Stool specimens were collected from a patient hospitalized in Taipei Veterans General Hospital with a CRKP infection and CRKP colonization in 2013 (Pan et al., 2013). The K. pneumoniae strains present in the specimens were isolated on Eosin Methylene Blue (EMB) agar plates. Their capsular type was determined by wzc sequencing (Pan et al., 2013) and confirmed by wzy PCR genotyping (Pan et al., 2015). Primer pairs are listed in Table 1. PCR amplifications were performed with the Long and Accurate PCR system (Takara, Tokyo, Japan). The cycling program was 96°C for 3 min, followed by 30 temperature cycles of 96°C for 30 s, 46°C for 15 s, and 72°C for 3 min. Drug susceptibility was assessed using a Vitek 2 system.

Additionally, we tested mice colonized with human microbiota free of K. pneumoniae then added K64 CRKP at the beginning. The results were similar. Therefore, the latter model was not used.

Bacteriophages were isolated from untreated water or soil. The agar overlay method was used for phage isolation, to obtain a pure preparation, and for titer determination. A high-titer phage preparation was generated by amplification and purification using the confluent lysis method as previously described (Lin et al., 2014).

In Taiwan, K64 was the most prevalent capsular type of CRKP (Pan et al., 2015). Therefore, we chose K64 for this decolonization test. To isolate a phage specific for K64 K. pneumoniae from the natural environment, a K. pneumoniae strain with K64 type capsule was co-inoculated with sewage collected in Taipei City in LB broth overnight. After centrifugation, the supernatant was filtered using a 0.45 μm filter and spotted onto LB plates overlaid with the K. pneumoniae K64 strain to detect phage plaques. An agar overlay method was used for the isolation of a pure phage preparation and to determine phage titer. Single plaque isolation, elution, and re-plating were performed repeatedly. The pure isolate of this phage lyzed the K64 type of K. pneumoniae strains efficiently and named as φK64-1 (Pan et al., 2017). Capsule type was determined by PCR using primers specific for the K64 wzy gene and confirmed by sequencing. The efficiency of bacterial killing was assessed by using a phage killing assay. Survivors that were resistant to this phage were isolated. Then, the efficiency of bacterial killing by bacteriophage mixtures was examined as previously described (Pan et al., 2017). Actually, all CRKP strains with the K64 capsule type were sensitive to the φK64-1 that was isolated in this study.

We had repeatedly screened phages using wild-type K. pneumoniae strains with K1 and K64 capsule types yielded only phages targeting the capsule. To obtain phages that targeted other bacterial surface structures, we constructed a capsule-negative strain by generating a magA deletion mutant of NTUH-K2044 on which O1 type LPS was exposed. Screening for phages that infect this mutant yielded a phage, φKO1-1 (Hsieh et al., 2014), that targets O1 LPS.

Phage killing assays were done as previously described with minor modifications (Mateus et al., 2014). Briefly, bacteria (1 × 10⁴ CFU) were incubated with phage (1 × 10⁵, 1 × 10⁶, 1 × 10⁷ PFU; i.e., MOI = 10, 10², 10³) at 37°C for 30 min. Then, the surviving bacteria were enumerated by plating serial dilutions and determining the CFU, and the survival ratio was calculated.

Fecal samples were collected from a patient carrying capsular type K64 CRKP. Then, samples were aliguoted in fresh conditions and stored at -80°C until use. To mimic possible future human inoculations (Sarker et al., 2016), this human-derived fecal microbiota was diluted with saline ten times and orally administered to 7-week-old germ-free C57BL/6 mice (Hong et al., 2016). The colonization of capsular type K64 CRKP in the mice was confirmed by PCR (Table 1) and real-time PCR (Figure 1).

The mice were bred and housed in germ-free isolation operation boxes within the animal care facility of the National Taiwan University College of Medicine (NTUCM) and the Laboratory Animal Center at the National Laboratory Animal Center (NLAC). The time course of animal experiments is shown in Figure 2.

DNA was extracted from stool samples using the QIAamp DNA Stool Mini Kit (Qiagen) according to the manufacturer’s protocol. K. pneumoniae was detected by PCR, real-time PCR, and confirmed by culture. Real-time PCR amplifications were performed with Power SYBR^® Green Master Mix (REF: 4367659). The cycling program was 95°C for 30 s, followed by 45 temperature cycles of 95°C for 5 s and 60°C for 30 s. Primer pairs that targeted wzy and wzc genes are listed in Table 1. The composition of the microbiota in the established mouse model was determined by the 16s rDNA sequencing. Briefly, the full-length16S gene was amplified by PCR using the V1–V9 region of the 16S rRNA gene primers 27F/1492R (Johnson et al., 2019). Then, the amplicons were purified by using the GeneJET Gel Extraction Kit (Thermo Scientific) and quantified using the Qubit dsDNA HS Assay Kit (Qubit) on a Qubit 2.0 Fluorometer (Qubit). Amplicons were purified using PCR purification kits (Qiagen, Hilden, Germany) and 1 μg of DNA was used for the SMRTbell 1.0 Template Prep Kit (Pacific Biosciences, Menlo Park, CA). SMRTbell-adapted sequences were run on the Pacific Biosciences (PacBio) RS II platform using P6C4v2 chemistry. Consensus sequences longer than 1,600 bp were discarded. K. pneumoniae colonization was assessed by PCR, real-time PCR, and confirmed by culture. We tested stool samples at three time points: (1) 1 day before phage inoculation, (2) 14 days after phage inoculation, and (3) 56 days after phage inoculation.

Germ-free mice were orally inoculated with human GI-tract microbiota containing K64 K. pneumoniae (diluted 100-fold). The mice were divided into five groups 1 week after the models were established: the control group (no phage treatment, group I, n = 4); group II (n = 4), which was inoculated orally with φK64-1; group III, which was inoculated orally with φKO1-1 (n = 4); group IV, which was inoculated orally with both φKO1-1 and φK64-1 (n = 28); and group V, which was inoculated orally and injected intraperitoneally with both φKO1-1 and φK64-1 (n = 16). The mice were treated with 1 × 10⁹ PFU/mice of bacteriophage (φKO1-1, φK64-1, or both) via intragastric injection one time a day for 3 days (day 1, day 3, and day 5). Since mice are coprophilous animals, each mouse was housed in a separate individually ventilated cage (IVC). No cage floors were used in this study. We collected fecal samples just before phage treatment (day 1) and at 7, 14, 21, 28, 35, and 56 days after the first day of phage treatment. DNA isolated from the stool samples was used to detect K64 K. pneumoniae by PCR and real-time PCR, and φKO1-1 and φK64-1 were detected by real-time PCR.

Carbapenem-resistant K. pneumoniae was detected in 13 of the 28 mice at 3–5 weeks after treatment (Table 2). We used the selective medium (DHL agar, Nissui. Code. 05040) with added Ampicillin (100 μg/ml) to isolate the colony (Enterobacteriaceae). After the mouse feces were diluted 10 times and 100 times, we used the spread-plate method and incubated at 37°C overnight, and the colonies were selected to confirm whether they were CRKP by PCR. Isolated CRKP (200 μl) were treated with φK64-1 (100 μl) and incubated at 37°C for 15–20 min. The liquid was added into the 3 ml of top agar (LB Broth with 0.7% agar) and was mixed well, before pouring it on LB agar, waiting for the top agar to solidify. It was incubated overnight at 37°C, and the spots were generated.

The fresh stool specimen of the patient colonized with K64 type K. pneumoniae was aliquoted and frozen at -80°C in 2013 until use. The humanized microbiota was established by inoculating diluted frozen aliquoted samples from the patient to germ-free mice (Vandeputte et al., 2017).

The multi-host K. pneumoniae bacteriophage φK64-1, which can infect K1, K11, K21, K25, K30, K35, K64, and K69 reference strains, was previously isolated. The genome sequence, which was determined by high-throughput sequencing, was 346,602 bp in length and contained multiple depolymerases (Pan et al., 2017). A second phage, φKO1-1, was isolated that infected acapsular K. pneumoniae strains which contain O1 LPS but not capsular strains (Hsieh et al., 2012).

The killing effects of φK64-1 and φKO1-1 alone and in combination on two K64-type K. pneumoniae strains, CB-0001 and S84, were first assessed in vitro. At a multiplicity of infection (MOI) of 1,000, 100 and 10, phage φK64-1 killed the majority of the K64 bacteria. The survivors after phage killing (i.e., insensitive to K64-1) were capsule negative strains that formed relatively non-mucoid colonies and were negative for alcian blue staining (data not shown). However, these non-capsulated survivors could be killed by φKO1-1, which targets the O1 antigen of LPS (Fang et al., 2004). Although φKO1-1 could not infect wild-type K64 strains with intact capsule, the combination of φK64-1 and φKO1-1 at MOIs of 1,000 and 100 killed all of the K64 bacteria. However, complete killing achieved only in half of the in vitro tests with MOI = 10 (Figure 3A). The survivors after in vitro phage killing were all non-capsuled. The wild type was susceptible to phage K64-1, while non-capsuled mutants were resistant to phage K64-1 and sensitive to phage KO1-1 (Figure 3B). The spot tests had been repeated for more than 5 times and showed the same picture.

The K. pneumoniae detection assay results by PCR in all study mice are shown in Table 2. K. pneumoniae was detected consistently in all untreated humanized microbiota model mice and all model mice treated with φKO1-1 alone. In mice treated with φK64-1 alone, K. pneumoniae was undetectable for 2 weeks, but was detected again at 3 weeks after phage treatment. In the 28 mice orally inoculated with both φK64-1 and φKO1-1, K. pneumoniae was undetectable from 2 to 4 weeks. However, K. pneumoniae was detected in 13 of the 28 mice at 3–5 weeks after treatment. The remaining 15 mice remained K. pneumoniae-free through the end of the study (56 days later). Among the 16 mice treated with φK64-1 + φKO1-1 both orally and intraperitoneally, 8 mice showed a transient loss of K. pneumoniae, whereas the remaining 8 mice were K. pneumoniae negative for 56 days.

The K. pneumoniae, φK64-1, and φKO1-1 detection assay results by real-time PCR in 6 mice are shown in Figure 1. The average number of K. pneumoniae in all untreated humanized microbial flora model mice was 3,480 ± 473 CFU/g (Figure 1A). Negative real-time PCR results also showed negative PCR results. Therefore, the results of PCR and real-time PCR were consistent.

After the treatment of φ K64-1 and φKO1-1, φK64-1 reached a peak on the 7 days after treatment, with an average of 26,669 ± 18,717 PFU/g, and the average number on the 14 days would drop to 753 ± 538 PFU/g, and it was almost undetectable at 56 days (Figure 1B). However, φO1-1 reached an average of 82,976 ± 9,553 PFU/g on the 7 days after treatment, increased to an average of 87,839 ± 6,224 PFU/g on the 14 days, and continued to increase thereafter, reaching an average of 125,125 ± 18,676 PFU/g at 56days (Figure 1C). No phage was detected from day 0 to 56 in the control group (not shown).

Results of 16S rDNA sequencing were uploaded to NCBI, PRJNA766775.¹

The prevalence of K. pneumoniae in humanized mice ranged from 0.01 to 0.09% before phage treatments, while the prevalence ranged from 0.04 to 0.08% 56 days after failed treatments.

Shannon’s diversity index is commonly used to characterize species diversity in a community. Germ-free mice were orally inoculated with human GI-tract microbiota, so Shannon’s diversity index was highest before treatment (SD0). The difference then become lower at SD14 and SD56 (Figure 4A).

The Firmicutes/Bacteroidetes (F/B) ratio is widely accepted to have an important influence in maintaining normal intestinal homeostasis. Increased or decreased F/B ratio is regarded as dysbiosis, whereby the former is usually observed with obesity, and the latter with inflammatory bowel disease (IBD; Stojanov et al., 2020). The F/B ratio was the highest before phage treatment (SD0). However, there was no significant difference between SD14 and SD56, and it fell below 0.5 at D56, indicating that the GI-tract microbiota might gradually return to balance (Figure 4B).

Carbapenem-resistant K. pneumoniae were isolated from mouse feces after 56 days of phage treatment failed mice, and were confirmed as K64 by PCR (data not shown). We observed that, in contract to the in vitro killing assays, the isolated CRKP could still be infected by φK64-1 by plaque assay (Figures 5A,B). The amount of φK64-1 was the highest after 7 days of phage treatment, but it was decreased by more than 1 × 10⁴ times after 14 days of phage treatment (Figure 1B).