A pilot study of rapid whole-genome sequencing for the investigation of a Legionella outbreak.

OBJECTIVES: Epidemiological investigations of Legionnaires' disease outbreaks rely on the rapid identification and typing of clinical and environmental Legionella isolates in order to identify and control the source of infection. Rapid bacterial whole-genome sequencing (WGS) is an emerging technology that has the potential to rapidly discriminate outbreak from non-outbreak isolates in a clinically relevant time frame.
METHODS: We performed a pilot study to determine the feasibility of using bacterial WGS to differentiate outbreak from non-outbreak isolates collected during an outbreak of Legionnaires' disease. Seven Legionella isolates (three clinical and four environmental) were obtained from the reference laboratory and sequenced using the Illumina MiSeq platform at Addenbrooke's Hospital, Cambridge. Bioinformatic analysis was performed blinded to the epidemiological data at the Wellcome Trust Sanger Institute.
RESULTS: We were able to distinguish outbreak from non-outbreak isolates using bacterial WGS, and to confirm the probable environmental source. Our analysis also highlighted constraints, which were the small number of Legionella pneumophila isolates available for sequencing, and the limited number of published genomes for comparison.
CONCLUSIONS: We have demonstrated the feasibility of using rapid WGS to investigate an outbreak of Legionnaires' disease. Future work includes building larger genomic databases of L pneumophila from both clinical and environmental sources, developing automated data interpretation software, and conducting a cost-benefit analysis of WGS versus current typing methods.

Epidemiological investigations of Legionnaires’ disease outbreaks rely on the rapid identification and typing of clinical and environmental Legionella pneumophila isolates in order to identify and control the source of infection.

Rapid bacterial whole genome sequencing (WGS) is an emerging technology that has the ability to identify and discriminate bacterial isolates.

We hypothesised that WGS could be used to discriminate outbreak from non-outbreak Legionella isolates in a clinically relevant time frame.

We retrospectively applied bacterial WGS to isolates cultured during a previous outbreak investigation, and were able to rapidly distinguish outbreak from non-outbreak isolates, and to identify the probable environmental source.

Our findings were consistent with those of previous epidemiological and microbiological investigations of the same outbreak.

This raises the possibility of conducting combined epidemiological and genomic outbreak investigations in real time.

We have demonstrated the feasibility of using rapid WGS to investigate an outbreak of Legionnaires’ disease.

Our study was limited by the small number of L pneumophila genomes available for comparison.

Future work includes the development of automated data interpretation software and a cost–benefit analysis of current typing methods compared with WGS.

Background

Legionella pneumophila causes outbreaks of respiratory infection in community settings and results in significant morbidity and mortality.1 The organism is common in aquatic environments and is spread by aerosol from a contaminated source, often cooling towers and other aerosol-producing devices. Nosocomial outbreaks that are related to contaminated water supplies have also been widely reported.2–4 The diagnosis of Legionnaires’ disease (LD) is based on a compatible clinical syndrome and detection of L pneumophila urinary antigen5 or isolation of the organism from respiratory specimens, which requires culture on selective media.6 Most cases of human infection are caused by L pneumophila serogroup 1. During Legionella outbreaks, clinical and environmental isolates are collected and sent to the reference laboratory for typing.7 Epidemiological investigations are dependent on the rapid identification and typing of the associated organisms in order to identify and control the source of infection. Current typing methods include phenotypic (monoclonal antibody subgrouping8) and genotypic (sequence-based typing9) methods, which typically take 1–2 days. High-throughput sequencing technology has the potential to rapidly provide information on organism identity and genetic relatedness and has been shown to provide a high degree of discrimination for a range of other bacteria such as methicillin-resistant Staphylococcus aureus,10 Mycobacterium tuberculosis,11 Escherichia coli 0104:H412 and Klebsiella pneumoniae.13 We hypothesised that WGS could be used to discriminate outbreak from non-outbreak isolates of L pneumophila in a comparable time frame, and with a higher level of discrimination, when compared with current typing methods. Therefore, we conducted a pilot study to determine the feasibility of using a rapid bench-top sequencing platform (Illumina MiSeq) to retrospectively investigate a Legionella outbreak.

Design

Objectives

The aim of this pilot study was to determine the feasibility of using bacterial WGS for the investigation of a previous Legionella outbreak.

Epidemiological and microbiological investigation

In 2003, an outbreak of LD occurred in Hereford, UK.14 The outbreak started with two community cases that presented with clinical features of infection within a few days of each other, one of whom died. Active case-finding identified two further cases in the local hospital and a formal outbreak investigation was carried out. Twenty-four further cases of LD were identified over the next three weeks. All cases had a positive L pneumophila urinary antigen test, and three patients’ samples were culture-positive for L pneumophila serogroup 1. Epidemiological and environmental investigations were undertaken to determine possible sources. A total of 142 environmental samples were collected from potential sources, which included 50 cooling towers on 11 premises. L pneumophila serogroup 1 was isolated from samples collected at three cooling towers at two different locations (sites A and B) and a domestic spa pool Clinical and environmental isolates were referred to the Respiratory and Systemic Infection Laboratory, Health Protection Agency, London, for L pneumophila monoclonal antibody (mAb) subgrouping followed by a three-allele DNA-sequence-based typing (SBT3) method then in use. The SBT3 profiles for two of the clinical isolates and isolates from two of the cooling towers were indistinguishable, suggesting that the cooling towers were the likely environmental source. The strains were subsequently re-examined using the current seven-allele SBT method,15 with the same outcome.

DNA extraction and whole genome sequencing

Seven L pneumophila isolates (three clinical and four environmental) were obtained from the reference laboratory where they had been stored at −80°C with minimal passage since the outbreak. DNA was extracted from each L pneumophila isolate (50 ng) and prepared for sequencing using the Nextera DNA Sample Prep Kit (Epicentre). Samples were pooled together and then run on a rapid whole-genome sequencing platform (Illumina MiSeq) at Addenbrooke's Hospital, Cambridge, generating 150 bp paired-end reads.

Bioinformatic analysis

Bioinformatic analysis was performed at the Wellcome Trust Sanger Institute and blinded to the epidemiological data. The sequencing data from the seven samples were mapped to a reference genome, L pneumophila-type strain Philadelphia-1,16 and compared with eight other publicly available L pneumophila genomes (table 1). Sequence reads were mapped onto the reference genome using the SMALT software programme. Regions containing phage or insertion sequence elements were excluded from the analysis. Single nucleotide polymorphisms (SNPs) were identified using a standard approach,17 by removing SNPs with low-quality scores and by filtering for SNPs that were present in at least 75% of the mapped reads. The minimum number of high-quality reads mapping to call a base was set to four, which is equivalent to a minimum coverage of four. Actual coverage ranged between 20× and 100× per isolate. A maximum likelihood phylogeny was estimated using the RA×ML software programme. The general time-reversible model with γ correction was used for among-site variation. Tandem repeats were not considered in the original analysis, although we did re-run the analysis excluding the 23 repetitive genes mentioned in the paper by Coilet al;18 the overall topology of the phylogenetic tree remained unchanged and would not have affected the interpretation of our data.

Table 1

Clinical, environmental and reference L pneumophila strains

Sample number	Accession number	Biological origin	Type of sample	Serogroup	Monoclonal antibody subgroup	Sequence type*
Reference genome
LP Philadelphia	AE017354.1	USA 1974	Clinical	1	Philadelphia	ST36
Published genomes
LP ATCC 43290	CP003192.1	USA	Clinical	12	NA	ST187
LP Alcoy	CP001828.1	Spain	Clinical	1	ND	ST578
LP Corby	CP000675.2	UK	Clinical	1	Knoxville	ST51
LP Lens	CR628337.1	France	Clinical	1	Benidorm	ST15
LP 130b	FR687201.1	USA	Clinical	1	Benidorm	ST42
LP Paris	CR628336.1	France	Clinical	1	Philadelphia	ST1
LP Lorraine	FQ958210.1	France	Clinical	1	ND	ST47
LPHL06041035	FQ958211.1	France	Environmental	1	ND	ST734
Outbreak investigation isolates
LP033	ERS166051	Patient 1	Clinical	1	Philadelphia	ST37
LP035	ERS166045	Patient 2	Clinical	1	Philadelphia	ST37
LP617	ERS166047	Patient 3	Clinical	1	Allentown/France	ST47
LP056	ERS166052	Site A cooling tower 1	Environmental	1	Philadelphia	ST37
LP427	ERS166050	Site A cooling tower 2	Environmental	1	Philadelphia	ST37
LP467	ERS166049	Domestic spa pool	Environmental	1	Philadelphia	ST37
LP423	ERS166048	Site B cooling tower 1	Environmental	1	Oxford/OLDA	ST1

*Sequence type was derived from the genome sequence data and was concordant with the results of the seven-allele sequence-based typing method.

NA, Not applicable; ND, not determined.

Results

Phenotypic and typing results

The microbiological characteristics of the L pneumophila isolates, included in this study, are summarised in table 1.

Genomic analysis

Whole genome phylogenetic analysis showed that two clinical isolates (LP033 and LP035) and three environmental isolates (LP056, LP427 and LP467) were closely related genetically, and accordingly clustered together on the tree (figure 1A). These five isolates were therefore considered to be the outbreak isolates, though it was not possible to obtain directional information from this analysis owing to the low number of SNPs differentiating isolates; in total, there were less than 15 SNP differences within the outbreak strain cluster (figure 1B). Furthermore, the genetic variability between isolates from two cooling tower isolates on site A, and the observation that these intermingled with the clinical isolates on the tree, suggested that some diversity existed in the source population before the onset of the outbreak. Sequence types were derived from the genome sequence data and confirmed that all five isolates were ST37.

Figure 1

Phylogenetic tree of Legionella pneumophila strains. (A) Phylogeny of the species L pneumophila. Clinical, environmental and reference isolates are shown in red, blue and black, respectively. Inset (B) close-up phylogeny of the isolates involved in the outbreak. The branch leading to the reference strain Philadelphia has been truncated for clarity.

The two remaining isolates (LP423 and LP617) were situated ∼75 000 to 77 500 SNPs, respectively, from the outbreak cluster, and thus were not considered to be part of the outbreak. Sequence types were derived from the genomic data and the clinical isolate (LP617) was ST47 whereas the environmental isolate (LP423) was ST1.

The five outbreak isolates were compared to the nine published strains and found to be most closely related to the Philadelphia-1 strain (which is ST36, a single locus variant of ST37) and to the ATCC 43 290 strain (which is ST187) (figure 1A). Both of these isolates were ∼10 000 to 13 000 SNPs distant from the outbreak cluster. The LP617 isolate was 56 SNPs different from Lorraine strain (also ST47), and the LP423 isolate was 906 SNPs different from the Paris strain (also ST1).

Comparison of epidemiological investigation and genomic analysis

Two clinical isolates (LP033 and LP035) had been obtained from patients included in the outbreak. Both strains were located within the outbreak cluster in the phylogenetic tree. The third clinical isolate (LP617) was obtained from a patient who had initially been linked to the outbreak. The original epidemiological investigation found, however, that this patient was a lorry driver, who had passed through Hereford at the time of the outbreak, and had likely acquired his infection elsewhere. This isolate was located distant to the outbreak cluster on the phylogenetic tree, and was therefore not considered to be linked to the outbreak. Thus, for the clinical isolates, the genomic data supported the results of the previous epidemiological investigation.

Three environmental isolates were located within the outbreak cluster. Two of these (LP056 and LP427) had been collected from two cooling towers at the same location (Site A) while the third environmental isolate (LP467) had been collected from a spa pool in local domestic premises. Given the small number of SNP differences between these three isolates (figure 1B), it was not possible to determine which of these isolates represented the source of the outbreak using genomic data alone. The original epidemiological investigation had, however, concluded that the cooling towers on site A were the most likely source.

The fourth environmental isolate (LP423) was obtained from a cooling tower at a different site (site B), which was considered epidemiologically unlikely to be the source of the outbreak; a view supported by the typing data. This isolate was located away from the outbreak cluster and was most closely related (906 SNPs different) to the Paris strain (figure 1A).

Comparison of conventional typing and genomic analysis

We also compared the results of the conventional typing (monoclonal antibody typing and sequence-based typing) with WGS. All of the isolates included in this analysis were L pneumophila serogroup 1, apart from the ATCC 43 290 strain, which was serogroup 12. All of the outbreak strains belonged to the mAb subgroup ‘Philadelphia’, and were ST37. The clinical non-outbreak isolate belonged to the mAb subgroup ‘Allentown/France’ and was ST47, whereas the environmental non-outbreak isolate belonged to the mAb subgroup ‘Oxford/OLDA’ and was ST1. Thus, in this outbreak, the performance of WGS sequence was equivalent to conventional SBT in differentiating the outbreak from the non-outbreak strains. WGS was unable to distinguish the epidemiologically most likely source of the outbreak (site A cooling towers) from the domestic spa pool.

Discussion

Here, we have demonstrated the feasibility of using WGS to perform an investigation of a Legionella outbreak. Using genomic analysis, we were readily able to distinguish outbreak from non-outbreak Legionella isolates, and to identify probable environmental sources, thus supporting the findings of the previous epidemiological investigation. The main advantage of WGS over other typing techniques such as monoclonal antibody typing,8 amplified fragment length polymorphism,19 pulsed-field gel electrophoresis,3 and sequence-based typing9 is that it interrogates the whole genome, thus giving maximum resolution, even within individual sequence types. Current barriers to routine implementation of WGS include the inability to sequence directly from clinical specimens, the lack of availability of comprehensive open-access genomic databases to compare isolates to, the lack of automated data interpretation software to deliver clinically relevant information and the need for cost-benefit analyses of WGS versus the current typing methods.

We acknowledge several limitations to our study. The study was performed retrospectively and was hampered by the small number of stored L pneumophila isolates available for WGS. In the original investigation, we examined multiple isolates from each environmental sample to confirm their phenotype (species, serogroup and monoclonal antibody subgroup). Each sample (and source) contained a single phenotype—hence only a single colony for each sample was characterised genotypically and archived for later use. For the clinical samples, five colonies were taken from each positive patient sample and characterised phenotypically. Again, only a single phenotype was identified in each patient and hence only a single colony from each was characterised genotypically. This issue remains a challenge for contemporaneous outbreak investigations for two reasons. First, the diagnosis of LD is usually made by the detection of L pneumophila urinary antigen, and is often not confirmed by culture of the organism from clinical specimens, which takes 2–3 days. Second, environmental samples can take even longer to culture than clinical specimens, and are usually not processed in the same laboratory. Thus, the number of clinical and environmental samples available for typing from Legionella outbreaks is likely to be limited.

Our analysis was also constrained by the limited available information on the genetic variation and population structure of L pneumophila at the whole genome level. Environmental and clinical isolates are not evenly distributed in the environment, based on sequence-based typing observations, suggesting that clinical isolates are a distinct subpopulation of environmental strains. Humans are continuously exposed to environmental Legionellae and it is not clear why certain sequence types predominate in human disease.20 One hypothesis is that disease only occurs in those who have increased susceptibility to infection, for example, the elderly, and the immunosuppressed.21 Whenever a Legionella outbreak occurs, it usually reflects the breakdown of Legionella control measures, with human infections occurring as a consequence.

The genetic diversity of Legionella strains within an environmental source, as seen in this analysis, could potentially undermine our ability to link environmental and clinical isolates in an outbreak situation. Thus, a detailed epidemiological investigation accompanied by thorough environmental sampling, sequencing and comparison with patient isolates will continue to be required to confirm the likely source of an outbreak.

Despite these caveats, our work here demonstrates that this WGS approach can provide highly discriminatory information within a clinically relevant time frame, but requires a parallel epidemiological investigation to rule in or rule out potential environmental sources. This heralds the opportunity of conducting combined epidemiological and genomic outbreak investigations in real-time, as has been performed for other pathogens.18