProtKB database and were used as query to search for homologous proteins in each of the analyzed genome, using the NCBI BLAST tools. Identification of virulence factors using whole genome data was performed using the tool PathogenFinder (Cosentino et al., 2013).

As a first approach to exploit D. discoideum-Klebsiella interactions in the study of K. pneumoniae virulence, we adapted a social development assay previously described in a screening for Pseudomonas aueruginosa antivirulence compounds (Bravo-Toncio et al., 2016). This assay is based on that virulent bacteria delay the completion of D. discoideum social development, while attenuated or non-pathogenic bacteria allows its rapid conclusion (normally within 2–3 days in our experimental conditions). Thus, assessing the effect of a given bacterial strain over the social development is a way to estimate its virulence. Consequently, we compared the effect of feeding D. discoideum with the K. pneumoniae strains RYC492, 700603 and BAA-1705 over the social development of the amoeba. Additionally, Klebsiella aerogenes DBS0305928 was included as non-pathogenic control, since this strain is routinely used as a food to support D. discoideum growth in laboratory conditions (Fey et al., 2007). To this end, 24-well plates containing N agar were inoculated with the different strains and incubated overnight at 23°C to allow the growth of a thin bacterial lawn, which was of a similar prominence in all the strains tested (Supplementary Figure 1A). Moreover, the three K. pneumoniae strains showed a similar growth in N broth at 23°C (Supplementary Figure 1B), indicating that a comparable number of bacterial cells are present in the lawn formed by each strain before inoculating amoeba cells. Then, ~1 × 10⁶ D. discoideum cells were deposited over the bacterial lawns and incubated at 23°C for several days, monitoring the progress of the social development, which involves mainly three consecutive stages: aggregation, elevation and culmination (Figure 1A). The completion of the cycle is characterized by the formation of multiple spore-containing fruiting bodies, such as those observed normally after 2–3 days when amoebae were fed with K. aerogenes DBS0305928 (Figures 1B,C). At day 1, amoebae fed with K. aerogenes showed a complete phagocytosis plaque (where the whole bacterial lawn inside was engulfed), with aggregating cells forming early multicellular forms in the center. In contrast, incomplete phagocytosis plaques were observed after feeding amoebae with K. pneumoniae 700603 and BAA-1705 (Figure 1B). However, in the following days, no differences in the cycle progression were observed between these strains and the control (culminating between 2 and 3 days), indicating that Kp 700603 and Kp BAA-1705 are mildly virulent for D. discoideum. Conversely, Kp RYC492 showed a highly virulent phenotype causing a delay in the cycle, where at day 1 no clear phagocytosis plaque was observed, and the early elevation phase was reached after 3 days of co-incubation. Noteworthy, the fruiting bodies and earlier multicellular structures formed by amoebae fed with any of the K. pneumoniae strains tested showed an abnormal morphology. We observed mainly structures with an increased size, such as thicker stems and a bigger sorus, compared with that observed when fed with K. aerogenes. This suggests that K. pneumoniae somehow affects D. discoideum multicellular morphological plan, and that this could be further exploited as an additional read-out in this kind of virulence assays. Taken together, these results indicate that social development assays are a useful and straightforward way to compare virulence among distinct K. pneumoniae strains.

As described previously, a major determinant of K. pneumoniae virulence is its capacity to evade or resist macrophage-mediated phagocytosis and killing. Previous work showed that assays using axenic cultures of D. discoideum can be used to study invasiveness and intracellular survival in Salmonella Typhimurium (STm) (Riquelme et al., 2016). Remarkably, null mutants in genes directly linked to Salmonella virulence (i.e., demonstrated to be attenuated in murine models) also showed an impaired ability to invade or to survive intracellularly in D. discoideum, indicating that STm (and probably other bacteria) uses a common set of genes and molecular mechanisms to infect amoeba and animal host cells. Thus, we adapted the previously described assays using vegetative D. discoideum cells, to evaluate the invasiveness and intracellular survival of K. pneumoniae strains. For this purpose, we co-incubated both organisms during 1 h to let amoebae feed on bacteria. Then, amoeba cells were washed to remove most of bacteria from the medium, and incubated at 23°C. At different time points, aliquots were extracted and treated with gentamycin to kill extracellular bacteria. Then, D. discoideum cells were lysed with Triton X-100, and intracellular bacteria recovered from infected amoebae were titrated. As a control, the same procedure was performed using K. aerogenes DBS0305928. Measuring the amount of bacteria that infected amoebae after 1 h of co-incubation (time 0 of the assay), we observed that the different strains were internalized at distinct levels (Figure 2A). While Kp 700603 was internalized significantly more than the control, Kp RYC492 internalization was considerably lower. Following viable intracellular bacteria over time, we observed that titers of all strains declined during the first 6 h, being almost undetectable after 24 h in the case of Kp RYC492 and K. aerogenes DBS0305928 (Figure 2B). Conversely, Kp 700603 survived and even replicated inside the amoebae after 24 h, showing a ~200-fold increase in the viable cell count between 6 and 24 h post-infection. Meanwhile, Kp BAA-1705 was internalized at similar levels than the control, and a small population of bacteria was detected after 24 h.

To further support our observations, we followed the fate of intracellular K. pneumoniae through live-cell imaging using laser-scanning confocal microscopy. To this end, a similar experiment was performed using D. discoideum cells expressing a calnexin-GFP protein fusion (Müller-Taubenberger et al., 2001), along with bacteria constitutively expressing the red fluorescent protein mCherry from plasmid pFCcGi-TMP. The interaction of intracellular bacteria and amoebae was monitored at 0, 1.5, 3, 6, and 24 h post-infection (Figure 2C, Supplementary Figure 2). In agreement with the bacterial count, a large number of red fluorescent bacteria were detected within amoeba after 1 h co-incubation with Kp 700603, Kp BAA-1705, and K. aerogenes DBS0305928, indicating that all these strains were rapidly internalized by D. discoideum. In the case of Kp RYC492, a notably smaller number of bacteria was detected at this time, which correlated with the low percentage of internalization measured for this strain (Figure 2A). In all the following time points, very few or no Kp RYC492 cells were detected in association with amoebae. Intracellular Kp BAA-1705 and K. aerogenes DBS0305928 were observed only until 3 h after the removal of the extracellular bacteria. In contrast, intracellular Kp 700603 cells were observed at every time point evaluated. In fact, we were able to detect bacteria within D. discoideum after 24 h of co-incubation. These observations confirmed that Kp 700603 is able to survive and replicate intracellularly in D. discoideum. Additionally, some extracellular Kp 700603 were typically detected after 24 h post-infection, suggesting that this strain is able to escape from D. discoideum cells upon replication.

The small proportion of Kp RYC492 cells that were internalized by D. discoideum, as well as the notably fewer bacteria found inside amoeba cells observed by confocal microscopy during the infection assays, suggested that this strain could avoid or prevent phagocytosis. To further investigate this possibility, we adapted the infection assay using axenic D. discoideum cultures described above. In this version, we allowed amoebae and bacteria to interact during the whole 24 h-period monitored. At distinct times, aliquots of the mixed cultures were obtained and mounted for microscopic observation, and the viable extracellular bacteria were titrated. This was performed centrifuging the aliquots at low speed to sediment D. discoideum cells along with intracellular bacteria, and then diluting and plating the supernatant. Although with different kinetics, phagocytic activity of D. discoideum caused the reduction over time in the viable count of all the tested bacterial strains except for Kp RYC492, which remained constant (Figure 3A). As expected, D. discoideum rapidly phagocytized K. aerogenes DBS0305928. On the other hand, Kp BAA-1705 and Kp 700603 extracellular bacteria showed a similar reduction upon 3 h of co-incubation with amoebae, but roughly 10-fold more bacteria of the Kp 700603 strain were observed after 24 h. This could be attributable to the ability of Kp 700603 to replicate inside D. discoideum cells. In accordance with the viable cell counts, observation of the mixed cultures by confocal microscopy revealed the presence of a high number of Kp RYC492 bacteria that were not associated with amoeba cells during all the times monitored (Figure 3B, Supplementary Figure 3). Conversely, for all the other strains most bacteria were found associated with D. discoideum cells, suggesting that Kp RYC492 not only avoid phagocytosis but also the adhesion to phagocytic cells. Taken together, these results show that D. discoideum is a suitable model to study different aspects of the K. pneumoniae virulence. In particular, they allowed uncovering striking differences among strains of this species, in the context of their interaction with the host, including traits widely related to pathogenesis such as resistance to phagocytosis, invasion and intracellular survival.

Zebrafish larvae have shown to be a useful surrogate model to study the infective process of enteric bacterial pathogens (van der Sar et al., 2003; Clatworthy et al., 2009; Nguyen-Chi et al., 2014). To our knowledge, no studies have evaluated the virulence of K. pneumoniae in the zebrafish larvae, so we sought to develop an infection model suitable to generate pathological effects in the larvae, which could be quantified and compared among different bacterial strains, and also to evaluate the in vivo innate immune response of this host. In order to accomplish this, we tried two kinds of inoculation methods: static immersion and injection. Both have been successfully used to cause infections with different bacterial pathogens such as Edwardsiella tarda and S. Typhimurium (van Soest et al., 2011; Varas et al., 2017a). Regarding the injection method, we followed either a systemic infection or a localized infection approach, based on previous studies of bacterial pathogens using zebrafish embryos (Benard et al., 2012). Thus, attempting different routes of administration allowed us to set up a successful infection method, as well as to evaluate distinct host responses related to the bacterial entry mechanism, which can help discriminate between bacterial strains with diverse virulence potential.

To achieve a systemic infection, zebrafish larvae (3-dpf) were injected in the dorsal caudal aorta (Figure 4A) with 10,000 to 20,000 cells of mCherry-labeled Kp RYC492, Kp BAA-1705, Kp 700603, and E. coli DH5α (the latter as a non-pathogenic bacterial control previously validated in this host model; Varas et al., 2017a), or with PBS as an injection control. After daily monitoring of the injected larvae the survival was quantified and compared (Figure 4B). As expected, E. coli DH5α caused no lethality at any dose, condition, and time evaluated. Despite incubating the infected animals at 28°C (the optimal temperature for bacterial growth is 37°C), the larvae injected with K. pneumoniae strains showed an increased mortality over time. Zebrafish larvae injected with Kp BAA-1705 showed no significant lethality compared with the group injected with E. coli DH5α (Gehan-Breslow-Wilcoxon test, p = 0.155). On the other hand, Kp RYC492 caused the highest lethality, reaching 75% at 3 dpi, while Kp 700603 caused intermediate lethality, with 78.6% survival at the end of the experiment. Over time, larvae injected with E. coli cells completely cleared the infection, while severely affected fish infected with K. pneumoniae strains showed a marked bacteremia, spreading through the circulatory system and internal organs (Figure 4C; as control, a non-infected dead larva is shown in Supplementary Figure 4A). This occurred in all death events, regardless of the K. pneumoniae strain.

The otic vesicle (Figure 4A) was selected as the larval tissue to inject the bacteria and generate a localized infection event. This is a confined compartment devoid of macrophages and neutrophils, and can be used to study the directed migration of innate immune cells (Benard et al., 2012). Bacterial cells or PBS (control) were carefully injected in the otic vesicle of Tg(BACmpo:gfp) zebrafish larvae harboring GFP-labeled neutrophils, and neutrophil recruitment was quantified by measuring green fluorescence inside the vesicle. Bacterial infection in this compartment normally causes neutrophil recruitment as soon as 3 h after injection (Benard et al., 2012). We observed a marked inflammation of the otic vesicle after 8 h of injection, which was similar between all subjects, even those exposed to E. coli DH5α (data not shown). To determine whether the fish immune cells are able to clear the bacteria and resolve, we examined the otic vesicle of injected fish 24 h after injection (Figure 4D; as control, a larva injected with sterile PBS is shown in Supplementary Figure 4B), when normally the clearance of non-pathogenic bacteria has occurred. The fluorescence of green-labeled neutrophils (Figure 4E) and red-labeled bacteria (Figure 4F) were quantified inside the otic vesicle from injected larvae. The bacterial burden was significantly higher in the larvae injected with Kp RYC492 and Kp 700603, compared to the larvae injected with E. coli DH5α (Dunn's multiple comparisons test, p < 0.0001). This correlates with the neutrophil recruitment, where a higher count was observed after injecting these two Kp strains, but only with statistical significance in the Kp RYC492 injected larvae (Dunn's multiple comparisons test, p < 0.0001). This result indicates that the Kp RYC492 strain (and Kp 700603 in a lesser extent) triggered a higher neutrophil recruitment in the injected area, and that these strains also show an increased ability to persist and/or replicate in the site of infection compared with E. coli DH5α or Kp BAA-1705.

In order to explore whether K. pneumoniae can infect zebrafish through the natural entry route for this enteric pathogen, and based on previous studies (van Soest et al., 2011; Varas et al., 2017a,b), we adapted and optimized an immersion protocol for its use with this pathogen. The selected conditions included using Tg(fli1:egfp) zebrafish larvae (3 dpf), and co-incubating them with bacteria during a period of 48 h. This transgenic line has a GFP-labeled vasculature, and was chosen to better evaluate the occurrence of bacteremia after immersion, and to obtain a fluorescence signal from the whole larvae (green) to contrast the bacteria-derived fluorescence (red). Afterwards, the larvae were washed to remove unattached bacteria, and then monitored for 4 days after immersion (dpi). Washing was performed daily during the experiment. Unlike to that observed with injections into the circulatory system, no lethality was observed in larvae immersed in suspensions of any K. pneumoniae strain evaluated (data not shown). Since bacterial infections in zebrafish might cause consequences other than death, we monitored the fluorescence of the bacteria within larvae through live-cell imaging. Observation of the larvae under fluorescence revealed that the bacteria accumulated in the gastrointestinal tract. To evaluate and score the colonization, we artificially separated the gut into three major zones: anterior, middle and posterior (Figure 5A). The amount of fluorescence in each zone was observed and an intensity score was assigned: no colonization = 0, low colonization = 1, medium colonization = 2 and high colonization = 3. The sum of the values for each zone was calculated for each individual, representing the total intestinal colonization score (Figure 5B). Photographs of representative individuals of each condition are shown in Figure 5C. There were no clear differences between distinct K. pneumoniae strains when observing the pattern of colonized zones, although they appear to be prone to colonize the anterior and middle digestive tract rather than the posterior digestive system (Supplementary Table 1). The total intestinal colonization score determined for E. coli DH5α was significantly lower than that observed for all the K. pneumoniae strains assayed, both at 48 and 96 hpi (Two-way ANOVA, p < 0.0001). E. coli DH5α colonization significantly decreased at 96 hpi compared to 48 hpi, while the opposite was observed with the three K. pneumoniae strains, where an increased bacterial colonization was observed in a later time (Figures 5B,C).

In conclusion, while the immersion method allowed differentiating the non-pathogenic bacteria (E. coli DH5α) from the pathogenic bacteria (K. pneumoniae strains), the injection method allowed discriminating between strains from the same bacterial species with different degree of virulence (Kp RYC492 > Kp 700603 > Kp BAA-1705).

The assays presented above indicated that Kp RYC492 is virulent to both D. discoideum and zebrafish hosts, showing resistance to phagocytosis by amoeba cells, a great ability to colonize the zebrafish gastrointestinal tract, and to trigger a strong innate immune response in the otic vesicle injection model. In spite of the strong neutrophil recruitment unleashed, this immune response failed to reduce the bacterial burden in a normal time lapse, compared to non-pathogenic bacteria and to other K. pneumoniae strains. As described previously, the capsule is one of the most relevant virulence factors in this species, and indeed it has been previously related to resistance to phagocytosis, immunogenicity, biofilm formation, and tissue colonization. Furthermore, we observed that the macrocolonies formed on sheep-blood and LB agar plates by Kp RYC492 but not by Kp 700603 or Kp BAA-1705 showed a mucoid phenotype, which could also be related to the properties of its capsule. In view of this, we tested if Kp RYC492 possess the hypermucoviscous phenotype (highly associated with hypervirulent strains), which has been semi-quantitatively defined by a positive “string test.” This test is positive when a bacteriology inoculation loop is able to generate a viscous string of more than 5 mm in length by stretching bacterial colonies on an agar plate (Shon et al., 2013). However, after streaking and growing Kp RYC492 in sheep-blood agar and performing the test, it resulted negative, indicating that this strain is not hypermucoviscous (data not shown). Next, we sought to explore if a simple classical capsule staining using India ink could allow us to distinguish morphological peculiarities of the Kp RYC492 capsule compared with that of Kp 700603 and Kp BAA-1705. After staining, the samples were observed using confocal microscopy in a laser-mediated bright field mode, and several fields were photographed and analyzed. Figure 6A shows pictures of 3 representative capsule-stained individual cells from each strain. Additionally, an expanded set of pictures of this experiment is presented in Supplementary Figure 5, where stained cells were ordered by cell length to facilitate the comparison, and to allow appreciating the capsular morphology in all the variety of cell dimensions observed. While both Kp 700603 and Kp BAA-1705 strains showed a well-defined and regular round-shaped capsule, Kp RYC492 showed a notoriously thicker capsule, with an irregular morphology and a fuzzy aspect. Thus, the classical staining allowed us to easily differentiate the strains analyzed in terms of the capsular morphology, where a thicker and more irregular capsule in Kp RYC492 correlated to his more virulent phenotype.

To investigate if Kp RYC492 has an increased ability to form biofilms, we conducted a simple assay in 96-well plates to measure biofilm formation over an abiotic surface, previously used to the same end with other strains of K. pneumoniae (Bruchmann et al., 2015). As expected, Kp RYC492 showed a significantly higher ability to form biofilms than E. coli DH5α and then all other K. pneumoniae strains tested (Figure 6B). This correlates with its increased ability to colonize the zebrafish intestine, and with having a more prominent capsule.

In order to better understand the molecular basis of the differences observed among the K. pneumoniae strains in the multiple virulence phenotypes assessed, we analyzed the previously sequenced genome of Kp RYC492, Kp 700603, and Kp BAA-1705 in search for virulence factors, as well as to gain information regarding their determinants for capsule production. Since in a previous work we found that genomic islands (GIs) integrated into tRNA-coding genes (tDNAs) are an important source of virulence factors in this species (Marcoleta et al., 2016), we started identifying and characterizing the GIs present in the chromosome of the three K. pneumoniae strains (Table 2 and Supplementary Table 2). Kp RYC492 chromosome harbors 4 GIs from which GIE492, encoding the determinants for producing microcin E492 and salmochelin siderophore, has a suggested role in K. pneumoniae virulence (Marcoleta et al., 2016). No additional putative virulence factors were found in the rest of the GIs, except for a Fe³⁺/spermidine/putrescine transport system encoded in a GI integrated at a Ser tDNA. Besides, two GIs were identified in the Kp 700603 chromosome, corresponding to a prophage and a GI encoding many hypothetical proteins of unknown function. Kp BAA-1705 harbors five GIs in its chromosome. Three of them likely correspond to prophages, and the other two encoded many proteins of unknown function, none of them predicted as putative virulence factors. Thus, in agreement with its experimentally demonstrated virulence, only Kp RYC492 harbors putative pathogenicity islands integrated in its chromosome. Though, none of these factors could be clearly linked to phagocytosis resistance.

Next, we analyzed and compared the whole genome sequences of the three K. pneumoniae strains using different bioinformatics tools. General and K. pneumoniae-specific identification of virulence factors were performed using PathogenFinder (Cosentino et al., 2013) and Kleborate (Lam et al., in review), respectively. Kp RYC492 showed the highest number of predicted virulence factors (294), although many of them correspond to proteins which function has not been addressed experimentally (Table 2, Supplementary Table 3). Despite Kp BAA-1705 and Kp 700603 showed a considerable smaller number of putative virulence factors (130 and 50, respectively) they were predicted as human pathogens by PathogenFinder with a probability of 82.2 and 84.2%, respectively, while KpRYC492 showed a score of 88.7%. Additionally, in the genome of Kp 700603 we could not find genes encoding putative homologs to proteins participating in type-III secretion systems (T3SS), such as those encoded in the Salmonella pathogenicity islands SPI-1 and SPI-2, which have a demonstrated role in invasion and intracellular survival, respectively (Que et al., 2013; Jennings et al., 2017). Regarding adhesion factors such as fimbriae, no differences where observed among the three strains. All of them harbor the mrkA and mrkD genes previously associated with adherence to animal tissues, as well as the ecpRABCDE operon, also associated to tissue's adherence (Alcántar-Curiel et al., 2013).

Kleborate and Kaptive tools were used to determine the capsular genotype of each strain, which in some cases can be associated to a defined capsular serotype (Wyres et al., 2016; Lam et al., in review). Kaptive tool classified the capsule loci of Kp BAA-1705 and Kp 700603 as KL107 and KL53, respectively, although none of them have been associated with determined serotypes. Kleborate tool indicated that Kp RYC492 possesses genes for the production of two capsular serotypes (K22 and K37). Unfortunately, to our knowledge there are no experimental data regarding the virulence and resistance to phagocytosis of other K. pneumoniae strains (besides RYC492) harboring any of these serotypes. Performing similar assays with additional strains of these serotypes will allow us to further test if there is an association between K22/K37 capsule types and an increased resistance to phagocytosis.

In conclusion and in agreement with the experimental data, genomic analyses allowed the identification of putative pathogenicity islands only in Kp RYC492, which also harbored the highest number of predicted virulence factors in different positions of its genome. Nevertheless, it was not possible to associate any of these factors with the phagocytosis resistance phenotype displayed by this strain. Regarding Kp 700603, no clear genetic determinants could be associated with its ability to replicate intracellularly in D. discoideum.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The reviewer SD and handling Editor declared their shared affiliation.