Taking hospital pathogen surveillance to the next level

Abstract

High-throughput bacterial genomic sequencing and subsequent analyses can produce large volumes of high-quality data rapidly. Advances in sequencing technology, with commensurate developments in bioinformatics, have increased the speed and efficiency with which it is possible to apply genomics to outbreak analysis and broader public health surveillance. This approach has been focused on targeted pathogenic taxa, such as Mycobacteria, and diseases corresponding to different modes of transmission, including food-and-water-borne diseases (FWDs) and sexually transmitted infections (STIs). In addition, major healthcare-associated pathogens such as methicillin-resistant Staphylococcus aureus , vancomycin-resistant enterococci and carbapenemase-producing Klebsiella pneumoniae are the focus of research projects and initiatives to understand transmission dynamics and temporal trends on both local and global scales. Here, we discuss current and future public health priorities relating to genome-based surveillance of major healthcare-associated pathogens. We highlight the specific challenges for the surveillance of healthcare-associated infections (HAIs), and how recent technical advances might be deployed most effectively to mitigate the increasing public health burden they cause.

Introduction

Disease notifications are collected over a broad range of geographical scales, from local, regional, national and continental up to the global level, and consist of three classes – syndromic surveillance, clinical surveillance and laboratory-confirmed cases. Not all of these three pillars are required in all instances; some diseases require a diagnostic, laboratory-driven notification, whilst others require both clinical and diagnostic confirmation. For novel diseases, such as coronavirus disease 2019 (COVID-19), which emerged in late 2019, only syndromic surveillance is possible in the initial stages of an outbreak, an epidemic or a pandemic due to a lack of knowledge concerning the disease and/or pathogen. Other infections are non-notifiable for various reasons, providing an even leaner portfolio of available data for public health. Outbreak investigation has followed the same basic activities for decades: a patient presenting with a suspected infectious disease will elicit a battery of diagnostic tests combined with epidemiological investigations to identify likely sources of the disease and potential cases of onward transmission. Strain typing to elucidate putative outbreak scenarios and transmission routes has traditionally been a specialty that is not necessarily linked to the other activities and medical disciplines and is essentially performed by specialists in a small number of well-equipped and dedicated expert and reference laboratories.

The results of strain typing should confirm (or disprove) epidemiological hypotheses deduced from clinical and patient data, or alert clinicians to potential outbreaks before these are epidemiologically recognized. Strain typing comprises many different techniques and approaches, complemented by an increasing knowledge of bacterial populations as well as technical and analytical advances. Not all methods are appropriate for all pathogens, standardization is challenging or impossible for many techniques, and for gel-based methods typing results are mostly not transferable between laboratories. Certain methods, such as ribotyping, rapid PCR-based typing (RAPD), amplified-fragment length polymorphism (AFLP), multi-locus variable number of tandem repeats analysis (MLVA) and macrorestriction analyses in pulsed-field gel electrophoresis (PFGE) – just to mention the most prominent ones – have replaced or accompanied each other for some years [1–5].

The first change came with the application of multi-locus sequence typing (MLST) as an early sequence-based approach, which generated robust and digital data that were readily transferrable and comparable between laboratories [5]. However, resolution is variable according to the species/pathogen and limited by comparing only seven of the several hundred to thousands of genes. Despite these limitations, MLST typing has retained its utility as a typing tool for defining clonal lineages within species. For the most part, lineages defined as sequence types (STs), clonal groups or clusters of closely related STs have remained robust in the genomics era. This has meant that nomenclature based on ST labels are still valid and useful, and web-based MLST platforms such as Enterobase (https://enterobase.warwick.ac.uk/), PubMLST (https://pubmlst.org/) and BIGSdb-Pasteur (https://bigsdb.pasteur.fr/) have facilitated the global establishment of this nomenclature [6].

The advent of high-throughput sequencing introduced a ‘one-size-fits-all’ approach, which allows the highest discriminatory power by determining almost the entire genetic information of a pathogen. The introduction of accurate genomic sequencing techniques is associated with decreased turnaround times, lowered costs and multiplexing options, which has allowed outbreak analyses and phylogenetic comparisons to be accessible to a broader scientific and medical community, rather than being limited to a small number of well-equipped research groups or referral centres. This process started about 10 years ago and heralded a new era in outbreak detection and genomic pathogen surveillance for public health purposes, with the latter being in the focus of this paper.

Turning pathogen surveillance upside down

The introduction of an increasingly fast, highly reliable, accurate, more affordable and technically standardizable, genome-based typing method underpinned a revolution in pathogen surveillance. Genome sequencing and subsequent data analyses are more scalable and allow hypothesis-free data evaluation. It is no longer necessary to limit typing to a small number of isolates selected based on comprehensive and representative clinical, patient, epidemiological and diagnostic data; rather, high-throughput genomic sequencing itself provides the ability to generate hypotheses of transmission pathways and/or source attribution from large, prospective and transversal isolate collections [7].

There are numerous examples in the literature demonstrating the identification of previously unknown sources and protracted outbreaks closer to real time than ever before [8]. This has been especially noticeable in countries such as Denmark and the UK, where substantial investment was made early on to integrate this new technology into a routine public health application [9, 10].

Interdisciplinary working groups at national and continental levels attached to international agencies such as the European Centre for Disease Prevention and Control (ECDC), Centers for Disease Control and Prevention (CDC), United States Food and Drug Administration (FDA) and European Food Safety Authority (EFSA) published concept papers providing roadmaps for the scaling up of genomic pathogen and disease surveillance using high-throughput genomic sequencing technology [11–13].

Early systematic public health initiatives focused on genome-based surveillance for pathogens of food- and water-borne diseases (FWDs) such as Listeria , entero-haemorrhagic Escherichia coli (EHEC) and Salmonella , but also Mycobacterium tuberculosis and bacteria causing sexually transmitted infections (STIs) [11, 14–16]. The reasons for selecting these pathogens in particular were manifold, comprehensible and justifiable, but public health priorities vary between different nations and initiatives and not all national and global action plans (on antimicrobial resistance and/or on genomic pathogen surveillance) were congruent when compared to each other and across the different initiatives.

The transition to whole-genome sequencing (WGS) is not, however, free from challenges, and a bottleneck remains in the technical and analytical skills required to process and evaluate the samples. Studies such as EuSCAPE and the CCRE survey of the European Antimicrobial Resistance Genes Surveillance Network (EURGen-Net: https://www.ecdc.europa.eu/en/about-us/who-we-work/disease-and-laboratory-networks/EURGen-net) and the EU-funded research project COMPARE (https://www.compare-europe.eu/) have successfully endeavoured to establish and build capacity for genomic surveillance in recent years. A similar activity was instigated by the Global Microbial Identifier (GMI) initiative (https://www.globalmicrobialidentifier.org/).

In addition, novel technological approaches and achievements such as metagenomics sequencing for pathogen diagnostic and strain typing, machine learning for antimicrobial resistance (AMR) prediction, and large-scale environmental sampling to extrapolate the influx of AMR and pathogens from ‘One-Health’ sources were developed and successfully applied in several of these activities [17, 18].

The severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) pandemic further demonstrated the public health value of genomic epidemiology and has triggered international investments in sequencing capacity and digitalization of surveillance infrastructures. Long-standing scientific collaborative networks with respect to training throughout South America and Africa have also come to fruition, as evidenced by the genomic epidemiology of SARS-CoV-2 and the submission of sequencing data from these countries during the pandemic [19].

Hospital pathogen surveillance – it is mainly about resistance

For STI and FWD pathogens the elucidation of a clonal spread and/or the identification of a potential source are primary objectives, whereas AMR is analysed as a secondary insight. For healthcare-associated (HAI) pathogens, there is a primary focus on AMR determinants and their potential for transmission, as this has urgent implications for therapeutic interventions and infection control. The ability to distinguish between clonal expansion versus horizontal resistance gene spread is central. Of course, susceptible nosocomial pathogens also spread; however, from the current public health view the spread of vancomycin-susceptible Enterococcus faecalis or methicillin-susceptible Staphylococcus aureus is of minor therapeutic concern and public health importance. In essence, therefore, molecular surveillance of bacterial hospital pathogens is in fact AMR surveillance.

Resistance in healthcare pathogens spreads in narrower or wider patient and hospital networks [20, 21]. Depending on the pathogen, the extent of clonal versus horizontal spread may differ, as well as transmissibility potential. Multidrug- and methicillin-resistant staphylococci, either S. aureus or S. epidermidis, rather spread clonally, within a hospital, region, nation, or even globally [22, 23]. On the other hand, vancomycin-resistant enterococci (VRE) spread regionally, with differences at the national and global level and variable dynamics of vanA and vanB vancomycin resistance determinant transfer within Enterococcus faecium [20, 24]. The same holds true for multidrug-resistant Gram-negative pathogens, where ESBL- and carbapenemase-mediated resistance in the nosocomial setting show variable transmission routes – (i) clone-associated, as with KPC-2-producing ST258 Klebsiella pneumoniae or CTX-M-15-producing ST131 E. coli ; (ii) plasmid-mediated, as with IncL carrying bla _OXA-48 or IncX4 carrying mcr-1; and (iii) IS-mediated, as with ISCR1, IS26 or ISEcp1, which can mobilize a diverse set of resistance genes [25–28].

To investigate clonal spread, strain typing based on WGS data often uses core genome MLST (cgMLST) approaches, allowing a common nomenclature and a standardized data format. In fact, a recent proposal for strain typing nomenclature based on cgMLST and life identification number (LIN) coding promises full stability and continuity of strain taxonomy [29]. However, cgMLST typing also has some limitations, for instance, a limited discriminatory power for all public health demands [30, 31], and is highly dependent on the quality and breadth of the scheme used. More flexible and adaptable methods will be needed to overcome this, such as working with identifying unique sequences (kmer), as has been suggested recently, as well as other promising alternatives, especially for typing multidrug-resistant HAI pathogens [31]. Whole-genome MLST (wgMLST) could be another alternative; however, it is difficult to standardize and lacks a systematic nomenclature, as the number of genes used varies according to the strains included in the analysis.

Techniques and approaches for genomic AMR surveillance – from short-read to long-read sequencing

AMR determinants are usually plasmid-borne and are commonly flanked by highly repetitive sequences that are difficult to assemble using short-read sequencing data only. Antibiotic resistance plasmids and other mobile genetic elements (MGEs) such as composite transposons and integrative mobilizable or conjugative elements are complex, flexible and mosaic structures that are not easily reconstructed using short-read sequencing data [32]. Reference plasmids deposited in data archives and novel technological approaches to reconstruct MGEs ease data extraction from short-read data [33, 34]. Long-read sequencing can circumvent this drawback, but at the expense of higher costs and higher computational demands, mainly with Pacific Bioscience’s technology (https://www.pacb.com/).

Oxford Nanopore Technology (ONT; https://nanoporetech.com/) represents a huge advance in scalability, affordability and flexibility, offering possibilities from field applications (MinION) and establishment of rapid sequence-based surveillance in low-resource regions, to medium- and high-throughput options such as GridION and PromethION; however, this comes at the cost of lower sequencing accuracy.

Until recently, long-read sequencing applications in genomic AMR surveillance and public health were research-driven, but with the option of multiplexing several samples per run it is now possible to drastically reduce costs per sample and increase throughput, which now makes long-read sequencing attractive and applicable for genomic AMR surveillance and public health applications. For several population-based research applications, huge sets of short-read sequencing data were improved by long-read sequencing [35, 36]. Combining long- and short-read data in this way is the most promising approach at present for routine, public health-oriented, WGS-based hospital pathogen surveillance. New bioinformatic pipelines can utilize the advantages of both data types to generate fully closed, or nearly fully closed, assemblies for both chromosomes and plasmids. This offers exciting new opportunities for outbreak and phylogenetic analyses, AMR prediction and genomic hospital pathogen and AMR surveillance [37, 38].

Beyond sequencing – the fun starts after the sequencing has finished

As much as sequencing capacities and capabilities have increased, other sides of the genomics process need to be considered. The data produced should be scrutinized for quality before being used in any downstream analyses. The workflows that can be used to process and evaluate the data are numerous, depending on the user. Commercial solutions such as EPISEQ (https://www.biomerieux-diagnostics.com/biomerieux-episeq-cs) [39] or AREScloud (https://www.opgen.com/ares/) [40], or open-source applications such as the DTU suite of bioinformatics analyses, are available and provide easy access (http://www.genomicepidemiology.org/services/) [41]. Commercial software solutions may have some advantages in terms of a user-friendliness and workflow accessibility for non-specialists. However, they are particularly well suited to specific clinical and research applications and thus do not fulfil every requirement. Moreover, dependence on commercial software solutions comes with a risk, for instance, if the providing companies withdraw future support for their products due to competitive and internal considerations. In this way, efforts in creating an open access streamlined reporting workflows that are especially well suited for clinical and public health epidemiology can significantly boost genomic surveillance. Command line-based tools are often free to use but require a working knowledge of Unix and installation of software. Cloud infrastructure such as MRC-CLIMB (https://www.climb.ac.uk) [42] or Galaxy (https://galaxyproject.org) [43] can help with challenges related to installation and accessibility, and can provide an important entry point for more complex analyses.

In any case, the user reporting the results needs to be able to make informed inferences from the results and to identify artefacts resulting from contamination or software failures. Training of laboratory microbiologists and standardization of tools and pipelines are essential. Large research consortia such as COMPARE and European bodies such as the ECDC and societies such as the European Society of Clinical Microbiology and Infectious Diseases (ESCMID), with their several study groups (e.g. ESCMID Study Group on Epidemiological Markers, ESGEM) have undertaken and will always make great efforts in offering various educational and practical activities regarding high-throughput sequencing applications for a larger user community (https://www.escmid.org/profession_career/educational_activities/; https://www.ecdc.europa.eu/en/training). Future considerations will revolve around quality control and management requirements, external quality assessments and potential International Organization for Standardization (ISO) accreditation, which are essential but currently mainly lacking in the field of genomic sequencing for diagnostic and public health purposes [44].

Integrated data analysis and sharing of data – the changing position of national reference laboratories

With much more data generated at local and regional levels, the role of national reference laboratories is likely to expand to oversee long-term developments with respect to AMR and bacterial evolution. Rather than sending strains to be typed across the country, dedicated platforms are required that allow the – initially protected – upload of the raw data alongside the relevant metadata such as sampling date, place of isolation and clinical patient details. Active data and strain sharing therefore is essential and must be reinforced according to open science and FAIR principles (findable, accessible, interoperable, reproducible; https://www.openaire.eu/openaire-and-eosc; https://www.go-fair.org/fair-principles/).

The data then need to be processed to be accessible for a wider community to put local developments into context. The SARS-CoV-2 pandemic highlighted the power of this approach for addressing questions such as whether specific variants are locally restricted or form part of a larger transmission network. The answer to this question very much informs intervention strategies to manage their further spread. This will greatly enhance surveillance efforts if carried out in a sustainable manner. Well-advanced examples of interactive platforms already exist with, for instance, Pathogenwatch [45], BIGSdb [6] and Enterobase [46] focusing mainly on FWD and healthcare pathogens, ‘nextstrain’ built for viral surveillance [47] and ‘GenomeTrakr’/”GalaxyTrakr’ [48] targeting food-borne pathogens (https://nextstrain.org/; https://pathogen.watch/; https://www.fda.gov/food/whole-genome-sequencing-wgs-program/genometrakr-network).

Genome-based surveillance and data sharing for healthcare-associated bacterial pathogens will greatly benefit from SARS-CoV-2 genomic surveillance efforts put into place in many countries and regions around the world, complemented by informative visualization and imaging tools. Lastly, long-term data storage needs to be ensured, and may in part be solved by depositing the raw reads and associated metadata in public repositories to be used for broad population-based analyses.

Conclusion – ‘descriptive is good, predictive would be better’

In the last decade, financial, infrastructural and technical efforts have been made to build up genome-based, One-Health-oriented surveillance for FWD pathogens in several European and American countries and on a global scale (https://www.eurgen-reflabcap.eu) [49], whilst the SARS-CoV-2 pandemic provided the impetus for the generation, dissemination and analysis of worldwide genome-based surveillance over a time scale short enough to inform real-time political decision making.

Hospital pathogen surveillance goes far beyond simple outbreak analyses and has a much broader longitudinal dimension, building on patient networks and patient-to-patient transmission chains. Pathogen or AMR dissemination could go on for years if unnoticed or if the implemented infection prevention and control measures are less effective [50–52].

The focus of hospital pathogen surveillance from a public health perspective is to a large extent AMR surveillance, which needs to detect and distinguish both vertical and horizontal, real-time transmission of AMR determinants. This requires a broader adoption of long-read sequencing for more detailed characterization of plasmids and other MGEs and novel bioinformatics solutions to assemble genetic AMR structures of interest from WGS data.

Genes conferring resistance to last-resort antibiotics such as mcr-mediated colistin resistance in Enterobacterales and gene-mediated linezolid resistance show a strong link to and interconnection with sectors outside the hospital setting and, as such, their genomic surveillance requires a much wider, One-Health-oriented approach [53, 54].

SARS-CoV-2 genome-based surveillance combined with epidemiological modelling and novel visualization tools have already demonstrated the future of genomic pathogen surveillance. The challenge and benchmark will be to move from retrospective and descriptive studies to real-time and predictive pathogen surveillance that not only confirms epidemiological hypotheses, but acts as an early warning system by identifying possible hidden or unnoticed reservoirs and future, upcoming risks.