Techniques in bacterial strain typing: past, present, and future

Abstract

Purpose of review

The advancement of molecular techniques such as whole-genome sequencing (WGS) has revolutionized the field of bacterial strain typing, with important implications for epidemiological surveillance and outbreak investigations. This review summarizes state-of-the-art techniques in strain typing and examines barriers faced by clinical and public health laboratories in implementing these new methodologies.

Recent findings

WGS-based methodologies are on track to become the new ‘gold standards’ in bacterial strain typing, replacing traditional methods like pulsed-field gel electrophoresis and multilocus sequence typing. These new techniques have an improved ability to identify genetic relationships among organisms of interest. Further, advances in long-read sequencing approaches will likely provide a highly discriminatory tool to perform pangenome analyses and characterize relevant accessory genome elements, including mobile genetic elements carrying antibiotic resistance determinants in real time. Barriers to widespread integration of these approaches include a lack of standardized workflows and technical training.

Summary

Genomic bacterial strain typing has facilitated a paradigm shift in clinical and molecular epidemiology. The increased resolution that these new techniques provide, along with epidemiological data, will facilitate the rapid identification of transmission routes with high confidence, leading to timely and effective deployment of infection control and public health interventions in outbreak settings.

INTRODUCTION

The increasing interconnectedness of society has greatly influenced the transmissibility and subsequent diversification of bacterial pathogens [1], creating a need for improved methods of bacterial characterization and classification. Bacterial strain typing – the practice of microbial characterization used to discriminate between strains of a bacterial species – is a fundamental aspect of epidemiological surveillance and investigation. Strain typing can characterize and confirm epidemiological linkage in an outbreak setting and provide insights into bacterial population dynamics. However, traditional typing methods often target only a small portion of the bacterial genome, limiting the resolution and, thus, the scope of our understanding of the molecular epidemiology of clinically relevant pathogens.

There are a few traditional methods that remain primary choices for strain typing in many clinical and public health laboratories. The first of these methods is pulsed-field gel electrophoresis (PFGE), which uses alternating electric fields applied at differing angles within an agarose gel to separate large DNA molecules, creating size-dependent banding patterns, or ‘fingerprints,’ based on restriction enzyme cleavage sites [2]. This method is known as the ‘gold standard’ for subtyping and, until recently, was the primary method used by the Center for Disease Control and Prevention’s PulseNet database for tracking outbreaks of foodborne illnesses [3]. However, PFGE has important limitations, including a need for protocols that are standardized for individual pathogens, extreme sensitivity to the selection of restriction enzymes, a time-consuming and labor-intensive workflow, and relatively low throughput. Thus, many large-scale surveillance efforts have transitioned to the use of WGS-based characterization, such as multilocus sequencing typing (MLST), a sequencing-based method that uses allelic permutations of conserved ‘housekeeping genes’ loci to create MLST schemes known as ‘sequence types’ (ST) [4]. Although this method provides unambiguous results and allows for easier inter-lab comparisons through a centralized database (PubMLST; pubmlst.org), it can be cost-prohibitive, each species requires a different typing schema, and it lacks the ability to further discriminate relatedness within STs [4,5].

Strain typing methodologies have recently undergone a paradigm shift as whole-genome sequencing (WGS) has become cheaper and more accessible to clinical and public health laboratories. WGS provides unmatched resolution and discriminatory power for highly related strains, and it has significant potential for outbreak detection, epidemiological surveillance, and infection control strategies.

NEW TECHNIQUES IN BACTERIAL STRAIN TYPING

The increased resolution provided by new strain typing methodologies has enabled the distinction of bacteria differing at only a few genetic sites, which is a significant advancement from the discriminatory power of traditional strain typing methods (Table 1). The definitions of commonly used terms to classify genetic relatedness among bacterial strains are shown in Table 2. It should be noted that, although these terms are often used interchangeably to define sets of related isolates, this exchangeable use is due to a drift in the terminology used over time amongst scientists studying different pathogens. We have included the original definitions apart from ‘clonal group,’ which was previously described as an a priori defined cluster of bacterial organisms that shared (n) alleles of their associated STs [8]. This definition is ambiguous in practice and is often operationalized with the same definition as clonal complex, so we recommend the stricter definition as defined in Table 2.

Table 1.

Features of molecular strain typing methods for bacterial organisms

Method	Type of markers used for differentiation	Discriminatory power	Reproducibility	Bioinformatic knowledge needed	Cost
Pulsed-field gel electrophoresis (PFGE)	Number of bands depending on restriction enzyme	•	••	•	••
Multilocus sequence typing (MLST)	7–8 housekeeping genes	••	••••a	••	••
Core genome MLST (cgMLST)	Hundreds to thousands of core genes	•••	••••a	•••	•••
Whole genome MLST (wgMLST)	Hundreds to thousands of core plus accessory genes	•••	••••	•••	•••
Reference-based single nucleotide polymorphism (SNP) calling	Depends on organism of interest plus reference choice	••••	•••	••••	••••
Reference-agnostic/k-mer based SNP calling	Depends on organism of interest	••••	••••	••••	••••

• low, •• medium, ••• high, •••• very high.

^aGenerally high, but depends on organism of interest and chosen reference.

Table 2.

Definitions of terms commonly used to classify genetic relatedness among bacterial strains

Term	Definition
Clade	A group of organisms that contains a single ancestor and its descendants; a monophyletic group [6]
Clade	A group of isolates that are genetically indistinguishable [though not necessarily identical] based on a particular molecular typing method and are presumed to be descendants of a common ancestor [7]
Sequence type (ST)	Organisms that possess identical allelic profiles of fragments of predetermined housekeeping genes [4]
Clonal group	All isolates that belong to a particular ST [8,9]
Clonal complex	A cluster of bacterial organisms that originate from a common ancestor and generally share at least 6/7 alleles of their associated ST with another member of the group [8]
Strain	Isolate(s) that are distinct from other isolates of the same genus and species based on phenotypic and/or genotypic features [9]

Beyond multilocus sequencing typing

The increased use of WGS has enabled the expansion of traditional MLST methods based on 7–8 housekeeping genes to hundreds or thousands of genetic loci, greatly enhancing the precision and discriminatory power of typing and providing relevant clinical and epidemiological information. Here, the two newest expansions of MLST are described in detail.

Core genome multilocus sequencing typing

This approach is also referred to as the gene-by-gene approach. cgMLST is similar to MLST but utilizes a larger proportion of the genome, defined as the core genome (the set of genes that is found in nearly all strains of a species) to determine genetic relatedness. After WGS, a genome assembly is aligned to a reference-based ‘scheme’ of core genes, and each isolate is characterized based on allelic variations relative to the reference [10■,11■]. Along with delivering higher resolution relative to traditional and MLST typing methods, cgMLST provides the opportunity to investigate organism phylogeny through strategies that include the use of distance-based techniques to create nearest neighbor or minimum-spanning trees [12■,13■]. Since its inception, cgMLST has become a widely used alternative to MLST for those seeking greater resolution through a similar workflow, and typing schemes based on cgMLST have been published for a number of bacterial species, facilitating its use in outbreak investigations [13■,14-16]. In a recent example of cgMLST application, Hansen et al. utilized this typing method to identify an outbreak of vancomycin-resistant Enterococcus faecium by establishing epidemiological links between patients carrying isolates belonging to the clone ST80-CT993 and distinguishing this clone from similar circulating STs. This analysis identified affected wards, and a targeted infection control intervention was successfully implemented in these areas, saving time and resources with important implications for hospital epidemiology [17■■].

To date, there is no centralized or consistent naming system for cgMLST schemes. In fact, there are three distinct cgMLST schemes available for Pseudomonas aeruginosa [14,15,18], which may result in lack of reproducibility in future published data. To create a reliable central database, a large number of high-quality reference genomes would be needed for each species of interest, which is an expensive and computationally intensive undertaking. Most importantly, cgMLST only accounts for the conserved genes within a species and ignores the contribution of the accessory genome—the portion of the genome that varies between strains of a species—to overall intra-clonal diversity [19■].

Whole-genome multilocus sequencing typing

This approach is an extension of cgMLST that utilizes both the core and accessory genomes (the pangenome), theoretically providing higher resolution than cgMLST for closely related isolates than the cgMLST approach. In a retrospective investigation of listeriosis outbreaks in small ruminants, wgMLST uncovered a larger breadth of genomic diversity relative to cgMLST, supporting previous findings indicating that wgMLST should be the primary typing method when investigating highly related bacterial groups [20■]. However, several studies have not been able to demonstrate a significant difference in discriminatory power between these two typing schemes [21,22]. A study by Blanc et al. even found wgMLST inferior to cgMLST due to homologous recombination of a DNA fragment affecting phylogeny with no epidemiological significance [23■■].

wgMLST shares some limitations with cgMLST since the choice of high-quality references is essential for reliable discriminatory power, and a standardized method of classification is lacking. Additionally, wgMLST requires a higher level of bioinformatic expertise relative to other typing methods, and assembly and alignment of genomic short reads (the output of the most commonly used sequencing platforms) are not robust to permit the reconstruction of complex genomic structures such as mobile genetic elements (MGEs) and long repeat structures [19■,23■■].

Single Nucleotide Polymorphism-based methods

Although cgMLST and wgMLST-based methods of strain typing are currently being applied as higher-resolution replacements for traditional MLST and PFGE, there is a considerable amount of genomic variability that cannot be accounted for with these methodologies. Indeed, regions in the accessory genome are often not considered with these approaches, resulting in an inability to differentiate closely related outbreak strains for the detection of recent transmission events when paired with traditional epidemiological metadata. The identification of single-nucleotide polymorphisms (SNPs) between bacterial isolates is one of the most commonly used analyses on WGS data and can be performed with or without the use of a reference genome. Here, two methods and applications of SNP calling are discussed.

Reference-based single-nucleotide polymorphism calling

SNP calling is most often reference-based and involves the alignment (frequently referred to as mapping) of sequenced isolates to a closely related reference genome to detect SNPs and quantify the genetic relatedness between strains [24]. Though MGEs and regions of recombination are generally excluded, over 95% of the genome is accounted for in these analyses [25]. Reference-based SNP calling is particularly useful when a relatively small number of isolates are available for analysis. Indeed, Hoang et al. used this strategy to identify region-specific lineages of six Bacillus anthracis strains isolated in northern Vietnamese provinces, finding that all strains could be classified into a single lineage that has not been previously reported in Asia [26■■].

Reference-based SNP calling as a typing method is relatively straightforward and can yield highly accurate results, but it has an important limitation; the selection of a reference genome is of paramount importance, and a closed genome that is highly related to the sequences of interest is highly desired [25]. Thus, reference-based SNP calling becomes a difficult task when analyzing nonmodel organisms for which there are no well-established reference genomes. Varied reference choices and SNP calling workflows can greatly influence the number of SNP differences identified, limiting reproducibility and comparability across studies and laboratories and resulting in incorrect epidemiological inferences [25,27]. For example, a study by Valiente-Mullor et al. examined the effect of using different reference genomes for SNP calling and phylogenetic analyses of five bacterial species and found that the choice of the reference strain had an impact on all parameters considered, including SNP calling and phylogenetic tree construction [28■■].

Reference-agnostic single-nucleotide polymorphism calling

To circumvent the need for an appropriate reference strain, methods based on k-mer comparisons have been developed for WGS data. K-mers are defined blocks of nucleotides of length (k) that can be compared in a pairwise fashion between sets of genomes of interest to model intra-sample diversity and taxonomy [29■]. This approach has been applied to a number of retrospective outbreak investigations, including the first WGS-based characterization of Bacillus cereus isolates linked to a foodborne outbreak, where investigators compared a number of reference-based SNP calling tools using a reference-free, k-mer based approach (kSNP3) [30]. This investigation found that kSNP3 produced consistent results that were not affected by the choice of reference genome. These findings also support existing literature that reference-free methods are most reliable in suspected outbreak situations, or where isolates are expected to be relatively similar [31■■]. Another study by Cremers et al., that used this approach to study an outbreak of methicillin-resistant Staphylococcus aureus in a neonatal intensive care unit, found that a k-mer-based pairwise SNP analysis substantially outperformed other typing methods, including cgMLST and wgMLST. Maximizing the amount of genetic material utilized for comparison from both the core and accessory genomes produced highly precise insights into potential chains of transmission among neonates [32■■].

Other considerations

Though SNP-based typing methods arguably deliver the highest discriminatory power of all the methods described thus far, there is still considerable debate among the scientific community regarding SNP thresholds for genetic relatedness (‘clonality’) that impacts the interpretation of outbreak and infection control investigations. Determination of clusters and significance thresholds is often based on substitution and recombination rates along with quantification of SNPs and is therefore not universally applicable to all bacterial species [24]. However, several recent studies have made efforts to define thresholds of genetic relatedness that indicate transmission events in an outbreak setting for several organisms. These include cutoffs of 25 whole-genome/15 core genome SNPs for methicillin-resistant S. aureus [33■], and 6 and 21 core genome SNPs for E. faecium and carbapenem-resistant Klebsiella pneumoniae, respectively [34■,35■]. There is a need for further studies on this subject, as these cut-offs may be dynamic and dependent on host and environmental factors [36■■]. Lastly, the importance of epidemiological evidence and context should be taken into consideration, as genomic data alone is not sufficient for determination of outbreak transmission dynamics with full confidence [20■,36■■].

FUTURE DIRECTIONS

Defining the core, or conserved, regions of bacterial genomes is an important first step in most strain typing and phylogenetic analyses. However, the bacterial ‘mobilome’ (the repertoire of acquired MGEs) is a primary driver of adaptive evolution through horizontal gene transfer and a major determinant of bacterial resistance and virulence phenotypes [37■]. Indeed, the mobilome is generally disregarded in most strain typing methodologies, as MGE structures are difficult to reconstruct with widely used WGS methods like short-read sequencing, and it is difficult to distinguish between transient gene acquisition and stable assimilation into genetic lineages [38]. Thus, more extensive research is critical to enable proper consideration of the role of the mobilome in the context of genomic diversification and its impacts on strain typing and outbreak investigations.

The advent of long-read sequencing technologies, such as those from Oxford Nanopore Technologies (ONT) and Pacific Biosciences, allow for accurate reconstruction of complicated MGEs—particularly plasmids harboring AMR and virulence determinants—due to their ability to generate sequencing reads that span the entirety of these complex genetic structures. However, the high error rates often associated with these technologies pose a challenge for accurate genomic analysis, particularly SNP calling. To overcome the limitations of both short- and long-read sequencing, hybrid assemblies may be created using highly accurate short-read data to ‘polish’ the less-accurate long-read sequences, generating closed, reference-quality genomes [39]. Neal-McKinney et al. compared Illumina short-read and Pacific Biosciences long-read sequencing alone to hybrid assemblies generated by this technique in Campylobacter jejuni and found that the latter created the most contiguous assemblies and was the superior method for SNP typing and definitive isolate characterization [40■]. This technique was also used by Prussing et al. to identify the potential transfer of a plasmid harboring bla_KPC-2 across bacterial species in epidemiologically linked patients [41■].

Despite these advances in sequencing technology and strain typing methodologies, most WGS-based outbreak investigations and surveillance efforts are still performed retrospectively, limiting the impact these methods can have on clinical decision-making and infection control interventions at the time they are most needed. Thus, there is a critical need to place more emphasis on developing tools and workflows for real-time sequencing and data analysis. Currently, there is only one methodology available for such applications—the long-read nanopore sequencing platform developed by ONT (www.nanoporetech.com). Since the release of its first sequencer, the MinION, in 2014, ONT sequencing platforms have been increasingly utilized in environments ranging from small-scale research studies that have uncovered new classes of antibiotic resistance plasmids [42■] to the implication of contaminated detergent as the source of an extended-spectrum beta-lactamase- producing Klebsiella michiganensis outbreak in an Australian neonatal unit [43■■]. However, while this technology has enormous potential for advancing the fields of real-time bacterial identification, strain typing, and outbreak and surveillance efforts, there is much work to be done to optimize and standardize long-read sequencing library preparation and analysis workflows before this technology can be scaled to larger datasets.

IMPLEMENTATION IN THE CLINICAL AND PUBLIC HEALTH LABORATORY

WGS strain-typing workflows are increasingly being adopted by clinical and public health laboratories as these technologies become more accessible and cost-effective. Sequencing the entire genome of an infecting/colonizing organism provides not only unparalleled discriminatory power for highly related lineages, but also delivers insight into drug susceptibility and virulence potential, which would otherwise require a combination of laboratory methods and tools. Timely accessibility of this breadth of information is crucial for effective outbreak management and infection control efforts. Yet there remain barriers to widespread integration of WGS-based bacterial typing into clinical and public health laboratory workflows.

Standardization

A significant barrier to implementation of WGS methodologies is the lack of standardized workflows. Protocols and analysis methods (from the quality of DNA extracted to the choice of SNP calling pipeline) vary considerably between laboratories, resulting in differing interpretations, quality control issues, and decreased reproducibility [44■,45■]. In 2017, a Swiss trial of nine laboratories aimed at fostering harmonization of WGS-based bacterial strain typing found that, whereas MLST typing, phylogenetic tree construction, and cluster identification were relatively harmonious across laboratories, differing interpretations of sequencing data based on SNP counts led to diverse inferences regarding strain relatedness during outbreak investigations, highlighting the need for standardized definitions and interpretation criteria to reach reproducible conclusions [12■]. In order for WGS-based methods to become the new standard in strain typing, there must be full confidence in the accuracy and robustness of the data generated across different sequencing platforms and laboratories. Every step of the WGS workflow—sample preparation, sequencing, and downstream analysis and interpretation—needs to be standardized and validated with a variety of bacterial species against current ‘gold standard’ typing methods. Furthermore, analysis tools and pipelines must be version-controlled, and parameters used for each workflow must be standardized and validated. This is not a trivial task, as sequencing technologies and data analysis methods are constantly changing. It may be helpful to look to human genetics for insight, as this field has made considerable progress in the creation of well-established references and tools for widescale laboratory use [46].

Analysis training and expertise

Another barrier to integration of WGS in many laboratories is the absence of bioinformatics expertise needed to analyze WGS data. As bioinformatic analysis approaches are not commonly utilized in most diagnostic or public health laboratories, emphasis must be placed on developing tools that are user friendly, otherwise, laboratories would need to hire bioinformaticians to aid in interpretation of data. Lastly, there remains a critical need to train the next generation of clinical microbiologists in WGS and bioinformatics practices to meet these needs and further the advancement of WGS analysis tools.

CONCLUSION

The increasing accessibility and cost-effectiveness of WGS have catalyzed the innovation of new, higher-resolution bacterial strain typing methods that are likely to replace traditional typing methods as the new ‘gold standard’ in the coming years. However, significant work will need to be done regarding standardization of sequencing and analysis workflows, personnel training, and increasing cost-effectiveness before such methodologies can be widely implemented in clinical and public health laboratories.

KEY POINTS.

• Whole-genome sequencing has enabled a paradigm shift in bacterial strain typing methodologies.

• Single nucleotide polymorphism (SNP) calling provides the highest discriminatory power relative to other WGS-based typing techniques but is subject to important limitations that include the lack of standardization in thresholds to define relatedness in bacterial species.

• There remain important barriers to wide-scale implementation of WGS-based strain typing methodologies in clinical and microbiological labs – namely, an absence of harmonized workflows and appropriate analytic training.