Virulence-associated gene carriage in CRKP isolates

We screened for the presence of five hypervirulence-associated genes (rmpA, rmpA2, iroB, iucA and peg344) by PCR in all CRKP isolates^13,18,19 and classified 413 (40.6%) isolates carrying both iucA and rmpA/rmpA2 genes as hypervirulent CRKPs (hv-CRKPs)²⁰, including 31 isolates with all five hypervirulence-associated genes. The majority of hv-CRKPs were from K types KL64 (312/413, 75.5%) and KL47 (67/413, 16.2%). Notably, among the top ten K-type strains, KL60 contained the highest proportion of hv-CRKP strains (7/11, 63.6%), followed by KL64 (312/513, 60.8%), KL62 (5/12, 41.7%) and KL47 (67/263, 25.5%) (Fig. ​(Fig.2a2a).

Open in a separate window

Fig. 2

Distribution of AMR and hypervirulence-associated genes among CRKP strains.

a, The proportion of hv-CRKP in different K-type CRKP isolates in China during 2016–2020. The grey bars indicate strains with iucA and rmpA/rmpA2 genes, and the black bars indicate strains with all five virulence genes (rmpA, rmpA2, iroB, iucA and peg344). The number of CRKP strains of K type, n, is given below. b, Trends in the proportion of hv-CRKP in all CRKPs and those from KL64 and KL47. Annual and total hv-CRKP rates in KL64 and KL47 strains in different regions are shown below. Grey blocks indicate years with less than three KL64 or KL47 strains. c, Circular alignment of virulent plasmids pST11-Vir-2 (CP039526) and pST11-Vir-3 (CP048945) referring to pST11-Vir-1-KPC (AY378100) via BRIG. Representative genes are labelled in the outermost circle, with the hypervirulence-associated genes in red.

The proportion of hv-CRKP increased over the five years, from 28.2% in 2016 to 45.7% in 2020 (P value for trend=0.0061), which could be attributed to both the population expansion of KL64 and a threefold increase in hv-CRKP levels in KL47, from 10.1% in 2016 to 40.5% in 2020 (P value for trend=1.1×10⁻⁴). By contrast, the frequency of hv-CRKP in KL64 remained stable over the period 2016–2020 (P value for trend=0.61) (Fig. ​(Fig.2b).2b). High levels of hv-CRKP in KL64 were found in Henan (88.2%), Zhejiang (87.8%), Beijing (81.8%) and Jiangsu (80.7%), while most KL47 strains in Shanghai (81.3%) and Liaoning (69.0%) were hv-CRKP. Conversely, low hv-CRKP frequencies were found in both KL64 (28.2%, 21.2%) and KL47 (0%, 0%) isolates from Hubei and Sichuan throughout the study period. Notably, the prevalence of KL47 hv-CRKPs increased significantly in Liaoning (2016–2019, P value for trend=0.019) and Beijing (2016–2020, P value for trend=0.037) (Fig. ​(Fig.2b2b).

The varying levels of hv-CRKP in KL64 and KL47 strains were partially explained by their associated virulence plasmids (Fig. ​(Fig.2c).2c). Details of the levels of hv-CRKP in KL64 and KL47 strains with associated virulence plasmids are described in Supplementary note.

Genomic epidemiology of KL47 and KL64 CRKP in China

To obtain a phylogenetic overview of the two most common K-type CRKPs, we performed whole-genome sequencing on 207 KL47 and 319 KL64 isolates and found that >99% (523/526) of them belonged to sequence type ST11 (Supplementary Tables 5 and 6). These 523 ST11 isolates were integrated with additional 3,047 public ST11 genomes to form a global genomic collection of 3,570 isolates. After filtering recombination regions, a maximum likelihood phylogeny of the ST11 lineage was estimated based on the remaining non-repetitive, non-recombinant single-nucleotide polymorphisms in the core genome. We manually separated the tree into four major lineages, of which L2 and L3 were broadly consistent with clonal groups identified based on the cgLIN codes¹⁴ and L1 is a subclade of the L2 lineage that consisted of almost exclusively Chinese isolates (99.1%, 2,111/2,130). Based on kleborate predictions, we found a strong correlation between the 44 K types and 6 O types in ST11, with KL64:O2 and KL47:OL101 accounting for 61% (2,187/3,570) of the strains, especially those in L1. The most common carbapenemase in ST11 was bla_KPC-2, followed by bla_OXA-48 and bla_NDM-1 (Fig. ​(Fig.3a3a).

Open in a separate window

Fig. 3

The population structure of ST11 strains and transmission patterns of the L1 lineage.

a, The population structure of ST11 strains. The metadata is shown on the right-hand side. The grey boxes show the four lineages identified along the tree. CARB, carbapenemase; CGs, clonal groups predicted by cgLIN codes. b, The maximum likelihood tree of lineage L1 with branches recalibrated by BactDating. Dates next to particular branches indicate their estimated date of origin with the 95%CI in brackets. Branches are colour-coded according to their estimated regional origin with root nodes of KL47 and KL64 clades shown nearby. Additional geographic estimates of the nodes have been calculated using three models implemented in the APE package in R and visualized in Extended Data Fig. ​Fig.3.3. The major regional transmissions (≥8 isolates in the target region) in each of the KL47 (8 events; blue) and KL64 (12 events; orange) clades are shown as grey arcs with serial numbers on the outside. The outer rings show regions, O types and K types, carbapenemase-carrying plasmids, virulence plasmids and each of the five hypervirulence genes from inside to outside, each colour-coded according to the key. The star in the KL64 clade shows a cluster of hv-CRKP isolates, with their times of origin and 95%CI nearby. There were 19 clusters of 14 rare serotypes as predicted by kleborate in lineage L1, which were labelled accordingly. c, National distributions of the ST11 L1 strains. Each region is colour-coded according to the percentage of KL64, and the arrows indicate regions that received ≥5 inter-regional transmissions from Shanghai (red) or Beijing (blue), with the thickness proportional to the number of transmissions. Distinct patterns were found between Shanghai and Beijing transmissions. Shanghai strains were mainly transmitted to the eastern and central regions, whereas Beijing strains were transmitted to the northern and northeastern regions, in line with the patterns of inter-regional health seeking in China. d, The number of observed transmissions originated from each region extracted from the L1 tree in b. The two predominant sources (Shanghai and Beijing) are highlighted in red.

Spatiotemporal transmission of L1 CRKP within China

Most of the strains in lineage L1 were from China (2,111 strains), followed by the USA (7), Oman (3), Singapore (3), and four other countries. Temporal signals were found in lineage L1 based on regression of root-to-tip genetic distance against sampling time (R₂=0.18, P=8.2×10⁻¹²) and further demonstrated based on two date randomization tests (Extended Data Fig. 2a–c and Methods). Therefore, based on the phylogeny above, we estimated the spatiotemporal dynamics of lineage L1 using BactDating and inferred the fluctuation of its effective population sizes using skygrowth in R. The ancestral inter-regional transmissions and serovar changes were also inferred using both TreeTime and the Analyses of Phylogenetics and Evolution (APE) package in R. The most recent common ancestor of L1 was estimated to be present before 1983 (95% confidence interval (95%CI) 1979–1987) in serotype KL47:OL101, whereas the KL64:O2 serotype was estimated to emerge in 2003 (95%CI 2002–2004) as a subclade of L1 and experienced rapid national dissemination in the past decade. Notably, we also predicted 14 other K genotypes in L1, some of which formed local clusters, that is, KL25 in Sichuan and KL110 in Jiangsu (Fig. ​(Fig.3b3b and Supplementary Table 7).

Open in a separate window

Extended Data Fig. 2

The root-to-tip distances and date randomisation tests for Lineage L1.implemented in the ACE function in the APE package.

a: A scatter plot generated by TempEst that shows linear regression of root-to-tip distances against isolation dates of the strains. The coefficient of determination (R2) and P-value is shown as well. The residual mean square was calculated using two-sided test. b, c: The date-randomisation tests based on the ‘True data’ and 10 datasets with the isolation dates of 1,756 samples randomly permutated. Data are presented as mean values±SEM in part C. Significant greater R2 values of the root-to-tip tests were obtained from the true data than all the randomized data (B). And significantly greater substitution rates were also obtained based on the BactDating inferences (C).

The KL47 strains were estimated to originate in either Zhejiang or Beijing by different models, although all agreed on its long-term persistence in Zhejiang since 1990 (95%CI 1986–1993) (Fig. ​(Fig.3b3b and Extended Data Fig. ​Fig.3).3). By contrast, the KL64 clade was estimated to originate in Zhejiang by all models. We predicted a total of 308 cross-regional transmissions along the phylogeny, which were branches that connect nodes with different geographic inferences by TreeTime. We found that 53.6% (165) of these transmissions had parental nodes in either Beijing or Shanghai, the two major sources of transmissions (Fig. ​(Fig.3c).3c). Among them, Beijing was estimated to contribute 75 transmissions, mostly to northern and northeastern regions, and Shanghai was estimated to contribute 90 transmissions, mainly to eastern and central regions (Fig. ​(Fig.3d3d).

Open in a separate window

Extended Data Fig. 3

Ancestral state reconstruction of L1 Lineage using the maximum likelihood (ML) algorithm implemented in the ACE function in the APE package.

The inferences were run based on three different models of (a) ER, which assumes equal transmission rates across all regions, (b) SYM, which assumes different, yet symmetric rates for transmissions between different pairs of regions, and (c) ARD, which is an all-rates-different model. Branches were colour-coded according to their most likely regional origins in the inferences (see keys). And the estimated possibility of the top region in certain major branches were shown nearby.

Furthermore, the majority of the transmissions leave only one or a few isolates in the target regions. We, therefore, focused on 20 major events that were evidenced by ≥8 isolates in the target regions (Fig. ​(Fig.3b).3b). Many of the major transmissions in the KL47 strains involve Beijing and its neighbouring, northern regions such as Henan (event KL47.2), Liaoning (event KL47.2) and Tianjin (events KL47.2 and KL47.7), contributing to the predominance of KL47 clade in northern China (blue route; Fig. ​Fig.3c).3c). Conversely, KL64 strains were estimated to experience nine major transmissions from Shanghai to central and eastern China including Jiangsu (event KL64.1), Hubei (event KL64.2), Sichuan (event KL64.5) and other regions (red route, Fig. ​Fig.3c),3c), resulting in their increasing KL64 frequencies.

Non-capsulated CRKP exist among KL64 and KL47 isolates

All isolates identified as KL64 or KL47 by genotyping were subjected to serologic typing, and 7.4% (38/513) KL64 and 12.2% (32/263) KL47 isolates, respectively, did not react with the corresponding antisera in the Quellung reaction (Extended Data Fig. ​Fig.44 and Supplementary Table 8). These isolates appeared translucent, showed a significant reduction in polysaccharide capsule uronic acid (Fig. ​(Fig.4a)4a) and were confirmed as NEKp (non-capsulated K. pneumoniae) strains by electron microscopy (Fig. ​(Fig.4b).4b). The frequency of NEKp did not change significantly over the study period (Supplementary Table 9; P value for trend=0.97 and 0.99) nor between samples from different body sites (Fig. ​(Fig.4c;4c; P=0.47).

Open in a separate window

Fig. 4

Capsule deficiency in clinical isolates.

a, CPS of NEKp and selected paired EKp strains measured by uronic acid assay. Paired EKp strains were randomly selected from regions and samples similar to those of NEKp. Parental EKp strains and their isogenic Δcps mutants are indicated by arrows. The 20–445Δcps and 18–81Δcps mutants had a significantly lower amount of polysaccharide capsule uronic acid compared to WT strains. The number of EKp and NEKp strains of K type KL64 or KL47, n, is given below. Data are shown as the mean±s.e.m. Each point represents the mean value of three biological replicates of one strain. For clarity, individual error bars for each strain are not shown. Unpaired two-sided Welch’s t-test was performed for statistical analysis. b, Transmission electron microscopy images of representative EKp (KL64 strain 19–241 and KL47 strain 18–729) and NEKp (KL64 strain 19–242 and KL47 strain 18–737) strains. Each electron micrograph shows the representative result of three repetitions. c, Frequency of NEKp strains isolated from different sites of infection. The number of CRKP strains in different infection sites is shown below. Pearson’s chi-squared test was used to analyse the proportion difference of NEKp strains in different infection samples. d, cps-locus and rfaH gene variants in NEKp isolates. Mutations found in the capsule biosynthesis genes in KL64 and KL47 NEKp isolates. The black flag indicates the location of the mutation. Red, black and blue lines indicate ISs, single-nucleotide polymorphisms and deletions, respectively. n, the total number of mutations observed in this gene. e, Metagenomic analysis of cps-locus variants in unprocessed clinical samples (for details, see Supplementary Table 10). NS, not significant.

Source data

Open in a separate window

Extended Data Fig. 4

Capsule deficiency in KL47 and KL64 genotyped clinical isolates.

Quellung reaction of K. pneumoniae strains. K. pneumoniae strains were incubated with anti-K64 serum (or anti-K47 serum) or control serum as described in Methods. Compared with the control, the positive results showed capsular swelling under light microscopy with an oil immersion lens (magnification, 100). WT strains (KL64: 20–445, KL47: 18–81) and isogenic Δcps mutants showed Quellung reaction positive and negative, respectively. Quellung reaction of 20–445Δcps and 18–81Δcps was restored by complementation. Besides, Quellung reaction was restored by complementation with wcaJ or wbaP in a set of corresponding gene mutant clinical NEKp strains (KL64: 19–479, 19–501; KL47:16–30, 16–133). Each graph shows the representative result of three repetitions. The arrow points to the swollen capsule. Scale bar, 5µm.

Various mutations, including insertion sequences (ISs), point mutations and deletions, were found in the cps region or the rfaH gene of NEKp strains, mainly affecting the initial glycosyltransferase genes (wbaP for KL47 and wcaJ for KL64) (Fig. ​(Fig.4d4d and Supplementary Table 8). Capsule-deficient phenotypes were replicated in the mutant strains 18–81-KL47ΔwbaP and 20–445-KL64ΔwcaJ, and their capsule production could be restored upon complementation (Fig. ​(Fig.4a4a and Extended Data Fig. ​Fig.4).4). Complementation with wcaJ/wbaP also restored capsule production in clinical NEKp strains with corresponding gene disruptions, further supporting the link between wcaJ/wbaP disruptions and the capsule-deficient phenotype (Extended Data Fig. ​Fig.44).

To investigate whether such capsule deficiency occurs during in vitro culture in the laboratory, we cultured ten random strains for each of KL47 and KL64 in blood agar plate for 48h and found very low levels (<0.56%) of capsule-deficient mutants (Extended Data Fig. ​Fig.5).5). In addition, we metagenomically sequenced 49 unprocessed clinical samples with CRKP infections. Among the 11 samples with >5-fold coverage, three had disrupted cps genes, including nonsense mutations or ISKpn26 insertions in wcaJ, and other disruptions, demonstrating the presence of capsule-deficient mutations in vivo (Fig. ​(Fig.4e4e and Supplementary Table 10).

Open in a separate window

Extended Data Fig. 5

The mutation of capsule deletion rate in vitro culture.

The mutation of capsule deletion rate of KL64 and KL47 EKP isolates after 48hours of culture in blood plate. Data are shown as the mean±SEM. Each point represents the mean value of three biological replicates of one strain. For clarity, individual error bars for each strain are not shown.

Source data

Characterization of NEKp strains under infection pressures

We compared NEKp/capsule-deficient (isogenic Δcps) strains with EKp (encapsulated K. pneumoniae) strains in vitro and in vivo to assess the impact of capsule synthesis on fitness. Consistent with previous studies²¹, NEKp/Δcps strains showed increased biofilm production and higher susceptibility to macrophage phagocytosis (Fig. 5a–c). Mouse intraperitoneal infection experiments showed that the isogenic Δcps mutants showed reduced lethality and competitive fitness for dissemination to the lung, liver and spleen compared to the parental EKp strains (Fig. 5d–h). Surprisingly, the serum-killing assay revealed a higher proportion of serum-resistant strains in NEKp strains than in EKp strains (Fig. ​(Fig.6a;6a; KL47 P=0.018, KL64 P=2.0×10⁻⁴). Notably, the isogenic Δcps mutants did not show significant growth advantages over their parents in the Luria–Bertani (LB) media (Extended Data Fig. ​Fig.6),6), suggesting that reduced metabolic load due to abolished capsule synthesis is unlikely to be a causative factor. By comparing the antiserum capacity of parental EKp strains, capsule polysaccharide depolymerase-treated (capsule-stripped) EKp strains, isogenic Δcps and ΔcpsΔops mutants, we demonstrated that both CPS and O-antigenpolysaccharide (OPS) play a critical role in serum resistance (Fig. ​(Fig.6b6b and Extended Data Fig. ​Fig.7).7). In addition, we found that isogenic Δcps mutants (except O⁻ strains) showed increased serum resistance along with enhanced O antigen expression compared to the capsule-stripped EKp strains (Fig. ​(Fig.6b6b and Extended Data Fig. ​Fig.7).7). Transcriptomic and real-time PCR (RT-PCR) analyses also revealed the overexpression of LPS and peptidoglycan synthesis genes in Δcps strains compared to parental strains (Extended Data Fig. ​Fig.8a).8a). These results suggest that disruption of capsule synthesis enhances the synthesis of O-antigen and ultimately increases antiserum capacity. We speculated that the cessation of CPS synthesis might lead to reduced consumption of undecaprenyl phosphate (Und-P) pool, a limiting factor in the biosynthesis of glycans²². This could potentially redirect Und-P for excess O-antigen synthesis. Our observations suggest a potential link between altered Und-P dynamics and changes in serum resistance, as indicated in Extended Data Fig. ​Fig.8b.8b. Nonetheless, the exact mechanisms and precise Und-P flux alterations following CPS synthesis disruption need to be further investigated.

Open in a separate window

Fig. 5

Evaluation of the effect of capsule synthesis on fitness in vitro and in vivo.

a, Biofilm formation of NEKp strains and the selected paired EKp strains. Parental EKp strains and their isogenic Δcps mutants are indicated by arrows. Data are shown as the mean±s.e.m. Each point represents the mean value of three biological replicates of one strain. For clarity, individual error bars for each strain are not shown. Unpaired two-sided Welch’s t-test was performed for statistical analysis. b, Phagocytosis of EKp and NEKp strains by RAW 264.7 macrophages. c, Phagocytosis of EKp and NEKp strains by peritoneal macrophages. Peritoneal macrophages were collected from 7-week-old female C57BL/6J mice. In b and c, WT parental EKp strains were more resistant to phagocytosis than isogenic Δcps strains. Data represent the mean±s.e.m. of four biological replicates of one strain. Unpaired two-sided Student’s t-test with Welch’s correction was performed for statistical analysis in b and c. d–h, The capsule of KL47 (c,f–h) and KL64 (e,f–h) K. pneumoniae impose competitive fitness in the mouse intraperitoneal infection model. The KL47 (d) and KL64 (e) survival rate was evaluated after intraperitoneal infection of C57BL/6J mice with EKp strains or their isogenic Δcps mutants at the assigned dose (19–89 or 19–89Δcps: 3×10⁷ CFU; 16–312, 18–404, 18–729 and 16–140 or their Δcps mutants: 6×10⁷ CFU). Kaplan–Meier analysis and log-rank test was performed for statistical analysis. n=6 (18–404 and 19–89 or their Δcps mutants), n=5 (18–729, 16–140 and 16–312 or their Δcps mutants). Bacterial load was assessed after intraperitoneal infection of C57BL/6J mice with a 1:1 ratio of WT to isogenic Δcps mutant. WT and Δcps mutants in the lung (f), liver (g) and spleen (h) were quantified at 24hpi to calculate the competitive index value. Two-sided one-sample t-test was performed to compare the mean of log₁₀-transformed competitive index to a hypothetical value of 0 for each group. Each dot represents an individual mouse. n=5.

Source data

Open in a separate window

Fig. 6

NEKp strains have survival advantages under infection pressure.

a, Distribution of serum resistance levels of NEKp strains and selected paired EKp strains. Bacteria were incubated in 75% baby rabbit serum, and the antiserum levels were determined. Colours represent different levels of serum resistance. The mean rank difference of two groups was analysed by Cochran–Mantel–Haenszel test. n indicates the number of strains. b, Effect of capsule deficiency on O-antigen expression and its association with serum resistance. Serum-killing assays were performed by incubating the EKp strain 16–312, KL64 CPS depolymerase (Dep41¹⁵)-pretreated strain16–312 (capsule-stripped), 16–312Δcps, 16–312ΔcpsΔops and strains complemented with cps or ops with 75% baby rabbit serum for 1 or 2h at 37°C. The percentage of survival was determined as the number of surviving bacteria relative to the initial bacterial addition. Unpaired two-sided Student’s t-test with Welch’s correction was used for statistical analysis. Data are shown as the mean±s.e.m. of three biological replicates and represent one experiment of three independent experiments. LPS was prepared by hot phenol–water method and analysed by silver staining. Core-OS, core-oligosaccharide. c, SBA against parental EKp and isogenic Δcps strains by serum from the patients infected with KL47 or KL64 K. pneumoniae. K. pneumoniae strains were incubated with the heat-inactivated serum from patients for 1h at 37°C in the presence of baby rabbit complement. Survival ratios were normalized to the survival strains in the presence of baby rabbit complement alone. Data are the mean±s.e.m. of four biological replicates. Unpaired two-sided Student’s t-test was used for statistical analysis. d, Model for the surface polysaccharide switch affects K. pneumoniae fitness during infection. The de novo biosynthesis of Und-P can be inhibited by fosmidomycin and increased by overexpression of ispU. CPS synthesis gene mutation results in the inability to utilize Und-P, thereby probably increasing the flow of Und-P to O antigen synthesis, with the consequence of increased biofilm formation, maintained or even increased antiserum capacity and facilitated evasion of antibody-mediated killing, although deficient in dissemination. Gene ispU encodes the undecaprenyl-pyrophosphate synthase; G3P, glyceraldehyde-3-phosphate.

Source data

Open in a separate window

Extended Data Fig. 6

Capsule of KL47 and KL64 K. pneumoniae does not impose a fitness cost in vitro.

In vitro competition assays were performed between KL47 strains (16–140, 18–404 and 19–249) and their isogenic Δcps mutants, and between KL64 strains (16–312 and 19–89) and their isogenic Δcps mutants. Bacteria were inoculated into LB broth at a ratio of 1:1 for 24h at 37°C, and WT and Δcps mutants were quantified by differential plating to calculate the competitive index (CI) value. Two-sided one-sample t-test was performed to compare the mean of CI to a hypothetical value of one (asterisk) for each group. Data represents the mean (±SEM) of three biological replicates. NS, no significance.

Source data

Open in a separate window

Extended Data Fig. 7

Effect of capsule deficiency on serum resistance and O antigen expression.

a: Serum killing assays were performed by incubating the parental EKp strains (twelve strains with different levels of serum resistance), KL64 capsular polysaccharide depolymerase (Dep41) or KL47 depolymerase (Dep42) pretreated parental EKp strains, the engineered isogenic Δcps strains with 75% baby rabbit serum for 1hour at 37°C. Percent survival was determined as the number of surviving bacteria relative to the initial bacterial addition. b: LPS was prepared by hot phenol-water method and analyzed by silver staining. The results of the serum killing assay showed that all capsule-stripped parental strains exhibited decreased serum resistance compared to that of untreated parental strains (except for those serum-sensitive strains that barely survived 1hour post serum treatment) (A), demonstrating that K. pneumoniae capsules do play an important role in serum resistance. Six out of twelve isogenic Δcps mutants exhibited increased resistance to serum compared to the capsule-stripped parental strains (A). Increased amounts of OPS were observed in the isogenic Δcps mutants compared to the parental EKp strains (B). The remaining six isogenic Δcps mutants which showed no change in resistance to serum compared to the the depolymerase-treated parental strains, were all serum-sensitive strains with disruption of O antigen synthesis, as confirmed by LPS silver staining (B) and genome sequence analysis (A). N.D., rfb-locus mutation gene not detected in O⁻ strain 16–140. Each point represents the mean value of three biological replicates. For clarity, individual error bars for each strain are not shown.

Source data

Open in a separate window

Extended Data Fig. 8

Investigation of the mechanism underlying the increased antiserum capacity of NEKp strains.

a: RNA sequencing and qRT-PCR of WT and Δcps strains. RNA sequencing heat map showing the differential expression of LPS and PG synthesis genes between EKp strain 16–312 and 16–312 Δcps mutant. qPCR assay was used to determine the relative mRNA levels of LPS and PG synthesis-related genes in 16–312 and 16–312 Δcps using the 2^-ΔΔCT method. PG, peptidoglycan. Data represents the mean (±SEM) of three biological replicates of one strain. Unpaired two-sided Welch’s t-test was performed for statistical analysis. b: Effect of Und-P on serum resistance using either fosmidomycin to inhibit the Und-P de novo biosynthesis or ispU overexpression to increase the Und-P biosynthesis. Strains 16–312Δcps were grown in LB medium with or without FM (500ng/ml); strains 16–312Δcps /ispU and 16–312Δcps/ EV were grown in LB medium with 10mM arabinose. The serum killing assay was performed by incubating the strains with 75% baby rabbit complement for 1hour at 37°C. Percent survival was determined as the number of surviving bacteria relative to the initial bacterial addition. FM, fosmidomycin; EV, empty vector. Data represents the mean (±SEM) of four biological replicates of one strain. Unpaired two-sided Student’s t-test was performed for statistical analysis. Core-OS, core-oligosaccharide.

Source data

Furthermore, compared to the parental KL47/KL64 EKp strains, the isogenic Δcps mutants showed higher resistance to heat-inactivated sera collected from patients infected with K. pneumoniae KL47/KL64 in the serum bactericidal activity (SBA) assay, suggesting capsule depletion as a potential mechanism for evading humoral immune responses (Fig. ​(Fig.6c6c).

Collectively, NEKp strains showed a fitness advantage by increasing biofilm formation, serum resistance and anti-SBA but deficiencies in dissemination and anti-phagocytosis (Fig. ​(Fig.6d6d).

Strain collection and antimicrobial susceptibility testing

A total of 1,017 CRKP strains were non-duplicate sequentially collected from K. pneumoniae infected patients admitted to 40 hospitals in 26 Chinese cities every October 1–31 between 2016 and 2020 by the China Antimicrobial Surveillance Network (www.chinets.com). All isolates were identified by matrix-assisted laser desorption ionization–time of flight mass spectrometry (bioMérieux, France). Minimal inhibitory concentrations (MICs) were determined using the reference broth microdilution method recommended by the Clinical and Laboratory Standards Institute (CLSI), which was used to categorize K. pneumoniae as susceptible (S), resistant (R) or intermediate (I). The breakpoints of K. pneumoniae according to CLSI M100-ED31 were interpreted for imipenem (S≤1mgl⁻¹, R≥4mgl⁻¹), meropenem (S≤1mgl⁻¹, R≥4mgl⁻¹), CZA (S≤8/4mgl⁻¹, R≥16/4mgl⁻¹), cefepime (S≤2mgl⁻¹, R≥16mgl⁻¹), ceftazidime (S≤4mgl⁻¹, R≥16mgl⁻¹), ceftriaxone (S≤1mgl⁻¹, R≥4mgl⁻¹), cefoperazone–sulbactam (according to cefoperazone S≤16mgl⁻¹, R≥64mgl⁻¹), piperacillin–tazobactam (S≤16mgl⁻¹, R≥128mgl⁻¹), cefuroxime (S≤8mgl⁻¹, R≥32mgl⁻¹), cefazolin (S≤2mgl⁻¹, R≥8mgl⁻¹), amikacin (S≤16mgl⁻¹, R≥64mgl⁻¹), trimethoprim–sulfamethoxazole (S≤2/38mgl⁻¹, R≥4/76mgl⁻¹), aztreonam (S≤4mgl⁻¹, R≥16mgl⁻¹), ciprofloxacin (S≤0.25mgl⁻¹, R≥1mgl⁻¹) and levofloxacin (S≤0.5mgl⁻¹, R≥2mgl⁻¹)³⁴. Tigecycline MICs were interpreted using the US Food and Drug Administration MIC breakpoints for Enterobacterales (S≤2mgl⁻¹, R≥8mgl⁻¹) (https://www.fda.gov/drugs/development-resources/antibacterial-susceptibility-test-interpretive-criteria). European Committee on Antimicrobial Susceptibility Testing (41) (EUCAST) MIC interpretive breakpoints were used for polymyxin B (S≤2mgl⁻¹, R≥4mgl⁻¹). Escherichia coli ATCC 25922 and E. coli ATCC 35218 were tested as the quality control strains for antimicrobial susceptibility testing. The K. pneumoniae strains resistant to at least one of the carbapenem antibiotics (meropenem or imipenem) were defined as CRKP. Abbreviations and their expansions in this study are listed in Supplementary Table 11.

More than half of the patients were male (667; 65.6%). The mean age of the patients was 56.17 (±18.70) years. The most common specimen sources were sputum (458/1,017, 45.03%), blood culture (242/1,017, 23.80%) and urine (120/1,017, 11.80%). See details in Supplementary Table 1. Most isolates (41.00%) were collected from patients in medical intensive care units.

K-typing by wzi genotyping

To identify the K types of K. pneumoniae strain, the wzi allele was amplified by PCR and aligned with the wzi sequences deposited in the database of Institut Pasteur (http://bigsdb.web.pasteur.fr)³¹. KL14, KL64, KL15, KL51 and KL102 were further identified by PCR amplification of the variable sequence of the wzy gene^35–37. Besides, the Kaptive software was also used to identify the capsule synthesis loci (K-loci or KL)³⁸. The primers used for genotyping are listed in Supplementary Table 12.

PCR screening for carbapenem resistance genes and virulence-associated genes

The presence of common carbapenem resistance genes (bla_IMP, bla_NDM, bla_VIM, bla_KPC and bla_OXA-48-like) and virulence-associated genes (rmpA, rmpA2, iroB, iucA and peg344) was confirmed by PCR using the primers described in Supplementary Table 12. The positive PCR amplicons of carbapenem resistance genes were sequenced and aligned with the sequences released from GenBank (www.ncbi.nlm.nih.gov/blast/).

Whole-genome sequencing and bioinformatics analysis

We performed whole-genome sequencing on 207 KL47 and 319 KL64 isolates to obtain a phylogenetic overview of the two most common K-type CRKPs (strains were randomly selected for sequencing, and the selection criteria are shown in Extended Data Fig. ​Fig.1)1) and on non-carbapenemase-producing CRKP strains to analyse other mechanisms of carbapenem resistance. Genomic libraries were prepared using the Rapid Plus DNA Lib Prep Kit for Illumina (catalogue number RK20208) and sequenced using Illumina NovaSeq 6000. The obtained reads were trimmed by Trimmomatic v0.39, and strain assemblies were generated using EToKi assemble³⁹. The assembled genomes were annotated using PROKKA⁴⁰. ISs were predicted using ISfinder (http://www-is.biotoul.fr). AMR genes and virulence genes were predicted using Kleborate⁴¹.

Phylogenetic analysis

Three of the 526 genomic sequenced strains were not in ST11 and therefore not included in the phylogenetic analysis. The remaining 523 genomes, plus all public ST11 genomes, were aligned to a reference genome (GCF_011066505.1; sample from blood, Hong Kong, China, 2016) using the EToKi align module to obtain a multiple sequence alignment of the nonrepetitive core genomic regions shared by ≥95% of the genomes. The resulting alignments were subjected to a maximum-likelihood phylogeny using the EToKi phylo module; the recombinant regions were removed using RecHMM⁴² and were visualized using Phandango⁴³. Furthermore, we used EToKi MLSType to genotype all ST11 genomes based on the scgMLST_629S scheme hosted in Institut Pasteur (https://bigsdb.pasteur.fr/klebsiella/). All the obtained core genome multilocus sequence typing (cgMLST) profiles were separated into clonal groups using the LINcoding script (https://gitlab.pasteur.fr/BEBP/LINcoding) with a cutoff of 69.7933% allele similarity as described in https://bigsdb.pasteur.fr/klebsiella/cgmlst-lincodes/.

Inferences of population dynamics for lineage L1

The origin time of the ST11 population was estimated using BactDating⁴⁴. Parallel chains of 5×10⁶ samples each were run for each of the substitution models of ‘strictgamma’, ‘mixedgamma’, and ‘carc’. The first 50% of the chain (3×10⁶ samples) for each model was discarded as burn-ins, and the convergence of the run was determined by ensuring effective sampling sizes of >100 for all parameters. The results from all samples were compared based on their Bayes factors using the modelcompare function in BactDating, which indicated that the ‘strictgamma’ model was optimal. Finally, the dynamics of effective population sizes were estimated using the skygrowth package in R⁴⁵, and the geographic origins of each node in the tree were estimated using the maximum-likelihood algorithm implemented in TreeTime⁴⁶. In addition, we also estimated the ancestral geographic origins of each node using the Ancestral Character Estimation function in the APE package in R with three different models of ‘ER’, which assumes equal transmission rates across all regions, ‘SYM’, which assumes different, yet symmetric rates for transmissions between different pairs of regions, and ‘ARD’, which is an all-rates-different model. These models generated slightly different results, which are summarized in Extended Data Fig. ​Fig.3.3. The tree was visualized using iTOL v6⁴⁷.

Evaluation of temporal signal in lineage L1

The presence of a temporal signal in lineage L1 was tested using three approaches. The regression of root-to-tip distances and dates of isolation was estimated using TempEst with a correlation coefficient (R) of 0.18 and P value of 8.2×10⁻¹². We then randomly permutated the isolation dates of the strains 10 times and estimated their R values as shown in Extended Data Fig. ​Fig.2b.2b. The same datasets were also used for BactDating inferences as described above, and their substitution rates were compared with the rate from the actual data, which was also shown in Extended Data Fig. ​Fig.2c2c.

Identification of carbapenemase-encoding and hypervirulence-associated plasmids

To detect plasmids in the ST11 isolates, all 6,779 complete sequences of Klebsiella plasmids were downloaded from the National Center for Biotechnology Information (NCBI) RefSeq database (July 2022) and used as references. All 3,570 ST11 genomes were compared with the reference plasmid sequences using BLASTn; the alignment regions for each plasmid among the assemblies were summarized, and only plasmids that have >50% of their sequences aligned with >85% identities were kept. For each genome, we reordered its associated plasmid references according to the alignment scores and selected the hits using a ‘Best-first’ greedy algorithm that is similar to what is described in ref. ⁴⁸. Briefly, for each iteration, the reference plasmid with the greatest alignment score was selected, and all contigs associated with it were taken out. This process was repeated until there was no hit left. Furthermore, to identify hypervirulence-associated plasmids, we predicted the presence of the carbapenemase or the five hypervirulence-associated genes (iucA, iroB, rmpA, rmpA2 and peg344) in the selected contigs based on BLASTn alignments. Only the plasmids that carry at least one of these functional genes were reported. The identified plasmids were then aligned onto the NCBI RefSeq database again to obtain similar plasmids and had their incompatibility and MOB types determined using the MOB-Typer software⁴⁹.

Capsular Quellung assay for genotyped KL47 and KL64 K. pneumoniae strains

The traditional serologic method, the Quellung assay, was performed to further confirm the K-type results for these KL47 (263 strains) and KL64 (513 strains) clinical strains in this study¹⁵. Rabbit K47-antiserum and K64-antiserum were prepared as described previously¹⁵. Bacteria in 1× phosphate-buffered saline (PBS) were mixed with the equal volume of rabbit K64-antiserum (titre of 1:1.5×10⁷) or K47-antiserum (titre of 1: 6.1×10⁶) or control rabbit serum and incubated with the methylene blue solution (Sangon Biotech, catalogue number A610622) on a glass slide for 15min. Finally, the mixture was observed under a light microscope.

Capsule extraction and uronic acid quantification

The capsule polysaccharide of strains was extracted and quantified as previously described^21,50. Briefly, 1×10⁹ colony forming units (CFU) bacteria were pelleted for capsule uronic acid extraction. Bacteria were mixed with 1% Zwittergent 3–14 detergent (Sangon Biotech, catalogue number A610552) in 100mM citric acid (Sangon Biotech, catalogue number A610055). Then uronic acid was precipitated by ethanol (Sinopharm Chemical Reagent, catalogue number 10009218), incubated with 0.0125M sodium tetraborate/sulfuric acid (tetraborate, Sigma-Aldrich, catalogue number 221732; sulfuric acid, Sinopharm Chemical Reagent, catalogue number 10021608) and 0.125% carbazole absolute ethanol (carbazole, Sangon Biotech, catalogue number A600269), finally measured at 530nm and correlated to a standard curve of d-glucuronic acid.

Transmission electron microscopy

Bacteria of 5×10⁸ CFU were pelleted and fixed in 2.5% glutaraldehyde (Sangon Biotech, catalogue number A600875) and 4% polyformaldehyde (Servicebio, catalogue number G1101) for 2h at room temperature. Then the ultra-thin sections were prepared, and the image from transmission electron microscopy was obtained with the Hitachi transmission electron microscope H-9500 (Tokyo, Japan) at the Electron Microscopy Imaging Laboratory of Shanghai Jiaotong University School of Medicine (Shanghai, China).

Estimation of the mutation rate of capsule deletion in vitro

A single colony of the EKp strain was plated on blood agar plate and incubated at 37°C for 48h. All bacterial colonies were then collected and resuspended in 1× PBS. The mutation rate was obtained by counting the proportion of strains that appeared translucent over the total colonies.

Metagenomic sequencing and analysis

Clinical specimens were obtained from Ruijin Hospital Affiliated to Shanghai Jiaotong University School of Medicine. The sputum, urine, bile and pus used for metagenomic sequencing were collected from K. pneumoniae infected patients, 83.3% (10/12) of whom were male. The mean age of the patients was 66.00 (±11.59) years (for details, see Supplementary Table 10). Sputum and pus specimens were incubated with 4% sterilized sodium hydroxide (Sinopharm Chemical Reagent, catalogue number 10019718) for 1h at room temperature to liquefy. Urine and bile specimens were centrifuged to obtain enriched specimens pellet. All specimens were incubated at 83°C for 30min for sterilization, then incubated with 5mgml⁻¹ lysozyme (Vazyme, catalogue number DE103) and 32Uml⁻¹ lysostaphin (Sigma-Aldrich, catalogue number L7386) for 4h at 37°C and incubated with 0.1mgml⁻¹ protease K (LABLEAD Biotech, catalogue number K0510) for 2h at 65°C. Finally, bacterial DNA was further extracted and purified using the TIANamp Bacterial DNA Kit (TIANGEN Biotech, catalogue number DP302). Paired-end libraries with insert sizes of ~300bp were prepared using Illumina’s standard genomic DNA library preparation procedure (VAHTS Universal DNA Library Prep kit for Illumina V3) and sequenced on an Illumina NovaSeq 6000 using the S4 reagent kits (v.1.5) according to the manufacturer’s instructions. Low-quality bases and adapters in the metagenomic reads were screened and removed using BBDuk (https://sourceforge.net/projects/bbmap/). The presence of K. pneumoniae in the samples was determined using both KRAKEN (https://ccb.jhu.edu/software/kraken/) and SPARSE (https://github.com/zheminzhou/SPARSE). Sequences of the K- and O-antigen gene clusters obtained from Kaptive were used as references to identify sequence variations. All reads in each sample were aligned to these reference sequences using minimap2 and visually examined with the help of samtools v.1.2. The long-range deletions and insertions of transposons were identified by recognizing multiple reads encompassing the conjugative sites. The frameshifts and nonsense mutations in the coding sequences were recognized by comparing all sequence variations with gene annotations.

Deletion clones and complementary plasmid construction

The K. pneumoniae cps (wcaJ, wbaP or wbaPwcaA), ops (wbbN or wbbL) and cpsops (wcaJwbbN) mutants were constructed using the λ Red-dependent recombination system and the flippase (FLP)/FLP recognition target (FRT) recombination system according to the methods reported previously^51,52. First, the donor DNA comprising 500bp upstream of the target gene, the hygromycin-resistant gene and 500bp downstream of the target gene was constructed via Gibson. The hygromycin-resistant gene flanked by the FRT was PCR-amplified from pUC19-Hph plasmid. Then the thermosensitive plasmid pKOBEG harbouring the apramycin-resistant gene was introduced into the competent cells derived from the wild-type (WT) strain using electroporation technology. The pKOBEG carried the λ phage red operon controlled by a promoter under the induction of arabinose. The donor DNA was then electroporated into WT pKOBEG. Ultimately, the cps or ops gene in the WT strain was replaced by the FRT-flanked hygromycin-resistant gene. The mutants were selected on LB agar plates containing apramycin (Sangon Biotech, catalogue number A600090) and hygromycin (Rhawn, catalogue number R053999) and confirmed by PCR and Sanger sequencing. The 16–312cpsops double knockout mutant was conducted by first replacing cps (wcaJ) using the hygromycin-resistant gene, then removing the hygromycin resistance gene by FLP /FRT recombination method and finally further deleting the ops (wbbN) gene by the hygromycin-resistant gene replacement. Plasmid pKOBEG was removed by growth at 42°C with shaking.

For complementary plasmid construction, the cps or ops genes of KL64 or KL47 strains were amplified from K. pneumoniae strains 16–312 and 19–249, respectively. They were then cloned into the pACYC184 or pBAD33 vector, and the recombinant plasmids were transformed into the cps or ops mutant strains. In addition, the 16–312Δcps::cps (16–312wcaJ::wcaJ) strain was generated by the λ-red-mediated in situ complementation method. Briefly, donor DNA containing 500bp upstream of the cps gene, the cps gene, the hygromycin resistance gene and 500bp downstream of cps gene was electroporated into 16–312Δcps strain harbouring pKOBEG vector. Mutants were then selected on LB agar plates containing apramycin and hygromycin. Expression of cps or ops genes was confirmed by the Quellung reaction assay or silver staining, respectively. Primers used for deletion clone construction and complementation and details of the engineered mutants are listed in Supplementary Tables 12 and 13 The gene in cps-locus or rfb-locus and their putative gene products are listed in Supplementary Table 14.

Macrophage phagocytosis assay

The phagocytosis assay was performed using the mouse macrophage cell line RAW 264.7 (ATCC catalogue number TIB-71) and peritoneal macrophages as previously described⁵³. Peritoneal macrophages were collected from 7-week-old female C57BL/6J mice on the 5th day after intraperitoneal injection of 1ml 5% thioglycollate broth (BD Difco, catalogue number 211716)⁵⁴. Briefly, 5×10⁵ cells were infected with bacteria at a multiplicity of infection of 50 for 1h. The cells were then washed three times with 1× PBS and treated with 500μl Dulbecco’s modified eagle medium (Servicebio, catalogue number G4515) containing 300μgml⁻¹ gentamycin (Sangon Biotech, catalogue number A506614) or amikacin (Solarbio, catalogue number A9660) for 1h to eliminate the extracellular bacteria. After washing three times with 1× PBS, the cells were treated with 1ml of 1% TritonX-100 (Sangon Biotech, catalogue number A110694) for 10min to release the intracellular bacteria. Then intracellular bacteria were counted by serial dilution agar plating method.

In vitro bacterial competition assay

Each WT strain (1×10⁵ml⁻¹) was mixed 1:1 with the isogenic Δcps mutant in LB medium and incubated at 37°C for 24h. Bacteria were then serially diluted and plated on LB and LB containing hygromycin (Rhawn, catalogue number R053999) plates. The ratio of cps (the number of colonies on hygromycin LB plates) to WT (the number of colonies on LB plates minus the number of colonies on hygromycin LB plates) was calculated to obtain the competitive index value.

In vivo bacterial competition assay

Groups of five 7-week-old female C57BL/6J mice were infected intraperitoneally with an assigned dose of WT or Δcps mutant to evaluate the systemic infectivity of the bacteria. Survival curves were monitored for 10days. Differences in mouse survival were determined by Kaplan–Meier survival analysis with log-rank test. The in vivo bacterial competition assay was performed in 7-week-old female C57BL/6J mice (n=5). Each WT strain was mixed with the isogenic Δcps mutant at a ratio of 1:1 in 200µl 1× PBS and injected intraperitoneally into mice. After 24h post infection, mice were euthanized, and the burdens of WT and Δcps mutant in lung, liver and spleen were counted by serial dilution on LB agar plates and LB agar plates containing hygromycin (Rhawn, catalogue number R053999), respectively. The ratio of mutant:WT was calculated to obtain the competitive index value.

Biofilm formation assay

The ability of K. pneumoniae strains to form biofilms was determined as described previously^21,55. Briefly, bacteria were incubated in tryptic soy broth medium (Hopebio, catalogue number HB4150) to optical density (OD)=0.4. The bacterial suspension was incubated in a 96-well microtitre plate at 37°C for 48h, then stained with 0.4% crystal violet (Macklin, catalogue number C805211) at 37°C for 45min and destained with 33% acetic acid (Sinopharm Chemical Reagent, catalogue number 10000208) for 30min. Finally, the sample was read at a wavelength of 595nm.

Serum resistance assay

The serum resistance ability of the strains was estimated using the serum resistance assay. About 5×10⁴ CFU bacteria were incubated with 75% baby rabbit complement at 37°C for 1, 2 and 3h. Bacterial counts were determined to enumerate the survival rate. The survival rate was calculated using the following formula: survival rate 1 (SR1)=(CFU surviving at 1h post-complement treatment/CFU surviving in initial)×100; SR2=(CFU surviving at 2h post-complement treatment/CFU surviving in initial)×100; SR3=(CFU surviving at 3h post-complement treatment/CFU surviving in initial)×100. Each isolate was classified as previously described⁵⁶. Briefly, serum resistance level 1 is SR1 and SR2<10%, and SR3<0.1%; level 2 is SR1 and SR2 between 10% and 100%, and SR3<10%; level 3 is SR1>100%, SR2 and SR3<100%; level 4 is SR1 and SR2>100%, and SR3<100%; level 5 is SR1, SR2 and SR3>100%, and SR2>SR3; level 6 is SR1, SR2 and SR3>100%, and SR1<SR2<SR3. Isolates belonging to levels 1 and 2, levels 3 and 4, and levels 5 and 6 were defined as ‘sensitive’, ‘intermediate’ and ‘resistant’, respectively.

To compare WT/Δcps strains with depolymerase-treated strains, bacteria were incubated with their corresponding K-type depolymerase (KL47 strains with Dep 42⁵⁷ and KL64 strains with Dep 41¹⁵, at a final concentration of 100μgml⁻¹), or K-type-irrelevant depolymerase (Dpo7¹⁷, which were specific for KL19 CPS) for 1h before incubation with 75% baby rabbit serum. The percentage of survival was determined as the number of surviving bacteria relative to the initial bacterial addition.

Furthermore, fosmidomycin and ispU overexpression were performed to evaluate the effect of Und-P on serum resistance. To inhibit the synthesis of Und-P, 16-312Δcps mutants were cultured overnight in LB medium with or without 500ngml⁻¹ fosmidomycin (Rhawn, catalogue number R047510) and subcultured to OD=0.7. To increase the synthesis of Und-P, a recombinant pBAD33-ispU plasmid with an arabinose-induced promoter was constructed and transformed into Δcps mutants. Bacteria were grown overnight in LB medium containing 10mM arabinose and 50µgml⁻¹ apramycin and subcultured to OD=0.7. Finally, bacteria were incubated with 75% baby rabbit complement at 37°C for 1h. Bacterial viability was determined by the serial dilution agar plating method.

LPS isolation and electrophoresis

A total of 10¹⁰ CFU bacteria were pelleted for LPS extraction. Capsular strains of the depolymerase treatment group were treated with the corresponding K-type depolymerase or K-type-irrelevant depolymerase at 37°C for 2h. Then, 100µgml⁻¹ deoxyribonuclease (Roche, catalogue number 4716728001), 50µgml⁻¹ ribonuclease (TIANGEN Biotech, catalogue number RT405) and 200µgml⁻¹ proteinase K (LABLEAD Biotech, catalogue number K0510) were used to eliminate nucleic acids and protein. LPS was extracted by the hot phenol/water method⁵⁸. Finally, samples were run on 15% sodium dodecyl sulfate–polyacrylamide gels and visualized by silver staining.

RNA sequencing

RNA sequencing was performed as previously described with slight modification^59,60. K. pneumoniae strains 16–312 and 16–312Δcps mutant were grown in LB to an OD₆₀₀ of 0.7. Bacteria were pelleted by centrifugation, snap frozen in liquid nitrogen and cryopreserved at −80°C. Total RNA from bacteria was extracted and purified using the RNAprep Pure Cell / Bacteria Kit (TIANGEN Biotech, catalogue number DP430). RNA integrity was assessed using the Agilent 2100 Bioanalyzer. RNA sequencing was performed at the Shanghai Personal Biotechnology (Shanghai, China). P<0.05 was defined as significant difference. The LPS and peptidoglycan synthetic genes and their putative gene products are listed in Supplementary Table 14.

Quantitative RT-PCR

Purified RNA was reverse transcribed to complementary DNA using the Thermo Scientific RevertAid RT Kit (Thermo Scientific, catalogue number M1631). Transcript levels of wbbM, LpxD, LpxM, LpxA, mraY and murA were detected by quantitative RT-PCR. The relative abundance of LPS- and peptidoglycan-synthesizing genes was calculated by the 2^−ΔΔCT method using the 16S ribosomal RNA gene as a reference. Primers for quantitative RT-PCR are listed in Supplementary Table 12.

Serum bactericidal activity

The SBA assay was determined using a method previously reported with slight modification^61,62. WT strain and isogenic Δcps mutant were incubated with 10% heat-inactivated serum from patients (infected with KL47 or KL64 K. pneumoniae strains) in the presence of 15% baby rabbit complement or 15% baby rabbit complement alone for 1h. The percentage of survival was calculated by dividing the viable counts of patient serum plus complement by complement alone. The serum of infected patients was obtained from Ruijin Hospital Affiliated to Shanghai Jiaotong University School of Medicine; 62.5% (5/8) of patients were male. The mean age of the patients was 51.13 (±13.78) years.

Statistical analyses

Unpaired two-sided Student’s t-test was used to compare the means between two groups. The Welch’s correction was used when variation was unequal. Pearson’s chi-square was used to compare groups for proportion data. The Cochran–Mantel–Haenszel test was used to compare the mean rank difference of two groups in serum resistance assay. The Cochran–Armitage test for trend was used to analyse the temporal trend in the proportions of the K types, hypervirulent strains and NEKp strains from 2016 to 2020. Differences in mouse survival were determined by Kaplan–Meier survival analysis with log-rank test. Two-sided one-sample t-test was used to compare the mean of log₁₀-transformed competitive index or competitive index to a hypothetical value of 0 or 1 in competition assay. No comparison between groups was involved in the competition assay. P<0.05 was considered statistically significant.

We gratefully acknowledge the contribution of the members of the CHINET group for collection of the isolates in this study. We thank the Electron Microscopy Imaging Laboratory in Shanghai Jiaotong University School of Medicine for helping with transmission electron microscopy imaging. We thank Y. Song at Shanghai Jiaotong University for helping with statistical analyses. We thank Y. Wang at China Agricultural University for sharing the plasmid pACYC184. We thank H. Ou at Shanghai Jiaotong University for sharing the plasmid pBAD33. This work was funded by the National Natural Science Foundation of China (grant number 81971896 (P.H.), 82172311 (F.H.), 32141002 (F.H.), 32170003 (Z.Z.) and 32370099 (Z.Z.)), the National Key Research and Development Project of China (2022YFA1304300) (P.H.), the China Antimicrobial Surveillance Network (Independent Medical Grants from Pfizer, 2018QD100) (F.H.) and Shanghai Antimicrobial Surveillance Network (3030231003) (F.H.).