Overview of the Molecular Epidemiology of Carbapenemase-producing Bacteria in the DOD

From 2009 through 2015, ~35,000 clinical bacterial isolates were centralized tested and archived at the MRSN. Of those, ~27,000 were MDR Gram-negatives, of which ~1500 were phenotypically carbapenem-resistant. Of those, 225 were confirmed to contain one or more of bla_KPC, bla_IMP, bla_NDM, bla_OXA48-like, and bla_VIM genes based on CarbaNP testing, PCR, and NGS (Table 1). Gene content by PCR and NGS were concordant. These carbapenemase-encoding gene (CEG)-containing isolates were found in 15 different species and in 146 inpatients from 19 facilities. 84% (189) were infecting isolates, and the remainder were surveillance/colonizing isolates (Table 1). The isolates were collected from various sites: 31% urine, 23% wound, 21% respiratory, 17% blood, and 8% other. The most common mechanisms of carbapenem resistance were bla_KPC (66%) and bla_NDM (23%) in Enterobacteriaceae, and bla_OXA23-like variants in Acinetobacter spp. CEGs were very rarely found in Pseudomonas aeruginosa; Pseudomonads that were negative for CEGs likely encode mutations in OMP, efflux, or AmpC regulatory genes. bla_KPC (and bla_OXA23 in Acinetobacter spp.) was the most prevalent mechanism in the contiguous U.S., while the other mechanisms were more commonly found in isolates originating outside the contiguous U.S.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Overview of Carbapenemase-producing Genotypes in the Health and Surveillance Systems of the U.S. Department of Defense.

https://doi.org/10.1371/journal.pone.0155770.t001

There are two general categories of patients treated at overseas U.S. military hospitals, i) DOD healthcare beneficiaries (active duty, dependents, and retirees); and ii) non-U.S. coalition forces and host nation patients (in the contiguous U.S., only the former category is treated). Demographic and outcome data is consistently available only for the first category. Of the DOD healthcare beneficiaries, patients from whom a CEG-containing isolate was isolated were: 53% male; an average of 36.1 years old (range 0.5–76 years); with a mean of 9 diagnoses (range 2–30), and no infection-related deaths. However, there were reports of fatal outbreaks in host nation patients in some of the host nation hospitals that participated / submitted isolates to our network.

Detailed genomic epidemiology of these isolates is complicated with emerging and retreating clones or sub-populations depending on the species and gene, and is too large for a single manuscript, and beyond the scope of this report. Detailed of outbreaks or pseudo-outbreak reports have been provided to the individual or collectively involved facilities. In summary, genetically related CEG-containing isolates were found in more than one hospital, providing molecular evidence in support of nosocomial transmission via patient transfer to higher levels of care. Some occurrences or outbreaks of CEG-containing strains were clonal, but limited to a single facility, while others were not clonal either within or between facilities. Some occurrences and outbreaks of CEG-containing bacteria involved genetically related plasmids, and others involved several unrelated plasmids. De-duplicating on the basis of one isolate per patient per species per gene, the incidence of CEG-containing isolates peaked in 2013 (n = 65) and then decreased in 2014 (n = 22) remaining low through November 2015 (n = 8). However, based on the ratio of isolates confirmed to harbor a CEG (numerator) to the total number carbapenem-resistant phenotypes isolates tested with CarbaNP^® and/or PCR (denominator), the proportional prevalence was highest in 2013 (80/243; 32%), decreasing in 2014 to 18% (62/339), then increasing to 25% (43/175) in 2015. During this reporting period, the monthly average number of multidrug-resistant isolates submitted to the network has remained stable (~300) or slightly increased, while the intensity or consistency of surveillance has improved. No CEG-containing isolates were detected in the Northwest or Southwest regions of the DOD health system.

Treatment Implications- Phenotype vs. Genotype

The most active antimicrobial agents against the CEG-carrying isolates were tigecycline (69% susceptible) and colistin. The least active agents were sulbactam (2% susceptible) and piperacillin-tazobactam (2.4% susceptible). Using provisional susceptibility breakpoints of </ = 4 for arbekacin and <2 for colistin, 44% and 94% of the CEG-containing isolates were susceptible, respectively. At a minimum inhibitory concentration (MIC) of 16, 71% of CPB were susceptible to arbekacin. These findings also illustrate the incremental value that mechanistic determinations add to the phenotypic antibiogram (MIC alone) for empiric treatment of these pathogens. For example, we compared the antibiograms of CEG-carrying isolates to the antibiograms of a similar-sized group of the same species that were phenotypically carbapenem resistant, but that were CEG-negative based on PCR or NGS. Based on the antibiograms of the carbapenem resistant phenotypes, 80% of CRE retained susceptibility to aminoglycosides, and 70% retained susceptibility to fluoroquinolones. However, among the organism confirmed to contain a CEG by genotype, fewer than 25% retained susceptibility to aminoglycosides or fluoroquinolones.

Capacity and Turnaround Time

The sequencing pipeline provides a throughput for the above functions and annotations on 25 genomes per week, with a proven surge capacity of ten times that. A separate work flow for outbreak and routine is no longer needed. Additionally, by manipulating the chemistries of the library preparation kits size with computer processor time, we further reduced analytical turnaround time such that a SNP-based dendrogram and drug-resistance gene content report can be consistently provided to the submitting facility within 48–72 hours after we receive the specimens (11).

Availability of Sequence Data, Isolates, and Database Structure

The third sub-task was to make isolates, associated sequence and antibiogram data, and the database architecture available to others. To accomplish this, we continue to upload sequence data to three public databases -GenBank, the NCBI's Genome Tracker & Pathogen Detection System, and the nascent interagency MDR-sequence database. Additionally, 'pathogen panels' consisting of 50 isolates of each species of MDR-ESKAPE organisms selected on the basis of genetic diversity (Pulsed Field Gel Electrophoresis (PFGE)/MLST) and anatomic site (10–15 each from blood, urine, wound, respiratory, and surveillance) are freely available to other government agencies and non-profit organizations through a relatively straight-forward material transfer agreement. The database structure is also freely available through the Walter Reed Army Institute of Research’s Business and Agreements Office.

In addition to the above isolates, the following have also been sequenced as part of separate and/or ongoing investigations: ~1200 Acinetobacter baumannii isolated along the primary casualty evacuation route from theater to the two main military referral centers in the National Capital Region between 2003 and 2014 (of these approximately 87% carried bla_OXA-23); 15 colistin-resistant A. baumannii; 20 extremely drug resistant (XDR)-highly virulent A. baumannii (fatal outbreak), 400 Staphylococcus spp, 200 MDR-E. coli; 100 MDR- or tigecycline resistant Klebsiella, 50 C. difficile, 10 vancomycin-resistant Enterbactereiacea (VRE); 10 drug-resistant Salmonella/Shigella, and 80 bacteriophage.

Generating Data

NGS platforms differ in their sequencing chemistries, read-lengths, and throughput. Not all perform comparably for MDR-bacteria. Lefterova et al. offer a helpful review of platform types and specifications, along with specific applications to clinical bacteriology [17]. Very briefly, hydrogen ion detection (Ion Torrent) and reversible terminator (Illumina) platforms are based on ‘sequencing by synthesis’ (SBS) and use bridge amplification to amplify a single molecule into a cluster. The cluster contains ~1000 copies of the original molecule. All 1000 copies are sequenced together, one base at a time. Occasionally, the chemistry becomes dis-synchronous—as one of the 1000 copies may accidentally have two bases incorporated in a single cycle instead of just one. With repeated cycles, the dis-synchrony amplifies into what is referred to as ‘phase’ error. This is the primary error mode of amplified SBS chemistry and the main limit of achieving very long read lengths.

We found that hydrogen ion detection platforms did not provide the quality of sequence or read-length needed for many of our outbreak investigations of MDR-bacteria. Reversible terminator-based platforms provided better sequence quality and read length, but increasingly we found that the read lengths of that platform insufficient for: a) describing novel resistance mechanisms (i.e. an insertion element inside a promoter or gene amplification); b) closing certain bacterial genomes, and describing novel plasmids or species; c) determining transmission patterns involving small mobile genetic elements); and d) vetting or confirming certain results through peer-reviewed publication.

Single molecule technologies like PacBio and Oxford Nanopore are not affected by phasing error resulting in the generation of very long reads (generally limited only by the sample preparation process). The larger reads are easier to assemble due to more overlap between reads. Importantly, they also span repeats in the DNA sequence, allowing unambiguous location of the read on a reference, or use overlaps between reads to reconstruct the original DNA sequence. However, the cost and logistical support required for the ultra-long read / single molecule real time sequencing are daunting for many research laboratories, and prohibitive for most clinical laboratories. Less technically and logistically demanding, and less expensive alternatives are needed. Alternatively, greater access to and shorter turnaround times from the facilities that have this capability would be helpful. (During this study, we contacted all the facilities in the eastern half of the U.S. that had this technology, including the National Institute of Health and the Centers for Disease Control and Prevention, but none were able to assist us).

Another consideration relevant to generating data is purchasing supplies and equipment with public / government funding- especially the with the category of operations and maintenance funds that most clinical or hospital laboratories are funded with, which have to be spent in a single fiscal year (FY). This precludes the use of flexible or agile procurement vehicles, such as cooperative agreements, and also mandates that all funds must be obligated before the end of the FY. This has downstream effects such as being required to accept a years’ worth of consumables (space), or being unable to purchase supplies during the period between the end of the FY and the arrival of next FY funds. This limitation, in conjunction with abiding by existing federal acquisitions laws, results in lengthy pre-approval processes prior to contract award. The procurement process is not aligned to keep up with scientific innovation: on several occasions, better performing and less expensive platforms or kits were released shortly after the contract was awarded, making it too late for a course correction. For example, immediately after a government contract was announced for an RSII sequencer, the vendor announced the release of a much improved, smaller, less expensive version (Sequel), but was unwilling to re-negotiate. To this end, vendors should alert contracting officers and their representatives to impending major advances, and be more flexible. Governments should allow more agile procurement options and/or consider revising federal acquisition regulations for bio-medical efforts.

Analyzing Data

Even with increasing automation, a multidisciplinary team of dedicated, highly trained, and experienced molecular microbiologists, laboratory technologists, bioinformaticians, and clinicians remain the most valuable asset for transforming the genomic data into biological and clinically useful knowledge. We suggest the creation of a number of fulltime bench and bioinformatic positions in order to consistently offer an analytic turnaround time of 48–72 hours for prioritized outbreak investigations, while routinely sequencing 300–400 isolates per month (Table 2).

Bioinformatic and information management remains one of the greatest challenges for implementing NGS across healthcare organizations or networks [7, 18]. Some of the best and most useful software applications are open source, and as such they require lengthy approval processes before installation on secure government computer networks (Table 1). Both http://www.mil-oss.org/ and http://www.forge.mil/, collaborative environments for civilian and military open source software and hardware developers, have been suggested as alternatives or solutions to the lengthy governances and the highly restrictive.mil environment [19]. However, they are neither feasible nor useful, especially for scientific and genomic research. Currently only very basic project tracking application are posted / available at both locations. Additionally, the time it takes for newly developed software to appear and be available on the site is even longer than obtaining individual governances and approvals.

Sharing Data

Insufficient bandwidth, lack of standardized sample preparation and analysis protocols, and definitions of NGS-based clonality limit the ability to send large data files or compare results between laboratories. For example, Fig 2A–2C depicts the same bla_oxa23 carrying Acinetobacter isolates from the same outbreak event clustered by three equally valid methods—Genious, SNP and PFGE. Fig 2A–2C depicts the same bla_kpc-carrying Klebsiella from the same outbreak event similarly clustered. Different conclusions regarding relatedness can be reached for the same outbreak event depending on which software was used, highlighting the need for standardized definitions. Further complicating this are the “challenge of the supragenome” and the high degree of genome plasticity. Most bacterial pathogens have a three component genome (core genes, distributed genes, and unique genes). The relative content and combinations of each change often- even in situ during a single chronic infection event [20]. This inherently affects clonality/relatedness. An outbreak can be analyzed and clustered based on either the core or the accessory genomes, or both.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Phylogeny of Carbapenemase-producing Klebsiella and Carbapenemase-producing Acinetobacter.

A-C represent the same Klebsiella isolates from the same outbreak event, and Fig 2D–2F represent the same Acinetobacter isolates from a separate same outbreak event. However, depending on the clustering method and software program used, different results can be obtained. This highlights the need for standardized approaches.

https://doi.org/10.1371/journal.pone.0155770.g002

Having laboratories provide raw-sequence data files to public databases (e.g. NCBI’s Genome Tracker) would facilitate standardized comparisons, but this is currently limited by secure file transfer protocols and bandwidth. Raw data files are very large and not amenable to emailing as attachments. Even for small projects, DVDs and external hard drives are currently the only available option (albeit insecure and inefficient) for transferring sequence data.

Reference materials for both proficiency testing and performing quality control facilitate laboratory accreditation and enhance comparability of results regionally, nationally, and internationally. To that end, we concur with others who advocate participation in the standardization and accreditation working groups [17, 21–23].

Other organizations have called for NGS standards, quality controls, and the availability of reference materials for proficiency testing [17, 21, 23], and we concur. To that end, our goal is to have our sequencing platforms accredited by the College of American Pathologists in 2017. In addition to achieving CAP accreditation for NGS, we will also participate in the Global Microbial Identifier's 2016 Proficiency Test to validate our processes and support the GMI Initiative [23].

Storing Data

Inexpensive and cloud-based options may make data storage seem inconsequential. However, referral laboratories or those sequencing isolates from numerous facilities can quickly accumulate terabytes of data, and over the typical 5–10 program lifecycle, hundreds or thousands of terabytes. Sequencing 300–400 isolates/month, we generate ~3–6 Tb monthly.

Managing Expectations

Although some of the problems discussed above are pathogen or government-peculiar and require similarly specific solutions, many are not. By keeping in mind the above considerations and challenges, clinicians, program managers, and other stakeholders can facilitate large-scale implementation of NGS, minimize delays, and maximize the chance of matching customer and stake-holder expectations with similar results and outcomes. Additionally, as ransomware attacks on healthcare organizations increase, [24] healthcare systems may choose or be forced to implement very tight internet and information management restrictions. Civilian organizations would then face the same IM challenges or constraints government and military networks face.

In summary, we use events encountered during the system-wide NGS-characterization of a broad collection of carbapenemase-producing bacteria to illustrate the considerations and challenges of large scale implementation of that technology to help others avoid potential difficulties, pitfalls and unmet expectations.