The strains, plasmids, and oligonucleotide primers used in this study are listed in Supplementary Table 1 . E. coli and K. pneumoniae strains were cultured in LB medium at 37°C on a shaker at 200 rpm. The following antibiotics were added to agar plates or liquid medium as needed: kanamycin (Kan, 50 μg/mL), chloramphenicol (Cm, 34 μg/mL), and ampicillin (Amp, 100 μg/mL).

To enable genome-wide functional screening, we constructed a high-density Mariner transposon library in Kp ATCC 43816. This was achieved through biparental conjugation using the suicide vector pKana_Mariner, following established protocols (Paczosa et al., 2020). Briefly, the diaminopimelic acid (DAP)-auxotrophic E. coli donor strain WM3064 (OD₆₀₀= 0.5) and the Kp recipient (OD₆₀₀ = 1.0) were harvested, combined at a donor-to-recipient ratio of 15:1 (300 μL:20 μL), and concentrated onto a 0.45-μm filter membrane. The membrane was placed on LB plates supplemented with DAP (57 μg/mL) and incubated at 37°C for 18 h to allow conjugation. Transconjugants were then selected on kanamycin-containing plates, resulting in a library of approximately 60,000 unique mutants. The library was preserved in 50% glycerol at −80°C.

The G. mellonella larvae were obtained from Ruiqingbait (Chongqing, China). Prior to infection, the larvae were surface disinfected with 75% ethanol (v/v). Bacterial suspensions for infection were prepared from overnight cultures of Kp, which were subcultured at a ratio of 1:100 in fresh LB broth. The cultures were grown to OD₆₀₀ = 1.0, harvested by centrifugation, and washed twice with sterile saline. Each larva was then injected with 10 μL of the bacterial suspension, delivering approximately 10⁵ CFU. To distinguish mortality due to infection from trauma induced by the injection procedure, a control group injected with 10 μL of sterile saline was included in all experiments. Following infection, the larvae were incubated in the dark at 37°C and monitored daily for survival over a seven-day period (Insua et al, 2013; Gebhardt et al., 2015; Bruchmann et al., 2021). Larvae were considered dead if they exhibited no response to physical stimuli or characteristic black pigmentation, and the time of death was recorded.

The Kp transposon-mutagenized library was cultured in LB medium containing Kan until OD₆₀₀ = 0.95. Then, a bacterial suspension containing 1 × 10⁶ CFU was injected into individual G. mellonella larvae (250–350 mg). Four hours after infection, the larvae were homogenized, and the larval homogenate was uniformly plated onto LB agar supplemented with Kan and Amp. Amp selectively inhibited the native microbiota of G. mellonella. The plates were then incubated at 37°C for four hours to allow for bacterial recovery, after which all bacterial cells were collected (in vivo output pool).

Concurrently, an in vitro control pool was generated by inoculating with the same infection dose in LB medium and incubated at 37°C for four and 24 hours. Bacterial cells from the in vitro cultures were harvested by centrifugation. The bacterial cells were then collected via centrifugation (in vitro group), and genomic DNA was extracted using a genomic DNA extraction kit (TIANGEN Biotech, China; cat# DP302). Library preparation and analysis were conducted using the previously described barcoded Tn-seq method (Van Opijnen and Camilli, 2012) ( Supplementary Figure 1 ). The library preparation and sequencing were performed on the Illumina HiSeq PE150 platform (BGI Genomics, Beijing, China).

Tn-seq data analysis was performed as previously described (Van Opijnen and Camilli, 2012). After Illumina sequencing, the raw reads were sorted into groups based on their barcodes using Fastp (Chen et al., 2018). Sixteen-nucleotide fragments from each read, corresponding to the ATCC 43816 genomic sequence flanking the transposon, were mapped to the ATCC 43816 genome (Accession Number: CP194565) using Bowtie 2 (Van Opijnen and Camilli, 2012). The gene IDs mentioned in the text are original sequencing IDs, which may be reprocessed when uploaded to the database. The correspondence between IDs can be found in the Supplementary Excel File 1 . To identify significant fitness genes in G. mellonella and LB medium, fitness values were calculated as log₂ [(post-infection reads + 1)/(pre-infection reads + 1)], and a Z-test was applied to all gene mutants whose log-transformed output-to-input ratios were significantly different from the overall distribution across all biological replicates. To minimize false-negative results at insertion sites, genes with fewer than two unique insertion sites or fewer than three counts per million reads were excluded from subsequent analyses. Consequently, only genes with a q-value less than 0.01 from the Z-test were considered significantly altered from the input (Gao et al., 2017a). For the complete Tn-Seq dataset, see the Supplementary Excel File 2 .

Gene knockout mutants were generated in ATCC 43816 using allelic replacement via the pDS132 suicide vector system. Flanking regions (~1 kb) of target open reading frames and a kanamycin resistance cassette were PCR-amplified with primers ( Supplementary Table 1 ) and cloned into pDS132. Recombinant plasmids were introduced into ATCC 43816 via biparental conjugation with E. col S17-1 λpir. Double-crossover recombinants were selected on LB agar supplemented with Kan and 10% (w/v) sucrose, utilizing sacB-mediated counter-selection to eliminate the chloramphenicol-resistant plasmid backbone. Transformants grown on Kan-containing LB plates but not on Cm-containing LB plates were isolated, and allelic replacements were confirmed by colony PCR. The genomic fragments containing the coding sequence were cloned into the low-copy vector pACYC184 via the BamHI/SphI restriction sites. The recombinant plasmids were then electroporated into the respective mutant strains. Transformants were selected using Cm selection and verified via PCR and sequencing ( Supplementary Table 1 ).

In vitro growth kinetics: Overnight Kp cultures were diluted 1:100 into 7 mL fresh LB medium and incubated at 37°C. Bacterial growth was monitored hourly by quantifying viable counts (CFU/mL) via plating serial dilutions onto LB agar. In vivo bacterial load: To quantify bacterial proliferation within the host, infected larvae were homogenized individually at two time points: at the time of infection (t = 0 hours) and four hours post-infection (t = 4 hours). The homogenates were suspended in 2 mL of sterile saline. The homogenates were serially diluted and plated onto LB agar supplemented with 5 μg/mL chloramphenicol, which selectively recovers Kpn while suppressing the growth of the G. mellonella microbiota. Bacterial growth in vivo was expressed as the ratio of CFUs per larva at t = 4 hours relative to CFUs per larva at t = 0 hours.

Overnight bacterial cultures were diluted 1:100 into fresh LB broth and incubated to OD₆₀₀ = 1.0. The mutant and WT cells were then mixed at a 1:1 CFU ratio. Serial dilutions were then performed to determine the initial inoculum densities (CFU/mL) for each strain. The mixed culture was diluted 1:100 into 5 mL of fresh LB medium, then co-cultured competitively at 37°C with shaking at 200 rpm for four hours. Post-incubation, the cultures were diluted with saline and then plated separately on antibiotic-free LB plates to count the total CFUs and on Kan-containing LB plates to determine the mutant CFUs. The WT CFUs were calculated by subtracting the mutant CFUs from the total CFUs. Additionally, in the in vivo competition assay, 10 µL of the saline-washed mixture (input) was infected with the G. mellonella, incubated at 37°C for 4 hours, and then diluted after homogenization before inoculating onto 5 μg/mL Cm-containing LB plates for screen Kp (total output CFUs), and onto Kan- and Cm-containing plates to assess the mutant CFUs (mutant output CFUs). Competition index (CI) = (mutant_output CFU/WT_output CFU)/(mutant_input CFU/WT_input CFU) (Gao et al., 2017a).

Hypermucoviscosity sedimentation assays were prepared as previously described (Mike et al., 2021). Briefly, the OD₆₀₀ value of the overnight bacterial cultures was measured. Then, 1 mL of the cultures was centrifuged at 1,000 × g for 5 minutes to pellet cells, and 500 μL of supernatant was collected for OD₆₀₀ measurement. The hypermucoviscosity sedimentation assay is evaluated by calculating the ratio of the OD₆₀₀ value of the supernatant to that of the overnight bacterial cultures.

Animal experiment protocols were approved by the Institutional Animal Ethics Committee of Xiamen University (Approval Number: XMULAC20190029). Bacteria incubated overnight were diluted 1:100 and reincubated to OD₆₀₀ = 1.0. The bacteria were then centrifuged at 8,000× g for three minutes, and the pellet was resuspended in sterile PBS. 20 g female ICR mice (Charles River Laboratories) were anesthetized via intraperitoneal injection of sodium pentobarbital and subsequently infected intranasally with 1 × 10⁵ CFU of Kp bacteria. 24 hours later, the mice were euthanized by spinal dislocation. Each gram of tissue was homogenized with 9 mL of sterile PBS, spread on plates for counting, and the CFUs were determined. In the competitive experiment, the WT and mutant strains were combined at a 1:1 CFU ratio and infected for 24 hours. The homogenized lung tissue was serially diluted in saline and inoculated onto antibiotic-free LB plates and Kan-containing LB plates. The plates were incubated at 37°C overnight, and the CFUs were counted. The competition index (CI) was calculated using the previously mentioned method, and the detection limit was set at 100 CFU/mL. Samples with no detectable CFUs were assumed to contain 99 CFUs/mL for analysis. Mortality was monitored at 24-hour intervals for 7 days following Kp infection to assess murine survival rates.

We constructed a Mariner-based transposon insertion library in K. pneumoniae ATCC 43816 using a method previously applied to create Campylobacter jejuni mutant libraries (Gao et al., 2017a). We assessed the quality of the mutation library through Tn-seq analysis ( Supplementary Figure 1 ). A total of 49,786 unique insertion sites were identified across the entire ATCC43816 genome ( Supplementary Table 2 ). Next, we conducted a comprehensive analysis of the Tn-seq data to identify genetic determinants associated with infection by G. mellonella.

The G. mellonella model is a dose- and time-dependent animal model ( Supplementary Figure 2 ). G. mellonella were infected with a dose of 10⁶ CFU of Kp per larva at 37°C, as this dose resulted in 50% survival of the larvae after four hours of infection ( Supplementary Figure 2 ). Four hours post-infection, the larvae were recovered, homogenized, plated, and cultured for 24 hours to prepare Tn-seq DNA samples. We performed high-throughput sequencing to assess genome-wide adaptive changes in Kp genome four hours post-infection. As shown in Figure 1A , each point in the scatter plot represents the calculated fitness value for a specific gene insertion mutant. This fitness value is defined as log₂ [(post-infection reads + 1)/(pre-infection reads + 1)] and quantifies the relative change in mutant abundance after infection compared to the pre-infection baseline. The key observation is the significant dispersion of points away from the central baseline (fitness = 0, indicating no change, Figure 1A ). We defined reads change exceeding a 4-fold increase or decrease (corresponding to fitness value = ± 1.3) as indicative of a biologically significant adaptive change. The high density of points falling outside the |fitness| = 1.3 threshold (both above and below zero) across numerous gene IDs provides direct visual evidence of widespread adaptive changes in the Kp genome four hours post-infection. Furthermore, the reproducibility of the Tn-seq results across biological replicates was confirmed (Pearson correlation r = 0.7716, Figure 1B ).

In vitro, equal amounts of mutant libraries were inoculated in LB medium and collected after 4 and 24 hours of incubation. We found that bacteria undergoing adaptive changes in G. mellonella exhibit similar correlations in LB medium (r = 0.7618, Figure 1C ), indicating that genes essential for proliferation in vitro often overlap with adaptive requirements in vivo (Ibberson et al., 2017). The reliability of the model in identifying adaptive and virulence factors is supported by the reduction in the reads of genes in the CPS biosynthesis operon post-infection, accompanied by a decrease in fitness values (Z-test, p < 0.01) ( Figure 1D , Supplementary Table 3 ). The connection between bacterial adaptation and the retention of virulence factors further demonstrates the strength of this model in clarifying the trade-off between pathogenicity and adaptability. Tn-seq analysis revealed genes with low insertion density, such as wzb, wza, and wzy, which were attributed to transposase insertion bias ( Figure 1D ). To avoid false negative results at insertion sites, genes with fewer than two unique insertion sites were excluded from subsequent analyses.

To systematically characterize the biological functions underlying bacterial adaptation in vivo, we performed a Clusters of Orthologous Groups (COG) enrichment analysis on adaptive-related genes identified in G. mellonella (Padj < 0.01). This approach enabled us to map these genes to conserve functional categories and pinpoint critical pathways for host adaptation. COG enrichment analysis revealed that adaptive-defect genes significantly enriched in specific COG functional terms compared with the input database ( Figure 2A , Supplementary Table 3 ). Notably, these genes were enriched in the functional categories of “cell wall/membrane/envelope biosynthesis,” “replication, recombination, and repair,” and “carbon source transport and metabolism.” These results demonstrate that structural integrity, via cell envelope biogenesis, and metabolic flexibility are key adaptive strategies employed by Kp during G. mellonella infection.

To distinguish genes specifically required for infection from those essential for general bacterial growth, we conducted a Venn analysis comparing fitness-defective genes (p < 0.01) in vivo (G. mellonella) versus in vitro (LB medium). This comparative design aimed to identify infection-specific vulnerabilities that could serve as targets for therapeutic intervention. The Venn analysis identified 76 infection-specific genes ( Figure 2B ). Of these, 86.2% (50 out of 58) have well-established roles in the pathogenicity of bacteria (including Kp) (excluding 18 uncharacterized ORFs; Supplementary Table 3 , gray background). This high proportion of known virulence-associated genes within this subset validates our model’s ability to uncover genuine host adaptation mechanisms and highlights potential targets for disrupting infection-specific pathways.

To identify novel virulence factors for bacterial survival during infection, we validated seven candidate genes associated with adaptive costs ( Table 1 ), based on transposon libraries obtained. Growth curves indicated that deleting these seven genes alone did not affect the growth rate of Kp in LB medium ( Supplementary Figure 3 ). However, when WT and mutant strains were inoculated at a 1:1 CFU ratio into LB medium for a competitive growth experiment in vitro, the ΔptsI and ΔGM2628 mutant strains displayed severe adaptive (mean CI = 0.09) and moderate adaptive defects (mean CI = 0.275) ( Supplementary Figure 4A ), which were reversed by ptsI and GM2628 genetic complementation ( Supplementary Figure 4B ). These results indicate that PtsI and GM2628 are critical for the rapid proliferation of Kp under G. mellonella host conditions. Competitive fitness assays in vivo revealed that five mutant strains (ΔptsI, ΔserC, ΔGM2628, ΔgcvA, and ΔfabR) exhibited significant adaptive defects compared to the WT in G. mellonella ( Figure 3A ). In contrast, deletion of ulaR or mocB had no effect on competitive fitness in vivo ( Figure 3A ). Consistently, G. mellonella infection experiments demonstrated that the same five mutants displayed attenuated virulence, as evidenced by increased larval survival ( Figures 3B-F , Supplementary Figure 5 ). Similar to the competitive fitness results, ulaR and mocB deletions did not significantly impact virulence ( Figures 3A, G, H ). By linking these genes to their functions, this study uncovered that carbohydrate metabolism (ptsI), amino acid metabolism (serC, and gcvA), and fatty acid biosynthesis (fabR) are determinants of survival and adaptability in host-pathogen interactions. These findings underscore the importance of metabolic flexibility as a vital survival strategy during host adaptation.

To investigate the association between hypermucoviscosity (HMV) and the virulence of hvKp, we employed centrifugation-based assays to quantify HMV. In these assays, viscous cells remain suspended, while non-viscous strains form compact pellets. Our findings revealed that under centrifugation conditions of 1,000 × g, the HMV of the ΔptsI, ΔGM2628, ΔfabR, ΔulaR, and ΔmocB mutant strains was significantly higher than that of the WT strain ( Figure 3I ). However, in the G. mellonella infection model, the virulence of the ΔptsI, ΔGM2628, and ΔfabR mutant strains was significantly diminished, while the virulence of the ΔulaR and ΔmocB strains remained unchanged ( Figure 3 ). These results suggest that HMV does not directly correlate with enhanced virulence in the G. mellonella model.

Further analysis of the relationship between HMV and adaptive defects showed that, although the ΔserC and ΔgcvA mutants-maintained HMV levels like the WT strain ( Figure 3I ), their adaptability was significantly reduced in G. mellonella ( Figure 3A ). In contrast, the ΔulaR, ΔfabR, and ΔmocB mutants exhibited increased HMV but no enhanced fitness ( Figures 3A, I ). These results suggest that adaptive defects can directly lead to reduced virulence independently of HMV changes. In summary, HMV is not a reliable marker of HvKp virulence in the G. mellonella model, and bacterial adaptation may regulate virulence in this model through HMV-independent pathways.

The G. mellonella infection model has become a popular tool for assessing pathogenicity. However, intrinsic physiological differences exist among various host models. Therefore, we investigated whether the fitness cost incurred by Kp genes during G. mellonella infection also occur in a mammalian host (mice), and whether fitness cost impact virulence. We used the K. pneumoniae mutant strain ΔptsI for cross-model validation. In murine lung infection experiments, mice were inoculated with 1.5 × 10⁴ CFU of WT or ΔptsI strain. The ΔptsI mutant maintained lung bacterial loads comparable to the WT strain 24 hours post-infection, indicating intact pulmonary colonization capacity ( Figure 4A ). However, its splenic bacterial load was significantly reduced (p < 0.01), suggesting impaired systemic dissemination ( Figure 4A ). Competitive experiments conducted 24 hours post-infection with a 1:1 CFU ratio of WT and mutant strains revealed differences in bacterial load ( Figure 4B ), with significant fitness defects observed in the lungs ( Figure 4C ). The competitive index in the lungs was 0.2 for ΔptsI ( Figure 4C ). Consistently, ΔptsI-infected mice 40% survival at seven days versus 100% lethality in WT controls ( Figure 4D ), indicating attenuation of virulence in ΔptsI. These results recapitulate the adaptation and virulence defects observed in G. mellonella, demonstrating conserved infection dynamics between the insect and murine pneumonia models.