A High-Throughput Short Sequence Typing Scheme for Serratia marcescens Pure Culture and Environmental DNA

ABSTRACT

Molecular typing methods are used to characterize the relatedness between bacterial isolates involved in infections. These approaches rely mostly on discrete loci or whole-genome sequencing (WGS) analyses of pure cultures. On the other hand, their application to environmental DNA profiling to evaluate epidemiological relatedness among patients and environments has received less attention. We developed a specific, high-throughput short sequence typing (HiSST) method for the opportunistic human pathogen Serratia marcescens. Genes displaying the highest polymorphism were retrieved from the core genome of 60 S. marcescens strains. Bioinformatics analyses showed that use of only three loci (within bssA, gabR, and dhaM) distinguished strains with a high level of efficiency. This HiSST scheme was applied to an epidemiological survey of S. marcescens in a neonatal intensive care unit (NICU). In a first case study, a strain responsible for an outbreak in the NICU was found in a sink drain of this unit, by using HiSST scheme and confirmed by WGS. The HiSST scheme was also applied to environmental DNA extracted from sink-environment samples. Diversity of S. marcescens was modest, with 11, 6, and 4 different sequence types (ST) of gabR, bssA, and dhaM loci among 19 sink drains, respectively. Epidemiological relationships among sinks were inferred on the basis of pairwise comparisons of ST profiles. Further research aimed at relating ST distribution patterns to environmental features encompassing sink location, utilization, and microbial diversity is needed to improve the surveillance and management of opportunistic pathogens.

IMPORTANCE Serratia marcescens is an important opportunistic human pathogen, often multidrug resistant and involved in outbreaks of nosocomial infections in neonatal intensive care units. Here, we propose a quick and user-friendly method to select the best typing scheme for nosocomial outbreaks in relating environmental and clinical sources. This method, named high-throughput short sequence typing (HiSST), allows to distinguish strains and to explore the diversity profile of nonculturable S. marcescens. The application of HiSST profile analysis for environmental DNA offers new possibilities to track opportunistic pathogens, identify their origin, and relate their distribution pattern with environmental features encompassing sink location, utilization, and microbial diversity. Adaptation of the method to other opportunistic pathogens is expected to improve knowledge regarding their ecology, which is of significant interest for epidemiological risk assessment and elaborate outbreak mitigation strategies.

INTRODUCTION

Interactions between patients and the hospital environment have gained attention in epidemiological studies aimed at identifying origins of nosocomial outbreaks. For instance, sink environments are recognized as a source of opportunistic pathogens in several health care-associated infections (HAI) (1–3). In preventive or outbreak investigations, molecular typing methods are commonly used to examine the relatedness of environmental or clinical isolates. Many typing techniques are available to achieve this goal (4–10), mostly based on multilocus sequence typing (MLST) methodologies initially developed by Maiden et al. (11). Democratization of high-throughput sequencing technologies have contributed to expand public genome databases, providing an unprecedented portrait of microbial diversity. This has led to the realization that the pangenome of bacterial species displays a mosaic landscape supporting the metabolic flexibility necessary to ensure species resistance and resilience toward disturbances. Such plasticity of microbial genome highlights the need to update and revisit conventional MLST schemes, typically relying on housekeeping genes. In some instances, these genes are not specific enough for accurate molecular typing of investigated strains (12, 13). In addition to lack of the resolution, sometimes the loci employed in the MLST schemes have been affected by recombination and do not properly genotype the isolates in question (14).

The genus Serratia consists of Gram-negative bacteria classified as members of the Enterobacteriaceae that are ubiquitous in water, soil, plants and different hosts, including insects, humans and other vertebrates (15, 16). Among Serratia species, Serratia marcescens is the most important opportunistic human pathogen, often multidrug resistant and involved in outbreaks of HAI in neonatal intensive care units (NICUs) (17–27). No MLST scheme exists for the molecular typing of S. marcescens, but other typing techniques have been used during previous epidemiological studies, such as pulsed-field gel electrophoresis (17, 19, 23), ribotyping (28) or more recently whole-genome MLST (29, 30). Even though these techniques were proven efficient to distinguish strains, they are not tailored to epidemiological surveys involving a large sample size because they are technically demanding due to upstream cultivation and isolation efforts.

This study introduces a new molecular typing approach, which we called high-throughput short sequence typing (HiSST), to detect and identify S. marcescens relying on culture-dependent and culture-independent applications. The HiSST method was developed based on whole-genome sequences of S. marcescens available in public databases then validated with reference culture collections, clinical isolates and environmental DNA samples.

RESULTS AND DISCUSSION

Design of the HiSST scheme.

Thirty-two out of the 3,301 genes of S. marcescens pangenome were identified as the most variable with 29 to 30 alleles per gene. Only the most specific and discriminatory genes were kept, after stepwise alignment, examination of polymorphism among S. marcescens genomes and specificity check (Fig. 1). This led to the selection of 15 loci for subsequent analyses listed in Table S4. The minimal number of loci included in the HiSST scheme was selected by topology data analysis of concatenated HiSST loci and genome similarity. We found that the dendrogram built from three loci has a similar topology to the cognate clustering analysis comprising the 15 most discriminatory loci, according to the downward loci selection procedure. Indeed, Pearson's Chi-squared test with simulated P value shows a significantly similar cluster (P < 0.001) between HiSST SNPs and core genome MLST dendrograms (Fig. 2). Also, cophenetic correlation (0.55) and Goodman-Kruskal-gamma index (0.93) confirm that the two phylogenetic trees are similar, however not identical (Table S3). This led to the selection of the three loci located in genes gabR, bssA and dhaM for the HiSST scheme. PHI tests (57) show no significant recombination within the sequence set for each selected locus (P > 0.10), implemented with the PhiPack software and based on compatibility of parsimoniously informative sites (Table S3). Each locus was discriminant, with more than 20% of nucleotide dissimilarity between S. marcescens strains and other species (Fig. S1). The locus gabR is more specific to S. marcescens species than bssA, followed by dhaM, with, respectively, more than 26%, 17% and 14% of nucleotide dissimilarity with S. ficaria, the closest relative species of S. marcescens for these loci.

FIG 1.

Step-by-step approach of the method used to develop the S. marcescens HiSST scheme. (A) A pan-genome allele database was created with the web service tool PGAdb-builder. (B) The 30 most variable and discriminating loci were selected by sorting loci with the highest number of different alleles using Excel. (C) Selected loci were aligned with the software BioEdit, and short gene fragments (<350 bp) displaying the highest variability were kept. (D) Three successive steps were necessary to define the trade-off between the number of different loci and specificity of the assay by comparison of concatenated HiSST loci and genome similarity. (E) Finally, oligonucleotides comprising 18- to 22-mers with either a single or no substituted base were designed to target discriminant internal loci and tested with the software tool Primer-BLAST.

FIG 2.

Convergent classification of S. marcescens strains based on the HiSST scheme and core genome MLST. Complete linkage dendrograms based on maximum distance of concatenated loci selected for the HiSST scheme (A) and the core genome MLST difference to discriminate strains of Serratia marcescens (B). Label colors correspond to the optimal number of clusters for HiSST SNPs, according to Rousseeuw's Silhouette quality index. Labels in the shaded red box correspond to the cluster of four strains isolated in this study, suggesting that the environmental BD1b-2wD strain and clinical isolates (ED3957, ED3958, and ED3959) descend from a single cell.

The HiSST scheme based on these 3 loci differentiates most S. marcescens strains better than the average nucleotide identity (ANI) score based on their whole genomes (Fig. 3). The pairwise ANIb genome similarity score is over 94% in its ability to distinguish S. marcescens strains from other Serratia species while only a few strains of Serratia spp. have more than 70% nucleotide identity with the three selected loci of S. marcescens. Classification of S. marcescens strains based on the HiSST scheme is congruent with classification scheme relying on complete genome sequences (Fig. 2). Some strains sharing an ANI score of more than 99.9% are not properly clustered by the HiSST scheme compared to those with a complete genome. This difference is expected due to the overall limitation of molecular methods based on few marker genes. At such a high level of similarity, the threshold delineating species, strains or clones is empirical, depending on the examined species. For example, P. aeruginosa has high genomic plasticity mainly due to frequent horizontal gene transfers (58, 59), while S. marcescens has a higher genetic diversity at the sequence level according to PGAdb-builder results, with also genome flexibility (60). Additional factors to consider include the study context (e.g., the precautionary principle for epidemiological studies tends to identify highly similar but not identical strains as a nonclonal strain) and the method used (i.e., depending on the sensitivity of the molecular typing method and the sequencing platform used, the evolution of the technology, and knowledge). As a whole, the minimal similarity threshold among S. marcescens strains is 89% for the three loci (Fig. S2 and based on BLAST results).

FIG 3.

Discrimination of Serratia spp. based on the HiSST scheme and whole-genome sequences. The heat-map reports the ANIb score of the three concatenated loci of the HiSST scheme (A) and genome similarity (B). S. ficaria (n = 1), S. quinivorans (n = 2), S. proteamaculans (n = 1) and S. liquefaciens (n = 5) were included as the outgroup. The more cells turn from blue to red, the higher the ANIb score, and the more the strains are genotypically similar.

Validation and application of the HiSST scheme.

Specificity of primers targeting gabR, bssA and dhaM loci was first confirmed by BLAST searches against the RefSeq database of nonredundant protein sequences. The efficacy and the specificity of the PCR assays was further confirmed with reference strains (Fig. S3). A PCR amplicon of the correct size was observed for all S. marcescens strains (n = 15), but not for other Serratia species or other species of gammaproteobacteria. An accuracy test of the HiSST scheme, including the bioinformatic procedure utilized to assign alleles to SST, was realized with S. marcescens reference strains Db11 and Db10. Genomic DNA from both strains was subjected to PCR amplicon sequencing with an average allocation of 1,000 reads per library. The HiSST profile (sequence type 1 [ST 1]) of both strains corresponded to the expected profile with a single amplicon sequence variant (ASV) for each gene, supporting the accuracy of the HiSST procedure and parameters utilized in sequence quality control (Fig. 4).

FIG 4.

Minimum spanning trees based on SNP analysis of S. marcescens and eDNA, using S. marcescens Db10 as a reference. The distance labels represent the number of discriminating SNPs between neighboring genotypes. Each pie chart label refers to SST identifier of the corresponding locus. Reference genomes are represented in gray and isolates or sampled sinks are represented by the color legend in pie charts. Dominant SSTs of eDNA are represented with red font in the legend, whereas gray font corresponds to the SST of eDNA in low abundance.

The method was applied to two different case studies conducted in the same NICU. The first case study sought to compare the ST profile of a strain isolated from the sink-drain environment (BD1b-2wD) with three clinical strains (ED3657, ED3658, and ED3659) from patients admitted in that NICU, where an outbreak occurred. The infection prevention and control team determined their relatedness based on the PFGE profiles. Molecular typing of the environmental strain BD1b-2wD and the clinical strains revealed very close relatedness, with all four having an identical HiSST profile, ST 47 (Fig. 5A). WGS was done for each strain to challenge the HiSST scheme result. Pairwise comparison of contiguated genomes confirmed the high degree of similarity between each strain (ANIm > 99.7%). In principle, bssA and gabR are sufficient to ensure diversity coverage of ST represented in the genome database (Fig. 5B), but inclusion of dhaM in the HiSST scheme is included to prevent false-negative results (i.e., in the case where the targeted gene is absent or subject to unknown mutations) and allows the distinguishing of an environmental or clinical origin of S. marcescens strains for culture-based diagnostics (Fig. 5A). These results suggest that the environmental BD1b-2wD strain and clinical isolates descend from a single cell, while providing supplementary experimental evidence supporting the specificity of the HiSST scheme (Fig. 2).

FIG 5.

Relationship among the ST profiles of reference strains and isolates and diversity coverage of the HiSST scheme. (A) A minimum spanning tree based on MLST analysis of the HiSST scheme is represented with distance labels corresponding to the number of discriminating alleles and pie chart labels referring to the ST identifier of the HiSST scheme. Orange nodes correspond to clinical isolates, the red node to the isolate from the NICU sink drain, and the green nodes to environmental isolates. In the legend, strains represented by red font correspond to unknown clinical (ED3957, ED3958, and ED3959) and sink drain (BD1b-2wD) isolates from this study. (B) Cumulative frequency of ST depending on the number of loci included in the HiSST scheme.

The second case study was conducted to explore the diversity of S. marcescens by applying the HiSST method to environmental DNA (eDNA). PCR amplicon sequencing of each locus was done to report the diversity of each “locus short sequence type” number (locus-SST) separately for a culture-independent epidemiological investigation. All retrieved ASV sequences were specific to S. marcescens. Diversity among the 19 sinks was modest, with 11, 6 and 4 different alleles of gabR, bssA and dhaM found, respectively (Fig. 4). A single allele was dominant in each sample, with a relative abundance of 70–100% (Table S2). Either a single or two allele(s) per sample were observed for gabR and dhaM loci, whereas gabR was represented by up to three different alleles per sample. For the three HiSST loci, rare alleles differ from the dominant allele in the same sample by 1 to 6 SNPs, suggesting the presence of other strains in the drain. Artificial inflation in diversity caused by sequencing errors is less likely due to the stringent filtering process of sequences (cf. Materials and Methods) and the low error probability of incorrect base-call for short sequences (61). The intercomparison of ST profiles among the 19 sinks of the NICU was done to infer potential epidemiological links (Fig. 6). The most straightforward link between sink environments is the case where ST profiles are identical. This situation was observed in sinks #72 and #73 for dominant ASVs (bssA-SST 36, gabR-SST 18, and dhaM-SST 2) that are likely colonized by the same S. marcescens strain. This link is supported by the proximity of both sinks in the NICU, with the same drain connection and interconnection through handwashing (62, 63). The sink PLM shared two loci-SST detected in sinks #72 and #73 (bssA-SST 36 and dhaM-SST 2) and two loci-SST in sink #80 (gabR-SST 2 and dhaM-ST 2). This result suggests an epidemiological link between the four sinks related to one another by the sink PLM (that is used for the initial handwashing at the NICU entrance). Finally, the HiSST profile (ST 2) of sink #80 is identical to that of S. marcescens 95 and BWH-35 strains included in the reference genome database, suggesting the colonization by a taxonomically closely related strain. S. marcescens 95 and BWH-35 were isolated from sputum in a Boston hospital (Massachusetts, USA) and are most likely variants of the same strain.

FIG 6.

Survey of Serratia marcescens in the sink drains of a NICU. A scheme of the surveyed NICU is depicted (A) along with an UPGMA dendrogram (B) based on the Jaccard distance computed with the HiSST profiles of gabR, bssA and dhaM loci among sink drains that showed positive PCR amplifications. Sink drains sampled in this study are symbolized by a rectangle with their identification in black letters. Each sink color corresponds to a HiSST genotypic profile, as seen in the dendrogram (B). If the color is identical between two sinks, then there is a higher probability that these sinks are colonized with the same strain(s), which indicates a potential relationship between these sink drains.

A limitation of the method was noticed in sink #55 where no PCR detection of dhaM was observed with positive amplification of gabR and bssA genes. Although this can be explained by the low level of S. marcescens in this sink combined with different amplification efficiencies between the three reactions, examination of future genome sequences deposited in public databases will be necessary to confirm the ubiquitous distribution of dhaM in S. marcescens.

These case studies illustrate the strengths of the HiSST scheme to identify clones and its broad applicability for epidemiological investigations. Beyond the conventional application of the method to genotype isolates, examination of eDNA offers a complementary tool for the source tracking of opportunistic pathogens. This could be done by the monitoring of bacterial succession in the NICU environment and patient samples through HiSST eDNA profiling. Under that framework, a convergence of HiSST profiles along spatial or temporal sampling sequences would provide strong evidence of opportunistic pathogen transfer across different environments.

In contrast to conventional application for isolate identification, HiSST profile analysis from eDNA is less prone to misinterpretation or aborted analysis for samples displaying no signal for certain genes. Indeed, the pairwise comparison of HiSST bacterial profiles can be expressed as a pairwise Jaccard distance computed with presence or absence score for detected or nondetected loci-SST, respectively. Downstream clustering and multivariate analyses offer options to correlate loci-SST distribution patterns with environmental features encompassing sink location, utilization, and microbial diversity (Fig. 6). Although this approach is a gold standard in microbial ecology, the second case study presented in this article is the first culture-independent application of loci-SST profile analysis of opportunistic pathogens for epidemiologic survey.

In conclusion, a combination of in silico analyses led to the development of a powerful HiSST assay to identify isolates of S. marcescens strains. The approach relying on pangenome examination rather than selection of conventional housekeeping genes contributed to the method specificity. For instance, conventional MLST schemes for P. aeruginosa and S. maltophilia are less specific than the HiSST method developed here for S. marcescens. Application of the procedure presented in this article to these other opportunistic pathogens of environmental origin led to more robust HiSST schemes (T. Bourdin, E. Déziel, and P. Constant, unpublished data). Despite the precision of the method presented here, specificity and coverage of the HiSST scheme will require regular validation and updates with the addition of new genome sequences in public databases. The bioinformatic pipeline implemented here or alternative methods (64–66) will facilitate regular updates of the HiSST scheme. This fact holds true for any molecular classification tool. Even though comparison of whole genomes appears to be the most robust method (12), public genome databases contain contaminations that may introduce biases for the identification of highly similar strains (67). In addition, the high proportion of similar or identical genes in whole genomes hides some dissimilarities between isolates, while HiSST highlights the most discriminating alleles. On the other hand, HiSST and other molecular typing methods (based on PCR products) cannot achieve the accuracy of whole-genome sequencing. Thus, a combination of whole-genome sequencing and a high discriminatory molecular typing method is recommended for culture-dependent epidemiological investigation (68). Beside isolate identification, the HiSST method proved efficient for loci-SST comparison and source tracking purposes of S. marcescens in eDNA samples without the need for culture.

Based on these results, the following epidemiological interpretations for molecular typing of isolates when using the HiSST scheme are proposed: (i) isolates that are identified by at least 2 of 3 loci-SST are confirmed as S. marcescens, (ii) isolates with an identical ST (i.e., identical gabR-SST, bssA-SST and dhaM-SST) are most likely clones and belong to the same genotype, and (iii) isolates that differ by 2 or 3 loci-SST are mostly unrelated and do not belong to the same genotype. For an epidemiological survey on eDNA samples when using the HiSST scheme described here, the following interpretations are proposed: (i) eDNA samples with loci-SST corresponding to the HiSST scheme indicate the presence of S. marcescens, (ii) eDNA samples with several loci-SST of one locus indicate the presence of several S. marcescens strains, and (iii) samples with an identical ST (i.e., identical gabR-SST, bssA-SST and dhaM-SST) contain very closely related strains and the sampling environment is most likely linked.

MATERIALS AND METHODS

Development of the HiSST scheme.

A pan-genome allele database was assembled from 60 complete genomes of S. marcescens retrieved from the NCBI GenBank database (last updated in July 2020) with the Build_PGAdb module available on the online tool PGAdb-builder (31). Conserved genes showing the highest number of alleles were selected as the most variable and discriminant. Alleles of each selected genes (n = 32) were aligned, nonoverlapping ends were removed and the sequence identity matrix was computed with the software BioEdit (32). Gene fragments (<350 bp) displaying the highest variability were chosen and aligned against the NCBI database with the Basic Local Alignment Search Tool (BLAST) to assess specificity. The 15 loci (i.e., nucleotide sequences of internal fragments of the previously selected genes) showing the highest variability and with the most specific nonoverlapping ends were selected as candidates for the HiSST scheme. A trade-off between the number of different loci and specificity of the assay was achieved by qualitative visual comparison and tree topology tests of concatenated HiSST loci and genome similarity (see below). Three successive steps were necessary to implement the approach relying on the 60 complete genomes of S. marcescens (Table S1) and 9 other strains of non-marcescens Serratia spp. available in GenBank. First, ANIb (33) analyses were performed on the 69 complete genomes with the BLAST+ alignment tool (https://ftp.ncbi.nlm.nih.gov/blast/executables/blast+/LATEST/) with the Python package pyani (https://github.com/widdowquinn/pyani) (34). Second, a stepwise approach was implemented to assemble concatenated loci alignments. A backward selection procedure was applied on the 15 most discriminant loci until only three loci remained. This led to multiple concatenated alignments comprising either 15, 7, 4 or 3 gene fragments. ANIb scores were calculated on the concatenated alignments of loci. Third, the discriminating power of the four different HiSST schemes was validated by topology data analysis of concatenated loci and complete genome trees. The topology of the different complete linkage dendrograms, based on maximum distance (i.e., supremum norm), were compared with R version 4.0.4 (35) using the packages pvclust (36), dendextend (37), vegan (38) and tidyverse (39). Topological distance was measured by Robinson and Foulds distance, and then tree topology tests were performed by calculating cophenetic correlation and Goodman-Kruskal-gamma index. A contingency table of the two topologies was created using k-mean partitioning approach, repeated 1,000 times. Thus, a Pearson's Chi-squared test was performed to verify the relationship between the two dendrograms, as described in Borcard et al. (40).

Further, ANIb values of the selected HiSST loci and complete genomes were compared using the packages circlize (41) and ComplexHeatmap (42), which help to visualize the difference of the discriminatory power between the HiSST scheme and the whole genome of S. marcescens. The R scripts we developed are available on GitHub (https://github.com/TBourd/R_scripts_for_HiSST_scheme). Following these analyses, the three gene fragments of gabR (HTH-type transcriptional regulatory protein), bssA (benzylsuccinate synthase alpha subunit) and dhaM (PTS-dependent dihydroxyacetone kinase, phosphotransferase subunit) were retained for the further development of the HiSST scheme, as described below.

Primer design and PCR amplification of gabR, bssA and dhaM internal loci.

Oligonucleotides comprising 18- to 22-mers with either a single or no substituted base were designed to target discriminant internal loci in the gabR, bssA and dhaM genes (Table 1). In silico tests of primers were performed with the software tool “Primer-BLAST” (43), using the RefSeq nonredundant proteins database to assess specificity and the Serratia marcescens subset RefSeq database to verify the coverage of primers for this species. The reaction was carried out in 25 μl of master mix containing 2.5 U/μl Fast-Taq DNA polymerase (Bio Basic Inc., Markham, Ontario, Canada), 1× of Fast-Taq buffer (Bio Basic Inc., Markham, Ontario, Canada), 200 μM dNTPs, 0.4 mg/ml bovine serum albumin (BSA), 0.4 μM each primer, and 2 ng/μl extracted DNA. A solution of 0.5× band sharpener (Bio Basic Inc., Markham, Canada) was included for the gabR mixture only. PCR conditions were optimized for each primer set with genomic DNA of S. marcescens strains as the template (Table 1).

TABLE 1.

HiSST locus specific primers sequences and PCR cycle conditions^a

Locus	Primer sequence (5′-3′)	PCR amplicon length (bp)	PCR cycle conditions
gabR	Forward: GAGCATCTGCGYAATATGCG	318	Initial denaturation at 95°C for 5 min, followed by 40 cycles at 95°C for 20 s, 58°C for 40 s, 72°C for 30 s and a final extension period of 5 min at 72°C.
	Reverse: CAGCGCGYTGAACACCTG
bssA	Forward: CGCAGTTTCTCAACGCYATCG	242	Initial denaturation at 95°C for 5 min, followed by 35 cycles at 95°C for 20 s, 58°C for 40 s, 72°C for 30 s and a final extension period of 5 min at 72°C.
	Reverse: CGAATGGCCGTTGGATTCGATC
dhaM	Forward: GGCGTCCAGCATYGCCTT	279	Initial denaturation at 95°C for 5 min, followed by 35 cycles at 95°C for 20 s, 60°C for 40 s, 72°C for 30 s and a final extension period of 5 min at 72°C.
	Reverse: GACGTGCGCGACATGCTG

^aIllumina linker sequences were added at each 3’-end sequence of primers: 5′-TCGTCGGCAGCGTCAGATGTGTATAAGAGACAG-3′ for forward and 5′-GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAG-3′ for reverse primers.

Validation of the HiSST scheme with reference strains.

Validation of primers was done with 28 reference strains of various origins (Table 2). Selected strains comprised S. marcescens (n = 15), Serratia rubidaea (n =1), Serratia liquefaciens (n = 1), Serratia plymuthica (n = 1), Pseudomonas aeruginosa (n = 3), Klebsiella pneumoniae (n = 1), Stenotrophomonas maltophilia (n = 4), Stenotrophomonas acidaminiphila (n = 1) and Stenotrophomonas nitritireducens (n = 1). The strains were purified on Trypticase soy broth (TSB) (Difco Laboratories, Sparks, MD, USA) with agar (15 g/liter) (Alpha Biosciences, Inc., Baltimore, MD, USA) at 30°C for 48 h. A single colony of each strain was inoculated in 2 ml TSB and grown for 48 h at 30°C for subsequent genomic DNA extraction.

TABLE 2.

Reference strains utilized as a positive or negative control for HiSST scheme validation^h

Species and strains (no. of strains or additional designation)	Lab collection no.	Isolation origin (country)	Provided by
Serratia marcescens (n = 15)
PCI 1107 (ATCC 14756, LMG 13576)	ED3691	Fort Detrick, Maryland (USA)	BCCM-LMGc
BD1b-2wD	ED4305	Drain water from a NICU (Canada)	Lab strain
Db11	ED3837	Insect isolate, Drosophila melanogaster (France)	A. Brassingaa ; J. J. Ewbankb
Db10	ED3838	Insect isolate, Drosophila melanogaster (France)	A. Brassingaa ; J. J. Ewbankb
BS 303 (ATCC 13880, LMG 2792)	ED3696	Pond water (Czech Republic)	BCCM-LMGc
L00128734	ED3957	Human clinical specimen (Canada)	LSPQd
L00128736	ED3958	Human clinical specimen (Canada)	LSPQd
L00128737	ED3959	Human clinical specimen (Canada)	LSPQd
L00128966	ED3960	Human clinical specimen (Canada)	LSPQd
L00128967	ED3961	Human clinical specimen (Canada)	LSPQd
L00129585	ED3962	Human clinical specimen (Canada)	LSPQd
L00130169	ED3963	Human clinical specimen (Canada)	LSPQd
L00133794	ED3964	Human clinical specimen (Canada)	LSPQd
L00134617	ED3965	Human clinical specimen (Canada)	LSPQd
L00085643	ED3966	Environmental (Canada)	LSPQd
Serratia rubidaea (n = 1)
FB299	ED3693	Environmental (USA)	Roger Laurent Bernieri
Serratia liquefaciens (n = 1)
ID150497	ED3967	Human clinical specimen (Canada)	LSPQd
Serratia plymuthica (n = 1)
ID157970	ED3968	Human clinical specimen (Canada)	LSPQd
Stenotrophomonas maltophilia (n = 4)
560 (ATCC 13636, LMG 961, NCTC 10258)	ED3699	Human, cerebrospinal fluid (USA)	BCCM-LMGc
L00083595	ED3969	Human clinical specimen (Canada)	LSPQd
L00092250	ED3970	Human clinical specimen (Canada)	LSPQd
L00124341	ED3971	Human clinical specimen (Canada)	LSPQd
Stenotrophomonas acidaminiphila (n = 1)
L00129488	ED3979	Human clinical specimen (Canada)	LSPQd
Stenotrophomonas nitritireducens (n = 1)
ATCC BAA-12 (LMG 22074, DSM 12575)	ED3701	Laboratory scale biofilter (Germany)	BCCM-LMGc
Pseudomonas aeruginosa (n = 3)
UCBPP-PA14	ED1	Human clinical specimen (USA)	69
PAO1	ED956	Human, wound (Australia)	Sylvie Chevaliere
FKS4A	ED0129	Human, cystic fibrosis (USA)	Luke Hoffmanf
Klebsiella pneumoniae (n = 1)
ATCC 4352 (LMG 3128)	ED3692	Cow's milk	ATCCg

^aAnn Brassinga, University of Manitoba, Winnipeg, MB, Canada.

^bJonathan J. Ewbank, University of Aix-Marseille, Marseille, France.

^cBelgian Coordinated Collections of Microorganisms, University of Ghent, Ghent, Belgium.

^dLaboratoire de Santé Publique du Québec, Sainte-Anne-de-Bellevue, Québec City, Québec, Canada.

^eSylvie Chevalier, University of Rouen-Normandie, Rouen, France.

^fLuke Hoffman, Seattle Children's Hospital, Seattle, WA, USA.

^gAmerican Type Culture Collection, Rockville, MD, USA.

^hTen strains of Serratia spp. were included to verify the specificity for S. marcescens of each selected locus, with S. ficaria (n = 1), S. liquefaciens (n = 5), S. quinivorans (n = 2), and S. proteamaculans (n = 2) also downloaded from the NCBI GenBank database. These Serratia spp. have the most similar nucleotide sequences of the selected locus according to the results of the BLAST run.

ⁱGift of Roger Laurent Bernier (Burlington, CA, USA).

Validation of the HiSST scheme with environmental DNA.

Biofilm and 50 ml of water from 10 different sink drains were sampled in April 2019 during an outbreak of S. marcescens in a neonatal intensive care unit (NICU) in a Montreal Hospital (Québec, Canada). The same day, samples were inoculated on a semiselective DNase test agar (44) supplemented with ampicillin (5 μg/ml), colistin (5 μg/ml), cephalothin (10 μg/ml), and amphotericin B (2.5 μg/ml) incubated for 48 h at 30°C. Colonies were purified on TSB with agar at 30°C for 48 h. During a second sampling campaign, biofilm and water (50 ml) samples from sink drains were collected twice from 19 sinks in January 2020. Samples were kept on ice during their transportation to the laboratory. Genomic DNA from isolated strains and environmental samples was extracted by a procedure combining mechanical and chemical lysis, using bead beater and ammonium acetate treatment, as previously described (45), prior to PCR amplicon sequencing. The two successive PCR amplifications necessary for the preparation of gabR, bssA and dhaM sequencing libraries were conducted with the AccuPrime Taq DNA polymerase system, high fidelity (Invitrogen Ltd., Carlsbad, CA, USA). PCR conditions and reaction mixtures were adapted following manufacturer instructions (Table 1). The first PCR was performed using modified gabR, bssA and dhaM primers, including Illumina linker sequences (Table 1) and 2 ng/μl of template DNA. PCR products were purified with AMPure XP beads (Beckman Coulter Inc., Brea, CA, USA). Purified PCR products were subjected to a second PCR performed for library preparation using barcoded primers (Table S2) supplied by Integrated DNA Technologies Inc. (Mississauga, Ontario, Canada). Purified PCR amplicons were quantified using the Quant-iT PicoGreen dsDNA assay kit (Invitrogen Ltd., Carlsbad, CA, USA), diluted and pooled into 75 μl comprising a final concentration of 1.5 ng/μl of DNA before shipping for sequencing. PCR amplicons were sequenced with the Illumina MiSeq PE-250 platform at the Centre d'Expertise et de Services Génome Québec (Montréal, Québec, Canada). Raw sequencing read processing included primer sequence removal with the software Cutadapt v2.10 (46), followed by quality control, paired ends merging and chimera check using the default parameters specified in the package dada2 v1.8.0 (47) that include packages ShortRead v1.48.0 (48) and Biostrings v2.58.0 (49). Reads containing a mismatch in the primer region were deleted (R script available at https://github.com/TBourd/R_scripts_for_HiSST_scheme). Filtered sequences were clustered into amplicon sequence variants (ASVs) displaying 100% identity. ST assignation of chimera-free ASV was done using gabR, bssA and dhaM reference databases with a 100% identity cutoff (Table S1). The proportion of reads remaining after each step of the bioinformatics pipeline is provided in Table S2.

SNP and HiSST profile analyses.

SNPs of each locus were analyzed from all unique nucleotide sequences for references strains and environmental DNA (Table S3). The SNP matrix and HiSST profile were analyzed by the geoBURST algorithm using the PHYLOViZ platform v2.0a (50), creating minimum spanning trees using default software settings. The coverage of the HiSST scheme was visualized in a chart representing the cumulative frequency of ST depending on the number of cumulative loci, based on 60 reference strains of S. marcescens (Table S1).

Validation of molecular typing by whole-genome sequencing (WGS).

Environmental strain BD1b-2wD and clinical strains ED3957, ED3958, and ED3959 were subjected to whole-genome sequencing (WGS) with the Illumina NextSeq 550 platform at the Microbial Genome Sequencing Center (Pittsburgh, PA, USA). A quality control of the WGS data was checked with FASTQC (https://www.bioinformatics.babraham.ac.uk/projects/fastqc). Illumina adapter clipping and quality trimming were performed with Trimmomatic v0.39 (51), by specifying an average quality required greater than 30. Then, genomes were assembled from a FASTQ file of paired-end reads using SPAdes de novo assembler (52) and Bandage (53) to visualize SPAdes graphical output. Contigs obtained from SPAdes output were aligned, ordered and oriented with the closest reference genome (S. marcescens AR_0122) to create a contiguated genome by using the tool ABACAS (54). Pairwise comparison of contiguated genomes of DB1b-2wD, ED3957, ED3958 and ED3959 was performed using ANIm scores (calculation of ANI based on the MUMmer algorithm, which is more adapted to comparing genomes with a high degree of similarity) (55, 56).

HiSST nomenclature and assignation.

The following nomenclature is used to identify HiSST loci and HiSST profiles: a “locus short sequence type” number (locus-SST) is assigned for each ASV of the individual loci, and the combination of multilocus SST of the overall HiSST profile is defined by a “sequence type” (ST) number. The HiSST identification of an isolate is performed with an R script, and new HiSST profiles for unknown isolates are added to the HiSST database using the same script. The HiSST scheme database and the R script called “HiSST-Assignation” are available on GitHub at https://github.com/TBourd/R_scripts_for_HiSST_scheme.

Data availability.

Raw sequencing reads have been deposited in the NCBI Sequence Read Archive under BioProject no. PRJNA729113. Raw sequencing reads of isolates are deposited under BioSample no. SAMN19117128 to SAMN19117139, and eDNA raw sequencing reads are under BioSample no. SAMN19110658 to SAMN19110711. Assembled genomes of the environmental strain BD1b-2wD and clinical strains ED3957, ED3958, and ED3959 have been deposited under the same BioProject accession no., PRJNA729113, and under BioSample accession no. SAMN19232018 to SAMN19232021.