肺炎克雷伯菌复合体成员感染的主要危险因素是先前的肠道定植，而感染通常是由定植菌株引起的。尽管肠道作为传染性肺炎克雷伯菌的储存库非常重要，但人们对肠道微生物群与感染之间的关系知之甚少。为了探讨这种关系，我们进行了一项病例对照研究，比较了肺炎克雷伯菌定植的重症监护患者和血液/肿瘤患者的肠道群落结构。病例是被肺炎克雷伯菌定植菌株感染的患者（N = 83）。对照是肺炎克雷伯菌定植且无症状的患者 (N = 149)。首先，我们对肺炎克雷伯菌定植患者的肠道群落结构进行了表征，且与病例状态无关。接下来，我们确定肠道群落数据对于使用机器学习模型对病例和对照进行分类非常有用，并且病例和对照之间的肠道群落结构有所不同。肺炎克雷伯菌的相对丰度是一种已知的感染风险因素，具有最大的特征重要性，但其他肠道微生物也提供了信息。最后，我们表明，肠道群落结构与细菌基因型数据的整合增强了机器学习模型区分病例和对照的能力。有趣的是，纳入患者临床变量未能提高机器学习模型区分病例和对照的能力。这项研究表明，将肠道菌群数据与肺炎克雷伯菌衍生的生物标志物结合起来，可以提高我们对肺炎克雷伯菌定植患者的感染进行分类的能力。重要性定植通常是具有致病潜力的细菌发病机制的第一步。此步骤提供了一个独特的干预窗口，因为给定的潜在病原体尚未对其宿主造成损害。此外，定植阶段的干预可能有助于减轻随着抗菌素耐药性上升而导致治疗失败的负担。然而，为了了解针对定植的干预措施的治疗潜力，我们必须首先了解定植的生物学原理，以及定植阶段的生物标志物是否可用于对感染风险进行分层。克雷伯菌属细菌包括许多具有不同程度致病潜力的物种。肺炎克雷伯菌复合体的成员具有最高的致病潜力。肠道中被这些细菌定植的患者随后被定植菌株感染的风险更高。然而，我们不知道肠道微生物群的其他成员是否可以用作预测感染风险的生物标志物。在这项研究中，我们表明，发生感染的定植患者与未发生感染的定植患者之间的肠道微生物群存在差异。此外，我们还表明，将肠道微生物群数据与细菌因素相结合可以提高对感染进行分类的能力。令人惊讶的是，患者临床因素对于单独或添加到基于微生物群的模型中对感染进行分类没有用。这表明细菌基因型及其所在的微生物群落可能决定感染的进展。当我们继续探索定植作为预防潜在病原体定植个体感染的干预点时，我们必须开发有效的方法来预测和分层感染风险。

The primary risk factor for infection with members of the Klebsiella pneumoniae species complex is prior gut colonization, and infection is often caused by the colonizing strain. Despite the importance of the gut as a reservoir for infectious K. pneumoniae, little is known about the association between the gut microbiome and infection. To explore this relationship, we undertook a case-control study comparing the gut community structure of K. pneumoniae-colonized intensive care and hematology/oncology patients. Cases were K. pneumoniae-colonized patients infected by their colonizing strain (N = 83). Controls were K. pneumoniae-colonized patients who remained asymptomatic (N = 149). First, we characterized the gut community structure of K. pneumoniae-colonized patients agnostic to case status. Next, we determined that gut community data is useful for classifying cases and controls using machine learning models and that the gut community structure differed between cases and controls. K. pneumoniae relative abundance, a known risk factor for infection, had the greatest feature importance, but other gut microbes were also informative. Finally, we show that integration of gut community structure with bacterial genotype data enhanced the ability of machine learning models to discriminate cases and controls. Interestingly, inclusion of patient clinical variables failed to improve the ability of machine learning models to discriminate cases and controls. This study demonstrates that including gut community data with K. pneumoniae-derived biomarkers improves our ability to classify infection in K. pneumoniae-colonized patients.IMPORTANCEColonization is generally the first step in pathogenesis for bacteria with pathogenic potential. This step provides a unique window for intervention since a given potential pathogen has yet to cause damage to its host. Moreover, intervention during the colonization stage may help alleviate the burden of therapy failure as antimicrobial resistance rises. Yet, to understand the therapeutic potential of interventions that target colonization, we must first understand the biology of colonization and if biomarkers at the colonization stage can be used to stratify infection risk. The bacterial genus Klebsiella includes many species with varying degrees of pathogenic potential. Members of the K. pneumoniae species complex have the highest pathogenic potential. Patients colonized in their gut by these bacteria are at higher risk of subsequent infection with their colonizing strain. However, we do not understand if other members of the gut microbiota can be used as a biomarker to predict infection risk. In this study, we show that the gut microbiota differs between colonized patients who develop an infection versus those who do not. Additionally, we show that integrating gut microbiota data with bacterial factors improves the ability to classify infections. Surprisingly, patient clinical factors were not useful for classifying infections alone or when added to microbiota-based models. This indicates that the bacterial genotype and the microbial community in which it exists may determine the progression to infection. As we continue to explore colonization as an intervention point to prevent infections in individuals colonized by potential pathogens, we must develop effective means for predicting and stratifying infection risk.