Stenotrophomonas maltophilia is an opportunistic, non-fermentative, Gram-negative, and ubiquitous Gammaproteobacterium. This bacterium has environmental origins (water, soil), but is known to cause a diverse range of infections in humans (Denton and Kerr, 1998; Ryan et al., 2008; Brooke, 2012). It is usually considered as a pathogen with reduced virulence, but it can cause a broad spectrum of nosocomial infection complications with a considerable mortality rate of up to 37.5% (Falagas et al., 2009). The most common clinical manifestations caused by S. maltophilia include respiratory tract infections (Nseir et al., 2006; Nicodemo and Paez, 2007; Weber et al., 2007) and bacteremia (Lai et al., 2004; Chang et al., 2014).

Stenotrophomonas maltophilia exhibits high levels of intrinsic resistance to multiple classes of antibiotics (e.g., carbapenems and aminoglycosides). In addition, several mechanisms are able to contribute to acquired antimicrobial resistance (e.g., against fluorchinolones, polymyxins) resulting in a multi-drug resistance profile (Crossman et al., 2008; Ryan et al., 2009; Brooke, 2012). S. maltophilia is recognized by its ability to form biofilms on abiotic surfaces including glass and plastics like polystyrene (Brooke, 2012; Vidigal et al., 2014), as well as on host tissues such as bronchial epithelial cells (Pompilio et al., 2010). These characteristics make S. maltophilia an emerging global multi-drug-resistant opportunistic pathogen (Brooke et al., 2017).

In addition to nosocomial infections in hospitalized patients, S. maltophilia is able to chronically colonize the airways of CF patients in a manner similar to Pseudomonas aeruginosa or Staphylococcus aureus. It was shown that chronic S. maltophilia infection is an independent risk factor for pulmonary exacerbation in CF patients, associated with decline in FEV1 and a threefold increased risk for death or lung transplantation within the CF patient population (Waters et al., 2011, 2013; Vidigal et al., 2013).

Molecular typing studies uncovered a high degree of genetic diversity between most S. maltophilia strains from hospitalized individuals as well as between those from CF patients (Denton and Kerr, 1998; Valdezate et al., 2004; Gülmez and Hasçelik, 2005; Kaiser et al., 2009; Wu et al., 2011; Vidigal et al., 2014; Youenou et al., 2015). The molecular typing methods used so far include amplified fragment length polymorphism (AFLP), restriction fragment length polymorphism (RFLP), 16S rRNA sequencing, multi locus sequence typing (MLST), repetitive element palindromic (rep)-PCR, and gyrB sequencing, each producing varying results with no commonly accepted standard or defined nomenclature (Adamek et al., 2011).

In contrast to these classical typing techniques that only interrogate a small part of the genome, next generation sequencing technologies allow for the near complete analysis of a bacterial genome, i.e., whole genome sequencing (WGS). This method provides a single-base-pair resolution between isolates, making it the ultimate molecular typing method for bacteria (Snyder et al., 2013). Recent studies have demonstrated that analysis of bacterial genomes based on single nucleotide polymorphisms (SNPs) enables the definition of relatedness between epidemiologically associated isolates and tracking of bacterial evolution over time, including in long-term S. maltophilia colonization of CF patients (Jelsbak et al., 2007; Bragonzi et al., 2009; Lewis, 2010; Eppinger et al., 2011; Chung et al., 2017; Esposito et al., 2017). A very recent study found that the bacterial adaptation of S. maltophilia to the CF lung is associated with consistent genotypic and phenotypic heterogeneity (Esposito et al., 2017). Here, we performed WGS based phylogenomic analyses of a diverse collection of human isolates (mainly from CF and intensive care unit patients), and 22 environmental isolates. The collection represents the known diversity of S. maltophilia and Stenotrophomonas rhizophila, including representatives of phylogenetic groups previously defined by traditional genotyping methods (Adamek et al., 2011). The phylogenomic classifications have been linked with underlying diseases, phenotypical assays for biofilm formation, extracellular proteolytic activity and lipopolysaccharide (LPS) pattern characterization.

In total, 89 S. maltophilia and S. rhizophila isolates were sequenced successfully, with at least 240 Mbp sequence data generated per sample. Mapping to the K279a genome resulted in a mean coverage depth of 77.0-fold (range 35.6–222.4) of covered regions, and a mean coverage breadth of the reference sequence of 79.0% (range 41–100%). Fully finished genomes of five strains were added as controls and references K279a (NC_010943.1), D457 (NC_017671.1), JV3 (NC_015947.1), R551-3 (NC_011071.1), and S. rhizophila DSM14405 (CP007597.1). For the full analysis of all 94 datasets of this study, 1,928,889 bp out of the 4,851,126 bp of the K279a reference genome complied with the chosen thresholds and were used to derive a set of 408,860 SNP positions for the calculation of phylogenetic trees.

Group definition with a threshold of a maximum distance of 50.000 distinct SNP positions to the nearest group member resulted in 12 groups/clades in the S. maltophilia and S. rhizophila phylogeny (designated groups 01–12), which correspond well with GyrB, AFLP, rep-PCR (with the notable exception of rep-3), and MLST typing results (Figure 1 and Table 2). The 12 groups are separated by large SNP differences in line with independent environmental acquisition (Figure 2). While both, a genome-wide genome alignment comparison of the five fully completed genomes (S. maltophilia K279a, D457, JV3, R551-3, and S. rhizophila DSM14405) and a NeighborNet splitstree built from the detected SNP positions of the 94 strains, indicate some genomic rearrangement, the distinction into separate groups remains stable in the NeighborNet tree (Supplementary Figure S2). Strikingly, the phylogenetic groups we defined almost exclusively comprise either isolates from the environment or human. Indeed, 9 out of the 12 groups were comprised entirely of human or environmental isolates, and the remaining three groups contained at least 75% either human or environmental isolates (Table 2).

In addition to a clear separation of human and environmental isolates in most groups, the overall majority of human derived strains could be grouped close to the K279a branch (K279a-like) in the S. maltophilia and S. rhizophila phylogeny (Figures 1, 2). Notably, neither proteolytic activity levels nor capabilities to form biofilms on polystyrene correlated with the phylogenetic separation (Figure 1 and Table 1).

To get a first indication on group-specific pathobiological characteristics, we mapped biofilm and protease production on the tree (Figure 1). In general, the majority of the isolates formed biofilms on a plastic surface in microtiter plates. Only 19 strains failed to form detectable biofilms. Surprisingly, a large fraction of the environmental isolates [11 (50%)] did not form biofilms. This is in contrast to the clinical isolates, from which only 8 strains (12.5%) failed to form biofilms under the same experimental conditions. With respect to protease secretion, we noticed that only 9 isolates failed to produce significant amounts of protease, but no correlation with respect to the SNP analysis or their origin was detected.

To obtain a deeper understanding of subgroup clonality, we performed a targeted analysis of 73 strains from which at least 75% of the K279a genome sequence could be recovered. This allows for an increased resolution of the resulting SNP phylogenies with 649,312 SNP positions detected and 3,061,984 bp out of the 4,851,126 bp reference sequence taken into account for the comparison (Figure 3). Overall, the isolates are separated by large SNP distances, reflecting their environmental occurrence. However, excluding full genome datasets, we identified four groups of strains for which group members were at a maximum distance of 25 distinct SNP positions from each other, possibly indicating transmission between patients: G1 (ICU02, ICU06, ICU98, ICU119), G2 (U7, U8), G3 (Sm527, Sm530), and G4 (Sm297, Sm454). In the cases of the ICU strain cluster (G1) and the CF groups (G3, G4), a spatio-temporal association was present, indicating that these patients were infected by the same strains presumably based on patient-to-patient or environment-to-patient transmission.

The LPS profiles of S. maltophilia and S. rhizophila were compared to identify identical ladder-like LPS patterns (Supplementary Figure S3). In spite of a significant heterogeneity of the LPS banding patterns, we have been able to assign 46 strains to a total of 16 LPS groups on the basis of electrophoretic mobility of the main bands, with at least two identical profiles within a group. Isolates with apparent unique LPS patterns or rough-type LPS lacking O-repeating units were ignored. Furthermore, the S. maltophilia and S. rhizophila strains were screened with a polyclonal antiserum against the O-antigen of S. maltophilia K279a. In order to obtain an antibody specifically directed against the O-antigenic chain of S. maltophilia K279a, a ΔrmlBACD deletion mutant of the strain lacking the gene cluster for dTDP-rhamnose synthesis was constructed. The polyclonal antibody against heat-inactivated S. maltophilia K279a wild-type cells was then adsorbed with heat-killed cells of the ΔrmlBACD mutant. Both the loss of O-substituted LPS in S. maltophilia K279a ΔrmlBACD and the O-specificity of the adsorbed immune serum was confirmed by SDS–PAGE and immunoblot analysis (Supplementary Figure S4). Immunoblot analyses of Stenotrophomonas spp. indicated serological cross-reactivity of the O-specific K279a antiserum with the O-antigens of 10.6% (10, including S. maltophilia K279a) of all isolates and 13.7% of isolates herein assigned to the group of 73 K279a-like strains, which suggested that these isolates expressed a structurally similar or even identical O-specific polysaccharide (Supplementary Figure S4). The serological cross-reactivity of the members of LPS-profile groups 02 and 03 was in good agreement with the phylogenetic tree (Figure 1), for which positive detection was restricted to clustered isolates closely related to K279a of LPS group 02, with the sole exception of strain DSM50170 displaying a unique LPS banding pattern. We also found that each group of K279a-like strains with a maximum distance of only 25 distinct SNP positions could be classified into separate LPS groups, each with an identical banding pattern to derive LPS groups 09 (ICU02, ICU06, ICU98, ICU119), 08 (U7, U8), 12 (Sm527, Sm530) and 03 (Sm297, Sm454), with the LPS of Sm297 and Sm454 showing, in addition, serological cross-reactivity with the O-specific K279a antibody.

Stenotrophomonas maltophilia is becoming increasingly relevant as a nosocomial pathogen. We used WGS to dissect the phylogenetic relationship of a set of 89 S. maltophilia and S. rhizophila isolates from human and environmental sources. Our phylogenetic analysis revealed a highly diverse population with deep branches separating groups of strains. Notably, these groups correspond well to those identified by traditional genotyping methods. Despite a generally conserved genome structure (Supplementary Figure S1), groups were distinguished by thousands of distinct SNP positions, with a reduced diversity within a “supergroup” of most human-derived K279a-like isolates (Figure 1). Interestingly, 9 out of the 12 groups consisted exclusively of either human (6 groups) or environmental isolates (3 groups). Out of the three remaining groups, two reflected the total ratio (75% human to 25% environmental isolates), and one comprised mostly environmental strains (80%). These data are in line with a previous MLST study from Germany which also found overrepresentation of some genogroups among human and environmental strains (Kaiser et al., 2009). Taken together, these results suggest a correlation between association with human diseases and genetic background. The identified groups may also differ with regard to their pathogenic potential, e.g., a higher pathogenic potential in strains closely related to the group defined by reference strain K279a. Alternatively, the strains within the very same lineage might have a higher potential to colonize humans and cause infectious diseases in comparison with others. The K279a reference strain also used in other studies was originally isolated from a septic patient’s blood sample (Crossman et al., 2008).

In contrast to our findings, Youenou et al. (2015) detected that genomic data from environmental and clinical S. maltophilia isolates did correlate neither to their origin nor to their antibiotic resistance profile. However, this study was limited to a small number of isolates (n = 14 in total; n = 5 from clinical origin). Another recent study also addressed the question of specific phylogenetic branches (Lira et al., 2017). A total of 10 environmental and 10 clinical isolates were sequenced and a large core genome with similar distributions of genes including virulence determinants was found. No specific evolutionary branches were detected. It was postulated that the evolution of S. maltophilia is based on the strain-specific acquisition of genes during adaptation to different microniches (Lira et al., 2017). Chung et al. (2017) revealed that, even within the lungs of one individual CF patient, the S. maltophilia isolates separated into several lineages with distinct phenotypes distinguished by adaptive mutations (Chung et al., 2017). This study also showed that mutations within certain strains were associated with specific pulmonary locations indicating selection driven by the micro-environment.

Another recent study investigated 91 S. maltophilia isolates from 10 CF patients over a period of 12 years using genotypic and phenotypic methods (Esposito et al., 2017). SNP-based and pangenome analysis showed three major lineages whereas phenotypic results (growth rate, biofilm formation, motility, mutation frequency, antibiotic susceptibility, and in vivo virulence) did not correlate with genomic data.

While our findings confirm a large genome diversity among environmental and human S. maltophilia isolates, their capacity to colonize CF-patients appears to be higher for certain defined groups. Furthermore, at least in few instances, we detected likely events of nosocomial transmission or acquisition from the same origin as already observed for Mycobacterium abscessus (Bryant et al., 2016) This warrants further investigations into transmission of S. maltophilia in the community and hospital setting to determine scale and universality of this phenomenon, as well as its importance for disease severity. Furthermore, it is tempting to speculate that the unequal distribution of human isolates in the S. maltophilia phylogeny is based on special virulence features of strains from groups associated with human infection.

One potential virulence factor is the O-specific polysaccharide that is the most variable part of the LPS molecule and determines the serological specificity of LPS and Gram-negative bacteria. It has further long been recognized that the O-polysaccharide is a fairly stable phenotypic marker that can be used to distinguish between groups of strains within a given species (Rimler, 1990; Mills et al., 1992; Aucken and Pitt, 1993; Jürgens and Fehrenbach, 1997; Santamaría et al., 1998). The enormous variability of O-polysaccharide structures has been attributed to evolutionary mechanisms that enabled alterations of the most exposed surface structure to maintain advantageous adaptations to particular environmental niches or evade niche-specific selection pressures or escape-specific antibody-based immunity. In fact, the structural diversity of O-polysaccharides reflects an extensive polymorphism at the genetic level, including gene mutations, genetic exchange events between different microorganisms, as well as recombination processes (Cunneen and Reeves, 2011). Varying numbers of repeating units in the O-specific polysaccharide result in unique ladder-like banding patterns of LPS separated by SDS–PAGE. Thereby, comparative LPS profiling and O-serotyping are valuable tools for epidemiological studies with high discriminatory power (Aucken and Pitt, 1993). Here, we show that cross-reactivity of an antiserum raised against the O-antigen of K279a correlates well with our phylogenetic analysis, with only two subgroups of closely related strains expressing the K279a-like O-side chain antigen, apart from one isolate, DSM50170, with a distinctly different gel profile.

However, the serological cross-reaction with DSM50170 indicates that the O-specific polysaccharide of K279a and reactive K279a-like isolates is similar to the structure of the O8-antigen of DSM50170, which consists of a branched tetra-saccharide with three rhamnopyranosyl residues in the main chain and 3-O-methylxylose as the substituent (Neal and Wilkinson, 1982; Knirel, 2011). Of note, a total of 31 different S. maltophilia O-serotypes and 16 O-antigenic structures have been described thus far, showing that most of the O-antigen repeating units of the S. maltophilia LPS have been branched tri- and tetra-saccharides of rhamnose, fucose, xylose, and glucose (Ryan et al., 2009; Knirel, 2011). While the LPS-based approach aimed to complement our WGS data has identified a number of closely related isolates, we herein confirm previous results on the variability of O-repeating unit structures within the genus Stenotrophomonas, which in fact reflects the versatility of the bacterium to adapt to various environmental niches.

The capability to form biofilms as well as to secrete proteolytic enzymes differed between individual strains but failed to correlate with the distinct groups, and was also not particularly high in groups containing only human isolates. These results are in accordance with a previous study including CF isolates (Esposito et al., 2017). The distribution of both properties, present or absent, across all groups suggests that both phenotypic traits can be universally present in the species complex but are not relevant for the pathogenic potential of a strain. However, both biofilm formation and proteolytic activity can be of relevance for isolates living in the environment or the human host depending on the individual niches.

A previous study combining comparative genomics, transcriptomics and physiological approaches with S. rhizophila DSM14405, a plant-associated S. maltophilia strain, and the human S. maltophilia K279a found habitat-specific genes in human vs. environmental strains (Alavi et al., 2014). Therefore, a systematic comparison of the pangenome of the highly diverse S. maltophilia and S. rhizophila species complex would be interesting. A prerequisite and ideal tool for this comparison would be the construction of a whole genome MLST scheme. Thereby, potential virulence factors could also be identified, which likely correlate either with the human or with the plant host.

In this context, it should be noted that the strain collection may have introduced a systematic bias since the majority of the human isolates analyzed were obtained at one site (University Hospital Essen, 41 out of 67 isolates). Similarly, the number of environmental isolates tested in this study was rather small, with just 22 sequenced strains, compared to the number of clinical isolates. In summary, WGS provided a high-resolution phylogeny of the S. maltophilia and S. rhizophila species complex. Our results indicate the presence of branches of strains adapted to the human host, predominantly CF patients, with the potential of human-specific virulence and pathogenicity. The reason for increased virulence, however, remains elusive, and needs to be investigated in future studies.

The datasets supporting the conclusions of this article are available in the European Nucleotide Archive (ENA) repository of the European Molecular Biology Laboratory (EMBL). All accession numbers are stated in Supplementary Table S1.

All clinical samples of the present study were analyzed after performing conventional microbiological diagnostics. The study did not include patient’s details and did not result in additional constraints for the patients. All data were anonymously analyzed without patients’ consent due to the retrospective nature of the study. All analyses were carried out in accordance with approved guidelines.

Study design: JS, US, SN, and TK. Carrying out experiments: UM, EA, WS, and TK. Writing: JS, UM, WS, and TK. Data analysis, final revision, and approval of final version: JS, UM, EA, LK, WS, US, SN, and TK.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.