ABSTRACT
Listeria monocytogenes causes foodborne outbreaks with high mortality. For improvement of outbreak cluster detection, the German consiliary laboratory for listeriosis implemented whole-genome sequencing (WGS) in 2015. A total of 424 human L. monocytogenes isolates collected in 2007 to 2017 were subjected to WGS and core-genome multilocus sequence typing (cgMLST). cgMLST grouped the isolates into 38 complexes, reflecting 4 known and 34 unknown disease clusters. Most of these complexes were confirmed by single nucleotide polymorphism (SNP) calling, but some were further differentiated. Interestingly, several cgMLST cluster types were further subtyped by pulsed-field gel electrophoresis, partly due to phage insertions in the accessory genome. Our results highlight the usefulness of cgMLST for routine cluster detection but also show that cgMLST complexes require validation by methods providing higher typing resolution. Twelve cgMLST clusters included recent cases, suggesting activity of the source. Therefore, the cgMLST nomenclature data presented here may support future public health actions.
INTRODUCTION
Listeria monocytogenes is a zoonotic pathogen causing disease in humans, with symptoms ranging from gastroenteritis to septicemia, meningoencephalitis, and abortion in pregnant females. Pathogen transmission occurs mainly through food, such as fish, meat, and dairy products but also contaminated fruits and vegetables (1, 2). Although the numbers of Listeria infections in Europe are low compared to those associated with other zoonotic gastrointestinal pathogens, case numbers have been increasing substantially during recent years (3). Moreover, disease fatality rates ranging from 6 to ∼30% are among the highest of all foodborne infectious diseases (4–6).
L. monocytogenes causes large and protracted outbreaks, many of which cannot be traced to a specific source. Therefore, timely typing of the pathogen is essential to detect disease clusters and incriminated sources as early as possible. Several methods are employed for subtyping of L. monocytogenes. Serotyping of somatic and flagellar antigens was used previously (7) but has been replaced by PCR-based molecular serotyping (8, 9). Typing resolution is further increased by multilocus sequence typing (MLST) using a scheme based on seven housekeeping genes (10). As the method of choice for further strain differentiation, genome restriction by two different restriction enzymes (AscI/ApaI) and fragment separation by pulsed-field gel electrophoresis (PFGE) have been used (11, 12). Recently, whole-genome sequencing (WGS) combined with core-genome MLST (cgMLST), whole-genome MLST, or single nucleotide polymorphism (SNP) calling has been introduced, further facilitating strain discrimination as well as data comparability between laboratories (13–17).
One of the cgMLST schemes developed uses a core genome set of 1,701 loci present in the majority of L. monocytogenes isolates (14). Based on allelic similarity (i.e., number of identical versus the number of different alleles), this method allowed clustering of outbreak isolates and separation of unrelated strains. Importantly, the analysis scheme provides an automatically curated cgMLST nomenclature facilitating communication between laboratories (14). The cgMLST approach was first described for investigation of a cluster of listeriosis cases occurring in Austria and Germany between 2011 and 2013. Here, serovar 1/2b L. monocytogenes isolates from human cases and food isolates were indistinguishable after PFGE analysis; however, cgMLST identified two separate disease clusters, showing the usefulness and the higher discriminatory power of the cgMLST approach (18).
In our study, we compared molecular serotyping, PFGE, 1,701-locus cgMLST, and SNP calling using a set of more than 400 recent human isolates collected in Germany. We present 34 unrecognized disease clusters and the associated cgMLST nomenclature as well as four known clusters which had previously been assigned to a food source. We found that strain discrimination usually progresses from PFGE to cgMLST to SNP-based analysis. Our data highlight that comprehensive and timely WGS-based subtyping of L. monocytogenes is necessary to optimize and intensify identification of disease clusters and their sources on both the national and global levels. Moreover, our analysis of allelic and SNP distances in cgMLST complexes and verified outbreak clusters helps to define cutoffs useful for cluster definition in epidemiological investigations using WGS.
MATERIALS AND METHODS
Bacterial strains and growth conditions.Strains of L. monocytogenes were routinely grown in brain heart infusion (BHI) broth or on BHI agar plates at 37°C. All L. monocytogenes strains are listed in Table S1 in the supplemental material.
Determination of molecular serogroups and PFGE.Molecular serogroups were determined by multiplex PCR (8, 19, 20). PFGE of all strains was performed using the PulseNet protocol (12). Restriction patterns were analyzed with BioNumerics software, version 7.1 (Applied Maths BVBA, Sint-Martens-Latem, Belgium).
Isolation of chromosomal DNA, library preparation, and genome sequencing.DNA was extracted using a GenElute Bacterial Genomic DNA kit (Sigma-Aldrich), and concentration was determined on a fluorescence microplate reader using a Quant-iT PicoGreen double-stranded DNA (dsDNA) assay kit and lambda-DNA as a standard (Thermo Fisher Scientific). One nanogram of genomic DNA was used for library generation by a Nextera XT DNA Library Prep kit (Illumina). Sequencing was performed by using a MiSeq Reagent kit (version 3) cartridge (600-cycle kit) on a MiSeq sequencer in paired-end mode with a 2× 300-bp read length or by using an Illumina HiSeq 1500 sequencer and a HiSeq PE Rapid Cluster kit (version 2) for cluster generation with a HiSeq Rapid SBS (version 2) sequencing kit for generating 2× 250-bp paired-end reads in a dual flow cell run.
MLST and cgMLST.Sequencing reads were mapped against an MLST scheme based on seven housekeeping genes (10) and the 1,701-locus cgMLST scheme (18), using the Burrows-Wheeler aligner (BWA) mapping algorithm of the Ridom SeqSphere software (Münster, Germany) and L. monocytogenes EGD-e chromosomal DNA as the reference sequence (NC_003210.1) (21). Sequence types (STs) and cluster types (CTs) were determined after automated allele submission to the cgMLST server for L. monocytogenes (http://www.cgmlst.org/ncs/schema/690488/). Calculation of allelic distances and minimal spanning trees was performed using the in-built distance matrix and minimum spanning tree functions, respectively, in the mode “pairwise ignore missing values.”
Read mapping and SNP calling.Sequences were reconstructed from raw read data using a custom-made pipeline which applied (i) read trimming using Trimmomatic (version 0.32) with default parameters (22), (ii) alignment of trimmed reads to the reference sequence using BWA-MEM (maximal exact match) with default parameters (0.7.10-r789), (iii) SAM file-to-BAM file conversion using SAMtools (version 0.7.10-r789) (23), (iv) pileup using SAMtools mpileup (without probabilistic realignment for the computation of base alignment quality), (v) variant calling using VarScan (version 2.3; parameters: min-coverage, 10; min-reads, 8; min-avg-qual, 20; min-var-freq, 0.8; min_freq-for-hom, 0.8; P value, 0.01; strand-filter disabled) (24), and (vi) consensus sequence creation. To build the consensus sequences, insertions were excluded, and base calls were considered only if supported by at least 80% of the reads (otherwise N has been called). SNPs were filtered using a previously published SNP filter (25) with an exclusion distance of 300.
Accession number(s).Raw sequence files for all strains were submitted to the European Nucleotide Archive (https://www.ebi.ac.uk/ena) under study accession number PRJEB24496.
RESULTS
Selection of L. monocytogenes isolates for WGS.The German part of the German-Austrian binational consiliary laboratory for L. monocytogenes collects 300 to 450 L. monocytogenes isolates from human infections per year. These isolates are routinely analyzed by PCR-based molecular serotyping (8, 9) and PFGE (11) for cluster detection. To further improve this, WGS combined with cgMLST was introduced in 2015. Since then, L. monocytogenes isolates that formed clusters according to their PFGE profiles and selected sporadic isolates were subjected to WGS and analyzed by a published 1,701-locus cgMLST scheme with an automatically curated nomenclature (14). By spring 2017, 424 genomes of L. monocytogenes (381 belonging to PFGE clusters and 43 sporadic isolates) isolated in 2007 to 2017 had been sequenced. Most of these had been received in 2015 (20.3%) and 2016 (55.2%) (Fig. 1A) and belonged to the major prevailing serogroups typically associated with human infections, i.e., molecular serogroups IIa (42.5%), IIb (12.7%), IIc (0.2%), and IVb (44.1%). Only two isolates comprised serogroup IVb-v1. Such isolates are closely associated with serotype 4 strains (Fig. 2A) (20). Altogether, the sequenced isolates belonged to 21 distinct MLST sequence types (ST), and ST8 (21.0%), ST6 (18.4%), and ST1 (16.3%) were the three most frequent (Fig. 1C; see also Fig. S1 in the supplemental material).
Basic epidemiological characteristics of the 424 L. monocytogenes strains sequenced in this study. Distribution of isolates according to their year of isolation (A), their molecular serogroup (B), and their MLST sequence type (C) is shown. Isolates for which no ST could be determined (e.g., due to missing alleles or sequencing problems) were collectively grouped into a separate category labeled with a question mark.
Population structure of human L. monocytogenes isolates as determined by cgMLST. (A) Minimum spanning tree (MST) showing relatedness of 424 human L. monocytogenes isolates. Individual isolates were labeled according to their CTs. Putative complexes (with ≤10 different alleles between neighbored isolates) are highlighted in gray. Molecular serogroups are indicated. (B) Size (number of isolates) and diversity (number of genotypes) of the 38 cgMLST complexes.
Cluster detection using core-genome MLST.The 424 selected human L. monocytogenes strains belonged to 171 different cgMLST cluster types (CTs) and formed 38 complexes (Fig. 2A and Table 1) with ≤10 different alleles between a pair of neighboring isolates. This threshold was introduced previously for cgMLST complex definition, but its practical usefulness still needs to be validated (14) since this procedure would also group more distantly related clones into the same complex in the presence of suitable intermediate isolates. However, a similar concept is used in classical MLST, where isolates with ≤1 different allele (out of 7 alleles in total) are grouped together into clonal complexes, and these clonal complexes were clearly distinct from each other (10). More than 50% of the cgMLST complexes identified here comprised isolates from several years, and 12 included recent cases (Table 1), suggesting activity of the source. Food sources have been identified for only four of the complexes thus far. The complexes contained between 2 and 54 isolates (median, 3.5 isolates) and between 2 and 22 distinguishable genotypes (i.e., distinct allelic profiles; median, 3 genotypes) (Fig. 2B). Two-thirds of all isolates (297 out of 424) were part of such complexes. While the majority of these complexes consisted of five or fewer isolates (27 complexes containing 77 isolates), almost one-third of all isolates (122 isolates) belonged to the three largest complexes.
Putative outbreak complexes detected by cgMLST
The largest complex was complex 1 with 54 CT1248 isolates. This complex represents a large listeriosis outbreak in southern Germany that occurred between 2012 and 2016 (26). Genetic diversity within this complex was low as only 22 distinct genotypes (0.4 genotypes per isolate) with ≤4 allelic differences to the closest related isolate were observed (Fig. 2A). A German meat processing company was the source for this outbreak (27), which came to an end after shutdown of the company. The second-largest complex, complex 2, contained 40 isolates from 2014 to 2017 and included CT2789 (38 isolates), CT3844 (1), and CT3920 (1). With up to six allelic differences between neighboring genotypes, this complex had a somewhat bigger isolate diversity. A food source associated with this outbreak has been identified recently (28). The third-largest complex, complex 3, contained 28 CT1121 isolates from 2011 to 2016 and 22 different genotypes (0.8 genotypes per isolate). Allelic difference was ≤4 between neighboring isolates. Here, no source of infection has been identified so far. Thus, the above-presented examples show that the 1,701-locus cgMLST scheme reproduces outbreak clusters.
Allelic distances in verified outbreak complexes.The definition of allelic distance cutoffs for the identification of outbreak complexes is an urgent need in next-generation sequencing (NGS)-based pathogen surveillance. The maximum allelic distance between any two isolates of complex 1 was 9 alleles, and it was 16 alleles in complex 2. Complex 7, which is made up of 13 CT1114 strains, represents the third outbreak with a known food source (unpublished data), and the maximum allelic distance in complex 7 was 10 alleles. Furthermore, two representatives of the 2009/2010 multinational listeriosis outbreak, associated with Quargel cheese (29), differed in nine alleles (Fig. 2A, complex 27). Maximal allelic distances in all cgMLST complexes ranged between 0 and 25 alleles (median, 4 alleles). These observations support a distance cutoff close to 10 alleles for assignment of outbreak complexes even though higher values may occur in more diverse complexes. This is in good agreement with work of Moura et al., who showed that epidemiologically linked isolates differed in ≤7 out of 1,748 loci in their cgMLST schemes (15).
To relate this cutoff to the resolution obtained with other typing techniques available for L. monocytogenes, maximum allelic distances between any two isolates of a distinct serogroup, ST, PFGE profile, and CT were calculated. This analysis revealed that the maximal allelic distance within a serogroup could be as large as 1,362 alleles (serogroup IIa), as large as 118 alleles within a certain ST (ST1; median, 43 alleles), and as large as 94 alleles for strains with the same PFGE profile (median, 32 alleles) (Fig. 3). In contrast, the maximal allelic distance between any two isolates of the same CT was 13 alleles in the case of CT90 strains (median, 4 alleles) (Fig. 3). In conclusion, MLST and PFGE do not reach the required typing resolution necessary for a reliable identification of outbreak clusters, whereas the allelic distances among isolates with the same CT were close to the proposed cutoff value of 10 alleles (14).
Maximum allelic differences of L. monocytogenes typing methods. Box plot showing the maximum number of allelic differences after 1,701-locus cgMLST observed between any two isolates of a certain molecular serogroup, MLST sequence type, PFGE profile, or cgMLST CT. Values were determined from allele distance matrices obtained from pairwise comparisons of cgMLST typing data.
Confirmation of outbreak clusters by SNP calling.The 1,701-locus cgMLST scheme considers only ∼60% of the available genome sequence, leaving the possibility of false-positive identification of outbreak clusters. For confirmation of cgMLST complexes, sequences of all 424 isolates were mapped against the L. monocytogenes EGD-e genome and aligned. Genomes covering less than 80% of the reference genome were discarded, generating an alignment consisting of 413 isolates with 22,655 SNPs and a phylogenetic tree (Fig. 4A) consistent with the population structure of L. monocytogenes: isolates clustered into two of the four major phylogenetic lineages, i.e., lineage I (including serogroups IIb, IVb and IVb-v1) and lineage II (serogroups IIa and IIc), and all isolates with distinct STs formed individual branches (Fig. 4A). A total of 32 of the 38 cgMLST complexes were grouped into distinct branches according to their cgMLST complex assignments (Fig. 4A), but 6 cgMLST complexes (complexes 4, 5, 6, 9, 11, and 25) were split into two to four subbranches by SNP calling (Fig. 4A, complex numbers labeled in red). These congruence limitations likely result from alignment length reduction through SNP filtering or from neglecting accessory genome parts in cgMLST comparisons. Thus, cgMLST complex-specific alignments were generated and processed individually. In order to remove regions with accumulated SNPs, which may result from recombination events or mobile elements, an SNP distance filter with an exclusion distance of ≤300 nucleotides was included (25). The median number of SNPs between any two isolates within a complex then ranged between 0 and 6 (Fig. 4B), and differences of up to 15 SNPs were observed with a single isolate in complex 1. Complex-specific SNP calling further differentiated isolates within 18 out of 38 cgMLST complexes (Fig. 4B, isolates shown as red circles) but all with a difference of ≤7 SNPs. In the remaining 20 complexes, no SNPS were detected at all. This shows that the 1,701-locus cgMLST scheme reproduces clusters of closely related isolates with high precision.
Reconstruction of L. monocytogenes population structure by SNP calling. (A) Unrooted neighbor-joining tree illustrating the population structure of the L. monocytogenes isolates sequenced in this study. Sequencing reads of all isolates were mapped against the L. monocytogenes EGD-e reference genome, and consensus sequences were aligned using an in-house pipeline. Isolates for which the genome sequences covered less than 80% of the reference sequence were removed from the alignment. Nucleotide positions that were invariable, containing gaps or ambiguities, were stripped from the alignment before tree calculation, which was performed using the Geneious Tree builder (Biomatters, Ltd.). Isolates are color coded according to their molecular serogroup, their MLST sequence type, and their cgMLST complex number. Complexes that do not cluster in unique branches by the SNP-calling approach are labeled in red. (B) Box plot showing the number of SNPs between any two isolates within each of the 38 cgMLST complexes. Median values for SNP distances are indicated. Isolates within the cgMLST complexes that are further differentiated within the complex by the SNP-calling approach are shown in red.
Typing resolution of PFGE versus cgMLST.PFGE differentiated 83 profiles (discriminatory power, D, of 0.9579), which is roughly half of the number of CTs discriminated by cgMLST (171 CTs; D = 0.9648) but still more discriminatory than MLST (21 STs; D = 0.8567). Thus, discriminatory power of cgMLST is clearly superior to PFGE. In good agreement with this finding, 73% of all PFGE profiles with ≥2 isolates could further be differentiated by cgMLST (30 out of 41 PFGE profiles). In contrast, only 40% of the CTs with ≥2 isolates were further discriminated by PFGE (16 out of 40 CTs). In some cases, strains with the same PFGE profiles were subdivided into as many as 15 different CTs (PFGE type 23/37) (Fig. 5A and B). The 16 different CTs that were further separated by PFGE differentiated into four (CT90) or two (CTs 69, 1247, 1248, 1263, 2198, 2521, 2752, 2789, 3816, 3819, 3926, 3997, 4076, 4083, and 4246) PFGE profiles and contained 157 isolates in total (37%) (Fig. 5A). Genomic variations outside the core genome must account for this variability in strains with identical CTs.
Differentiation of PFGE profiles by cgMLST and vice versa. (A) Box plot showing the number of different CTs that can be detected within each of the 83 distinct PFGE profiles (PFGE → CT) and the number of different PFGE profiles that are observed among isolates with one of the 171 different CTs (CT → PFGE). Relevant PFGE profile designations and CTs are indicated. (B) Minimum spanning tree showing further discrimination of L. monocytogenes strains with the identical PFGE type AscI/ApaI 23/27 by cgMLST into 15 different CTs. Isolates are colored according to their CTs, which are also indicated.
PFGE outperforms cgMLST in subtyping of CT90 strains.CT90 strains belonging to the four PFGE profiles 3/8var2 (6 isolates), 3/23 (7 isolates), 3/142 (1 isolate), and 3/287 (1 isolate) (Fig. 6A) were identical after AscI digestion (data not shown) but distinguishable using ApaI (Fig. 6B). Specifically, (i) the ∼380-kb band in profile 3/8var2 shifted to ∼430 kb in the 3/23 profiles and (ii) the ∼250-kb band in the 3/8var2 profile shifted to ∼300 kb, which represents a double band in the 3/23 profiles based on staining intensity (Fig. 6B). PFGE profiles 3/142 and 3/287 were subtypes of 3/23 (data not shown). For identification of the underlying genomic rearrangements, sequencing reads of representative 3/8var2 (16-01401) and 3/23 (16-01911) strains were de novo assembled using the A5 pipeline (30), and the contigs obtained (19 per isolate with mean N50 length of 425,199 bp) were arranged into circular pseudochromosomes using the Ragout assembly tool (31) and EGD-e chromosomal DNA as the reference. The resulting genome sequences were manually curated by comparing predicted restriction fragments and laboratory-deduced restriction fragments, all located within the deduced contigs. A virtual ApaI digest of both pseudochromosomes confirmed that a 378-kb fragment in the 3/8var2 strain was shifted to 426 kb in the 3/23 strain and that a 243-kb fragment in the 3/8var2 strain was shifted to 284 kb in the 3/23 isolate (Fig. 6C). Both shifted fragments contained prophages in the 3/23 strain (Fig. 6D). The prophage present in the 284-kb ApaI fragment was 41.3 kb in size and inserted into the tRNALys gene. This prophage shows similarity with the Listeria phage LP-030-3 (96% identity), originally isolated from the serotype 4b L. monocytogenes outbreak strain F2365 (32). The prophage present in the 426-kb ApaI fragment was 48.1 kb in size and integrated into lmo1263, encoding a putative transcriptional regulator. This phage does not show similarity to described Listeria phages but is also present in other sequenced L. monocytogenes strains, including strain CFSAN008100 (GenBank accession number NZ_CP011398.1). Apparently, PFGE can further discriminate L. monocytogenes isolates beyond the CT level, especially when mobile genetic elements are picked up or lost or when other rearrangements occur in the accessory genome.
Further discrimination of CT90 isolates by PFGE. (A) Minimum spanning tree of 15 L. monocytogenes CT90 isolates. Isolates are colored according to their PFGE profiles. (B) Pulsed-field gel electrophoresis of CT90 isolates with PFGE profiles 3/8var2 and 3/23 after ApaI macrorestriction. Bands that are shifted in size due to phage insertions are boxed. (C) Virtual ApaI digest of pseudochromosomes obtained from assembled sequencing reads of strain 16-01401 and 16-01911 chromosomal DNA. (D) Position of prophages (labeled in yellow) in the 426-kb and 284-kb ApaI fragments of the AscI/ApaI 3/23 strain 16-01911, which are absent in the corresponding 378-kb and 243-kb fragments of the AscI/ApaI 3/8var2 strain 16-01401.
DISCUSSION
We present a collection of genome-sequenced L. monocytogenes strains from listeriosis patients in Germany over 10 years. It represents on average 16.8% and up to 56.9% (2016) of all human L. monocytogenes isolates collected by the consiliary laboratory. Typing resolution increased from that obtained with MLST (21 STs) to that with PFGE (83 PFGE profiles) and again to that with cgMLST (171 CTs). cgMLST clustered all isolates according to their molecular serogroups and was in absolute congruence with MLST (see Fig. S1 in the supplemental material). SNP calling confirmed all cgMLST complexes with a median SNP number between 0 and 6 SNPs. The mutation rate of L. monocytogenes, 2.5 × 10−7 substitutions/site/year, is astonishingly low (15). Due to this remarkable sequence stability, the number of SNPs observed between epidemiologically linked isolates is expected to be low but may increase with outbreak duration. This view is supported by two known cases of maternal-neonatal transmission included in our collection, which could not be differentiated by cgMLST or SNP calling in one case or by one different allele (lmo0857) in the other (data not shown). Because the low mutation rates contrast with the high number of different alleles occasionally observed within a single cgMLST complex, additional confirmation of cgMLST complexes by SNP calling involving suitable SNP filters seems necessary for cluster assignment. While the exclusion of the accessory genome is necessary to allow robust comparisons of distantly related isolates, it leads to loss of typing resolution in closely related strains. CT90 isolates were further differentiated by PFGE due to phage insertions in the accessory genome, and SNP calling reproduced their separation (Fig. S2). Thus, CT90 is another example emphasizing the necessity to validate cgMLST clusters by methods with higher resolution, such as SNP calling. The use of reference genomes from close relatives during read mapping would increase the chance that SNPs/regions contributing to strain discrimination are covered by the reference genome and do not get lost during mapping and subsequent SNP filtering. The most ideal reference genome, however, would be that of a selected isolate from the central node of a given cgMLST complex. Sequencing technologies allowing simple and cheap reconstruction of closed bacterial genomes from raw sequencing data would be required to generate such complex-specific reference genomes for routine validation of cgMLST complexes. Until then, PFGE can still be useful to discover such rearrangements in the accessory genome.
One advantage of cgMLST is a simplified data exchange in the form of CT designations between laboratories. While this supports the assignment of isolates with identical CTs to the same epidemiological context, isolates allocated to different CTs are not necessarily unrelated. We have identified five complexes that contain isolates with up to three different CTs, and their genetic relationship is not obvious from their CT designations. This limitation is due to cgMLST nomenclature rules and might be solved by implementation of a hierarchical nomenclature comparable to SNP addresses (33) in the future.
Identification of food vehicles has been difficult in the past, mainly because of the long incubation period of listeriosis of up to 70 days (6, 34), and they are known for only 4 of the 38 cgMLST complexes (27, 28, 35). Food isolates associated with cluster 1 and cluster 2 had been identified by routine screening of food isolates by the German national food safety agency after cgMLST implementation (27, 28). This illustrates that comprehensive WGS-based typing of food and human isolates in real time facilitates association of food isolates with human disease clusters in order to stop transmission of L. monocytogenes.
ACKNOWLEDGMENTS
The project was funded by the Intensified Molecular Surveillance Initiative of the Robert Koch Institute.
We thank Ute Strutz, Gerlinde Bartel, and Monique Duwe for technical assistance and Sandra Simon, Stefan Fiedler, and Jennifer Bender for DNA sequencing support. We acknowledge Christina Lang for bioinformatic support. S.H. thanks Susanne Halbedel for helpful discussions.
FOOTNOTES
- Received 19 January 2018.
- Returned for modification 5 March 2018.
- Accepted 24 March 2018.
- Accepted manuscript posted online 11 April 2018.
Supplemental material for this article may be found at https://doi.org/10.1128/JCM.00119-18.
- Copyright © 2018 American Society for Microbiology.
