A total of 76 clinical MDR K. pneumoniae isolates were collected from different patients between May 2020 and January 2022 (Supplementary Table 1). Isolates were plated on blood agar plates and cultured at 37°C for 24 h to isolate the bacterial clones. A VITEK 2 Compact System (bioMérieux, France) was used to identify the positive culture strains. All isolates were stored in 40% (v/v) glycerol broth at –80°C until use. A total of 4,972 isolates of K. pneumoniae in China as of 2021 were collected from the NCBI database (including whole genome sequences and whole genome shotgun sequences).¹

Antibiotic susceptibility of all strains was performed by broth microdilution as recommended by the Clinical and Laboratory Standards Institute (CLSI). K. pneumoniae was cultured overnight in LB liquid medium at 37°C and 220 rpm. 0.5 μL of K. pneumoniae solution was streaked onto LB plates and incubated at 37°C for 24 h. Several monoclonal strains were selected to adjust the concentration of the bacteria in MH (Mueller-Hinton Broth) medium (Wiegand et al., 2008). The final inoculum size for broth dilution was 5 × 10⁵ colony-forming units (CFUs)/well, after inoculating into MH medium with various concentrations of Polymyxin B. Each concentration gradient was divided into three parallel groups and grown at 37°C and 220 rpm with shaking for 24 and 48 h. The experiment was repeated three times independently.

Klebsiella pneumoniae growth curves were established in LB medium manually. Overnight cultures were diluted to an OD₆₀₀ of 0.02 and grown in 96-well plates at 37°C and 220 rpm with shaking. The absorbance of the culture solution at 600 nm (OD₆₀₀) was measured every 0.5 h until it had peaked and was flat.

K. pneumoniae viscosity was determined using the string test (Shon et al., 2013). Strains that formed strings 5 mm or longer after stretching with the tip of a sterile inoculation loop were considered to have a hypermucoviscosity phenotype. K. pneumoniae was cultured overnight in LB liquid medium at 37°C and 220 rpm. The cultures were diluted the following day to an OD₆₀₀ of 1 and centrifuged at 2,350 g for 5 min, and the OD₆₀₀ of the supernatant was measured every minute.

The virulence of K. pneumoniae isolates was evaluated using the G. mellonella infection model. Larvae (0.3–0.4 g) were stored in the dark and used within 3 days after shipment (Tianjin Huiyude Biotechnology Co., Ltd.). Prior to injection, the bacterial pellet was washed with sterile saline and diluted to 1 × 10⁸ CFU/mL. Using a 1 mL insulin syringe (Shanghai Kindly Ent Dev), 10 μL of the bacterial suspension was injected into the center of the second gastropod of the larvae. A group of 10 larvae were randomly selected for injection. Each treatment was performed in triplicate with 30 larvae. After injection, the larvae were incubated at room temperature and survival was monitored daily for 3 days. Death was recorded when the larvae no longer responded to touch. The larvae without or injected with 10 μL sterile saline were used as negative controls. In all cases, no dead larvae were observed in the negative control groups.

Peripheral blood was collected from healthy volunteers (the author himself) and stored in heparin sodium anticoagulant tubes. Logarithmic phase K. pneumoniae was collected and suspended in PBS at 10⁸ CFU/mL. Then, 10 μL of bacteria and 90 μL of blood were mixed in a 96-well plate and incubated for 1 h at 37°C in 5% CO₂. After 1 h, the blood cells were lyzed using cell lysis buffer (10 mM EDTA, 0.25% Triton-X100 in PBS) and plated on TSB agar. The following day, the surviving bacterial cells were determined using a CFU assay.

Human neutrophils were isolated from the venous blood of healthy volunteers using a human peripheral blood neutrophil isolation kit (Solarbio, China) according to the manufacturer’s instructions (Gao et al., 2019). Logarithmic phase K. pneumoniae was collected and suspended in PBS at 10⁸ CFUs/mL. Then, 20 μL of bacteria were mixed with 100 μL of neutrophils (10⁶ cells/ml) in a 96-well plate. The total volume was adjusted to 150 μL with PBS and they were incubated for 1 h at 37°C in 5% CO₂. Neutrophils were lyzed after 1 h using cell lysis buffer and plating on TSB agar. The following day, the surviving bacterial cells were assessed by CFU assay.

A total of 39 isolates were sequenced, 7 isolates were sequenced by the third-generation sequencing, and the remaining 32 isolates were sequenced by the second-generation sequencing. Briefly, whole genome sequencing of K. pneumoniae was performed using a PacBio RS II and an Illumina HiSeq 4000 platform at the Beijing Genomics Institute (BGI, Shenzhen, China). Four SMRT cells Zero-Mode Waveguide arrays for sequencing were used with the PacBio platform to generate the subread set. PacBio subreads (length < 1 kb) were removed. The Pbdagcon program² was used for self-correction. Draft genomic unitigs, which are uncontested groups of fragments, were assembled using the Celera Assembler against a high-quality corrected circular consensus sequence subreads set. To improve the accuracy of the genomic sequences, GATK³ and SOAP tool packages (SOAP2, SOAPsnp, SOAPindel) were used to make single-base corrections. De novo hybrid assembly of short Illumina reads and long PacBio reads was performed using Unicycler v0.4.8 (Wick et al., 2017) and annotated using the rapid prokaryotic genome annotation tool, Prokka 1.14.6 (Seemann, 2014). The plasmid map was drawn using BRIG 0.95 and Easyfig 2.2.5 (Alikhan et al., 2011; Sullivan et al., 2011).

Acquired antimicrobial resistance genes (ARGs) were identified using ABRicate version 1.0.1⁴ by aligning the genomic sequences to the ResFinder database and NCBI database (Zankari et al., 2012). The virulence factors of the isolates were identified using Kleborate and ABRicate by aligning the genomic sequences to the VFDB database (Zankari et al., 2012; Wyres et al., 2016a). Multilocus sequence typing (MLST) was performed by MLST 2.1⁵ (Jolley and Maiden, 2010). Capsule typing was performed by Kleborate (Wyres et al., 2016b; Lam et al., 2021). The HarvestTools kit (Parsnp, Gingr, and HarvestTools) and BacWGSTdb were used to perform a comparative genomic analysis and phylogenetic analysis on the different isolates and to construct phylogenetic trees based on Single nucleotide polymorphisms (SNPs) in all isolates using the maximum likelihood method. The interactive tree of life (iTOL) v5⁶ was used to draw a phylogenetic tree (Treangen et al., 2014; Letunic and Bork, 2019; Feng et al., 2021). SNPs were called from the completed genome of 21,072,329 using Snippy⁷ and the genes in which SNPs were located were annotated and functionally classified using eggNOG 5.0 (Huerta-Cepas et al., 2019). Whole genome sequencing data were deposited in the NCBI database⁸ and are publicly available under BioProject: PRJNA669320, PRJNA685326, PRJNA823907, and PRJNA838703.

All analyses were performed using Prism software (GraphPad Software, La Jolla, CA, United States). Error bars represent SEM. All experiments were repeated at least three times.

We performed MLST and capsular typing tests on 76 MDR isolates (Supplementary Table 1), they were resistant to at least three antimicrobial classes (Magiorakos et al., 2012). And all isolates were resistant to cefuroxime axetil, cefuroxime sodium, and levofloxacin (Supplementary Table 2). A total of six MDR-hvKp isolates were detected (Supplementary Table 3) and they were resistant to ceftriaxone, amoxicillin/clavulanic acid, ceftazidime, cefoperazone/sulbactam, cefuroxime axetil, cefuroxime, ertapenem, cefepime, cefoxitin, imipenem, levofloxacin, trimethoprim/sulfamethoxazole, piperacillin/tazobactam. Four isolates that were resistant to polymyxins included two ST11 isolates [FRPDR and 21,091,025, the minimum inhibitory concentration (MIC) of polymyxin were 16 and 8 μg/mL, respectively]. Of these, ST11 was the most abundant with 15 isolates, followed by ST15 and ST307 with 13 and 10 isolates, respectively (Supplementary Table 4). Of the ST11 isolates, nine were KL64, five were KL47, and one was KL15. All ST15 isolates were KL19, whereas all ST307 isolates were KL102. We also identified an isolate which was a single locus variant of ST11, KP2000557 (ST751-KL64). Compared with ST11, it contained a mutation in the 1,716 DNA sequence of rpoB (c. 1716T > G, p. Ile572Met), suggesting that it may be derived from an ST11 mutation.

To evaluate the virulence of these isolates and to identify hvKp, we performed growth curves, Galleria mellonella infection model, and mucoviscosity assay (Walker et al., 2019) using NTUH-K2044 (the first classical hypervirulent K. pneumoniae) as a hvKp control (Wu et al., 2009). NTUH-K2044 exhibited the highest growth rate (Supplementary Figure 1). The growth rate of the ST307 isolates was similar to that of NTUH-K2044, that of the ST11 and ST15 isolates were lower compared with that of NTUH-K2044. Eleven ST11 isolates exhibited high lethality toward G. mellonella and were similar or more lethal than NTUH-K2044 (Figure 1). However, only two ST15 isolates and one ST307 isolate had higher lethality compared with NTUH-K2044 against G. mellonella (ST15: 21,073,066, 21,091,216; ST307: 21,072,123) (Supplementary Figure 2). The OD₆₀₀ of the isolates was adjusted to 1 and centrifuged at 2,350 g for 5 min (Walker et al., 2019). The supernatant of the classical hvKp NTUH-K2044 was completely transparent after centrifugation for 3 min (Figure 2 and Supplementary Figure 3), whereas that of 21,072,329, 21,080,534, and KP2000557 remained opaque, and it was difficult to centrifuge them completely (Figures 2A–C). However, in the string test, only KP2000557 could form strings above 1 cm. Nonetheless, we identified K. pneumoniae isolates carrying rmpA and rmpA2 and having a higher or similar lethality to G. mellonella than NTUH-K2044 as hvKp. All 6 isolates of MDR-hvKp were ST11 isolates; thus, we focused our study on ST11. We performed human whole blood survival assays and human neutrophil bactericidal assays. Of note, 10⁶ CFU of MDR-hvKp KP2000557 did not survive in healthy human whole blood but resisted human neutrophil killing (Supplementary Figure 4). 10⁶ CFU of 21,072,928, 21,111,661, and KP2000557 were unable to survive in healthy human whole blood.

Whole genome sequencing revealed a total of 414 resistance genes in 39 isolates (ST11, ST15, and ST307) (Figure 3). The ESBL gene that was most frequently detected and carried by each isolate was bla_SHV, with 43 in total (bla_SHV–12: 8, bla_SHV–106: 23, bla_SHV–158: 9, bla_SHV–187: 3). Other detected ESBL genes included bla_CTX–M (bla_CTX–M–14: 1, bla_CTX–M–15: 17, bla_CTX–M–65: 4). In addition to ESBL genes, other β-lactamase encoding genes were also detected including bla_TEM (bla_TEM–1: 25), bla_KPC (bla_KPC–2: 18), bla_OXA (bla_OXA–1: 14), bla_LAP (bla_LAP–2: 6), bla_DHA (bla_DHA–1: 1) and bla_SHV–11 (bla_SHV–11: 1) (Liakopoulos et al., 2016). The rmpA and rmpA2 were only detected in ST11 isolates and were located on plasmids. Five ST11 isolates carried both rmpA and rmpA2 (21,080,237, 21,080,534, 21,080,937, FRPDR, and KP2000557). All isolates (ST11 and ST15), except ST307, carried ybt, which encodes yersiniabactin. Only MDR-hvKp 21080937 carried iro, which encodes salmochelin. The iuc encoding aerobactin was only detected in ST11 isolates and all rmpA-carrying isolates carried iuc. The clb encoding colibactin was absent in all isolates (Supplementary Table 5).

The generation of most MDR hvKp is due to the strains acquiring hypervirulence genes and MDR genes. These genes can be spread between strains via plasmids. Therefore, we analyzed the surrounding sequences of hypervirulence genes and MDR genes (ESBL, carbapenemase-encoding genes, and other β-lactamase encoding genes) carried by MDR hvKp and their plasmids. We performed third-generation sequencing on the 6 MDR-hvKp isolates and two polymyxin-resistant isolates. IncR plasmid p21072329_1 carried bla_CTX–M–65, bla_KPC–2, and bla_TEM–1 (Supplementary Figure 5A). p21080237_2 and p21091025_1 belonged to the IncFII/IncR plasmid and carried bla_KPC–2, bla_SHV–12, and bla_TEM–1 (Supplementary Figures 5B,C). pFRPDR_2 belonged to the IncFII plasmid and carried bla_CTX–M–65, bla_KPC–2, bla_SHV–12, and bla_TEM–1 (Supplementary Figure 5D). pKP2000557_3 belonged to the IncFII/IncR plasmid and carried bla_CTX–M–65, bla_KPC–2, bla_SHV–187, and bla_TEM–1 (Supplementary Figure 5E). In addition, with the exception of 21,080,937, each isolate also carried a plasmid containing bla_LAP–2 (Figure 4A and Supplementary Figure 6). All isolates carried three sequences surrounding bla_TEM–1, namely bla_TEM–1 alone (p21080237_2), other drug resistance genes-bla_CTX–M–65-bla_TEM–1 (p21072329_1), and bla_CTX–M–65-bla_TEM–1 (p21091025_1, pFRPDR_2, pKP2000557_3) (Figure 4B and Supplementary Figure 7A). All isolates had bla_KPC–2 co-existing with the surrounding ISs. All isolates had bla_SHV–12 co-existing with upstream TnAs1. And some isolates had bla_KPC–2 appearing upstream of bla_SHV–12 (p21080237_2, p21080534_1, p21091025_1, pFRPDR_2) (Supplementary Figure 7B). Chromosomally located bla_SHV appeared to be site-specific, with sequences inserted downstream of the lac operon (Supplementary Figure 7C). The bla_LAP–2 of all isolates formed the structure of TnAs1-ftsI- bla_LAP–2 –ISAs17-qnrS1-hin-ISKpn19 (Supplementary Figure 7D).

The sequences around rmpA2 carried by all isolates were completely identical, namely IS630-iucA-iucB-iucC-iucD-iutA-rmpA2-ISEc16 (Figures 5A–C). Meanwhile, the RmpA2 of all isolates were identical (Supplementary Figure 8A). The rmpA on p21080937_2 inserted a guanine (G) into the 286th nucleotide sequence, resulting in a frameshift mutation of rmpA (Supplementary Figure 8B). And a missense mutation (418C > A, Leu140Ile) occurred in rmpA on pKP2000557_2. There were mainly two types of sequences around rmpA in each isolate, the presence or absence of the downstream salmochelin operon iro (Figure 5D). By comparing the gene clusters of different isolates, we found that the original rmpADC gene cluster was rmpA-rmpC-rmpD-iroN-iroD-iroC-iroB. Because of the insertion of ISKpn26 into iroN, the sequence of the gene cluster was changed, resulting in the insertion mutation. ISs-mediated transfer may lead to rmpADC detachment from iroBCDN, resulting in rmpADC gene cluster sequences on p21080237_1, 21080534_4, pFRPDR_1 and pKP2000557_2 (Supplementary Figure 9).

Through comparison, we found that some plasmids carried by hvKp have extremely high similarity. Compared with p21080237_2, p21091025_1 carried an ESBL gene, bla_CTX–M–65, which was absent from p21080237_2 (Figure 4C). NCBI database comparison revealed that p21072329_3 exhibited highly similar to the pMTY12128 from Japan (Supplementary Figure 10A). pMTY12128 had only an additional insertion of ISKpn8 than p21072329_3; otherwise, they share 100% homology. Furthermore, p21072329_4 shared 100% homology with p21080237_6 (Supplementary Figure 10B). All isolates, except 21,080,237, contained a ColRNAI plasmid, p21072328_5 (5,596 bp), which carries genes of unknown function (Supplementary Figure 11C).

To determine the relatedness of our isolates to other Chinese isolates, we collected 4,973 whole genome sequences of all K. pneumoniae isolates from China contained in the NCBI database as of 2021. Of these isolates, 2,524 (50.75%) ST11 isolates were detected. We conducted a comparative genomics analysis of the ST11 and ST751 isolates (Supplementary Figure 12A). Fifteen ST11 isolates and one ST751 isolate constituted three clades based on the phylogeny which also corresponds to the K-types (Supplementary Figure 12B). The lowest SNP among all ST11 strains was 119 (21,080,237 and 21,080,631) and the highest was 3,797 (21,072,329 and 21,101,834) (Supplementary Figure 13). 21,080,237 and 21,080,631 shared high homology and they belonged to the same clade as the isolates from Zhejiang. 21,080,534 and 21,083,181 were similar to the isolates from Nanjing (KPN 02 and KPN08, etc.).

Among the MDR-hvKps, only 21,072,329 showed a hypervirulence and hypermucoviscosity phenotype without rmpADC and rmpA2. Furthermore, we did not detect mcr-like genes, such as mcr-1, in the genome or plasmid of the ST11 polymyxin-resistant isolates, FRPDR, and 21,091,025. In addition, mgrB had no IS insertion.

Using 20,172,329 as a reference genome, we identified 6,283 non-repetitive mutations. Of these, 5,508 mutations were in coding sequences (CDS), 773 mutations were in non-coding regions, and 2 mutations were in tRNA coding regions. Among the mutations in the CDS, there were 4,015 synonymous mutations, 1,390 missense mutations, 64 frameshift mutations, 30 nonsense mutations, and 4 intragenic mutations. The putative functions of these mutated CDSs were classified using the eggNOG database (Figure 6A). Although “cell wall/membrane/envelop biogenesis” exhibited the highest number of mutations, most of the mutations were synonymous mutations (276) and there were only 83 missense mutations, less compared with “carbohydrate metabolism and transport” (138) and “transcription” (99). In addition, there were many genes related to “cell wall/membrane/envelop biogenesis” (260) (Figure 6B), which does not indicate that this function is prone to mutation.

Isolates 21,091,025 and FRPDR did not carry mcr-like genes, we hypothesized that the mutation may be located in the chromosome-mediated polymyxin resistance genes arnT, eptA (pmrC), mgrB, phoPQ, pmrAB, pmrD, and ugd. The SNPs of colistin-related resistance genes were counted and 28 SNPs were found, including 4 missense mutations and 24 synonymous mutations (Figure 6C). In 21,091,025, the 572nd nucleotide of phoP was mutated from adenine (A) to cytosine (C), resulting in a mutation of the 191st amino acid of the encoded PhoP from Asp to Ala. Both phoE and pmrA of FRPDR were mutated. The 273rd nucleotide of phoE is mutated from cytosine (C) to adenine (A), causing the mutation of the 91st amino acid of its encoded PhoE from Ser to Arg. The 157th nucleotide of pmrA is a guanine (G) mutated to a thymine (T), resulting in a mutation of the 53rd amino acid of PmrA encoded by Gly to Cys. These mutations may lead to polymyxin resistance. In addition, the 115th amino acid of PhoP of 21,120,240 was also mutated (Leu115Pro), however, this did not result in polymyxin resistance. We analyzed the SNPs of 21,072,329 to explore the reasons for its hypervirulence. K. pneumoniae with different capsule types differs greatly in the gene cluster. Thus, we only analyzed the mutations present in 21,072,329 compared with other ST11-KL47 isolates, including ST11-KL47 K. pneumoniae 19PDR22 that we previously obtained (He et al., 2022). Through a comparison of different ST11-KL47 isolates, we found two SNPs that were present only in 21,072,329 and located in wzc (221T > G, Leu74Arg) and gene 23445 (845G > T, Gly282Val) (Figure 6D). Wzc is a tyrosine kinase involved in capsular polysaccharide (CPS) synthesis and assembly, and 23445 encodes the LysR family transcriptional regulator. However, it remains to be determined whether their mutation leads to the generation of a hypervirulence phenotype.