KPC-3-, GES-5-, and VIM-1-Producing Enterobacterales Isolated from Urban Ponds

Carbapenems are antibiotics of pivotal importance in human medicine, the efficacy of which is threatened by the increasing prevalence of carbapenem-resistant Enterobacterales (CRE). Urban ponds may be reservoirs of CRE, although this hypothesis has been poorly explored. We assessed the proportion of CRE in urban ponds over a one-year period and retrieved 23 isolates. These were submitted to BOX-PCR, PFGE, 16S rDNA sequencing, antibiotic susceptibility tests, detection of carbapenemase-encoding genes, and conjugation assays. Isolates were affiliated with Klebsiella (n = 1), Raoultella (n = 11), Citrobacter (n = 8), and Enterobacter (n = 3). Carbapenemase-encoding genes were detected in 21 isolates: blaKPC (n = 20), blaGES-5 (n = 6), and blaVIM (n = 1), with 7 isolates carrying two carbapenemase genes. Clonal isolates were collected from different ponds and in different campaigns. Citrobacter F6, Raoultella N9, and Enterobacter N10 were predicted as pathogens from whole-genome sequence analysis, which also revealed the presence of several resistance genes and mobile genetic elements. We found that blaKPC-3 was located on Tn4401b (Citrobacter F6 and Enterobacter N10) or Tn4401d (Raoultella N9). The former was part of an IncFIA-FII pBK30683-like plasmid. In addition, blaGES-5 was in a class 3 integron, either chromosomal (Raoultella N9) or plasmidic (Enterobacter N10). Our findings confirmed the role of urban ponds as reservoirs and dispersal sites for CRE.


Introduction
The spread of antimicrobial resistance and the upsurge in multidrug-resistant bacteria over the last decades led to an unprecedented impact on public health systems all around the globe, causing an ever-rising death toll accompanied by a significant economic impact [1][2][3]. Moreover, therapeutic options used to treat infectious diseases are becoming increasingly scarce, and even last-resort antibiotics are losing their efficacy [4].
Carbapenems, which are among the most active and potent agents against multidrugresistant Gram-negative pathogens, are becoming threatened by the global dissemination of carbapenem-resistant bacteria, in which carbapenem-resistant Enterobacterales (CRE) play an important role [5]. As a result, the World Health Organization (WHO) has included CRE in the list of critical targets for research and development of new antibiotics [6]. The two main mechanisms behind carbapenem resistance in CRE consist of the expression of cephalosporinases combined with mutation-derived permeability defects and, even more important, in carbapenemase production, enzymes that are able to efficiently hydrolyze carbapenem antibiotics [7]. KPC appears to dominate over other carbapenemases in Portugal, being associated with significant outbreaks of both KPC-2 and KPC-3 variants [8][9][10]. Carbapenemase-producing bacteria have also been isolated from several different Portuguese environmental settings, namely river water (IMP-, VIM-, Metricel ® ) with a 0.45 μm pore size. The filter membranes were then transferred to m-Fecal Coliform agar (m-FC) (VWR) plates supplemented with imipenem (4 μg/mL) and incubated at 37 °C for 18 h. To estimate the proportion of imipenem-resistant bacteria, m-FC without antibiotic was used. Coliform colonies were counted, and imipenem-resistant colonies were purified and stored in 20% glycerol at −80 °C.

Molecular Typing and Identification
Whole-cell suspensions were used as the DNA template for each isolate in all PCR analysis. Briefly, one bacterial colony of each isolate was resuspended in 20 μL of sterile distilled water, and 1 μL of this suspension was used as template in each PCR reaction. BOX-PCR (primers and conditions as described in Tacão et al., 2012 [38]) was used as a first discriminatory molecular-typing method for imipenem-resistant isolates. Isolates recovered from each sampling site exhibiting different BOX-profiles were selected for amplification and analysis of the 16S rRNA gene using the primers listed in Table S1. Additionally, considering the lack of accuracy of most phenotypic tests in the identification of Raoultella species, all isolates affiliated with the genus Raoultella were further identified using matrix-laser desorption ionization mass spectrometry (Bruker MALDI Biotyper IVD, Portugal) in order to obtain a reliable species identification [39].
CRE isolates with different BOX profiles or collected in different sampling campaigns and from different ponds were further analyzed by PFGE of XbaI-digested DNA, using the CHEF-DR II System (Bio-Rad Laboratories, Hercules, CA, USA) as previously described [40]. PFGE patterns were analyzed using GelCompar II software (v 6.1) with the criteria established by Tenover and colleagues [41].

Molecular Typing and Identification
Whole-cell suspensions were used as the DNA template for each isolate in all PCR analysis. Briefly, one bacterial colony of each isolate was resuspended in 20 µL of sterile distilled water, and 1 µL of this suspension was used as template in each PCR reaction. BOX-PCR (primers and conditions as described in Tacão et al., 2012 [38]) was used as a first discriminatory molecular-typing method for imipenem-resistant isolates. Isolates recovered from each sampling site exhibiting different BOX-profiles were selected for amplification and analysis of the 16S rRNA gene using the primers listed in Table S1. Additionally, considering the lack of accuracy of most phenotypic tests in the identification of Raoultella species, all isolates affiliated with the genus Raoultella were further identified using matrixlaser desorption ionization mass spectrometry (Bruker MALDI Biotyper IVD, Portugal) in order to obtain a reliable species identification [39].
CRE isolates with different BOX profiles or collected in different sampling campaigns and from different ponds were further analyzed by PFGE of XbaI-digested DNA, using the CHEF-DR II System (Bio-Rad Laboratories, Hercules, CA, USA) as previously described [40]. PFGE patterns were analyzed using GelCompar II software (v 6.1) with the criteria established by Tenover and colleagues [41].

Mating Assays and Plasmid Analysis
Mating assays were performed for CRE harboring carbapenemase genes that exhibited unique PFGE patterns, as previously described [47]. In short, donor strains and the sodiumazide-resistant strain E. coli J53 (recipient strain) were grown overnight in Luria-Bertani broth (LB) at 37 • C and 180 rpm. Donor and recipient strains were mixed at a 1:1 ratio on solid media using plate count agar (PCA) and left overnight at 37 • C. Putative transconjugants were selected on PCA plates supplemented with imipenem (4 µg/mL) and sodium azide (100 µg/mL). Transconjugants identity was confirmed with BOX-PCR typing and PCR confirmation of the acquired carbapenemase gene. Minimum inhibitory concentration (MIC) gradient strips (Liofilchem, Italy) were used to evaluate the susceptibility of donors, transconjugants, and the recipient strain to imipenem, meropenem, ertapenem, ceftazidime, and cefotaxime, following the EUCAST MIC clinical breakpoints [42].
Plasmid DNA was extracted using the NZYMiniprep kit (NZYTech, Portugal) following the manufacturer's protocol and analyzed by electrophoresis. Plasmid replicon typing (PBRT) was performed to detect replicons belonging to 18 incompatibility groups, as previously described [48]. We also evaluated the presence of pBK30661-and pBK30683-like plasmids, as described before [49].

Whole-Genome Sequence Analysis
Three CRE isolates were selected for whole-genome sequence (WGS) based on molecular typing and both antibiotic-resistance phenotype and genotype. Genomic DNA was extracted from each isolate using the Wizard ® Genomic DNA Purification Kit (Promega, Madison, WI, USA) according to the manufacturer's protocol. Paired-end libraries were created using the Illumina HiSeq 2500 platform. The raw reads quality was verified using FastQC software and submitted to the trimming process in order to exclude those that contained a phred quality score below 20 using Trimmomatic v.0.36 [50]. The genomes were assembled using the SPAdes version 3.14.0 program [51]. Draft genomes were annotated using Rapid Annotation using Subsystem Technology (RAST) [52]. The rRNA were predicted using the RNAmmer 1.2 Server [53], and tRNA were predicted using tRNAscan-SE 2.0 [54]. In order to confirm species identification, ANIb and ANIm were calculated by using the JSpeciesWS tool [55], and dDDH was calculated against representative genomes by using the Genome-to-Genome Distance Calculator v2.1 [56]. We also considered the G + C% divergence in the overall genome-relatedness analysis [57]. Sequence types were attributed through a multilocus sequence typing analysis using MLST v2.0 [58] and PubMLST v1 [59]. ARGs were screened using CARD [60], and genomic islands were predicted using the Genomic Island Prediction Software (GIPSy) [61]. In silico screening of plasmid replicons was performed using PlasmidFinder v2.1 [62]. The CRISPR Finder tool [63] and the PHASTER tool [64] were used to assess the presence of clustered regularly interspaced short palindromic repeats (CRISPRs) and prophage sequences. Putative virulence factors were inspected using the Virulence Factors of Pathogenic Bacteria (VFDB) database [65]. The pathogenicities of all three CRE isolates were also accessed using PathogenFinder 1.1 [66].

Prevalence and Diversity of CRE in Urban Ponds
Colony counts on mFC varied between 0.5 and 3.70 log 10 CFU/mL ( Figure S1). Globally, P1 exhibited the highest average mean mFC counts/mL (3.02 log 10 ), while P4 exhibited the lowest (1.05 log 10 ). The February 2019 campaign was associated with overall higher mFC counts/mL, followed by December and March 2019.
Imipenem-resistant bacteria were detected in ponds P1, P2, and P3, and also in the estuarine canal E1. P1 and E1 were the only aquatic systems in which imipenemresistant colonies were detected in all sampling campaigns. On the other hand, in pond P2, imipenem-resistant colonies were detected only in February and March 2019, and in pond P3 only in November 2019. Overall, the percentage of imipenem-resistant CFUs was low (0.021% on average), with the highest percentage (0.16%) occurring in site P1 in September 2019 (Table S2). A collection of 23 CREs was obtained in this study, isolated from sites P1 (n = 19), P3 (n = 2), and E1 (n = 2). Isolates were affiliated with the genera Citrobacter (n = 8), Raoultella (n = 11), Enterobacter (n = 3), and Klebsiella (n = 1) ( Table 1). The MALDI-TOF analysis of the Raoultella isolates showed that they all were affiliated with Raoultella ornithinolytica. The 16S rRNA gene sequence analysis of the imipenem-resistant bacteria found in Pond 2 demonstrated their affiliation with the Shewanella genus, which harbor intrinsic carbapenem-resistance mechanisms [67], and were consequently excluded from the CRE collection.

Antibiotic Susceptibility, ARGs and MGEs Content
The majority of the CRE collection was classified as multidrug resistant (approximately 87%), with the only exceptions being Enterobacter isolates (n = 3). The isolates were resistant to all β-lactams (Table 1), with the exception of the Klebsiella isolate, which was susceptible to cefepime and aztreonam. On the other hand, the CRE isolates were susceptible to tigecycline, with the exception of isolates Citrobacter F1 and F4. Additionally, the Enterobacter isolates were susceptible to all non-β-lactam antibiotics tested, excluding Enterobacter N10, which was resistant to ciprofloxacin. Within the Raoultella genus, all isolates were susceptible to tetracycline, tigecycline, and chloramphenicol.
With the exceptions of Enterobacter N11 and Enterobacter N15, CRE isolates harbored at least one of the inspected carbapenemase-encoding genes ( Table 1); bla KPC was the most frequently identified (20 out of 23), being detected in all Raoultella and Citrobacter isolates, and also in Enterobacter N10. Furthermore, the Enterobacter N10 and Raoultella isolates (n = 5) collected in November 2019 coharbored both bla KPC and bla GES. The sequencing analysis revealed the presence of the bla GES-5 variant, which has been proven to exhibit carbapenemase activity [68]. The bla VIM gene was only detected in Citrobacter N5, which also harbored bla KPC . The single Klebsiella isolate in the collection harbored bla GES-5 . Both bla CTX-M and mcr-1 genes were not detected.
The CRE collection was also inspected for the presence of integrons. The gene IntI1 was detected in all CRE isolates, except for Enterobacter N11, Enterobacter N15, and Klebsiella N14. IntI3 was identified in Klebsiella N14. Moreover, with the exception of Raoultella N1, all isolates that were collected in November 2019 in which both bla KPC and bla GES-5 were detected also coharbored both integrase-encoding genes IntI1 and IntI3. Class 2 integrons were not detected ( Table 1).
The genetic context of the bla VIM gene identified in Citrobacter N5 was determined by PCR and sequence analysis, revealing the presence of this gene in a class 1 integron containing other ARGs encoding resistance to aminoglycosides (aacA4; aadA1), chloramphenicol (catB2), and sulfonamides (sul1). Further PCR analysis also revealed the presence of the bla GES-5 gene cassette as part of a class 3 integron in all isolates in which this gene was detected. Citrobacter P3 Enterobacter H --a Determined by BLAST using 16S rRNA gene. b PRL, piperacillin; TZP, piperacillin-tazobactam; FEP, cefepime; CTX, cefotaxime; CAZ, ceftazidime; IPM, imipenem; ETP, ertapenem; MEM, meropenem; ATM, aztreonam; TE, tetracycline; TGC, tigecycline; AK, amikacin; CIP, ciprofloxacin; C, chloramphenicol; SXT, trimethoprim-sulfamethoxazole; CN, gentamicin (dark grey, resistant; white, susceptible). c BOX profiles are designated with letters and PFGE profiles with Roman numerals; A hyphen is displayed whenever no replicon, carbapenemase gene, and integrase gene were detected.

CRE Clonal Relatedness
Clonal relationships among the CRE isolates were first assessed using BOX-PCR (Table 1). Among the Citrobacter genus, the same BOX profile was observed within a group comprising isolates collected in February 2019 and N6 (collected in November 2019). In regard to the Enterobacter genus, both isolates N11 and N15, collected from Pond 3, shared the same BOX profile; while isolate N10, collected from Pond 1, exhibited a singular BOX profile. Within the Raoultella genus, two clusters exhibiting two distinctive BOX profiles were observed, one composed of all Raoultella isolates collected from Pond 1 in September and October 2019, and the other composed of all Raoultella isolates collected in November 2019 from sampling sites P1 and E1. Although we adopted the criteria of different BOX profiles and/or sampling campaigns/sites in the selection of CRE for PFGE analysis, isolates obtained from sampling site E1 (N12 and N13) were both selected due to the PCR detection of pBK30683-like plasmid exclusively on Raoultella N12.
The PFGE analysis further divided Raoultella isolates into two different clusters ( Figure S2). One of the clusters comprised isolates collected in November 2019, with isolate N12 exhibiting a slightly different PFGE profile. The other cluster was composed of Raoultella isolates collected in October 2019 and isolate S1 from the September campaign. Regarding the Citrobacter genus, PFGE analysis resulted in two singletons represented by N5 and F5 ( Figure S2), and one cluster comprising two separate clades with isolates collected in February and November. Together, the results obtained with BOX and PFGE confirmed the presence of clonal isolates in different ponds (Raoultella N1, N8, and N9 in pond P1 and Raoultella N12 in pond E1) and collected in different campaigns (Raoultella S1 and O1 in September and October 2019, respectively; Raoultella N1 in November 2019 and Raoultella N8, N9, and N12 in November 2020; and finally Citrobacter F1 and N6 in February 2019 and November 2020, respectively).

Plasmid Analysis and Transfer of Carbapenem Resistance
The presence of plasmids was evaluated in all CRE isolates ( Figure S3), and 10 distinct plasmid profiles were identified. Within the Citrobacter genus, four distinctive plasmid profiles were detected. Regarding the Raoultella genus, all isolates collected in November exhibited the same plasmid profile. Raoultella O5 exhibited a singular plasmid profile, while Raoultella O1 and all isolates collected in September shared the same plasmid profile. Among the Enterobacter isolates, two distinct plasmid profiles were identified. The single Klebsiella isolate (N14) also showed a unique plasmid profile.

WGS Analysis
Three CRE isolates representing the three most represented genera were selected for WGS and further analysis: Citrobacter F6, R. ornithinolytica N9, and Enterobacter N10. The general genomic features of all three genomes are described in detail in Table S3.
The C. freundii F6 draft genome was 5,117,230 bp in size and organized in 230 contigs, with 51.7% in terms of G + C content, an N50 of 88,228, and 4997 predicted coding sequences (Table S3). C. freundii F6 belonged to the sequence type ST270 (Table S3), and based on WGS analysis, a total of 142 virulence factors were predicted; they were distributed in a total of 16 virulence categories, with most allocated to secretion systems (n = 33), antiphagocytosis (n = 22), and adherence (n = 17) functions ( Figure S5).
In fact, a BLAST search revealed that the contig in which qnrS1 was located (3259 bp in size) exhibited 100% similarity in terms of nucleotide identity with part of a 249.6 Kb plasmid previously described in a Salmonella enterica isolate (CP039857). Additional ARGs, namely bla OXA-1 (encoding β-lactam resistance), catB3 (chloramphenicol resistance), aacA4cr (aminoglycoside resistance), mphA (encoding macrolide resistance), and sul1 (encoding sulfonamide resistance) ( Table 3) were found within a class 1 integron (In1387) previously described in the IncN3 broad-host-range plasmid pTRE-132 [70], exhibiting the following gene cassette array: 5 CS IntI1|aacA4-cr|bla OXA-1 |catB3|qac∆1|sul1. Botts and colleagues captured the pTRE-132 plasmid from wetland bacteria into an E. coli recipient strain that was able to further transfer it through conjugation, resulting in resistance to ampicillin and ticarcillin, as well as decreased susceptibility to other classes of antibiotics (fluoroquinolones, chloramphenicol, aminoglycosides, and sulfonamides) [70]. This plasmid also harbors other ARGs, namely qnrB and an additional copy of sul1 [70]. In terms of MGEs, six replicons were detected in C. freundii F6, namely pKPC-CAV1321, IncFIA (HI1), IncFII (K), IncM1, and IncN (Table 3). Further genomic analysis led to the identification of 14 putative pathogenicity islands, 6 putative resistance islands, 2 CRISPR sequences, and 5 intact prophage regions ( Figure S4). The sequence analysis of these genomic islands identified coding sequences that were part of a mercury-resistance operon, a type IV secretion system, and multidrug-efflux transporters responsible for the transport of metals.
Next, bla KPC-3 was identified by GIPSy as being part of resistance island RI5 ( Figure S4A). In addition to bla KPC-3 , RI5 also harbored a gene coding for a multidrug-efflux transporter. The carbapenemase-encoding gene was flanked by two insertion sequences, ISKpn6 and ISKpn7, which in turn were part of a Tn4401b transposon (Figure 2).
The Tn4401b isoform was confirmed, as no deletions were observed in the 386 bp region [71]. However, both ISKpn6 and ISKpn7 were fragmented (Figure 2). The bla KPC-3harboring contig (10,172 bp) revealed the same nucleotide sequence (100% identity; query coverage of 98%) with the corresponding region of an IncN plasmid described in a clinical E. coli strain (KU295134), and 99.7% of nucleotide sequence similarity (query coverage of 97%) with one other IncN plasmid described in several K. pneumoniae clinical strains [72], and also with a IncFIB/FII plasmid identified in a K. pneumoniae clinical strain (CP054266) [73]. these genomic islands identified coding sequences that were part of a mercury-resistance operon, a type IV secretion system, and multidrug-efflux transporters responsible for the transport of metals.
Next, blaKPC-3 was identified by GIPSy as being part of resistance island RI5 ( Figure  S4A). In addition to blaKPC-3, RI5 also harbored a gene coding for a multidrug-efflux transporter. The carbapenemase-encoding gene was flanked by two insertion sequences, IS-Kpn6 and ISKpn7, which in turn were part of a Tn4401b transposon (Figure 2). The Tn4401b isoform was confirmed, as no deletions were observed in the 386 bp region [71]. However, both ISKpn6 and ISKpn7 were fragmented ( Figure 2). The blaKPC-3-harboring contig (10,172 bp) revealed the same nucleotide sequence (100% identity; query coverage of 98%) with the corresponding region of an IncN plasmid described in a clinical E. coli strain (KU295134), and 99.7% of nucleotide sequence similarity (query coverage of 97%) with one other IncN plasmid described in several K. pneumoniae clinical strains [72], and also with a IncFIB/FII plasmid identified in a K. pneumoniae clinical strain (CP054266) [73].

R. ornithinolytica N9
The R. ornithinolytica N9 draft genome was 6,252,962 bp in size and organized in 420 contigs, with 55.2% in terms of G + C content, an N50 of 109,201, and 6014 predicted coding sequences (Table S3). A total of 130 virulence factors were predicted for R. ornithinolytica N9 across 12 virulence categories, with adherence (n = 44), iron uptake (n = 34), and secretion systems (n = 20) being the most represented ( Figure S5).

R. ornithinolytica N9
The R. ornithinolytica N9 draft genome was 6,252,962 bp in size and organized in 420 contigs, with 55.2% in terms of G + C content, an N50 of 109,201, and 6014 predicted coding sequences (Table S3). A total of 130 virulence factors were predicted for R. ornithinolytica N9 across 12 virulence categories, with adherence (n = 44), iron uptake (n = 34), and secretion systems (n = 20) being the most represented ( Figure S5).
The analysis of the bla GES-5 gene context revealed its location in a class 3 integron, followed by part of an aacA4 gene cassette. However, the contig ended at this cassette, and it was not possible to clarify the full composition of this integron. A gene encoding an ORN β-lactamase, bla ORN-1 , intrinsic in R. ornithinolytica chromosome [74], was also detected on the same contig of bla GES-5 .

E. kobei N10
The predicted E. kobei N10 draft genome was 5,397,864 bp in size and organized in 431 contigs, with 54.4% in terms of G+C content, an N50 of 65,421, and 5128 predicted coding sequences (Table S3). E. kobei N10 was assigned to the new sequence type ST1378, as a result of the new alleles detected in fusA, leuS, and rpiB genes (Table S3). A total of 152 virulence factors were predicted for E. kobei N10, distributed over 15 virulence categories and with secretion systems (n = 45), iron uptake (n = 26), and antiphagocytosis (n = 24) being the most represented ( Figure S5).
Next, bla GES-5 was identified as part of a putative resistance island (RI17), as well as the fosA coding for fosfomycin resistance (RI7). We also found that bla GES-5 was present in a class 3 integron, which in turn was located in a contig (8366 bp) approximately 99.8% identical in terms of nucleotide identity to a corresponding region in plasmids pHPRU11 (MN180807), described in the clinical strain E. cloacae Ecl-w [75]; pJF-707 (KX946994), found in the clinical strain K. oxytoca H143640707; and pUL3AT (HE616889), found in E. cloacae LIM73 isolated from hospital effluent [76] (Figure 3).
On the other hand, the bla KPC-3 gene was identified in a Tn4401b transposon, which in turn was located in a contig (10.17 Kb) that was 100% identical in terms of nucleotide sequence to the same region of the C. freundii F6 plasmid described above (Figure 3). These two isolates also had in common the presence of the ARGs aacA4, mphA, and sul1.

Discussion
Although we are witnessing an increase in the global levels of carbapenem resistance, the number of studies investigating the occurrence of CRE in urban lakes and ponds is still limited. In Portugal, reports on the occurrence of CRE in aquatic environments are still very scarce, and are limited to rivers [11][12][13]76] and wastewater effluents [14,19,77].

Discussion
Although we are witnessing an increase in the global levels of carbapenem resistance, the number of studies investigating the occurrence of CRE in urban lakes and ponds is still limited. In Portugal, reports on the occurrence of CRE in aquatic environments are still very scarce, and are limited to rivers [11][12][13]76] and wastewater effluents [14,19,77].
Urban lakes and ponds can serve as significant reservoirs of antibiotic-resistant bacteria and ARGs, and contribute to their spread due to the likely exposure of humans and animals; for example, during recreational activities [26,78]. On the other hand, in some cases these systems are physically connected to other aquatic systems in which the spread of antibiotic resistance can occur by other routes, such as estuarine channels. In this study, we reported the presence and characterization of CRE collected from five urban ponds and an estuarine canal located in the Aveiro city center.
Overall, the colony counts on mFC media ( Figure S1) never exceeded the value corresponding to a fair classification of superficial waters (third-best classification possible out of five categories that varied from excellent to very poor) in terms of total coliforms, according to the criteria used by the National System of Information on Hydric Resources (SNIRH) to classify superficial water (https://snirh.apambiente.pt, accessed on 13 December 2021). The low percentage of imipenem-resistant Enterobacterales (between 0 and 1%) was not surprising, as carbapenems are considered last-resort drugs that are used exclusively in clinical settings in Portugal [76]. Pond 1 stood out from the other sampling sites as the only site where imipenem-resistant Enterobacterales were detected in every sampling campaign (Table S2). In fact, roughly 84% of the CRE collection was obtained from Pond 1 ( Table 1). This pond is located in Aveiro Municipal Park, where several contamination sources could be responsible for the higher CRE prevalence. CRE present in the fecal droppings from companion animals have been previously described [79,80], although with low prevalence. On the other hand, runoffs from surrounding areas or leaks in wastewater pipelines also represent plausible contamination sources [14,81], including from hospital wastewater [82], considering that the hospital is located approximately 600 m from Pond 1, and the hospital's wastewater pipeline system is also located in close proximity. However, further research is needed to clarify the contamination source. Molecular typing revealed that Citrobacter and Raoultella isolates exhibiting the same BOX and PFGE profiles could be detected in Pond 1 over time spans of 10 months and 1 month, respectively. The presence over a long period of the same strains may indicate the existence of a persistent contamination source that is responsible for periodic discharges. At the same time, CRE occurring in these ponds are not submitted to any streams or currents, in contrast to rivers and larger aquatic systems. Therefore, without a significant water flow regime, CRE are not exposed to the resulting dilution effect, and tend to accumulate [78]. The occurrence of Raoultella isolates with the same PFGE pulsotype in Pond 1 and in the adjacent E1 canal indicated that P1 may be contributing to the dissemination of CRE to the canal and ultimately to Ria de Aveiro, in which the occurrence of Enterobacterales harboring integrons and β-lactamase genes has been previously reported [83]. This hypothesis points to an even more worrying scenario, since Ria de Aveiro is used as a food source, therefore providing an additional route through which humans can be exposed to CRE.
The β-lactam resistance observed in the CRE collection was mainly due to the presence of the carbapenemase genes bla KPC-3 and bla GES-5 . The high frequency of bla KPC harboring CRE reflected the epidemiology observed in Portuguese hospitals, since it was the most frequent carbapenemase gene among CRE clinical isolates [9,84,85]. Furthermore, KPC was also the most common carbapenemase reported from environmental CRE in Portugal [11,12,14]. In contrast, Araújo and colleagues established a CRE collection obtained from raw wastewater in which bla  was detected in all isolates, including Citrobacter and Enterobacter isolates [19]. In addition, the occurrence of a GES-5-producing K. pneumoniae isolate recovered from an urban water stream has been described in Portugal [15]. As hypothesized by Araújo and colleagues [19], GES-5-producing CRE may originate from nonhuman sources.
Citrobacter species are known for their capacity to colonize humans, animals, plants, and the environment, and also are often associated with high-risk transferable ARGs such as carbapenemases [86]. The occurrence of environmental bla KPC -harboring Citrobacter has been previously described in rivers [25,87] and hospital effluents [88][89][90]. On the other hand, the copresence of bla KPC and bla VIM in Citrobacter has only been described once in a clinical C. freundii isolate in Spain [91]. In both the Spanish C. freundii clinical isolate and Citrobacter N5, bla VIM-1 was inserted in a class 1 integron containing five other ARGs: aacA4-dfrB1-aadA1-catB3-sul1, thus contributing to its MDR phenotype. C. freundii F6 belonged to sequence type ST270, which was first assigned to an isolate obtained from a diarrheal patient in China [92], showing that this ST is not confined to clinical settings, as it is disseminated across different continents.
In C. freundii F6, R. ornithinolytica N9, and E. kobei N10, the bla KPC-3 gene was present as part of a Tn3-based mobile transposon Tn4401 exhibiting two different isoforms: Tn4401b in C. freundii F6 and E. kobei N10; and Tn4401d in R. ornithinolytica N9, specifically in the pBK30683-like plasmid pN9KPC [49] (Figure 2). The mobile genetic platform bla KPC-3 -Tn4401 appears to be widespread in CRE isolated from Portuguese hospitals [10,93,94], and has also been detected circulating in the environment [11,14], associated with both IncN and FIA/FII plasmids. The Tn4401 transposon has also been proven to have an important role in the dissemination of bla KPC -type genes within a large range of plasmid structures and bacterial species [18,95,96]. Although it was not possible to determine the type of plasmid structure in which the bla KPC-3 gene was located in C. freundii N6 and E. kobei N10, the replicons (e.g., IncN, IncX5, and pKPC-CAV1321) identified by the in silico analysis of both these isolates have already been associated with bla KPC in both clinical [91,96,97] and environmental isolates [35,98], including in Portugal [8,11,94].
CRE harboring pBK30683-like plasmids have already been reported in Portugal, namely in clinical K. pneumoniae isolates [94] and in environmental K. pneumoniae isolated from a highly polluted river [11]. However, to the best of our knowledge, this is the first time that such plasmids have been described in Raoultella isolates.
Raoultella species are closely related to Klebsiella spp. and can be found ubiquitously in the environment, although they have already been associated with several cases of human infections linked to production of carbapenemases [99][100][101][102]. In addition, Raoultella isolates harboring carbapenemases have also been described in environmental settings, although never in Portugal [103][104][105][106]. Gomi and colleagues described the occurrence of bla GES-5 situated in a class 1 integron in a R. ornithinolytica strain isolated from wastewater [104]. In R. ornithinolytica N9, although this gene was likely present in chromosomal DNA, it was associated with a class 3 integron, which may play a potential role in the dissemination of this gene. R. ornithinolytica, previously known as K. ornithinolytica, is often misidentified as K. oxytoca by conventional biochemical identification methods [101,107]. Due to this common misidentification, the significance of Raoultella spp. as a nosocomial pathogen may be underestimated, as well as their association with carbapenemase production [108].
The role of the Enterobacter genus in carbapenemase gene flow between environmental, animal, and human settings has also been recognized [86,109]. Enterobacter isolates harboring bla KPC and bla GES genes have already been described in both clinical [110][111][112][113] and environmental settings [90,104].
In addition, in Portugal the bla GES-5 gene has already been found in association with IntI3 in clinical K. pneumoniae isolates [92], and also in two Citrobacter environmental isolates [11,19]. The occurrence of bla KPC-3 in Tn4401b transposons in both C. freundii F6 and E. kobei N10 highlighted the importance of these MGEs in the dissemination and acquisition of carbapenemase genes and other ARGs among different Enterobacterales species [114]. This was also pointed out by the occurrence of plasmids harboring the bla GES-5 gene that exhibited a high similarity to the one detected in E. kobei N10, in different Enterobacterales hosts isolated from environmental [11,19] and clinical settings [20] (Figure 3). Furthermore, the association of the bla GES-5 gene described in E. kobei N10 with a class 3 integrase and a genomic island enhances the chances of horizontal gene-transfer events [115,116].
The results obtained here highlighted that, although urban ponds are usually smaller and more confined than other aquatic systems, they are associated with bacteria that have a large variety of ARGs and also MGEs, which contribute to the dissimilation of these genes. The important role of conjugative plasmids was made clear, as these structures were identified in all CRE isolates, including previously described plasmids that were associated with dispersion of carbapenemase genes in clinical settings. In addition to plasmids, a considerable variety of MGEs were also identified, including genomic islands, transposons, and integrons. The importance of integrons was evidenced here, as they were linked with MDR phenotypes (in strains R. ornithinolytica N9, C. freundii F6, and Citrobacter N5) and the dispersion of carbapenemase genes (in strains R. ornithinolytica N9 and E. kobei N10).

Conclusions
In conclusion, our study supported the evidence that urban landscape ponds can serve as hotspots for multidrug-resistant bacteria that are also resistant to last-resort antibiotics. It is also worrisome that the acquired carbapenemase genes described here were located in conjugative plasmids and at the same time were associated with other mobile genetic elements such as integrons and genomic islands.
The fact that CRE were detected in every sampling campaign and that the same strains were able to persist within these ponds over time represents a serious public health threat, and highlights the importance of adopting a One Health approach, in which the environment is viewed as a transmission vehicle of antimicrobial resistance.
Further investigation should focus on elucidating the source(s) of contamination with CRE, particularly in Pond 1, from which most of the CRE isolates were obtained. Moreover, the transmission of resistance from these urban aquatic systems to humans and the underlying clinical implications are still largely underexplored and should also be attained.

Supplementary Materials:
The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/ijerph19105848/s1, Table S1: PCR primers used in molecular typing, 16S-rRNA-based affiliation, and screening of ARGs and integrons [76,[117][118][119][120]. Table S2: Percentage of imipenem-resistant CFUs in the sampled ponds and estuarine canal in February and March 2019, and from September 2019 to February 2020. No imipenem-resistant colonies were obtained from Pond P4 and P5. Hyphens are exhibited whenever no imipenem-resistant CFUs were detected. Table S3: General features of C. freundii F6, R. ornithinolytica N9, and E. kobei N10 draft genomes, and WGS-based parameters used for phylogenetic affiliation. Figure S1: Average counting of typical coliforms in CFU/mL (log 10 ) collected in February and March 2019, and from September 2019 to February 2020, from five ponds and an estuarine canal. In both February and March 2019, samples were only collected from ponds P1 and P2, while in September 2019, no samples were collected from E1. Figure S2: Dendrogram with the PFGE restriction patterns of genomic DNA from the eight Raoultella (A) and the seven Citrobacter (B) selected strains digested with XbaI. The dendrogram was constructed using GelCompar II version 6.1 (Applied Maths, Belgium) with a Dice coefficient and UPGMA clustering (Opt. 2.00%; Tol 1.0%; H > 0.0%; S > 0.0%). Figure S3: Plasmid DNA fingerprint of the entire CRE collection comprising 23 isolates. The molecular weight marker 1.5 Kb DNA NZYDNA Ladder VI (NzyTech, Portugal) is represented by the letter M. Figure S4: Circular genome representation of C. freundii F6 (A), R. ornithinolytica N9 (B), and E. kobei N10 (C). BRIG performed the alignment using a local BLAST+ with the standard parameters (50% lower-70% upper cut-off for identity and e-value of 10). The ring color gradients correspond to varying degrees of identity of BLAST matches. Circular genomic maps include information on GC skew, GC content, putative pathogenicity islands (PIs), putative resistance islands (RIs), prophage sequences (PS), and the location of ARGs and CRISPRs within the clockwise and counterclockwise CDS strands. Figure  S5: Category of bacterial virulence factors found in C. freundii F6 (a), R. ornithinolytica N9 (b), and E. kobei N10 (c) according to the VFDB analysis. Data Availability Statement: All relevant data are within the manuscript and its Supporting Material files. Additionally, the whole-genome sequences of Citrobacter freundii F6, Raoultella ornithinolytica N9, and Enterobacter kobei N10 can be found in the DDBJ/ENA/GenBank database under the accession numbers JAJBJE000000000, JAJDEY000000000, and JAKKZG00000000, respectively.

Conflicts of Interest:
The authors declare no conflict of interest.