Protracted Regional Dissemination of GIM-1-Producing Serratia marcescens in Western Germany

ABSTRACT

The metallo-beta-lactamase GIM-1 has been found in various bacterial host species nearly exclusively in western Germany. However, not much is known about the epidemiology of GIM-1-positive Serratia marcescens. Here we report on a surprisingly protracted regional dissemination. In-hospital transmission was investigated by using conventional epidemiological tools to identify spatiotemporal links. Strain typing was performed using pulsed-field gel electrophoresis (PFGE) and whole-genome sequencing (WGS). Bayesian phylogeny was used to infer the time axis of the observed occurrence. Thirteen S. marcescens strains from 10 patients from 6 different German hospitals were investigated. Suspected in-hospital transmissions were confirmed by molecular typing at a higher resolution by WGS than by PFGE. A detailed sequence analysis demonstrated the spread of one predominant strain variant but also provided evidence for transfer of the bla_GIM-1 gene cassette between different strains. A Bayesian phylogenetic analysis showed that the most recent common ancestor of the identified clonal cluster could be dated back to April 1993 (95% highest posterior density interval, January 1973 to March 2003) and that this strain might have already harbored the bla_GIM-1 at that time and, therewith, years before the first detection of this resistance gene in clinical specimens. This study shows a long-standing clonal and plasmid-mediated expansion of GIM-1-producing S. marcescens that might have gone unnoticed in the absence of a standardized and effective molecular screening for carbapenemases. The systematic and early detection of resistance is thus highly advisable, especially for the prevention of potentially long-term dissemination that may progress beyond control.

INTRODUCTION

Serratia marcescens is commonly found in water and soil and is recognized to be an important nosocomial pathogen that can cause various infections ranging from urinary tract infections to more severe infections like pneumonia or bloodstream infections (1). Numerous outbreaks, mostly in neonates, deriving from environmental sources and colonized humans serving as reservoirs have been described (2). Nearly half of the published S. marcescens outbreaks are reported from neonatal units, and at least two to three outbreaks can be expected on German neonatal intensive care units per year (2, 3). Treatment of infections caused by S. marcescens can be difficult due to intrinsic and acquired resistance mechanisms, such as chromosomal AmpC beta-lactamase or extended-spectrum beta-lactamases (ESBLs) (1). Moreover, S. marcescens is intrinsically resistant to colistin, an increasingly important antibiotic for the treatment of infections caused by multidrug-resistant bacteria (1). The emergence of carbapenemases, such as OXA-48 and KPC, and metallo-beta-lactamases (MBLs), such as IMP or VIM, in S. marcescens within the last decade is particularly alarming (1, 4, 5). To date, there have been only two reports of the MBL German imipenemase-1 (GIM-1) in S. marcescens (6, 7). Since its first description in a Pseudomonas aeruginosa strain isolated in 2002 (8), GIM-1 has not spread far beyond a confined region in western Germany; most reports are from the North Rhine-Westphalia region. Despite this somewhat restricted geographic mobility, the emergence of bla_GIM-1 in a number of species within the Enterobacteriaceae family has been demonstrated (6, 7, 9–12), and in addition, a new variant, GIM-2, was recently described (13). Here, we report on the local epidemiology of GIM-1-positive S. marcescens strains, their genomic evolution, and their estimated temporal distribution in western Germany.

RESULTS

Isolate collection, susceptibility testing, and resistome analysis.

Thirteen S. marcescens isolates were collected from 10 patients hospitalized in six medical centers from September 2008 to December 2015. Nearly half of the isolates were collected from hospital A (n = 6). Most isolates (n = 10) were characterized at the Institute of Medical Microbiology and Hospital Hygiene in Düsseldorf, Germany. One isolate was provided by the Medizinisches Versorgungszentrum Dr. Stein (Mönchengladbach, Germany), and two were provided by the National Reference Centre for Multidrug-resistant Gram-negative Bacteria (Bochum, Germany). Basic epidemiological and genetic data for the bla_GIM-1-positive isolates are shown in Table 1.

TABLE 1.

Characteristics of bla_GIM-1-positive Serratia marcescens strains ordered by isolation date

Strain no.a	Patient no.	Isolation date (mo-yr)	Hospital	Origin	No. of days in hospital at first isolation	PFGE typeb
SM-ID1	1	September 2008	A	Respiratory tract	28	A
SM-ID2	2	September 2008	A	Urine	82	B
SM-ID3	3	November 2008	A	Urine	82	A
SM-ID4	4	December 2008	D	Urine	NAc	A
SM-ID5	3	May 2009	A	Wound	Follow-up	A
SM-ID6	4	September 2010	A	Abdominal drainage	Follow-up	A
SM-ID7	5	June 2013	B	Urine	1	A
SM-ID8	6	August 2013	B	Screening (nose)	1	A
SM-ID9	7	August 2013	C	Urine	1	C
SM-ID10	5	March 2014	B	Urine	Follow-up	A
SM-ID11	8	January 2014	E	Skin (foot)	NA	D
SM-ID12	9	September 2015	F	Respiratory tract	NA	A
SM-ID13	10	December 2015	A	Screening (axilla)	2	A

^aSM-ID2 and SM-ID3 were called M9 and M11, respectively, in a previous study (7).

^bPFGE types were given letters.

^cNA, not available.

All 13 isolates were phenotypically resistant to several antibiotic classes (see Table S1 in the supplemental material), and resistome analysis displayed four different antibiotic resistance gene profiles, with the same profile being found in 10 isolates and distinct profiles occurring in isolates SM-ID2, SM-ID9, and SM-ID11 (Table S2). Resistance to the following beta-lactam antibiotics was observed in all isolates: piperacillin, piperacillin-tazobactam, ceftriaxone, and ceftazidime. A genetic in silico search for bla genes showed only narrow-spectrum bla_OXA genes, aside from bla_GIM-1. All isolates except isolate SM-ID2 were (intermediate) susceptible to cefepime and aztreonam, two antibiotics to which GIM-1 is known to have a low affinity (14). In isolate SM-ID2, resistance to these two drugs might point to another resistance mechanism, such as the presence of a derepressed AmpC beta-lactamase, porin loss, or the presence of an efflux pump. In the 13 isolates, the MICs for meropenem and ertapenem ranged from 0.5 to >32 μg/ml and the MIC for imipenem ranged from 1 to >32 μg/ml. Resistance to gentamicin and tobramycin in all isolates could be explained by the various aminoglycoside resistance genes; amikacin remained the sole aminoglycoside usable against the isolates. Fluoroquinolone resistance can be partly explained by the aac(6′)Ib-cr gene, found in all isolates. The possibility of the presence of additional mutations in the gyrA or parC gene cannot be excluded. SM-ID2 carried a tetracycline resistance gene, tet(41), encoding a specific efflux pump first described in S. marcescens (15). Interestingly, SM-ID2 was the only isolate nonsusceptible to tigecycline. All isolates carried the sulfonamide resistance gene sul1, and only one isolate (SM-ID9) possessed the trimethoprim resistance gene dfrA1. Thus, the observed resistance to sulfamethoxazole-trimethoprim in four isolates can be only partly explained. All isolates displayed a positive phenotypic MBL test result and a positive modified Hodge test result.

Conventional epidemiology and genotyping.

The timeline of the hospital stays and the times of isolation of GIM-1-positive S. marcescens strains are illustrated in Fig. S1 (which includes a detailed description). Looking at intra- and interhospital spatiotemporal linkages, we were able to establish one potential transmission chain between patient 1 and patient 3 and from patient 3 to patient 4 in hospital A in 2008. No other overlaps (at the ward level) were found.

Genotyping of all 13 isolates by pulsed-field gel electrophoresis (PFGE) revealed one cluster (PFGE type A) containing 10 isolates and 3 unrelated isolates (types B, C, and D; Table 1).

Phylogeny of GIM-1-positive Serratia marcescens isolates on the basis of whole-genome sequencing (WGS) analysis.

The base core genome of the 13 GIM-1-positive S. marcescens strains consisted of 4,268,119 bp. A total of 147,116 single nucleotide polymorphisms (SNPs) were detected; of those, 48 were parsim informative. The maximum-likelihood (ML) phylogeny of core genome SNPs demonstrated a high degree of genetic relatedness between all strains except SM-ID2, which appeared as an outlier with a different genetic backbone (Fig. 1A). In the following, we refer to these 12 highly related strains as a clonal cluster since they differed in only 198 SNPs and had a cluster core genome length of 5,180,303 bp. An ML phylogeny of the clonal cluster revealed a more detailed relational pattern (Fig. 1B). As expected, strains isolated from the same patient grouped together.

FIG 1.

Maximum-likelihood phylogeny of the core genomes of the 13 GIM-1-harboring Serratia marcescens strains from Germany. Both scale bars show the expected number of changes per site. The asterisks label branches with bootstrap percentages of <90%. he trees are unrooted. (A) Phylogeny of all strains. SM-ID2 proved to be only distantly related to the remaining isolates, which formed a clonal cluster. (B) Phylogeny of the clonal cluster excluding SM-ID2. Accessory genome groups are displayed by different color codes for the strain identifiers, namely, light brown, yellow, and green, showing an overlap the core genome tree clades.

Genomes from the clonal cluster were further investigated regarding their genomic gene structure. A total of 4,938 core genome genes and 624 accessory genome genes were identified. A Ward cluster analysis of the accessory genome revealed 3 subgroups that overlapped clades from the ML phylogeny (Fig. 1B). Subsequently, a BEAST analysis of this data set was performed to infer the evolutionary timeline (Fig. 2). This phylogenetic pattern was similar to that for the ML phylogeny, with strains isolated from the same patients clustering together. Again, accessory genome subgroups overlapped BEAST clades. Isolates SM-ID12 and SM-ID13 were particularly interesting since they formed one accessory genome subgroup, suggesting evolution in the same environment, despite their isolation at quite distantly located hospitals. Generally, the accessory genome subtypes observed are likely to indicate a geographically independent development potentially through the uptake of a gene pattern confined to the individual region, a characteristic that allows estimation of whether clonal strains are derived from the same area, irrespective of their original place of detection. This could turn out to be quite useful when aiming for a reconstruction of transregional transmission chains.

FIG 2.

Bayesian phylogenetic analysis of a clonal cluster including 12 GIM-1-harboring Serratia marcescens strains. The tree is constructed from the alignment of the core genome sequences of the strains in the clonal cluster. Node and branch colors show the known (leaves) and predicted (internal branches) locations of the place of strain isolation. The locations are displayed on a map of North Rhine-Westphalia by using the same location color code used in the tree. While the first strain from that cluster was detected in September 2008, the estimated emergence of the most recent common ancestor (MRCA) marks the temporal origin of the tree in April 1993. Accessory genome groups are indicated by the different colors of the strain identifiers, showing three distinct groups with significant genomic differences regardless of core genome variations. The asterisk labels a branch with a posterior probability of <0.9. NL, The Netherlands; B, Belgium.

A median mutation rate of 4.31 × 10⁻⁷ SNPs per site per year (95% highest posterior density interval [HPD], 1.83 × 10⁻⁷ to 6.97 × 10⁻⁷) or 2.23 SNPs per year (95% HPD, 0.95 to 3.61) was estimated. On the basis of this calculation, the emergence of the most recent common ancestor (MRCA) was dated back to April 1993 (95% HPD, January 1973 to March 2003), about 15 years before the first discovery of a GIM-1-positive S. marcescens strain in humans. Hypermutator phenotypes can severely disturb measurements of evolutionary timelines. Therefore, the DNA mismatch repair genes mutS, mutL, mutH, mutM, and mutY of the clonal cluster were investigated for the presence of variations compared to the sequences of these genes in nonhypermutator strain SM-db11. In SM-ID8, three synonymous nucleotide substitutions were detected in the mutS gene at gene positions 664, 669, and 672, but none were found in the other genes. Furthermore, six indels were present in comparison to the sequence of the SM-db11 reference genome, but all of those were found in isolates of the clonal cluster, thus not providing evidence of an inhomogeneous mutation rate (Table S3).

Genetic environment and localization of bla_GIM-1.

Since mobile genetic elements or their chromosomal integration can be difficult to assemble and assemblies from short reads are prone to errors, the differences in the genomic bla_GIM-1 environment were investigated by mapping high-quality sequencing reads against the sequence of an approximately 26-kb-long bla_GIM-1-carrying plasmid from Enterobacter cloacae (GenBank accession number KC511628). In this Enterobacter isolate, the bla_GIM-1 gene was embedded in the class 1 integron In770 and the Tn21 subgroup transposon Tn6216 of approximately 14 kb. The detailed integron structure of the plasmid with GenBank accession number KC511628 is displayed in Fig. 3A. Three different mapping patterns became apparent (Fig. 3B). For pattern 1 (for isolates SM-ID9 and SM-ID11), all reads covered the reference plasmid, indicating a complete conservation of these sequences within the genomes of both strains. The SM-ID2 genome formed the second pattern. The respective reads covered only about half of the reference plasmid and some parts with only a very low level of coverage, with the latter finding probably reflecting reads associated with distinct genomic regions. In general, this pattern showed only a moderate degree of conservation. While the class 1 integron was fully conserved, the transposon structure was not. The remaining members of the clonal cluster (isolates SM-ID1, SM-ID3 to SM-ID8, SM-ID10, SM-12, and SM-ID13) showed a third pattern in which large regions of the plasmid were conserved. The transposon and integron structures were fully preserved. Of note, the first 9 kb of the sequence of the plasmid with GenBank accession number KC511628 contained the relevant mobile genetic elements, including the transposon containing the class 1 integron and the resistance genes bla_GIM-1, aacA4, aadA1, bla_OXA-2, qacEΔ1, and sul1. All these genes were present in sequence reads of all strains; however, they were present to different degrees of coverage.

FIG 3.

Reconstruction and read mapping analysis of the plasmid with GenBank accession number KC511628. (A) Detailed reconstruction of the GIM-1-harboring plasmid with GenBank accession number KC511628. IR_L, IR_R, and IRt, left-hand inverted repeats, right-hand inverted repeats, and right-hand inverted repeats of the transposons, respectively. IRT_np and IR_i, left-hand inverted repeats of the transposon and integron, respectively. (B) Mapping of sequence reads against the sequence of the plasmid with GenBank accession number KC511628. The innermost ring represents the read for the reference plasmid. The second-innermost ring shows the GC content (black). The three following rings illustrate the coverage of sequence reads from isolates SM-ID1 (pattern 3, red), SM-ID2 (pattern 2, green), and SM-ID9 (pattern 1, gray) over each position in the reference plasmid. The chosen sequences are representative for each pattern. The height of the three rings represents the coverage depth. Regions with coverage of more than 1 standard deviation from the mean coverage from the analysis are shown in blue. Gaps are represented by white space. In pattern 1, the whole plasmid is covered, while there are large gaps within the second half of the plasmid in pattern 2, with only very low coverage being seen at some positions. Pattern 3 represents an intermediate coverage, showing a generally high coverage but with five major gaps in the second half of the plasmid. The outermost ring highlights annotated regions from the plasmid with GenBank accession number KC511628. This illustration was generated using the BRIG program.

A PlasmidFinder and a subsequent BLASTn comparison of the sequences with those of reference plasmids R69 (GenBank accession number KM406488) and R471 (GenBank accession number KM406489) (16) found evidence for an IncM plasmid fragment that was located on the bla_GIM-1-containing contig from SM-ID2 (99% sequence similarity), supporting previous results of S1 nuclease restriction analysis and in-gel detection with a ³²P-radiolabeled bla_GIM-1specific probe that demonstrated that the gene was located on a 140-kb plasmid in the strain (7). In the other 12 strains, we were not able either to demonstrate in silico plasmid structures with PlasmidFinder or to demonstrate the presence of the bla_GIM-1 gene on a plasmid by S1 nuclease analysis. In vitro conjugation experiments were not successful for any of the 13 strains.

Phylogenetic context of GIM-1-positive Serratia marcescens strains.

The sequences of 16 full or draft genomes from nine different countries were phylogenetically analyzed by comparison with those of isolates SM-ID9 (representative of the clonal cluster) and SM-ID2 to gain further insights into the potential worldwide distribution of both GIM-1-positive lineages (Table S4). A core genome with a length of 3,474,204 bp was constructed from these 18 strains. A total of 406,654 SNPs were detected. The mean ratio of SNPs within the predicted regions of recombination over SNPs outside such regions (r/m ratio) was 0.34 (standard deviation, ±0.44), indicating that a significant level of exchange of genetic material contributes to the overall evolution of S. marcescens lineages. The two lineages described in our study were identified to be members of two separate phylogenetic clusters (Fig. S2). SM-ID2 is part of a small distinct cluster with CAV1492, a multidrug-resistant strain isolated from the respiratory tract of a patient in Virginia, USA. The strain carries a bla_KPC-2 gene on the plasmid pKPC_CAV1492 (European Nucleotide Archive [ENA] accession number CP011639). Of note is the high number of strain-specific blocks of recombination in both strains from this cluster. These blocks could have recently been acquired in that particular phylogenetic branch and could thus indicate a highly adaptive strain lineage that might regularly exchange genetic material. SM-ID9 is part of a larger cluster and is closely related to strain SM39, which was recovered from the blood of a patient in Japan (17) and which carries the metallo-beta-lactamase gene bla_IMP-1 on plasmid pSMC1 (ENA accession number AP013064). This underlines the lineage's ability to gather and potentially exchange antimicrobial resistance genes.

DISCUSSION

The global spread of genes, such as those for carbapenemases, that confer broad-spectrum antimicrobial resistance constitutes a major threat to modern medicine (5). The metallo-beta-lactamase GIM-1 is, at this stage, mainly restricted geographically to a confined region in western Germany and represents a serious problem in hospitals in that area (6, 7, 9–12, 18). However, bla_GIM-1 has also been rarely reported from other regions of Germany (10, 19).

As more detailed epidemiological data are therefore needed, we investigated possible inter- and intrahospital transmissions of 13 GIM-1-positive S. marcescens strains isolated from 10 patients and six medical centers and provide data indicating that all isolates except one are highly related and exist in one clonal cluster. Conventional epidemiological data analysis for detection of spatiotemporal linkages of patient hospital admissions provides evidence of transmission between three patients in hospital A (between patients 1, 3, and 4) in 2008, though the strain in patient 4 was first detected in hospital D in 2008 and was detected in 2010 only in hospital A. This was confirmed by molecular typing, which showed that the isolates had both the same PFGE pattern and a distinct clustering in the ML phylogeny on the basis of WGS. Also interesting in this regard is the situation in hospital B, where patients 5 and 6 could not be epidemiologically linked. Whether a transmission between the patients occurred, for instance, through an unknown intermediate carrier, could not be resolved by PFGE, which showed that the PFGE patterns for the isolates from both patients were the same as those for most members of the clonal cluster. However, WGS-based core genome analysis illustrated that the variants of the strains from the two patients were fairly different, even though the strains belonged to the same clonal cluster (Fig. 1). This led to two conclusions: (i) that WGS-based methods have a higher discriminatory power for the typing of S. marcescens strains than PFGE and that this may be relevant for a detailed analysis of transmission. It has previously been shown for other bacterial species that WGS has a higher genomic resolution than PFGE (20). (ii) Although the possibility of a long-lasting presence and diversification of the clonal cluster strain within hospital B and, hence, a potentially indirect transmission cannot be fully excluded as an explanation for the observed phylogenic differences between the strains from the two patients, the combination of epidemiological and sequence data is highly indicative of the import of both strain variants into hospital B rather than in-hospital transmission, thus providing evidence for the circulation and evolution of such strains either in the community or in other hospitals in which the patients were previously admitted.

Despite the lack of hard evidence for a relevant degree of circulation of GIM-1-positive S. marcescens isolates within the community, a more detailed picture evolved when the sequence data for all 12 strains from the clonal cluster were analyzed. The occurrence of different clades in the core genome ML phylogeny as well as three accessory genome types from strains isolated in different regions within western Germany points to an independent strain evolution, potentially through the exchange of distinct genes in different environments. This supports the hypothesis that a strain is circulating within the community, though its magnitude is difficult to estimate.

Of note, in-hospital transmission was not the only mode of spread. WGS analysis of the SM-ID2 isolate from patient 2, who also stayed at hospital A, demonstrated an S. marcescens strain that was genetically very distant from the strains of the clonal cluster. The finding that the genomic environment of the bla_GIM-1 gene of SM-ID2 was similar to that of the clonal cluster with regard to the integron structure suggests that the spread of the bla_GIM-1 gene is not entirely due to the transmission of a successful outbreak strain but is also due to the transmission of mobile genetic elements containing bla_GIM-1, probably even between different species, since this particular integron (In770) was previously demonstrated in Klebsiella oxytoca and E. cloacae (7). The previous demonstration of a unique bla_GIM-1-carrying plasmid in SM-ID2 using an S1 nuclease approach and the demonstration here of an IncM plasmid fragment on the contig that harbored bla_GIM-1 further support this assumption (7). Such IncL/M plasmids have previously been described to carry MBL genes, e.g., the bla_IMP-34 gene in a K. oxytoca strain from Japan (21) and a bla_NDM-1 gene in a Klebsiella pneumoniae strain from the Sultanate of Oman (22). Horizontal gene transfer as a mode of transmission has the potential to aggravate the spread of GIM-1, possibly leading to a more widespread geographical distribution. While the findings from a verification of actual bla_GIM-1-carrying plasmid structures in the clonal cluster were less conclusive, there is also a good chance that a horizontal transmission event had previously taken place for that strain type due to the high similarity of its bla_GIM-1 gene-bearing transposon and integron environment with those of the plasmid with GenBank accession number KC511628 in all 12 isolates.

Another interesting aspect is the evolution of the clonal cluster over time and its extrapolation to the past. Since such an analysis might be greatly affected by the presence of hypermutating strains in the data set, we excluded this possibility by investigating the mismatch repair genes. The Bayesian phylogenetic analysis estimated the occurrence of the most recent common ancestor to go back to 1993. It is tempting to ask whether GIM-1 was already present in the bacterium at that time. Figure 2 shows that the divergence of SM-ID9 and SM-ID11 from the rest of the clonal cluster occurred at the time of MRCA emergence and during the turn of the millennium, respectively, and that both strains have developed independently since then, as seen by their sequence differences (ML analysis; Fig. 1) and distinct PFGE patterns (Table 1). If GIM-1 had not been present in these strains at that time, it must have been acquired more recently, most likely from strains of the same clonal cluster. Considering the possibility of a recent transfer, one would assume that the genetic environment of the bla_GIM-1 gene would have a high degree of similarity with that of one of the other members of the cluster. However, we have found two different mapping patterns in the cluster. Of note, the two strains SM-ID9 and SM-ID11, which diverged early from the rest of the clonal cluster, had a distinct mapping pattern. The mapping pattern of the rest of the cluster has, interestingly, been stable since the emergence of that branch in approximately 2002. Considering these observations, a recent transfer can be regarded to be unlikely, unless SM-ID9 and SM-ID11 acquired GIM-1 from another species or strain type during their independent development. However, only one GIM-1-positive S. marcescens isolate (SM-ID2) that is likely to have experienced a horizontal gene transfer over the study period has been found.

If GIM-1 was already present in the MRCA, these strains could have circulated within the community since the early 1990s and have remained undetected until recently. Due to the low MICs of carbapenems for some GIM-1-positive S. marcescens strains (resulting in nondetection of the resistance gene in isolated organisms) and the relatively low prevalence of S. marcescens as a cause of nosocomial infection, this is not an unlikely scenario. Lower MICs of imipenem and meropenem in the susceptible or intermediate range were previously described for GIM-1-harboring bacteria (6, 9, 10, 18). Nevertheless, even the study isolates demonstrating susceptibility to carbapenems would have not been missed, if surveillance were based on the EUCAST recommendations (23). Taken together, these findings reflects the necessity of a comprehensive hospital-based molecular surveillance for carbapenem-nonsusceptible strains to detect these threats and to implement infection control measures as early as possible before resistance elements and their host strains become widely disseminated. This demand becomes even more relevant because of the fact that both GIM-1-positive S. marcescens lineages in our study had a high degree of genetic relatedness to multidrug-resistant clinical isolates from Japan and the United States, underlining the potential of these lineages to cause severe infections and to gather genetic material that renders them nonsusceptible to commonly used antibiotics.

Our study has some limitations. It is impossible to be sure as to the date from which the MRCA has circulated in Germany. It might have existed in another country before and imported into Germany at a time point later than 1993. However, the divergence of SM-ID11 from the rest of the clonal cluster approximately during the turn of the millennium would support the hypothesis of a spread within the German population. Furthermore, the number of GIM-1-positive S. marcescens isolates might have been underestimated due to the lack of standard patient rectal screening and molecular strain analysis procedures. The true prevalence of carriage in the German population is difficult to assess, as rectal screening for multidrug-resistant Gram-negative organisms is recommended only for high-risk patients (24). However, in Germany the prevalence of carbapenemase-producing organisms is still very low. In a recent multicenter study of 4,376 patients, rectal carriage strains carrying VIM-1, NDM-1, and IMP-8 were detected in only 0.11% of patients on admission (25). Nevertheless, we are confident that the current study represents a fairly realistic and comprehensive picture of the epidemiology in the region studied, as the participating laboratories (including the reference center) provide microbiological services for large parts of the region in which GIM-1 has been detected and where a strict program for molecular screening for carbapenemases has been established. Finally, we were not able to resolve the exact genetic location of the bla_GIM-1 gene in the isolates of the clonal cluster, in which it seems to be of chromosomal origin. We were faced with inconclusive results by S1 nuclease analysis for isolate SM-ID4, in which bla_GIM-1 was previously reported on a small plasmid (6). Since S1 nuclease digestion and subsequent probing should detect small plasmids, like the plasmid with GenBank accession number KC511628, we suggest that the plasmid previously described either is not very stable during cultivation and storage or is difficult to detect by this method due to DNA degradation, known to occur in S. marcescens (28). Also, we were not able to determine the full sequence of the plasmid in SM-ID2. Future studies using long-read technologies will address this aspect.

In conclusion, the occurrence of GIM-1 in S. marcescens is rare and regionally restricted at present. The data presented here show a mostly clonal expansion of GIM-1-producing S. marcescens strains isolated in western Germany at different hospitals over a period of 8 years and may, in addition, herald a rise of GIM-1 in the community since the early 1990s.

MATERIALS AND METHODS

Isolate collection, identification, and susceptibility and phenotypic testing.

Clinical and screening isolates of the Enterobacteriaceae family with nonsusceptibility to carbapenems (imipenem and/or meropenem) were collected at the Institute of Medical Microbiology and Hospital Hygiene, Düsseldorf, Germany, from 2007 onwards and tested for the presence of carbapenemases. The institute provides a diagnostic service for a tertiary care center and two secondary care centers in and around Düsseldorf, the capital of the state of North Rhine-Westphalia in western Germany. Prior to 2013, interpretation of the susceptibilities of the clinical isolates was based on CLSI standards; thereafter, it was according to EUCAST methods and breakpoints, which recommends further testing of isolates with reduced susceptibility to carbapenems (a meropenem MIC of >0.12 μg/ml and an imipenem MIC of >1 μg/ml) (23). Carbapenemase detection was performed by real-time PCR (for detection of bla_IMP-1, bla_VIM-1-like, bla_VIM-2-like, bla_GIM-1, bla_NDM, bla_GES, bla_KPC, and bla_OXA-48) (7, 26) and phenotypically by an MBL combination disc test using imipenem with or without EDTA (12). Furthermore, the modified Hodge test was performed as previously described (27).

Additionally, GIM-1-producing S. marcescens isolates from two other participating laboratories were included: the Medizinisches Versorgungszentrum Dr. Stein, Mönchengladbach, Germany (where the inclusion criteria for carbapenemase screening is an imipenem and/or meropenem MIC of ≥1 μg/ml), which services more than 50 hospitals mostly in North Rhine-Westphalia, and the National Reference Centre for Multidrug-resistant Gram-negative Bacteria (Department of Medical Microbiology, Ruhr University Bochum), which receives carbapenem-nonsusceptible Gram-negative organisms from all over Germany. Previously published isolates SM-ID4 (6) and SM-ID2 and SM-ID3 (7) were included.

Identification was carried out using standard microbiological procedures, including tests with a Vitek 2 system and matrix-assisted laser desorption ionization–time of flight mass spectrometry (bioMérieux, Germany). Antibiotic susceptibility testing was performed with the Vitek 2 system (bioMérieux, Germany), and the MICs (MIC test strip; Liofilchem, Italy) of the following antimicrobial agents were obtained: piperacillin, piperacillin-tazobactam, ceftriaxone, ceftazidime, cefepime, imipenem, meropenem, ertapenem, aztreonam, gentamicin, tobramycin, amikacin, ciprofloxacin, levofloxacin, sulfamethoxazole-trimethoprim, and tigecycline. EUCAST breakpoints (version 6.0, 2016) were used for interpretation of susceptibility.

Conjugation experiments.

Conjugation experiments were carried out using both the donor bla_GIM-1 strains and the recipient strain (sodium azide-resistant Escherichia coli J53) on sheep blood agar at a recipient/donor ratio of 1:1. Thereafter, selection was carried out on medium containing 0.125 μg/ml meropenem and 100 μg/ml sodium azide. E. coli isolates growing on the selective medium were further screened as probable transconjugants for bla_GIM-1. As a positive control, we used a GIM-1-positive Citrobacter freundii strain (strain 2157) with a conjugative plasmid described previously (18).

S1 nuclease restriction and hybridization.

S1 nuclease restriction and in-gel detection using a ³²P-radiolabeled bla_GIM-1-specific probe were performed as previously described (7). E. cloacae strain M15, described to have an approximately 25-kb plasmid (GenBank accession number KC511628), was used as a positive control (7).

Conventional genotyping.

Genotyping of all S. marcescens isolates was carried out by PFGE after XbaI (Fermentas) restriction under the following conditions: 6 V/cm² for 20 h with pulse times of 5 s to 50 s. To overcome DNA degradation, two modifications were introduced: heating of the bacteria in EDTA and addition of thiourea to the buffer (28). Strain relatedness was assessed using the BioNumerics Tree and Network Inference module (version 6.5) with the band-based Dice similarity coefficient and the unweighted pair geometric matched analysis dendrogram (band matching tolerance, 1%; optimization, 0.5%). The cutoff value to define a PFGE cluster was set at 90%.

Setting and epidemiological investigations.

The University Hospital, Düsseldorf (hospital A), is a 1,200-bed tertiary care center. All other hospitals are secondary care centers. Hospitals B, C, D, and E are within a 30-km radius of hospital A, and hospital F is approximately 70 km from Düsseldorf. Epidemiological data for the affected patients were collected to detect potential transmission events.

Whole-genome sequencing analysis. (i) Genomic DNA extraction, library preparation, and whole-genome sequencing.

Genomic DNA from 13 GIM-1-positive S. marcescens strains was extracted using a QIAamp DNA minikit (Qiagen, Hilden, Germany) and sheared by use of a M220 focused ultrasonicator (Covaris, Woburn, MA, USA) to obtain 550-bp fragments. DNA libraries were prepared with a TruSeq Nano DNA LT kit (Illumina, San Diego, CA, USA) using the standard protocol. The quality of the barcoded libraries was assessed using a QIAxcel Advanced instrument (Qiagen, Hilden, Germany). All libraries were sequenced on an Illumina MiSeq sequencer (Illumina, San Diego, CA, USA) at 2 × 250 bp. High-quality reads were attained using a Trimmomatic trimming tool (29), with a mean of 1,294,548 read pairs being obtained per sample. Quality control was carried out with the FastQC tool (http://www.bioinformatics.babraham.ac.uk/projects/fastqc/).

(ii) Genome assembly, core genome identification, and variant calling.

The A5-MiSeq pipeline was run as an assembly tool (30). The progressiveMauve system (version 2.3.1) was selected to construct a full alignment of the 13 S. marcescens genomes using default settings (31). Locally colinear blocks (LCBs) of less than 1 kb in length were removed, and the remaining LCBs were framed into a core genome alignment. For the detection of single nucleotide polymorphisms (SNPs), one study isolate, SM-ID1, was chosen to be the reference isolate. Reads were mapped against the sequence of the reference isolate by use of the Burrows-Wheeler aligner (BWA) (32), and core genome SNPs were subsequently called by use of the SAMtools package (33). SNPs were required to have a minimum SNP quality of 30 and a minimum root mean square mapping quality of 40 (Phred scale).

(iii) Core recombination and phylogenetic analysis.

Variations based on recombination events were removed prior to phylogenetic analysis. The presence of regions of recombination within the core genome was investigated using BratNextGen software (34) with 100 permutations for significance estimation. Prophage regions were identified using the PHAST program (35).

Maximum-likelihood phylogenetic trees were constructed by use of the RAxML tool (version 8.2.8) (36) and the general time reversible (GTR) model as a nucleotide substitution model and the gamma distribution (G) to describe rate variation among sites. Branch support was estimated by the use of 1,000 bootstrap replicates.

To further investigate the regional dissemination and divergence dates, the BEAST package (version 1.8) was applied (37). A continuous-time Markov chain Monte Carlo method was utilized with a setup of 300 million steps. Samples were logged every 30,000th step. GTR + G was used as a substitution model, with a constant size coalescent being used as the tree prior. The location of strain isolation was investigated as discrete trait, while tip dates were described as the month and year of isolation, and a strict molecular clock was assumed. Parameter estimation was performed using Tracer software (version 1.6) after a 10% burn-in removal. The effective sample size was >7,000 for all parameters. The maximum clade credibility tree was constructed from 9,000 trees by use of the TreeAnnotator program (version 1.8) (37).

The sequences of SM-ID9, as a member of the clonal cluster, and SM-ID2 were further phylogenetically compared to full and draft genome sequences available from the NCBI GenBank database or the European Nucleotide Archive (date of search, 20 October 2016). Clinical strains were included when patient material and the location of detection were documented. A core genome alignment was built using the progressiveMauve system (version 2.3.1, default settings) (31). The Gubbins program (version 2.1.0) was used to identify genomic regions that had undergone homologous recombination. Subsequently, a maximum-likelihood phylogeny was constructed on the basis of mutations outside these regions using a maximum of 10 iterations and a GTR substitution model with a gamma distribution of rates (38).

(iv) Accessory genome analysis.

Differences in the accessory genome within the clonal cluster were examined by the generation of a presence-absence gene matrix using the Roary program (39). Accessory genes from each strain were extracted from this matrix, and a cluster analysis by the Ward method using the Pearson correlation coefficient was conducted by the use of Stata software (version 12.1; Stat Corp., College Station, TX, USA).

(v) Hypermutator analysis.

Divergence date estimation can be severely biased by the presence of hypermutator strains. The complete genome of strain SM-db11 (GenBank accession number HG326223) (17), an S. marcescens strain not reported to be a hypermutator, was used as a reference. SNP calling of the sequences of all strains from the clonal cluster against the sequence of SM-db11 was performed as described above. Indel calling was performed by use of the GATK software package according to the best practice recommendations (40). The concordance of the sequence of DNA mismatch repair genes from all strains from the clonal cluster with the sequence of both SM-db11 and the sequences of the strains within the clonal cluster was assumed to reflect a nonhypermutator phenotype.

(vi) bla_GIM-1 genomic environment investigation.

The genomic environment of bla_GIM-1-harboring contigs was annotated using Prokka software (41). Automated annotation was supplemented by manual BLAST approaches. Additionally, BWA was applied for mapping high-quality sequence reads against the sequence of the only fully sequenced bla_GIM-1-harboring plasmid that was originally extracted from an Enterobacter cloacae strain M15 (described by Wendel et al.; GenBank accession number KC511628). A minimum root mean square mapping quality of 30 (Phred scale) was used for the mapping procedure. Visualization of mapping was done using the BLAST ring image generator (BRIG) (42) and IGV (43) programs. A search for plasmid structures was performed by using the PlasmidFinder program (44). When plasmid structures were detected, the sequences were extracted and mapped against the sequences of the strain assemblies (BLASTn) to see whether the bla_GIM-1-harboring contigs contained these sequences.

(vii) Resistome analysis.

Acquired resistance genes on the assembled sequences were identified by use of the ResFinder program (version 2.1; with a threshold of 98% identity and a minimum length of 60%) (45). This search was augmented by mapping the sequence reads against the sequences in the ARG-ANNOT database (46) using the DIAMOND program (maximum expected value, 10⁻⁵; query cover, 75%) (47).

Accession number(s).

The sequence reads of all strains have been deposited as a project at the European Nucleotide Archive (ENA) under accession number PRJEB15351.