Threshold-free genomic cluster detection to track transmission pathways in health-care settings: a genomic epidemiology analysis

Abstract

Background: A crucial barrier to the routine application of whole-genome sequencing (WGS) for infection prevention is the insufficient criteria for determining whether a genomic linkage is consistent with transmission within the facility. We evaluated the use of single-nucleotide variant (SNV) thresholds, as well as a novel threshold-free approach, for inferring transmission linkages in a high-transmission setting.

Methods: We did a retrospective genomic epidemiology analysis of samples previously collected in the context of an intervention study at a long-term acute care hospital in the USA. We performed WGS on 435 isolates of Klebsiella pneumoniae harbouring the bla_KPC carbapenemase (KPC-Kp) collected from 256 patients through admission and surveillance culturing (once every 2 weeks) of almost every patient who was admitted to hospital over a 1-year period.

Findings: Our analysis showed that the standard approach of using an SNV threshold to define transmission would lead to false-positive and false-negative inferences. False-positive inferences were driven by the frequent importation of closely related strains, which were presumably linked via transmission at connected health-care facilities. False-negative inferences stemmed from the diversity of colonising populations that were spread among patients, with multiple examples of hypermutator strain emergence within patients and, as a result, putative transmission links separated by large genetic distances. Motivated by limitations of an SNV threshold, we implemented a novel threshold-free transmission cluster inference approach, in which each of the acquired KPC-Kp isolates were linked back to the imported KPC-Kp isolate with which it shared the most variants. This approach yielded clusters that varied in levels of genetic diversity but where 105 (81%) of 129 unique strain acquisition events were associated with epidemiological links in the hospital. Of 100 patients who acquired KPC-Kp isolates that were included in a cluster, 47 could be linked to a single patient who was positive for KPC-Kp at admission, compared with 31 and 25 using 10 SNV and 20 SNV thresholds, respectively. Holistic examination of clusters highlighted extensive variation in the magnitude of onward transmission stemming from more than 100 importation events and revealed patterns in cluster propagation that could inform improvements to infection prevention strategies.

Interpretation: Our results show how the integration of culture surveillance data into genomic analyses can overcome limitations of cluster detection based on SNV-thresholds and improve the ability to track pathways of pathogen transmission in health-care settings.

Funding: US Center for Disease Control and Prevention and University of Michigan.

Figure 1:. Endemicity of KPC-Kp in the LTACH throughout the 1-year study

(A) Isolates obtained through rectal surveillance culturing of patients in the LTACH once every 2 weeks. Grey boxes indicate the study-start (0 days) and every two 14-day surveillance periods (28 days) throughout the study. Bars indicate the KPC-Kp isolates collected at the beginning of the study (for which importation or acquisition status is not known; light grey, study-start), within 3 days of the patient first entering the facility (blue, importation), or after negative surveillance or more than 3 days after ever being in the LTACH during the study (dark grey, acquisition). (B) KPC-Kp prevalence (blue line) is defined as the number of patients presently in the LTACH who are or ever had been surveillance positive for at least one KPC-Kp isolate during the study, divided by the number of patients in the facility (daily census) throughout the 1-year study. KPC-Kp=Klebsiella pneumoniae harbouring the bla_KPC carbapenemase. LTACH=long-term acute care hospital.

Figure 2:. Detection of threshold-free transmission clusters within the LTACH

(A) Patient bed trace showing surveillance culture data for patients who tested positive for KPC-Kp strain ST258, at any point during the study. A plot illustrating other sequence types detected during the study is shown in the appendix (p 12). (B) Comparison of minimum pairwise SNV distances to the closest imported isolate for acquired and imported isolates. SNV distance versus density (ie, distribution was normalised so total area under curve is the same for acquired and imported distributions) of KPC-Kp isolates from ST258 is shown. Light grey bars indicate the minimum distance between isolates collected from patients who acquired KPC-Kp colonisation after being in the LTACH for more than 3 days during the admission-positive isolates. Dark grey bars indicate the minimum distance between isolates collected from patients who were positive on admission to the LTACH. The darkest grey colour indicates overlap between the two distributions. Patients who were-positive for KPC-Kp on the first day of the study and those who represent a mixture of recent and previous colonisation were considered admission-positive for this analysis so that acquisitions derived from those transmission chains could be linked. The two-sample KS test was used for differences in the distribution of pairwise SNV distances (for ST258, KS statistic=0·16 and p=0·28; for non-ST258, KS statistic=0·21 and p=0·39; for all sequence types combined, KS statistic=0·09 and p=0·81). (C) Distribution of isolates and patients in the 49 transmission clusters detected with the threshold-free approach. Each column represents isolates from one cluster. Admission-positive patients (pink) were patients whose isolate in the cluster was obtained within 3 days of first admission to the facility. Purple indicates isolates obtained from acquisition patients who first acquired KPC-Kp colonisation more than 3 days after first admission to the LTACH. Orange indicates isolates from admission-positive patients that were collected more than 3 days after admission to the LTACH, indicating either prolonged colonisation or secondary strain acquisition in the LTACH. Blue indicates patients who were first positive after being in the LTACH for more than 3 days, but from whom no negative swab was collected before first KPC-Kp detection. Grey indicates patients who were positive on the first day of the study. The bar across top of figure indicates sequence type of isolates. KPC-Kp=Klebsiella pneumoniae harbouring the bla_KPC carbapenemase. KS=Kolmogorov-Smirnov. LTACH=long-term acute care hospital. SNV=single nucleotide variant.

Figure 3:. Genetic diversity in transmission clusters, prolonged colonisation, and emergence of hypermutator strains

(A) Maximum pairwise SNV distance distinguishing isolates from the same cluster. Grey bars indicate the sequence type of the isolates in transmission clusters. (B) Phylogenetic tree indicating observed ST16 clusters with a pairwise genetic distance greater than 30 SNVs. (C) Observed frequencies in mutational classes across isolates included in each transmission cluster among clusters with a maximum pairwise SNV distance of at least 30 SNV. Bars on the right of each sequence-type group indicate the overall population frequency of mutational classes among members of that sequence type in the study population. (D) Maximum intrapatient, intracluster genetic diversity among admission-positive patients and acquisition patients. NA=not in one of the large SNV distance clusters. SNV=single nucleotide variant.

Figure 4:. Descriptive vignettes from transmission clusters detected through the integration of genomic and surveillance data

Patients are indicated on the y-axis and time is on the x-axis. Putative route of transmission within each cluster is indicated in the text above the cluster. Surveillance culturing information is indicated by the circles, and floor location in the LTACH is indicated by the coloured rectangles. (A) Transmission between admission-positive patient 1 to acquisition patient 144. Both patients were on the teal ward while patient 1 was positive and patient 144 was negative. (B) No admission-positive patient precedes several acquisition patients in this cluster; therefore, false-negative surveillance of a patient in the cluster or a patient not captured in the study is the probable source. (C) No spatiotemporal exposures between several patients indicate a missing intermediate source patient undetected by surveillance culturing. (D) Transmission between two patients who did not reside on the same ward indicates potential escape from cohort location, or transmission at a common location or via an unidentified common health-care worker source in the facility. (E) Multiple admission-positive patients and no spatiotemporal exposures in the LTACH indicates potential transmission outside of the facility before admission to the LTACH. All 49 transmission clusters detected are shown in the appendix (p 18). LTACH=long-term acute care hospital.