Empirical genomic methods for tracking plasmid spread among healthcare-associated bacteria

Collection of clinical bacterial isolates and corresponding patient data

The study took place at the University of Pittsburgh Medical Center (UPMC) Presbyterian Hospital and the University of Pittsburgh School of Medicine. All bacterial isolates analyzed in this study were collected through the Enhanced Detection System for Hospital-Associated Transmission (EDS-HAT) project, using previously published eligibility criteria for selection of isolates.^8,9,11 EDS-HAT uses a combination of WGS surveillance of selected hospital-associated pathogens and machine learning of the electronic health record to identify otherwise undetected outbreaks and the responsible transmission routes, respectively. Isolates were identified using TheraDoc software (Version 4.6, Premier, Inc, Charlotte NC). The University of Pittsburgh Institutional Review Board (IRB) approved the EDS-HAT project and classified it as being exempt from informed consent because direct contact with human subjects is not performed. Approval was also granted (STUDY20060252) to use data collected for the EDS-HAT project to study the horizontal transfer of mobile genetic elements in the hospital.

Short- and long-read whole-genome sequencing, processing, and assembly

DNA was extracted from overnight cultures of single bacterial colonies using a Qiagen DNeasy Blood and Tissue Kit (Qiagen, Germantown MD). Short-read WGS was performed using the Illumina platform (Illumina, San Diego CA), with libraries constructed with a Nextera DNA Sample Prep Kit with 150bp paired-end reads and sequenced on the NextSeq platform.

Trim Galore v0.6.1 was used to remove sequencing adaptors, low-quality bases, and poor-quality reads.¹² Bacterial species were identified from processed Illumina reads by alignment to Kraken v1.0 and RefSeq databases.^13,14 Genomes of strains sequenced only with Illumina technology were assembled using SPAdes v3.11.¹⁵ The quality of assembled genomes was then verified using QUAST.¹⁶ Assembled genomes were excluded if they failed to meet the following quality control parameters: genome-wide read depth of at least 40X, cumulative length of assembled genome within 20% of the expected length for the assigned genus, fewer than 400 contigs in the assembled genome, and an N50 value of greater than 50,000bp. Illumina sequencing data for all isolates is deposited at NCBI with accession numbers listed in Appendix 1.

Long-read sequencing and base-calling of select isolates were performed using Oxford Nanopore technology (Oxford Nanopore Technologies, Oxford, United Kingdom). Libraries were constructed using a rapid multiplex barcoding kit (catalog number SQK-RBK004).

Sequencing was performed using an Oxford Nanopore MinION device with R9.4.1 flow cells. Base-calling and read processing was performed using Albacore v2.3.3 or Guppy v2.3.1 (Oxford Nanopore Technologies, Oxford, United Kingdom) using default parameters. Hybrid assembly was performed for genomes for which both short- and long-read sequencing data were available and whose short-read only assemblies passed the aforementioned parameters, using Unicycler v0.4.7 or v0.4.8-beta.¹⁷

Genome annotation, plasmid assembly, characterization, alignment, and phylogenetic analyses

Assembled genomes and hybrid-assembled plasmids were annotated with Prokka v1.13 or v1.14.¹⁸ Multi-locus sequence types (STs) were assigned using mlst v2.16.1 with PubMLST typing schemes.^19,20 Antibiotic resistance genes were identified by BLASTn alignments of assembled genomes to the ResFinder v4.1 database.^21,22 Plasmid replicons were identified by BLASTn alignments of assembled genomes to the PlasmidFinder v2.1 database.²³ Other genomic features were identified from annotations by Prokka¹⁸ and by BLASTn alignments to the VFDB database.²⁴ Species were defined by grouping isolates with average nucleotide identity (ANI) of at least 95% to one another and less than 95% ANI to genomes of other species.²⁵ Whole-genome phylogenies were constructed from core genome alignments generated by Roary v5.18.2²⁶ using RAxML v8.0.26 with 1,000 bootstrap iterations.²⁷ Plasmid phylogenies were constructed from core genome alignments of reference plasmids made with snippy-core v4.4.5²⁸ using RAxML v8.0.26 with 1,000 bootstrap iterations.²⁷ Annotated plasmids were aligned to one another using EasyFig v2.2.2.²⁸ Plasmids were named according to the genus- or species-based identification code of the isolate from which they were first identified and the contig in the hybrid assembly corresponding to the closed plasmid sequence, separated by an underscore.

Sequence similarity analyses to establish metrics of plasmid transmission

Plasmids were identified from hybrid-assembled genomes of clonal isolates implicated in outbreaks within the same hospital.^10,11 Contigs were classified as plasmids if they were circularized during hybrid assembly,¹⁷ were 2-300kb in length, and possessed at least one replicon or conjugative plasmid gene identified by PlasmidFinder, Prokka, or RAST.^18,23,30 The proportion of plasmid gene content preserved in bacterial genomes (“sequence coverage”) was calculated by searching for plasmid sequences among the contigs of each assembled bacterial genome using BLASTn,³⁰ using a nucleotide sequence identity threshold of 95%. Total plasmid nucleotide sequence identity was quantified based on single nucleotide polymorphisms (SNPs) identified by snippy-core v4.4.5.²⁸

Resolving plasmids circulating among nosocomial bacterial pathogens in a single hospital

Eighty-nine closed reference plasmid sequences that met aforementioned sequence coverage criteria were selected from whole-genome bacterial sequences constructed using previously published methods (Appendix 2).^5,9,11 Plasmids were de-duplicated by aligning and visualizing closed reference plasmid sequences of similar lengths and distribution among STs and species using Mauve or EasyFig v2.2.2.^29,31 3,074 whole-genome sequences from clinical bacterial isolates were then screened for the presence of 89 plasmids by calculating sequence coverage by BLASTn-based contig mapping³² and sequence identity using snippy-core,²⁸ as described above (Supplementary Figure 1). Plasmids were considered potentially involved in vertical transmission and/or horizontal transfer if they were present at sufficient nucleotide similarity and sequence coverage in at least two bacterial genomes of the same or different bacterial species.

• Download figure
• Open in new tab

Supplementary Figure 1: Conceptual flow diagram of alignment-based pipeline to infer horizontal plasmid transfer

Different bacterial icons denote different species, and different colors denote different sequence types (STs). Figure constructed with icons made by Freepik from Flaticon.com.

Identification of geotemporal associations where horizontal plasmid transfer may have occurred

Geotemporal associations between plasmid-carrying isolates were identified by systematic review of patient electronic health record (EHR) data, using a previously published machine learning algorithm based on case-control methodology followed by manual review of the algorithm’s findings.³³ Geotemporal links highlighted by the algorithm were manually evaluated if (1) they included bacterial isolates of different STs, species, or genera; (2) at least one patient was infected during, after, or shortly before the date of culture of their plasmid-carrying isolate while in the location; and (3) at least one other patient later infected by another isolate with the same plasmid was exposed to the same location within 90 days of the date(s) of culture of their isolate(s). Admission records for roommates with overlapping stays were identified as shared exposures regardless of the length of cohabitation. Manual investigations of patient clusters of interest were performed by a board-certified infection preventionist (AJS), with the investigator blinded to genomic data regarding associated pathogens or plasmids.

Empirical thresholds of plasmid similarity to indicate potential horizontal transfer

While the epidemiology of plasmid transfer among hospital-associated bacteria has previously been described,^4,5,34,35 the field lacks consistent guidelines for determining when horizontal plasmid transfer among infection-derived bacterial isolates can be inferred. We hypothesized that a nucleotide sequence similarity-based threshold could be developed by first quantifying the sequence similarity of plasmids that were carried by bacterial strains known to have been transmitted among hospital inpatients. We first examined four plasmids carried by three clusters of carbapenemase-producing Klebsiella pneumoniae and vancomycin-resistant Enterococcus faecium that showed strong genomic and epidemiologic evidence of nosocomial transmission within our hospital.^10,11 We also examined plasmids present in pairs or groups of clinical isolates of identical species and sequence types that were isolated from the same patients at different dates during their hospital stays. These plasmids were presumed to be vertically transmitted during bacterial cell division and long-term carriage within each patient. In total, we performed 57 comparisons between 25 plasmids in 47 bacterial genomes; 15 of these comparisons were between same-patient isolates. Elapsed time between dates of culture of plasmid-linked strains ranged from zero days to 427 days (mean 119.5 days, median 82 days).

We next used a previously published BLASTn-based method to calculate the proportion of each plasmid’s sequence that was preserved in linked isolates.⁵ We also quantified the number of single nucleotide polymorphisms (SNPs) among the shared plasmid sequences present in each group of epidemiologically linked isolates. We refer to these metrics as “plasmid sequence coverage” and “plasmid sequence identity”, respectively (Figure 1a). Ninety-five percent of these comparisons had sequence coverage values of at least 93·7% (Figure 1b) and nucleotide sequence identity of at least 99·985% (i.e. fewer than 15 SNPs per 100kb of plasmid sequence, or1 SNP per 6·67kb) (Figure 1c). These results provided initial estimates of sequence similarity thresholds that were then applied for downstream analyses to detect horizontal transfer within the hospital environment.

• Download figure
• Open in new tab

Figure 1: Empirically derived metrics of plasmid sequence similarity to infer horizontal plasmid transfer

(A) Approaches used to calculate (top) plasmid sequence coverage and (bottom) plasmid sequence identity. (B) Sequence coverage of plasmids carried by bacteria engaged in transmission between patients or prolonged carriage in the same patient. Dotted line shows the 5th percentile of coverage among analyzed plasmids. (C) Sequence identity of plasmids carried by isolates engaged in transmission between patients or prolonged carriage in the same patient. Dotted line shows the 95th percentile of SNPs per 100kb of plasmid reference sequence.

Systematic identification of plasmid horizontal transfer from a single hospital

To systematically identify potential horizontal transfer of plasmids within our hospital, we applied our empirical thresholds described above to search for the presence of 89 reference plasmids in the genomes of 3,074 clinical bacterial isolates collected from 2,086 patients (Supplementary Figure 1) (Appendix 2). Twelve of these plasmids were previously described in a prior study of mobile genetic element diversity in our hospital.⁵ We identified and additional 34 plasmids carried by isolates from more than one patient that had sequence coverage ≥95% and sequence identity ≥99·985% compared to the reference plasmid sequence. The 46 shared plasmids were present in 336 of 3,074 (10·9%) bacterial isolates and in seven of 12 (58·3%) genera in our genomic dataset (Table 1). Bacterial isolates carrying shared plasmids were collected from 291 of 2,086 (14%) unique patients. Shared plasmids were most abundant among isolates belonging to Enterococcus (19 plasmids in 214 isolates, 85·6% of Enterococcus isolates in the dataset) (Figure 2a) and Klebsiella (16 plasmids in 77 isolates, 56·6%) (Figure 2b). Nine of the 46 shared plasmids were detected in isolates belonging to different species, and 23 others were found in isolates of different multi-locus STs within the same species, suggesting horizontal plasmid transfer (Appendix 2). The remaining 15 plasmids were found in isolates belonging to the same species and ST, suggesting they may have been vertically transmitted between patients along with the bacteria carrying them. Taken together, these results demonstrate the widespread distribution of numerous plasmids engaged in vertical transmission and horizontal transfer within a single hospital.

• Download figure
• Open in new tab

Figure 2: Plasmid distribution among nosocomial pathogens

Bootstrapped RAxML phylogenies of (A) vancomycin-resistant Enterococcus faecium and (B) Klebsiella spp. carrying shared plasmids. Color strips denote (A) sequence type (ST) of E. faecium or (B) species of Klebsiella.

Geotemporal associations among patients implicated in horizontal plasmid transfer

To investigate the epidemiology of horizontal plasmid transfer, we applied a previously published screening algorithm³³ to systematically review hospital electronic health record (EHR) data of patients infected with isolates belonging to different STs, species, or genera that shared plasmids with high nucleotide identity and high sequence coverage. Using this approach, we identified 18 groups of patients with geotemporal associations who were infected with isolates encoding eight different plasmids that were likely horizontally transferred (Table 2, Appendix 2). These associations fell into three major categories, in which previously uninfected patients were found to be culture-positive for plasmid-carrying pathogens after they were: (1) admitted to rooms previously occupied by infected patients, (2) admitted to units simultaneously occupied by at least one infected patient, or (3) admitted to units previously occupied by at least one infected patient. These findings suggest that plasmids are frequently horizontally transferred among pathogens infecting hospitalized patients independent from bacterial transmission.

Nucleotide sequence identity threshold improves identification of horizontally transferred plasmids

To investigate whether the methods we developed provided improved resolution over our previously published sequence coverage-based approach,⁵ we reanalyzed ten plasmids that were resolved and investigated in our earlier study. We found that a small ColRNAI plasmid (pKLP00155_6, 9·5kb) that was detected in a large number of isolates of multiple species was most likely two closely related but distinct plasmids that were co-circulating in our hospital (Figure 3a). Pairwise comparisons to the pKLP00155_6 reference sequence revealed a group of 13 isolates with >95% sequence coverage of the reference plasmid, but whose sequence identities ranged from 42 to 73 SNPs per 100kb. When the same sequences were compared to pCB00073_2, another reference plasmid that had previously been considered the same as pKLP00155_6 based on sequence coverage alone,⁵ their pairwise sequence identities ranged from 0 to 11 SNPs per 100kb. Separation of these two plasmids was further validated by phylogenetic analysis, which clearly delineated the plasmid sequences into two groups (Figure 3b). When we investigated geotemporal associations between patients infected with bacteria carrying these two plasmids (Table 2), we found that two patients infected by isolates of different bacterial species carrying pCB00073_2 were both cultured during overlapping stays on the same hospital unit (Figure 3c). Unit-based horizontal transmission was also observed for 16 of the 24 patients carrying pKLP00155_6 housed in three different units, with one patient linked to two of these units (Figure 3d).

• Download figure
• Open in new tab

Figure 3: Resolution of two closely related colicin-encoding plasmid variants

(A) Nucleotide sequence alignment with point mutations highlighted by red asterisks and an insertion sequence marked by a red box. (B) Bootstrapped RAxML phylogeny of core plasmid sequences, using the pCB00073_2 sequence as an internal reference. (C) Timeline of dates of culture (de-identified as study days) and inpatient stays (gray shading) of two patients on the same unit infected with bacterial isolates of different species carrying pCB00073_2. (D) Timeline of culture dates of three groups of patients on three different shared units whose nosocomial isolates all carried pKLP00155_6.

We also observed several cases of plasmid pairs that met the sequence identity threshold of ≥99·985% identity, but did not meet the ≥95% sequence coverage threshold. Sequence alignments confirmed that these plasmid pairs each shared a core or “backbone” plasmid sequence, with the longer plasmid of each pair carrying additional genetic cargo of transposases and/or IS6, IS26, and IS110-mediated insertion sequences (Supplementary Figure 2).

• Download figure
• Open in new tab

Supplementary Figure 2: Alignments of four pairs of plasmids with shared backbone sequences that met sequence identity and coverage thresholds

Grey shading indicates a BLASTn alignment of at least 2,000bp with an e-value of 0 at nucleotide sequence identities of at least 99.9%. Darker shading indicates higher nucleotide sequence identity.

Phylogenetic analysis of one of these pairs, pKLP00161_2 and pKLP00218_2, did not separate the plasmids into distinct clades defined by the presence or absence of the additional MGE, suggesting potential gain or loss of plasmid cargo during co-temporal circulation of the bacterial isolates carrying these plasmids (Supplementary Figure 3). We considered isolates with high sequence identity and high sequence coverage to one or the other of these plasmid pairs as encoding the same plasmid backbone for the purposes of assessing epidemiologic links between patients.

• Download figure
• Open in new tab

Supplementary Figure 3: Two recombining variants of Klebsiella spp. Plasmids

Bootstrapped RAxML phylogeny of core plasmid sequences conserved among isolates carrying either variant, using the pKLP00218_2 sequence as a reference.

We next performed extensive reviews of 19 patient care records associated with pathogens of different STs or species that carried four different plasmid variants with shared backbones. Ten patients infected with multiple STs of K. pneumoniae and one ST of K. oxytoca that all carried the pKLP00218_2 backbone sequence had potential exposures across four different hospital units (Figure 4a). Nine patients infected with five different STs of E. faecium carrying the pVRE32553_4 backbone sequence had potential exposures across three different hospital units (Figure 4b). Taken together, these findings demonstrate how empirically derived plasmid similarity thresholds enable tracking of plasmid circulation and horizontal transfer among bacteria. Moreover, these data highlight the potential plasticity of plasmid backbones as they circulate in a hospital.

• Download figure
• Open in new tab

Figure 4: Geographically associated patients infected by bacteria carrying plasmid backbones engaged in horizontal transfer

Plots show associations of patients infected by (A) multiple STs and species of Klebsiella spp. carrying plasmid pKLP00218_2, and (B) multiple STs of E. faecium carrying pVRE32553_4. Study days correspond to dates of culture of plasmid-carrying pathogens. Shapes and colors of data points correspond to bacterial species and ST, respectively. See Supplementary Figure 2 for plasmid alignments depicting additional mobile genetic element (MGE) sequences.