| Objective:Hypermucoviscous Klebsiella pneumoniae(HmKp)poses an emerging and highly pathogenic global health threat.The clinical diagnosis of HmKp is applicable in clinical settings,but there are few studies.This study aimed to investigate the clinical and genomic characteristics of HmKp isolates to better understand the virulence mechanisms of the hypermucoviscous phenotype.Methods: From May 2018 to August 2021,203 non-repeat K.pneumoniae isolates causing invasive infections were collected from a hospital in Beijing,China.Isolates were divided into HmKp(n=90,44.3%)and non-HmKp(n=113,55.7%)groups according to string test results.The clinical characteristics of the two groups were compared,and the factors associated with the infection of HmKp were examined by Logistic regression.Whole genome sequencing was performed on the strains.STs were determined via MLST,Capsular K serotypes were identified by Kleborate software.The presence of resistance and virulence genes was predicted by ABRicate.raxml-ng software was used for phylogenetic analysis,and roary software was used for pan-genome analysis.Functional annotation of genes was done on RAST.In additionally,gene sequences of K.pneumoniae were obtained through an extensive literature search.Spine was used to generate a core genome.Clust AGE was used to identify accessory sequences and divided into smaller elements-subelement.a model based on subelements were established to predict the hypermucoviscous phenotype.90% of samples are used to learn predictive models for the phenotype,and predictions are made for the 10% of samples held out during training.We experimented with support vector machines(SVMs),Lasso model and Random Forests as the learning method and the best-performing combination was then used to was used to select the hypermucoviscous phenotype associated genes.Results: There was no significant difference in gender and age between HmKp and non-HMKP(P>0.05).The annual proportions of HmKp isolates were similar over the four-year timespan.58.9%(53/90)HmKp infection were Hospital-acquired infection.The proportion of patients with fever was higher in the HmKp group than in the nonHMKP group(P=0.032).Multivariate regression showed that diabetes mellitus(odds ratio [OR]=2.20,95% confidence interval(CI): 1.20-4.06,P=0.011)and liver abscess(OR=3.00,CI 95%:1.32-7.29,P=0.011)were associated with HmKp infections.The results of drug susceptibility test showed that Rates of most antimicrobial resistance in HmKp group were lower than in non-HmKp group.K.pneumoniae was highly diverse,comprising 87 sequence types(STs)and 54 serotypes.Among HmKp isolates,ST23 was the most frequent ST(25/90,27.8%),and the most prevalent serotypes were KL2(31/90,34.4%)and KL1(27/90,30.0%).Most KL1 strains belonged to ST23(31/35,88.6%),KL2 isolates clustered in different sub-branches with different STs.The prevalence of 74 genes was statistically higher in HmKp;These included genes encoding siderophores,lipopolysaccharide(LPS)synthesis,the type VI secretion system and the gene cluster for capsule production and so on.Thirteen virulence genes were only located on the capsular polysaccharide synthesis region of KL1 strains.The results of comparative genomic analysis showed that 17 genes were used to predict HmKp with an accuracy of more than 85%,including rmp AC,7 iron-acquisition related genes,and pag O,which may promote liver abscess formation.Six hundred fifteen isolates were retrieved through literature search and 391 isolates in previous study were included in this study.Isolates were divided into HmKp(n=373,37.1%)and non-HmKp(n=633,62.9%).The core genome of the 1006 isolates was 5.32 Mb,and the accessory genome was 1077281 bp in length and divided into 210674 subelements.We tested the performance of SVMs,Lasso model and Random Forests as the learning method and found that Random Forests had the best overall performance: Accuracy of 0.76,sensitivity of 0.68,specificity of 0.81,positive predictive value(PPV)of 0.64,and F1 score of 0.66.Four of top ten subelements ranked by importance based on random-forest model were genes related with hemolysin secretion.4 of the top 10 subelements are annotated as iron acquisitionrelated genes fec A,a part of the two-component system Evg A/S Evg A and α-hemolysin operon genes Hly D.These genes were all related to hemolysin secretion.Conclusion: Patients with HmKp infection are more likely to have fever than those with non-HmKp infection,suggesting that attention should be paid to temperature monitoring in clinical diagnosis and treatment.Diabetes mellitus and liver abscess were associated with HmKp infections.We observed that HmKp exhibited antibioticsusceptible phenotypes but carried more SHV-type ESBL genes than non-Therefore,monitoring of antimicrobial resistance is required throughout an entire treatment course.The prevalence of genes encoding siderophores,lipopolysaccharide(LPS)synthesis,the type VI secretion system and the gene cluster for capsule production was statistically higher in HmKp.We find seventeen HmKp genes were highly associated with HmKp.The hypermucoviscous phenotype may be related to the transfer of plasmids.The machine learning predicting hypermucoviscous phenotype based on accessory genomes showed that that there may be a correlation between the hypermucoviscous phenotype and hemolysin secretion. |
