Characterization of the Kp ter operon

The complete ter locus consists of two distinct, but highly conserved, operons encoded on opposite DNA strands: a set of 14 individual genes with potential biosynthetic functionality that includes terXYW, and a tellurium resistance operon consisting of terZABCDEF (referred to as the “ter operon”). This is followed by a predicted haloacid dehydrogenase (HAD) that is highly conserved among ter-encoding Kp isolates (Fig 1A and S1 Table). Computational prediction of protein structure and function based on amino acid sequence using I-TASSER [38–40] indicates that many of the ter operon proteins are involved in response to stress (S1 Table), which comports with previous studies [21]. The deletion of terC, which is predicted to be involved in transmembrane transport (S1 Table), from the pK2044 plasmid renders hypervirulent Kp strain NTUH-K2044 (clone Kp2259) exquisitely sensitive to TeO₃^-2 (Fig 1B); however, terC expressed in trans in a ΔterC strain did not restore phenotypic resistance to TeO₃^-2 (Fig 1B). Sequencing of a second clone (Kp2257) did not reveal any spurious mutations and demonstrated insertion into the terC locus while replicating the sensitive to TeO₃^-2 phenotype (S1A and S1B Fig); thus, this phenotype is attributable to insufficient terC expression from the plasmid or polar effects of the mutation. The induction of TeO₃^-2 sensitivity via polar inactivation of terDEF via mutation of terC is consistent with previous studies that demonstrate that terBCDE are necessary for resistance to TeO₃^-2 [28,34]. Expression of terZ-F in trans in a ΔterC mutant fully restored TeO₃^-2 resistance (Fig 1B). Furthermore, the expression of terZ-F in trans was sufficient to confer TeO₃^-2 resistance to the Escherichia coli strain MG1655 (Fig 1C). These results indicate that an intact ter operon is necessary for TeO₃^-2 resistance in Kp NTUH-K2044.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. The Kp terZ-F genes are sufficient for TeO₃^-2 resistance.

(A) The ter locus is organized in two operons, a putative biosynthetic cluster and a TeO₃^-2 resistance cluster. These sections are found on opposite DNA strands and are encoded bidirectionally. The representative ter locus from the hvKp strain NTUH-K2044 is shown. NTUH-K2044 containing the empty vector pACYC184, the isogenic ΔterC mutant (clone Kp2259) containing an empty vector, the pTerC, or the pTerZ-F plasmid (B), and the E. coli K12 strain MG1655 with or without the pTerZ-F plasmid (C) were grown on LB or LB containing 10 or 100 μM K₂TeO₃^-2 to visualize inhibition of growth (dilution series 10⁰−10⁻⁷ of overnight culture). Two representative clones (labeled #1 and #2) of NTUH-K2044ΔterC containing the pTerC or the pTerZ-F plasmid and MG1655 containing the pTerZ-F plasmid are shown.

https://doi.org/10.1371/journal.ppat.1009537.g001

The ter operon is a genetically independent factor, rather than a biomarker of hvKp

Next, we sought to determine if ter is genetically independent or a biomarker of hvKp, as has been previously reported [35,36]. To this end, we returned to the ter-encoding (ter+) Kp isolates from our previous study [10]. These isolates were highly diverse, as reflected by their sequence types (Fig 2A) and none were a hvKp sequence type previously associated with the ter operon. To determine virulence gene content, we sequenced ter-encoding plasmids from our patient isolates using long-read sequencing (Oxford Nanopore Technologies, Oxford, UK). ter-encoding plasmids were characteristically large (86.8–430.8 kb; S2 Fig) and some were derived from plasmid fusions (S2 Table). These plasmids displayed substantial sequence variation outside the ter locus (S3A and S3B Fig), indicating that the ter operon is not a marker for a widely circulating, highly conserved, plasmid. Additionally, these ter-encoding plasmids do not contain virulence genes associated with Kp hypervirulence, nor was a single antibiotic resistance gene highly present on these plasmids (Fig 2B and S2 and S3 Tables). This supports the premise that the ter operon is an independent fitness factor during infection, rather than a marker of a hypervirulence or antibiotic resistance-encoding plasmid. Finally, predicted open reading frames (ORFs) directly up- and downstream of the ter operon displayed minimal functional conservation (Fig 2C), further indicating that this operon is not tightly linked to a virulence factor. Next, we repeated this analysis using publicly available reference genomes of ter-encoding Kp isolates (n = 88). These isolates were not limited to hvKp sequence types associated with the ter operon (Fig 2D), and indeed, the ter+ plasmids displayed a high degree of sequence variability (S3C Fig). 42% of ter+ plasmids contained a rmpA/A2 homolog and an accessory iron acquisition system. The remaining 58% ter+ plasmids had no classical hypervirulence factor present (Fig 2E and S2 and S3 Tables), and again, the predicted up- and downstream ORFs displayed little functional conservation except for transposase activity (Fig 2F and S1 Table). Together, these results indicate that ter is a genetically independent factor, and is likely playing a direct, but yet undescribed role in Kp disease.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. The Kp ter operon is not exclusive to hypervirulence plasmids.

ter+ plasmids from Martin et al. mSystems, 2018 [10] (A-C) and reference strains from the NCBI database (D-F) were analyzed. (A,D) Relative frequencies of sequence types (ST) of Kp strains containing ter+ plasmids. HvKp sequence types previously associated with the ter operon are outlined in a dashed line. (B,E) Heat map of ter+ plasmid sequence similarity to genes known to influence infection and antibiotic resistance genes. Each row represents an individual plasmid in the order of S2 Table (Martin et al. mSystems, 2018 [10] index 1–14, NCBI reference strains index 15–102). The pK2044 hvKp plasmid is highlighted by the red box, and hypervirulent Kp sequence types (hvST) previously associated with the ter operon are indicated. (C,F) To determine if any neighboring gene was consistently associated with ter, the gene neighborhood of ter plasmids encoding the ter operon from Martin et al. mSystems, 2018 [10] was visualized (C) and the frequency of ORFs adjacent to the ter operon encoded on reference plasmids from the NCBI database was calculated (F).

https://doi.org/10.1371/journal.ppat.1009537.g002

To determine the geographical and ecological range of Klebsiella encoding the ter operon, 14,060 Klebsiella sp. genomes, including 1,989 containing terZ-F, and their associated metadata were extracted from the Pathosystems Resource Integration Center (S1 Data) [41]. These genomes were derived from Klebsiella sp. strains from 6 of 7 continents (S4A Fig), indicating a wide geographical range that corresponds to the environmental ubiquity of Klebsiella sp. Assessment of the specific source of Klebsiella sp. isolation indicates a wide variety of hosts, including both animals and plants, and a number of environmental sources (S4B Fig). Interestingly, we found that terZ-F containing isolates were evenly distributed amongst all isolation sources (~14% of all isolates, S4C Fig); however, humans are the most represented isolation source, which is not surprising given the over-representation of human isolates in bacterial genome repositories. Many ter-containing Klebsiella sp. were isolated from human gut, blood, and respiratory samples (S4B and S4D Fig), which both supports a role for the ter operon in the human gut and comports with our previous studies where we originally identified a strong association between the ter operon and infection in Kp colonized patients [10]. Overall, relatively few ter-containing Klebsiella sp. isolates came from liver abscesses (S4D Fig), which are traditionally associated with hvKp [36], although they were enriched in this infection site (S4D Fig). Overall, ter-containing strains have a wide geographical and ecological range and are found across multiple sites of human infection.

TerC is a microbiome-dependent gut fitness factor

We previously reported a strong association between the ter operon and Kp infection (pneumonia and bacteremia) in Kp colonized patients, yet also found that terC is dispensable in a murine model of pneumonia [10]. To determine if the ter operon is important for bacteremia, WT Kp and ΔterC were competed in a peritoneal injection model of murine bacteremia. terC was dispensable in all tissues with the exception of a modest defect in the brain (S5 Fig). We did not explore this finding further, as a role in meningitis would not explain the correlation between the ter operon and infections observed in patients. We then hypothesized that the ter operon may be required during gut colonization, which precedes infection [11,12]. Exposure to antibiotics was not associated with Kp colonization or subsequent infection in our intensive care unit patient population [10,42], indicating that Kp must contend with the indigenous microbiota to colonize the gut and cause infection. Therefore, C57BL/6J mice were sourced from two different housing sites at The Jackson Laboratory (barriers RB16 and RB07) to control for natural variations in the gut microbiota induced by housing conditions [17–20]. Mice were orally gavaged with 100 μL of a mixture of wild-type and ΔterC Kp (Fig 3A). Intriguingly, a fitness defect (median 5.8-, 4.7-, 8.9-, and 4.0-fold-defect on days 1–4, respectively) was observed consistently for ΔterC in the mice sourced from RB16 over several days (Figs 3B and S6A) but not from RB07, despite their genetic identity (Figs 3C and S6B). These data suggest that the fitness defect exhibited by ΔterC is dependent on the gut microbiota.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. TerC is a fitness factor during gut colonization.

(A) Three days prior to inoculation, male and female C57BL6/J mice sourced from barriers RB16 and RB07 were treated with 0.5 g/L ampicillin or regular drinking water. (B-E) NTUH-K2044 and the isogenic ΔterC mutant (clone Kp2259) were mixed 1:1 and approximately 5x10⁶ CFU were orally gavaged into mice (n = 9–18 per group). A fresh fecal pellet was collected daily from each animal, CFUs were enumerated, and log competitive indices (mutant:WT) were calculated (median and IQR displayed, *P < 0.05, **P < 0.005, ***P < 0.0005, one-sample t test compared to a hypothetical value of 0). (F) NTUH-K2044 and the isogenic ΔterC mutant containing an empty vector or the pTerZ-F plasmid were mixed 1:1 and approximately 5x10⁶ CFU were orally gavaged into mice sourced from barrier RB16 (n = 14–16). A fresh fecal pellet was collected 24 hours after inoculation, CFUs were enumerated, and log competitive indices (mutant:WT) were calculated (F, median and IQR displayed, ****P < 0.00005, one-sample t test compared to a hypothetical value of 0 or Student’s t test). Each data point represents an individual animal.

https://doi.org/10.1371/journal.ppat.1009537.g003

To begin to characterize the effect of the indigenous microbiota on the ΔterC mutant, mice were treated with antibiotics and the experiment was repeated. Consistent with previous studies [15], treatment with antibiotics increased overall Kp colonization density in mice from both barriers (S6C and S6D Fig); however, this treatment also restored the fitness of ΔterC in mice sourced from RB16 (Figs 3D and S6C). Conversely, antibiotic treatment of mice sourced from RB07 did not impact ΔterC fitness (Figs 3E and S6D). These data indicate that an antibiotic-susceptible member or members of the microbiota of mice sourced from RB16 are involved in reducing the fitness of the ΔterC mutant, as opposed to the microbiota of mice sourced from RB07 enhancing ΔterC fitness. Furthermore, complementation of the ΔterC mutant by expression of terZ-F in trans ameliorated its fitness defect in mice sourced from RB16 (Fig 3F). This finding was confirmed with the fully sequenced ΔterC clone (S1C Fig). To determine if the ter operon is required for colonization, mono-colonization studies were performed. These results show that both the wild type and ΔterC mutant were able to colonize the gut (S7 Fig), although there was mouse-to-mouse variation that may obfuscate an advantage of the ter operon during mono-colonization. Consistent with previous studies [43–46], the mice exhibited mortality throughout the duration of the experiment. However, the increased bacterial load due to antibiotics (S6 Fig) was not associated with increased mortality, and both strains exhibited equal virulence in this model regardless of barrier (S8 Fig). Collectively, these data suggest that the ter operon is a microbiome-dependent gut fitness factor.

Reduced fitness of the ΔterC mutant is associated with specific gut microbiota constituents

To determine if the composition of the gut microbiota of mice sourced from RB16 and mice sourced from RB07 differed, we performed 16S rRNA gene sequence analysis from fecal DNA collected throughout the course of these experiments (S9 Fig and S2 Data). To compare the microbiota between all groups (mice sourced from RB16, RB07, RB16+Abx, RB07+Abx) and all time points, θ_YC distances [47] were calculated, and principal coordinates analysis was used to visualize these distances. θ_YC dissimilarity accounts for both the number of shared and unique species as well as differential species abundance in a single metric [47]. As expected, microbiota differed between male and female mice (Fig 4A.i, axis 1, females cluster on left of graph) [48–50]. Despite sex-based differences, the fecal microbiota of RB16 and RB07 were significantly dissimilar on the day of inoculation (Fig 4A.i, axis 2, AMOVA P = 0.001; Fig 4B), suggesting that the results observed in Fig 3B–3C were attributable to differences in the microbiota of these mice. In addition, the fecal microbiota of antibiotic treated mice sourced from RB16 and RB07 were dissimilar from their untreated counterparts (Fig 4A.ii and iii, AMOVA P < 0.001), but not from one another (Fig 4A.iv, AMOVA P = 0.676). Assessment of the magnitude of dissimilarity indicated that intergroup dissimilarity was higher than intragroup dissimilarity, and antibiotic treatment resulted in the greatest dissimilarity (Fig 4B, higher values indicate higher dissimilarity). These findings were consistent across all time points (S10–S14 Figs). These data demonstrate an association between ter-dependent fitness in the gut and the composition of the gut microbiota and suggest that an individual or group of gut microbiota constituents in the mice sourced from RB16 underlies the observed loss of fitness.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. The fecal microbiota in which terC is (RB16) and is not (RB07) a fitness factor are distinct.

Fecal pellets collected from male and female C57BL6/J mice sourced from barriers RB16 and RB07 (n = 9–20 mice per group) on the day of Kp inoculation were subjected to 16S rRNA sequencing. Pairwise community dissimilarity values between the fecal microbiota communities of barriers RB16 and RB07 with or without three days treatment with 0.5 g/L ampicillin were visualized by Principal coordinates analysis (A, AMOVA) and individually (B, **P < 0.005, ****P < 0.00005, one-way ANOVA followed by Tukey’s multiple comparisons post-hoc test). (C) Diversity of the fecal microbiota was summarized by inverse Simpson index (blue points: RB16+Abx, orange points: RB07+Abx, **P < 0.005, one-way ANOVA followed by Tukey’s multiple comparisons post-hoc test). LEfSe was used to determine if specific bacterial families (D) or OTUs (F) were differentially abundant between the fecal microbiota of RB16 and RB07 (D, LDA ≥ 3.5 and P < 0.05 are shown). Differential bacterial families (E) or OTUs (G) relative abundance values were plotted (E, *P < 0.05, **P < 0.005, ***P < 0.0005, ****P < 0.00005, Student’s t test).

https://doi.org/10.1371/journal.ppat.1009537.g004

There were several differences between the microbiota in mice sourced from RB16 and RB07. The diversity of the fecal microbiota of mice sourced from RB16 was significantly higher than mice sourced from RB07 on the day of inoculation (Fig 4C), and this difference was present throughout the experiment (S15 Fig). We next sought to determine if a bacterial family or families differentiated the fecal microbiota of mice sourced from RB16 and RB07 through the use of linear discriminant analysis (LDA) effect size (LEfSe) [51]. On the day of inoculation, 7 bacterial families were found to be differentially abundant, 2 of which were more abundant in mice sourced from RB16 and 5 of which were more abundant in mice sourced from RB07 (Fig 4D and 4E). Only unclassified Clostridiales, associated with mice sourced from RB07, remained differential throughout the entire experiment (S16 and S17 Figs). We hypothesized that no family consistently distinguished mice sourced from RB16 from mice sourced from RB07 because variations in individual operational taxonomic unit (OTU) relative abundances obscured family-level analysis. As such, we used LEfSe to determine differentially abundant OTUs between these microbiotas. On the day of inoculation, 18 OTUs were found to be differentially abundant, 7 of which were more abundant in mice sourced from RB07, and 11 of which were more abundant in mice sourced from RB16 (Fig 4F). Intriguingly, 4 of the 11 OTUs associated with mice sourced from RB16 had 16S rRNA sequences most similar to Muribaculum intestinale (Fig 4F; 0008, 0009, 0018, 0019) [52–54]. These 4 OTUs, as well as 0030 which had 16S rRNA sequence most similar to Clostridium scindens, remained more abundant in mice sourced from RB16 than RB07 for the entire experiment (S18 and S19 Figs). Notably, the Porphyromonadaceae family, of which M. intestinale was considered a member in our dataset, was also enriched in mice sourced from RB16 on days 1–3 post-inoculation (S16 and S17 Figs). The identification of these species as differentially abundant in mice sourced from RB16 is notable, as M. intestinale belongs to family S24-7 which is suggested to play a role in gut inflammatory homeostasis [20,54–57], and C. scindens has been reported to be a clinically relevant probiotic candidate [58]. The only OTU that was more abundant in mice sourced from RB07 throughout the experiment is 0037, which is most similar to Olsenella profusa (Figs 4F and S18 and S19). OTUs associated with mice sourced from RB16 by LEfSe were mostly absent in mice sourced from RB07 (Fig 4G) throughout the course of the experiment (S18 and S19 Figs). If these OTUs are tightly associated with the observed ΔterC mutant fitness defect, they should also be sensitive to the antibiotic treatment that ameliorates the defect. Indeed, differential OTUs, as well as corresponding families, were sensitive to antibiotic treatment (S20 and S21 Figs) highlighting their potential role in reducing the fitness of the ΔterC mutant. Finally, we asked if the introduction of Kp impacted the relative abundance of these families and OTUs. At the family level, the microbiota of mice sourced from RB16 was only minimally modulated (S22A Fig) by Kp inoculation, and the only OTU that differentiated mice sourced from RB16 from RB07 that was modulated was 0045 (Mordavella massiliensis, S21B Fig). Further analysis of the gut communities of mice sourced from RB16 and RB07 demonstrated that these communities remained stable through the course of the experiment in antibiotic-naïve mice but shift drastically following antibiotic treatment (S22C Fig). These data indicate that the microbiota constituents of mice sourced from RB16 reduce the fitness of the ΔterC mutant and are highly stable during Kp colonization.

SCFA metabolism is predicted to be enriched in the gut of mice sourced from RB16

Next, we interrogated what biological processes may result in the reduced fitness of the ΔterC mutant. Specifically, we were interested in exploring biological processes that were characteristic of the microbiota constituents of mice sourced from RB16 associated with reduced fitness of the ΔterC mutant. To this end, the metagenomes of the fecal microbiota of mice sourced from RB16 and RB07 were predicted using PICRUSt2 [59]. The relative abundance of predicted metabolic pathways as annotated by MetaCyc [60] differed between the fecal microbiota of mice sourced from RB16 and RB07 (S23 Fig and S3 Data). A significant difference overall between predicted metabolic pathways of the fecal microbiota of RB16 and RB07 was detected on the day of inoculation and on multiple days post-inoculation although this was not readily detected in the first two principal components that account for the majority of variation in the data (S23 Fig). The subtlety of this finding was not surprising, as the microbiota of mice sourced from RB16 and RB07 are distinct based on specific OTUs (Fig 4) but share a large number of gut constituents (S9 Fig). Thus, we expected that a limited number of predicted metabolic pathways would differentiate microbiota of mice sourced from RB16 and RB07, and that some of these pathways would correspond to the microbiota constituents of mice sourced from RB16 associated with the reduced the fitness of the ΔterC mutant.

LEfSe was used to determine which specific metabolic pathways differentiated the predicted metagenomic profiles of the gut microbiota of mice sourced from RB16 and RB07 [51]. This analysis revealed several metabolic pathways that were enriched in the gut microbiota of mice sourced from RB16 on the day of inoculation and throughout the experiment, including gluconeogenesis, peptidoglycan biosynthesis, and fermentation of short-chain fatty acids (S24 and S25 Figs, S4 Table and S3 Data). The enrichment of SCFA metabolic pathways in the gut microbiota of mice sourced from RB16 may be explained by the presence of M. intestinale, which was strongly associated with the gut microbiota of mice sourced from RB16. In addition, antibiotic treatment of mice sourced from RB16 led to a decrease in OTUs that correspond to M. intestinale (0008, 0009, 0018, 0019), and a drastic reduction in the relative abundance of predicted SCFA metabolic pathways (S26 Fig). This corresponds to our 16S rRNA sequencing data, as SCFA pathways underpin the impact of M. intestinale on gut inflammatory homeostasis [20,54–57].

Exogenous SCFA administration reduces the fitness of the ΔterC mutant in the gut

SCFAs are known to have a wide variety of functions in the host, including increasing antimicrobial peptide production [61], intestinal epithelial barrier function [62], and acceleration of the immune response to Enterobacteriaceae [63]. Moreover, previous studies have identified a protective role for the SCFA acetate during Kp lung infection [64]. To determine if SCFA metabolism is responsible for the fitness defect exhibited by ΔterC, we first explored if SCFAs directly kill or inhibit the growth of Kp in a ter-dependent manner. Previous studies have shown that SCFAs are able to directly inhibit the growth of Kp under acidified conditions through disruption of respiration [65]. Thus, we grew WT Kp and ΔterC in the presence of SCFAs (acetate, butyrate, and propionate) in both neutral and acidified conditions. Similar to previous studies, high concentrations of SCFAs slightly inhibited growth of both strains in neutral conditions, but completely arrested growth in acidic conditions (S27A Fig). To determine if an individual SCFA was responsible for growth inhibition, WT Kp and ΔterC was grown in acetate, butyrate, and propionate individually and in combination in both neutral and acidified conditions. Acetate and butyrate had the largest impacts on Kp growth; however, growth inhibition was not dependent on the presence of terC (S27B Fig). Next, we assessed if growth inhibition occurs in a ter-dependent manner at lower concentrations of SFCAs. Again, titration of SCFAs in acidified media inhibited growth in a concentration-dependent manner, but not a ter-dependent manner (S27C Fig). To determine if WT Kp are able to antagonistically inhibit the growth of ΔterC in the presence of SCFAs, competitive growth assays were performed in the presence of SCFAs. No antagonism was observed between WT Kp and ΔterC in the presence of SCFAs (S27D Fig). Finally, we performed killing assays with the WT Kp and ΔterC strains to determine if SCFAs can kill Kp in a ter-dependent manner; however, no killing was observed for either strain (S27E Fig). Collectively, these data demonstrate that SCFAs can indeed inhibit Kp growth under acidified conditions as previously described [65], though not in a ter-dependent manner.

To assess the effect of SCFAs on Kp gut fitness in vivo, we repeated our competitive gut colonization experiments with mice sourced from RB16 and RB07, but also included mice sourced from RB07 treated continuously with a cocktail of SCFAs via drinking water. As previously observed, ΔterC Kp were less fit than WT Kp in mice sourced from RB16 but were equally fit in mice sourced from RB07 (Fig 5). Treatment of mice from RB07 with a SCFA cocktail led to significantly reduced fitness of the ΔterC mutant in the gut similar to that of what was observed in RB16 (Fig 5). Finally, we measured the presence of SCFAs in the fecal pellets of mice sourced from both barriers, as well as mice sourced from RB07 treated with exogenous SCFAs. Fecal SCFA quantification revealed higher SCFA concentrations in the feces of mice sourced from RB16 compared to those from RB07 (S28A Fig); however, this difference did not reach statistical significance. Treatment of mice sourced from RB16 with antibiotics, which restored the ΔterC fitness defect, also significantly reduced fecal SCFA levels (S29A Fig), specifically acetate and butyrate (S29C and S29D Fig). These data indicate that the reduced fitness of the ΔterC mutant is due to gut SCFA levels, which is associated with differences in specific indigenous gut microbiota constituents between mice sourced from RB16 and RB07.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Exogenous treatment of mice with SCFAs results in a terC fitness defect.

Seven days prior to inoculation, male and female C57BL6/J mice sourced from barriers RB16 and RB07 were treated with a SCFA cocktail or regular drinking water (sham). NTUH-K2044 and the isogenic ΔterC mutant (clone Kp2257) were mixed 1:1 and approximately 5x10⁶ CFU were orally gavaged into mice (n = 19 per group). A fresh fecal pellet was collected daily from each animal, CFUs were enumerated, and log competitive indices (mutant:WT) were calculated (median and IQR displayed, *P < 0.05, **P < 0.005, ***P < 0.0005, one-sample t test compared to a hypothetical value of 0 or Holm-Sidak multiple-comparison test following one-way ANOVA).

https://doi.org/10.1371/journal.ppat.1009537.g005

Ethics statement

This study was performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals [102]. The University of Michigan Institutional Animal Care and Use Committee approved this research (PRO00007474).

Materials, media, and bacterial strains

All chemicals were purchased from Sigma-Aldrich (St. Louis, MO) unless otherwise indicated. E. coli K12 strain MG1655, Kp strain NTUH-K2044 [103], and isogenic mutants were cultured in Luria-Bertani (LB, Becton, Dickinson and Company, Franklin Lakes, NJ) broth at 37°C with shaking, or on LB agar at 27°C (Thermo Fisher Scientific). The isogenic ΔterC mutant was generated as previously described [10]. Briefly, the λ-red mutagenesis system was used to inactivate the terC gene [104]. Electrocompetent NTUH-K2044 cells encoding the pKD46 plasmid were transformed with a kanamycin resistance cassette amplified from the pKD4 plasmid containing homologous overhangs to the terC locus. Transformed cells were recovered overnight at 30° C in SOC media, then selected in the presence of 40 μg/mL kanamycin. The isogenic ΔterC mutant was confirmed by whole-genome sequencing using the Illumina NexteraXT kit on the Illumina MiSeq using a 2x250 bp V2 kit. Illumina reads are available on the Sequence Read Archive (SRA) in BioProject PRJNA464397. To construct the pTerC and pTerZ-F complementation plasmids, PCR products derived from WT NTUH-K2044 containing the terC or terZABCDEF open reading frames were inserted into pCR 2.1 using TOPO TA cloning (Life Technologies, Carlsbad, CA) and directionally ligated into pACYC184 following digestion with Xbal and HindIII. The ligation mixture was transformed into NEB 10-beta Competent E. coli (New England Biolabs, Ipswich, MA) by heat shock. E. coli transformants were selected at 37°C on LB agar containing 30 μg/ml chloramphenicol, re-cultured, and confirmed by colony PCR. Single transformants were then grown in batch culture for plasmid extraction using the Plasmid Midi Kit (Qiagen, Germantown, MD). MG1655 and NTUH-K2044 competent cells were prepared as previously described [105], electroporated with the complementation plasmids or corresponding empty vector, and selected at 37°C on LB agar containing 30 (MG1655) or 80 μg/ml (NTUH-K2044) chloramphenicol. Following selection, transformants were re-cultured, and confirmed by colony PCR and by growth in presence of TeO₃^-2. All primer sequences can be found in S5 Table. For all subsequent experiments, complemented and control strains were grown in the presence of the appropriate antibiotic.

ter+ genome identification

For plasmid analysis, ter-encoding reference strains and plasmids from the National Center for Biotechnology Information (NCBI) nucleotide collection database were identified using BLAST [106], wherein the entire ter locus (S3 Table) was used as the query, and Klebsiella pneumoniae (taxid:573) as the subject (extraction date 03/27/2019). The ter operon was not identified on any Kp chromosomes. For identification of Klebsiella sp. encoding the ter operon, individual PATRIC Global Family annotations corresponding to the NTUH-K2044 terZABCDEF gene products were searched against the Pathosystems Resource Integration Center (PATRIC) genome database [41]. The resulting list of genomes and corresponding metadata (extraction date 11/16/2020) was then restricted to Klebsiella sp. and further refined by identifying genomes that have every NTUH-K2044 terZABCDEF gene product annotation and are of good quality as noted by PATRIC. Metadata was visualized in R (v.3.6.3) using the “ggplot2,” “ggmap,” “maps,” and “mapdata” packages and Prism 8 (GraphPad Software, La Jolla, CA).

Plasmid sequencing and analysis

To characterize ter-encoding Kp strains, the multi-locus sequence type (MLST) was assigned using the Bacterial Isolate Genome Sequence Database (BIGSdb) [107,108]. To characterize the ter-encoding plasmids from our previous study [10], genomic DNA was extracted from pure Kp cultures using the DNeasy PowerSoil Pro Kit (Qiagen, Hilden, Germany). Long-read genomic sequencing was performed using GridION X5 (Oxford, England) sequencing instrument. Each Nanopore sequencing library was prepared using 1 μg of DNA with the 1D ligation kit (SQK-LSK108, Oxford Nanopore Technologies) and sequenced using R9.4.1 flowcells (FLO-MIN106). MinKNOW software was used to collect sequencing data. Nanopore reads were called using Albacore v2.2.3 and assembled using Canu v1.7 [109]. Assemblies were corrected for ten rounds with Illumina reads using Pilon v1.22 [110] in conjunction with the bowtie2 v2.3.3.1 aligner [111]. Assembled plasmid sequences were circularized and annotated using Dfast [112] prokaryotic annotation pipeline. Pairwise alignments were performed using BLAST [106]. To assess the presence of hvKp and antibiotic resistance genes, reference sequences (S3 Table) were extracted, and BLAST [106] was used to align these reference sequences to ter encoding plasmid sequences. To study the genes conserved around the ter locus, gene level multiple sequence alignment (MSA) of the genes within 10 kbp upstream of the putative biosynthetic locus and downstream of the ter operon in all the plasmids. These loci were visualized by coding annotated genes using their 4-character gene names and unannotated hypothetical proteins using their gene cluster identifier as determined by CD-HIT software [113] for the MSA. An additional MSA was performed using MAFFT [114] in the L-INS-i mode and visualized the MSA using MSAviewer [115] to understand conserved genes around the ter locus. To predict protein structure and function of genes within and adjacent to the ter locus, ter encoding plasmids were annotated using the PATRIC RAST tool kit [116,117]. Unique annotation frequencies were calculated, and then unique predicted amino acid sequences were annotated predictively using I-TASSER [38–40]. S1 Table indicates reference amino acid sequences used for protein structure and function prediction. Only the highest scoring Predicted Biological Process, Predicted Molecular Function, and Predicted Pubchem Ligand Binding Site are reported. Finally, complete plasmid MSA was performed and visualized using Mauve (MegAlign Pro, DNASTAR Inc., Madison, WI). Illumina and Nanopore reads are available on the SRA in BioProject PRJNA464397.

Murine models of infection

Six- to 12-week-old C57BL/6J male and female mice from barriers RB07 and RB16 (Jackson Laboratory, Jackson, ME) were used for all murine models of infection. Gender was evenly distributed in all groups. For bacteremia studies, WT NTUH-K2044 and NTUH-K2044ΔterC were cultured overnight in LB, then bacteria were pelleted, resuspended, mixed 1:1, diluted in sterile PBS to the appropriate dose, and mice were inoculated intraperitoneally with approximately 5×10⁵ CFU in 100 μL of PBS. After 24 hours, mice were euthanized by CO₂ asphyxiation and blood, spleen, liver, and brain were collected. Solid organs were weighed and homogenized in sterile PBS, and whole blood and solid organ homogenates were plated on selective media. For oral inoculation studies, mice from both barriers were given regular drinking water, water containing 0.5 g/L ampicillin 3 days prior to inoculation and throughout the experiment, or water containing a SCFA cocktail (67.5 mM sodium acetate, 40 mM sodium butyrate, 25.9 sodium propionate) 7 days prior to inoculation and throughout the experiment. The SCFA dose and duration was chosen based on previous studies [64,118]. Antibiotic or SCFA-containing water was changed every 3 days. Kp strains were cultured overnight in LB in the presence of antibiotics when appropriate, then bacteria were pelleted, resuspended, mixed 1:1, diluted in sterile PBS to the appropriate dose, and mice were orally inoculated via oral gavage with approximately 5×10⁶ CFU in 100 μL of PBS. Single strain infections were performed as above without mixing the two strains. For four days post-inoculation, a fresh fecal pellet was collected from each mouse, weighed, and homogenized in sterile PBS, and homogenates were dilution plated on both LB agar containing 10 μg/ml ampicillin or 40 μg/ml kanamycin to determine Kp load. When complemented or empty vector control strains were used, plasmid maintenance was monitored by plating both on LB agar containing 10 μg/ml ampicillin or 40 μg/ml kanamycin to determine total Kp load, and on LB agar containing 10 μg/ml ampicillin and 80 μg/ml chloramphenicol or 40 μg/ml kanamycin and 80 μg/ml chloramphenicol to determine plasmid maintaining Kp load. In all models, mice were monitored daily for signs of distress (hunched posture, ruffled fur, decreased mobility, and dehydration) and euthanized at predetermined timepoints, or if signs of significant distress were displayed. No blinding was performed between experimental groups.

16S rRNA sequencing and analysis

Fecal DNA was isolated using the MagAttract PowerMicrobiome DNA/RNA Kit (Qiagen) and an epMotion 5075 liquid handling system. The V4 region of the 16S rRNA gene was amplified and sequenced as previously described [119]. Standard PCRs used 1, 2 or 7 μL of undiluted DNA and touchdown PCR used 7 μL of undiluted DNA. 16S rRNA gene sequence data was processed and analyzed using the software package mothur (v.1.40.2) [120,121]. Sequences were binned into OTUs based on 97% sequence similarity using the OptiClust method [122] following sequence processing and alignment to the SILVA reference alignment (release 128) [123]. θ_YC distances [47] were calculated between communities, and AMOVA [124] was used to determine statistically significant differences between experimental groups [47]. Principal coordinates analysis was used to visualize the θ_YC distances between samples. Taxonomic composition of the bacterial communities was assessed by classifying sequences within mothur using a modified version of the Ribosomal Database Project training set (version 16) [125,126], and diversity metrics, including inverse Simpson, were calculated. Finally, linear discriminant analysis effect size was used to determine if specific families and OTUs were differentially abundant in different groups [51]. Putative genus and species assignments were performed by comparing the representative 16S rRNA sequences from OTUs to the NCBI 16S ribosomal RNA sequence database. These assignments were confirmed using the Ribosomal Database Project (RDP) database, with the exception of OTUs assigned to Muribaculum intestinale based on NCBI, but to Porphyromonadaceae by RDP [125]. All other assignments were in agreement. 16S rRNA gene sequencing reads are available on the SRA in BioProject PRJNA464397.

PICRUSt2 metagenome prediction

16S rRNA gene sequence data processed using the software package mothur (v.1.40.2) [120,121] was used for metagenome prediction analysis using the PICRUSt2 pipeline [59]. 16S rRNA gene sequences were aligned using HMMER (v.3.3, hmmer.org), placed in the default 16S rRNA gene reference tree, which is comprised of 20,000 16S rRNA gene sequences in the Integrated Microbial Genomes database [127], using EPA-NG [128], and then the complete tree was constructed with GAPPA [129]. Following tree construction, unknown lineages were inferred, and KEGG pathway copy number was predicted using castor [130]. Finally, metabolic pathway abundances were inferred from MetaCyc using MinPath [131]. For analysis, metabolic pathway abundances were rounded to the nearest whole number and normalized across each sample to determine the relative abundance of each gene family and inferred metabolic pathway. Principle coordinate analysis was performed in R (v.3.6.3) using the “ggplot2” and “ggfortify” packages to visualize differences in inferred metabolic pathway relative abundances between experimental groups. AMOVA P values [124] were calculated using the “vegan” package and used to determine statistically significant differences between experimental groups [47].

SCFA growth and killing assays

SCFA containing LB was prepared by adding acetic acid, butyric acid, and/or propionic acid to LB at the appropriate concentration. The pH of SCFA-containing LB was adjusted to 7.5 or 5.75 with HCl or NaOH until the appropriate pH was achieved. Control LB lacking SCFAs was also pH adjusted to 7.5 or 5.75. Finally, pH adjusted media was sterile filtered using a 0.22 μM filter. For growth assays, Kp strains were cultured overnight at 37° C with aeration in LB in the presence of antibiotics when appropriate. Overnight cultures were diluted to an OD₆₀₀ of 0.02 in pH adjusted LB, then mixed 1:1 with pH adjusted LB with or without 2X SCFAs, to achieve a final 1X dilution of SCFAs and OD₆₀₀ of 0.01. For competitive growth assays, overnight cultures were mixed 1:1 before dilution. Cultures were incubated at 37° C with aeration and OD₆₀₀ readings were taken every 15 min using an Eon microplate reader with Gen5 software (Version 2.0, BioTek, Winooski, VT) for 24 hours. Area under the curve was quantified using Prism 8 (GraphPad Software, La Jolla, CA). For killing assays, 1 mL of overnight culture was pelleted, resuspended in pH adjusted LB with or without SCFAs, and cultured at 37° C with aeration. Cultures were sampled immediately, then every 2 hours and dilution plated on LB agar to quantify bacterial viability.

SCFA quantification

SCFAs were quantified as previously described [86]. Briefly, archived fecal pellets were suspended 1:2, 1:5, or 1:10 (weight:volume) in sterile PBS and homogenized. Homogenized samples were then centrifuged at 10,000 × g for 5 minutes to pellet the solid fraction and the supernatant was retained. The supernatant was then vacuum through a 0.22 μm filter prior to HPLC analysis. SCFA composition was measured using a Shimadzu HPLC (Shimadzu Scientific Instruments) equipped with an RID-10A refractive index detector. 30 μL injections were run on an Aminex HPX-87H column (Bio-Rad Laboratories, Hercules, CA) at 50° C with 0.01 H₂SO₄ mobile phase and a flow rate of 0.6 mL/minute. SCFA concentration was determined by interpolation from a 9-point standard curve containing acetate, butyrate, and propionate at concentrations between 0.1 mM to 40 mM, then normalized to dilution factor and tissue weight. Total SCFA concentration is a sum of acetate, butyrate, and propionate concentrations.

Statistical analysis

For in vitro studies, two-tailed Student’s t-tests or ANOVA followed by indicated post-hoc test was used to determine significant differences between groups. All in vitro experimental replicates represent biological replicates. All animal experiments were repeated at least twice with independent bacterial cultures. Competitive indices ((CFU mutant output/CFU WT output)/(CFU mutant input/CFU WT input)) [132] were log transformed and a one-sample t-test was used to determine significant differences from a hypothetical value of 0 and paired ratio t-test, or two-tailed Student’s t-test as indicated in figure legends was used to determine significant differences between groups. A P value of less than 0.05 was considered statistically significant for all experiments, and analysis was performed using Prism 8 (GraphPad Software, La Jolla, CA) unless otherwise indicated.