Status and potential of bacterial genomics for public health practice: a scoping review

WGS provides increased resolution for case ascertainment and linking possible sources to these cases during outbreak investigations of food- and waterborne pathogens compared to conventional typing methods where only a small fraction of the genome is used (e.g., pulsed-field gel electrophoresis [PFGE], multiple-locus variable-number tandem-repeat analysis [MLVA], and multi-locus sequence typing [MLST]). The discriminatory power of WGS allows heterogeneous clusters of isolates, often indistinguishable with these previously used typing methods, to be split up into smaller groups of cases that are more likely to originate from a common source [21‐31]. The increased resolution of WGS is particularly useful for clonal pathogens or serotypes that show little genetic variation [32‐35]. Investigations applying WGS during the course of an outbreak were able to identify the likely source of infection and rapidly implement control measures to stop further spread [36‐42]. On several occasions, WGS data combined with epidemiological investigations enabled food authorities to intervene based on strong evidence and the subsequent timely recall of potentially contaminated food [43‐45]. Moreover, the digital and universal nature of WGS allows data to be exchanged and analyzed between different countries during multi-national outbreaks [34, 44, 46‐49]. However, this requires internationally standardized protocols and nomenclature, as well as meaningful interpretation guidelines [32]. In addition to rapid source tracing, WGS provides insights into the virulome of certain pathogenic clades [50‐53]. For example, real-time WGS is able to generate timely information concerning the presence of virulence genes during a Shiga toxin-producing Escherichia coli (STEC) outbreak [51]. WGS was often used to guide nosocomial outbreak investigations. It was reported several times that transmission events that were suspected based on epidemiological data alone or using low-resolution strain typing methods like antibiogram profiles had been disproved by integrating WGS data [54‐57]. The ability to quickly exclude a patient or potential source during an outbreak investigation is equally important for infection control purposes as the confirmation of related isolates [39], thereby preventing inappropriate, costly, and ineffective control measures [55, 56].

The most highlighted issue is the fact that WGS, as is equally the case for conventional typing methods, cannot stand on its own and that epidemiological data (including time, place, and exposure data) should complement the WGS results to identify a common source or link cases during outbreak investigations [24, 58‐64]. False conclusions could be drawn from WGS data alone since it is possible that epidemiologically unrelated isolates are highly similar at the SNP level [58, 65‐67]. Another reported issue was the potential misinterpretation of isolate relationships given the diversity of isolates that can be found within a single host (e.g., following long term carriage) or environmental reservoir. It was stressed by several studies that it is important to account for this “cloud of diversity” by increasing the number of samples taken from the suspected source [55, 66‐73]. On the other hand, within-host diversity allows to identify long-term carriers [74].

Several public health agencies launched pilot projects to implement WGS in routine practice for control-oriented surveillance purposes, such as early outbreak detection, and evaluated its performance [75‐77]. In 2013, a multi-agency collaboration prospectively performed WGS on all available L. monocytogenes isolates collected from patients, food, and food processing environments in the USA. Implementation of WGS data into their surveillance activities led to the detection of an increased number of outbreak clusters. In addition, combining WGS data with robust epidemiological information solved more outbreaks compared to before with PFGE [77]. Retrospective comparisons during the transition period from traditional to WGS-based characterization were not considered an obstacle given the possibility to accurately extract traditional typing information for L. monocytogenes from WGS data [75]. Also for STEC O157, the added value of implementing WGS as a tool to inform national surveillance was demonstrated as it led to early and accurate outbreak detection [78], as well as the ability to extract information concerning important virulence determinants and monitor the emergence of hyper-virulent strains [79, 80]. Similarly, a prospective trial of sequencing all Salmonella Typhimurium isolates concurrently with the conventional MLVA typing technique in Australia demonstrated the higher resolution offered by WGS leading to better source attributions and more targeted epidemiological investigations [81]. However, several challenges related to the interpretation of WGS data remain. As for outbreak investigations, it was reported several times that WGS results should not be interpreted on their own. A single cutoff of the number of SNPs to assess relatedness cannot consistently predict whether isolates are epidemiologically linked [77, 81‐83]. However, field data (e.g., demographic data or exposure histories) are only valuable when organized in a standardized format, requiring a more systematic approach to epidemiological data collection [84]. Another implementation barrier reported was the limited capacity of the public health unit to understand and use WGS data, implying an increased need for collaboration and exchange of expertise between microbiologists, bioinformaticians, and epidemiologists [81].

NGS has been applied to monitor antimicrobial resistance of hospital-acquired infections. Genotypic prediction of resistance of S. aureus strains seems at least as reliable as routine phenotypic testing. However, phenotypic prediction based on the genotype cannot replace phenotypic testing, as the present understanding of the genetic basis of resistance and the associated databases are not comprehensive [85]. This limitation was shown during a WGS-based surveillance of antimicrobial resistant determinants in Klebsiella pneumonia where the phenotypically determined resistance was higher than the sequenced-based resistance [86].

WGS has proven to be a more reliable tool to predict epidemiological links between tuberculosis cases than the conventional variable number of tandem repeat (VNTR) genotyping that often lead to false cluster identification [87]. WGS as a tool for the identification of tuberculosis outbreaks may be particularly useful in settings where the genetic diversity is expected to be lower such as geographically restricted M. tuberculosis populations [88], genetically closely related genotypes imported from a high-incidence region [89], or for highly monomorphic M. tuberculosis lineages [90].

Several studies showed the value of WGS in understanding the impact of vaccination on circulating pathogen populations, potentially resulting in antigenic drift to escape vaccine-mediated immune selective pressure (i.e., strain replacement) [91‐100]. Gaining insights into this is achieved by comparing the incidence of infections caused by vaccine targeted serotypes before and after the introduction of the vaccine, and to potentially identify the proliferation of non-vaccine targeted strains. The adoption of WGS methods to monitor pathogen populations during immunization programs has proven to be useful and could potentially identify differential impacts on distinct serotypes [91]. Genomic surveillance provides the required resolution for the development of targeted interventions [92] and to predict the impact of implementing a vaccination program in a given population [101]. The routine use of WGS for surveillance purposes can also inform antibiotic stewardship. One advantage of introducing WGS to inform treatment guidelines is the ability to identify genetically linked resistance that can be co-selected by multiple drugs, as opposed to phenotypic resistance rates that consider each antimicrobial class as a discrete unit [102]. In addition, resistance rates can vary significantly by clone implying that monitoring changes in population structure using WGS is useful to guide antibiotic usage policies [103]. Besides informing vaccination programs and antibiotic stewardship, WGS can reveal a detailed understanding of the transmission dynamics within and between healthcare settings, the community, and individual households, to appropriately direct control programs and decolonization strategies [104‐112].

Several studies highlighted the public health benefits of WGS-guided surveillance to monitor the spread of multidrug-resistant isolates and mobile genes, including resistance-carrying transposons and plasmids that are able to transfer resistance between bacterial species [113‐118]. The zoonotic potential of clinically relevant multi-resistant bacteria stresses the importance of the ‘One Health’ approach. In general, WGS-informed surveillance including isolates from different hosts and settings can provide evidence for interspecies transmission and focus control efforts on important reservoirs [119‐126].

The main issue reported when using WGS data to detect and to confirm transmission between isolates was the difficulty, if not impossibility, to define with a single SNP/allele threshold how much genetic variation can exist within an epidemiologically related cluster [1, 23, 28, 57, 58, 65, 67, 159]. The number of SNPs within a cluster often depends on various factors, such as the genetic diversity within each species, its molecular clock, evolutionary forces, the nature of the outbreak (point-source, long-lasting, multinational, etc.), the extent of diversity in the background population, within-host diversity, the population bottleneck during transmission, the level of asymptomatic infections, the number of isolates included in the analysis, and the methods used for genomic analysis [1, 25, 67, 74, 81, 83, 147, 160‐162]. Many studies stress the fact that we cannot rely solely on genomic information during outbreak investigations or surveillance activities and that epidemiological data describing the temporal and spatial dynamics of infection should always be considered [59‐61, 63, 67, 77, 81‐83]. Contextual data should therefore be collected carefully and combined with WGS data for a proper interpretation, which is almost seamlessly linked to the challenge of data integration.

A useful interpretation of genomic data is highly dependent on the epidemiological and clinical metadata [1, 160, 163, 164]. The integration of laboratory and epidemiological data is often hampered by the incomplete and/or unstructured nature of the contextual data [13, 84, 128, 165]. For example, during a multi-country outbreak investigation, it is important to develop a codebook for uniform and standardized data entry between countries [34, 49]. To maximize the potential of WGS, public health professionals have to identify a minimum set of variables (such as time, place of infection, host characteristics, clinical presentation, and exposures) that should be incorporated within surveillance activities of a particular pathogen [165].

Although WGS data has the potential to support phenotypic predictions of virulence and resistance based on the genotype, phenotypic data will still be needed to identify new resistance/virulence mechanisms and to keep the databases up-to-date [12, 85, 86, 157, 166, 167]. Therefore, phenotypic testing results and clinical data have to be collected in a standardized manner alongside the sequence data to feed the databases from which associations between genotype and phenotype can be observed [160].

More recently, digital streams (also called “Internet of things”) are being used as an input for surveillance systems (i.e., digital epidemiology). Examples include search engines, social media, mobile phones, and health trackers. These novel data streams, generated outside public health, could potentially enrich epidemiology by providing information on natural and social phenomena [168].

One Health, the concept of structured collaboration and coordination between human, animal, and eco health systems, has become an emerging focus due to the increased understanding of how animal and ecological reservoirs significantly influence human health [169]. Therefore, the management of infectious diseases requires sampling from different hosts and sources. As indicated by Rantsiou et al., the development of WGS is currently not at the same level in the food industry as compared to public health agencies [170]. The outputs produced by the different sectors should remain comparable at any time to ensure the linkage of isolates.

An overview of data integration is presented in Fig. 3.

Following the previous section addressing the importance of data integration, it is clear that the switch to WGS requires an increase in multi-disciplinary working [9, 81, 148, 171]. In particular for data interpretation, expertise in bioinformatics and in biological, epidemiological, and microbiological sciences needs to be combined. Infectious disease epidemiologists implementing WGS data into their routine workflow might need training in genomics as well as skills in analyzing high-dimensional data sets. In addition to interdisciplinary and inter-sectoral (One Health) collaboration, the implementation of WGS should be coordinated at an international level as infectious diseases do not respect national boundaries [49].

As for any type of epidemiological study, genomics-informed surveillance activities should be based on robust sampling strategies defining the required sample size and the number of samples needed from the different sources. The sampling framework will vary depending on the type of public health application (e.g., investigating an explosive outbreak often requires dense sampling including multiple samples from single sources, as opposed to strategy-oriented surveillance requiring a representative coverage of the population) [171]. Selection bias can be introduced when the subset of samples selected for WGS is not representative [128, 163]. In order to efficiently assemble a representative sample, it is often needed to develop a stratified sampling scheme according to time, place, and person, or to perform normalizations to maintain original sampling fractions.

Genomic information must be interpreted and translated in a meaningful manner into both immediate public health action and longer term prevention programs [2, 57, 172]. Moving from low-resolution typing methods to WGS-based typing will lead to an increase in the detected number of hazards. Not only will the sensitive nature of WGS increase the number of clusters detected [76, 77], it will also provide additional information on the presence of resistance genes, virulence genes, etc. It will be important to filter out the hazards that are truly relevant from a public health point of view (i.e., separate the “signal” from the “noise”) and that subsequently require the initiation of a public health action.