Characterization of Carbapenemase-Producing Klebsiella pneumoniae Isolates from Two Romanian Hospitals Co-Presenting Resistance and Heteroresistance to Colistin

Klebsiella pneumoniae is a notorious human pathogen involved in healthcare-associated infections. The worldwide expansion of infections induced by colistin-resistant and carbapenemase-producing Enterobacterales (CPE) isolates has been increasingly reported. This study aims to analyze the phenotypic and molecular profiles of 10 colistin-resistant (CR) isolates and 2 pairs of colistin-heteroresistant (ChR) (parental and the corresponding resistant mutants) isolates of K. pneumoniae CPE sourced from two hospitals. The phenotypes of strains in the selected collection had been previously characterized. Antimicrobial susceptibility testing was performed using a Vitek 2 Compact system (BioMérieux SA, Marcy l’Etoile, France), the disc diffusion method, and broth microdilution (BMD) for colistin. Whole-genome sequencing (WGS) did not uncover evidence of any mobile colistin resistance (mcr) genes, although the mgrB gene of seven isolates appeared to be disrupted by insertion sequences (ISKpn25 or ISKpn26). Possible deleterious missense mutations were found in phoP (L4F), phoQ (Q426L, L26Q, L224Q, Q317K), pmrB (R256G, P95L, T157P, V352E), and crrB (P151S) genes. The identified isolates belonged to the following clonal lineages: ST101 (n = 6), ST147 (n = 5), ST258 (n = 2), and ST307 (n = 1). All strains harbored IncF plasmids. OXA-48 producers carried IncL and IncR plasmids, while one blaNDM-1 genome was found to harbor IncC plasmids. Ceftazidime–avibactam remains a therapeutic option for KPC-2 and OXA-48 producers. Resistance to meropenem–vaborbactam has emerged in some blakPC-2-carrying isolates. Our study demonstrates that the results of WGS can provide essential evidence for the surveillance of antimicrobial resistance.


Introduction
Recently, it has been reported that carbapenemase-producing Enterobacterales (CPE) isolates are expanding at an alarming rate around the globe [1,2]. These Gram-negative bacilli are responsible for causing a variety of hospital-acquired and community-onset human infections, with Klebsiella pneumoniae being the principal pathogen associated with significant mortality [2][3][4].
Antimicrobial resistance of K. pneumoniae strains continues to be problematic [5]. Resistance to carbapenems is frequently associated with resistance to multiple other classes of antibiotics, which leads to limited possibilities for the treatment of infections induced by these multidrugresistant (MDR) and extensively drug-resistant (XDR) pathogens [3,5,6]. Around the world, the geographical distribution of carbapenem resistance is reported to be heterogeneous [7].
The global propagation of carbapenemases is the result of a continuous interaction between the clonal dispersion of some efficient CPE lineages and the horizontal transfer of their resistance genes [9]. KPC-producing K. pneumoniae multilocus sequence type ST258/512 is a recognized "high-risk international clone" [2,10]. In Europe, ST258/512 and three additional clonal lineages, ST11, ST15, and ST101, along with their derivatives, are predominant [2,10], offering stability for carbapenemase genes [9]. The issues that remain unresolved are the distinct success of K. pneumoniae ST258 and the special relationship with epidemic resistance plasmids of incompatibility group F (IncF), which have diverse replicon types (FIA, FIB, and FII) [11,12]. The acquired antimicrobial resistance genes located on IncF can rapidly disseminate within species, while those situated on other types of plasmids, such as IncA/C, IncL/M, and IncN, can be transmitted between species [11,12]. The latter three types of plasmids are unusual in K. pneumoniae ST258 and have not been determined to contribute to international dominance [10]. IncL/M type plasmids are associated with bla OXA-48-like genes, and IncA/C and IncN types with bla NDM genes that encode New Delhi metallo-β-lactamases (NDMs) [9,11].
The worldwide escalation of colistin-resistant CPE isolates is of significant concern [15]. Colistin is a reserve antimicrobial agent with bactericidal action, principally as a consequence of interactions with the outer and inner membranes of Gram-negative bacilli [15,16]. Multiple chromosomal mutations constitute the main substrate of acquired resistance to colistin and promote alterations of the outer membrane lipopolysaccharide (LPS) with an abnormal positive charge, which decreases the affinity and action of colistin [17]. This strategy implies that the addition of 4-amino-4-deoxy-L-arabinose (L-Ara4N) and phosphoethanolamine (PEtN) molecules induces the covalent transformation of lipid A of LPS [17]. A panel of various genes modulates these modifications, essentially based on the two-component regulatory systems (TCRSs) phoPQ and pmrAB and the feedback regulator mgrB gene and the colistin resistance regulation (crrAB) operon, respectively [17,18]. Transposition of insertion sequences (ISs) such as ISL3 (ISKpn25), IS5 (ISKpn26), ISKpn14, and IS903B from plasmids into the mgrB gene mediates the inactivation and disruption of this regulator gene, conferring colistin resistance [19][20][21]. Antibiotic heteroresistance is considered to be the presence of subpopulations with increased resistance within the majority cell population in the same culture [22]. However, a coherent definition and global guidelines for antibiotic heteroresistance detection are lacking, which complicates its diagnosis [22]. Of particular concern is the emergence of transferable colistin resistance via plasmids [5,15,16,18], with the discovery of a series of mobile colistin resistance (mcr) genes, mcr-1 to mcr-10, with diverse variants [23] that encode PEtN, which is incorporated into lipid A [15,23].
Despite continuous research activity, there is still an incomplete understanding of the molecular mechanisms of colistin resistance and heteroresistance [16,24], along with limited data regarding their dissemination and impact [5,18,25]. Whole-genome sequencing (WGS) has become an essential modern tool for interrogating both existing and emerging mechanisms of antibiotic resistance, and it has excellent potential in infection control surveillance [26].
Consequently, the aim of the study is to perform a detailed phenotypic and genomic characterization of 10 colistin-resistant (CR) isolates and 2 pairs of colistin-heteroresistant (ChR) (parental and the corresponding resistant mutants) isolates of K. pneumoniae CPE, collected from two Romanian hospitals in 2017 and 2021, toward delineating the variety of molecular substrates of colistin resistance and heteroresistance, identifying the genetic determinants of resistance to other classes of antibiotics, confirming the presence of circulating high-risk clones and dominant plasmids, and exploring potential therapeutic options, including novel β-lactam/β-lactamase inhibitor combinations.

Phenotypic Antimicrobial Susceptibility Testing
All isolates presented resistance to multiple classes of antimicrobial agents, specifically meropenem, ertapenem, doripenem, aztreonam, cephalosporins, old generations of β-lactam/β-lactamase inhibitors, ciprofloxacin, and levofloxacin. The phenotypic results of antibacterial drug testing are given in Table 1.
Of the three aminoglycosides tested, gentamicin was the most active in vitro (n = 8 susceptible isolates), while total resistance was noted for tobramycin ( Table 1). Most of the isolates (n = 12) were susceptible to tigecycline and ceftazidime-avibactam (Table 1). Imipenem-relebactam exhibited potent activity against all KPC producers, while meropenem-vaborbactam showed activity against only five out of the eight KPC producers (Table 1).
Minimum inhibitory concentrations (MICs) of colistin obtained by Vitek 2 Compact software are given in Table 1.
Colistin susceptibility test results obtained with other phenotypic methods and reference broth microdilution (BMD) are given in Table 2. MIC values of colistin obtained by BMD ranged from 0.25 to >64 mg/L ( SPAdes assembly metrics for the sequenced genomes are presented in Table 3. They suggest that all the sequences are adequate for downstream analysis.

Genetic Determinants of Antimicrobial Resistance
All genomes belonged to MDR strains, with several defined resistance genes against different categories of antibiotics, along with other antibiotic resistance determinants, such as insertions, point mutations, or substitutions that lead to the disruption of genes, thus leading to resistance to at least six antibiotic classes.
(A) Molecular mechanism of resistance to various antibiotic classes, except colistin The highest heterogeneity of resistance-conferring elements was noted for β-lactams (n = 13), aminoglycosides (n = 11), and fluoroquinolones (n = 8) (Table 4). Isolates bla NDM -1-positive BC7_BM and BC1_TM presented the highest numbers of genetic determinants (n = 28 and n = 22, respectively), while the strain BC2_BM OXA-48 producer had the lowest (n = 12) (Table 4).    In agreement with the phenotypic results, the genotypic analysis confirmed the presence of carbapenem-hydrolyzing enzymes responsible for carbapenem resistance.
The oqxAB efflux pump genes and various chromosomal mutations in the quinolone resistance-determining region (QRDR) of gyrA and parC genes contributed to fluoroquinolone resistance in all the analyzed strains. Six strains were observed with the genotype profile of gyrA D87N and parC S80I mutations ( Table 4). All the isolates carried a gyrA S83 mutation, with gyrA S83Y identified in six strains and gyrA S83I in eight. Additionally, plasmid-mediated quinolone resistance (PMQR) genes were noted: aac(6 )-Ib-cr5 (n = 2) and qnrS1 (n = 1) ( Table 4).
A number of alterations in the principal outer membrane of porin-encoding genes ompK36 and ompK37, which potentially confer resistance to cephalosporins and carbapenems, identified by ResFinder, are summarized in Supplementary Table S1.
Similarly, using BioEdit for sequence alignment, the glycine-aspartic acid duplication (GD) (position 134-135 duplicated in 136-137) in ompK36 was visualized in five strains (BC2_BM, BC4_BM, BC5_TM, BC9_TM, and BC10_TM). The insertion of ISKpn26 (IS5) upstream of ompK36 (position -48 versus the start of ompK36 gene), which leads to ompK36 inactivation, was visualized in WGS data and further confirmed by Sanger sequencing (data not shown) in the BC3_TM and BC8_BM isolates. The BC1_TM strain possessed an N186K mutation, while BC7_BM carried several single nucleotide polymorphisms (SNPs), along with deletions, insertions, and substitutions. Similarly, the isolates BC3_TM, BC7_BM, and BC8_BM presented SNPs and substitutions in ompK37, while an insertion in ompK35 that produced a frame shift (FS) at amino acid 166 was noted in isolate BC6_BM. ClustalW (BioEdit) alignment of ompK35, ompK36, and ompK37 genes is shown in Supplementary Figure S1.
(B) Molecular determinants of colistin resistance The genomic results indicate the absence of plasmidial mcr genes. At least one colistin-resistance chromosomal determinant was noted in the mgrB gene or in the two-component regulatory systems phoPQ, pmrAB, and crrAB (Table 5).
As in the case of BC6_BM, the two ChR strains and their corresponding resistant mutants harbored ISL3 (ISKpn25) in the same nucleotide position and an R256G SNP in the pmrB gene, but with a supplementary presumptive intolerant L224Q SNP in the phoQ gene. Notably, only the resistant mutants BC12_TM_B_m and BC14_TM_C_m presented deleterious Q317K SNPs in the phoQ gene and P151S in the crrB gene (Table 5).
In contrast, no changes to pmrA and crrA genes resulting in functional amino acid substitutions were identified in any isolates; only changes encoding neutral amino acid substitutions were identified ( Table 5).
The Protein Variation Effect Analyzer (PROVEAN) scores for amino acid changes with a potentially harmful influence on the biological function of the analogous proteins are given in Table 6. The sorting intolerant from tolerant (SIFT) algorithm also confirmed the tolerant/intolerant classification.   In the case of strains BC3_TM and BC8_BM, despite being characterized by Vitek 2 Compact as being resistant to gentamicin, no corresponding molecular resistance element was noted (Tables 1 and 4). Similarly, the instrument indicated that BC1_TM, BC2_BM, BC4_BM, and BC10_TM were resistant to chloramphenicol, but no genetic resistance markers were detected (Tables 1 and 4). In these two situations, the disc diffusion method indicated susceptibility to the same antimicrobial agents in all cases. Strain BC9_TM was phenotypically susceptible to amikacin and trimethoprim-sulfamethoxazole, but resistance genes aac(6 )-Ib-cr5 and dfrA14 were detected. Two isolates, BC7_BM and BC9_TM, were categorized as susceptible to tigecycline, but tet(A) and tet(D) genes were identified. Two other strains (BC4_BM and BC10_TM) that showed intermediate susceptibility to tigecycline did not carry tetracycline resistance markers. Conversely, isolate BC14_TM_C_m was interpreted as resistant to chloramphenicol based on both testing methods, but no resistance genes were found upon genetic analysis (Tables 1 and 4). For the remaining antimicrobial compounds, except colistin, no disagreement between phenotypic profiles and molecular elements was noted (Tables 1 and 4).
Interestingly, in the case of BC3_TM, the colistin susceptibility profile based on Vitek 2 Compact revealed a discrepancy between the CMI results obtained concomitantly by AST-N222 and AST-XN05 cards, though molecular markers of colistin resistance were detected (Tables 1 and 5). In the case of BC2_BM, disagreement was noted between the Vitek 2 Compact results and the molecular analysis of colistin. All BMD colistin results correlated strongly with genotypic findings, except in the case of two ChR isolates.

Molecular Serotyping, Plasmid Replicon Identification, and Sequence Type Determination
Molecular serotyping determined that six strains were KL17;O1v1, five were KL112;O2v2, two were KL106;O2v2, and one isolate belonged to serotype KL102;O2v2 (Table 7). No gene characteristics of hypervirulence were identified by the Virulence Factor Database (VFDB) tool.
The identified replicons covered mainly Col, IncFIA, IncFIB, IncFII, IncC, IncL, IncR, and IncX3 type plasmids, with a total of 12 types (Table 7). Virulence-encoding plasmid replicon types IncFIB(K) and IncFII(K) were identified in 12 strains. More than half of the isolates were positive for ColRNAI. IncFIB(pQil) was observed, along with other replicon types, in six strains ( Table 7). The genome of all the OXA-48 producers carried IncL and IncR plasmids, while one bla NDM-1 genome was associated with the IncC plasmid ( Table 7). The correlation between ST258 clones and the IncF-carrying plasmid with FII(K) replicons was ascertained. The IncX3 plasmid was positive in two KPC-2-producing isolates. In the genome of strains BC6_BM, BC11_TM_B_hR, BC12_TM_B_m, BC13_TM_C_hR, and BC14_TM_C_m, two replicons from the same incompatibility class were identified, IncFIB(K) and IncFIB(pQil), suggesting the presence of multi-replicon plasmids along with other replicons.
In the mapping of raw reads of the two bla NDM-1 sequenced strains to 2014-2015 Târgu Mures , bla NDM-1 plasmid sequences, no similarity was found between reference plasmids and bla NDM-1 strain BC1_TM. However, the historical plasmid pNDM_18ES had 96% coverage with 99.89% identity by the reads of BC7_BM strain, and plasmids 1TM and 6TM had 76% and 73% coverage, respectively, with 99.9% identity.
Nine complex types were identified by core genome MLST (cgMLST). Five of them (CT5839, CT5840, CT5853, CT5854, and CT5848) were considered new founders by SeqSphere and were included in the K. pneumoniae database of the cgMLST.org Nomenclature Server.
Overall, ST101 isolates presented the highest number of distinct plasmid replicon types (n = 7 types). KL17 and KL112 serotypes were exclusively associated with ST101 and ST147, respectively.
The minimum spanning tree of the 14 CR and ChR strains revealed three clusters of closely related strains (

Discussion
In this study, we analyzed the phenotypic features, resistome, virulome, and plasmid content of 10 CR and 2 pairs of ChR (parental and the corresponding resistant mutants) isolates of K. pneumoniae CPE, collected from two Romanian hospitals between 2017 and 2021. This information is of critical epidemiological importance in the context of the continuous global expansion of high-risk MDR K. pneumoniae clones [1,2,9,10,27]. Additionally, the European Antimicrobial Resistance Surveillance Network (EARS-Net) received reports of an increasing trend of invasive infections due to K. pneumoniae CPE isolates at Romanian hospitals, from 32.3% in 2019 to 48.3% in 2020 [5,28]. The varying degrees of

Discussion
In this study, we analyzed the phenotypic features, resistome, virulome, and plasmid content of 10 CR and 2 pairs of ChR (parental and the corresponding resistant mutants) isolates of K. pneumoniae CPE, collected from two Romanian hospitals between 2017 and 2021. This information is of critical epidemiological importance in the context of the continuous global expansion of high-risk MDR K. pneumoniae clones [1,2,9,10,27]. Additionally, the European Antimicrobial Resistance Surveillance Network (EARS-Net) received reports of an increasing trend of invasive infections due to K. pneumoniae CPE isolates at Romanian hospitals, from 32.3% in 2019 to 48.3% in 2020 [5,28]. The varying degrees of resistance observed between European countries may be partially explained by differences in antimicrobial consumption, with the lowest rates of both resistance and use in Northern and Central Europe [10,28].
To the best of our knowledge, this is the first Romanian study that focuses on evaluating molecular determinants of resistance in CR and ChR K. pneumoniae CPE isolates. The results may also contribute to mitigating the spread of antimicrobial resistance in Romanian hospitals.
In line with previous reports [2,3,5], our phenotypic findings reflect the remarkable rates of resistance to several classes of antimicrobial agents; only a few pathogens remain susceptible to gentamicin and trimethoprim-sulfamethoxazole, but most exhibit susceptibility to tigecycline. Unsurprisingly, the two NDM-1-positive isolates were susceptible only to tigecycline. New compounds with targeted action against NDM producers have been developed, but none have yet been approved for clinical use [29]. Additionally, among the new combinations of β-lactam/β-lactamase inhibitors, ceftazidime-avibactam and imipenem/cilastatin-relebactam were found to be reliable treatment options for all of our KPC producers, while ceftazidime-avibactam exhibited supplementary in vitro activity against all bla OXA-48 -carrying isolates. In contrast, the resistance to meropenemvaborbactam in three out of the eight KPC producers represents an unwelcome occurrence in the context of patients who have not previously been exposed to this promising novel antimicrobial compound.
The resistome analysis confirms the existence of an extensive repertoire of antibiotic resistance genes, along with insertions or point mutations, and all our isolates possessed determinants of resistance to β-lactams, amikacin, tobramycin, fosfomycin, fluoroquinolones, colistin, and phenicol. Similarly, several previous reports described a great variety of resistance genes in K. pneumoniae isolates with the MDR or XDR phenotype [27,[30][31][32] and pointed out the potential flexibility of these pathogens to accumulate and exchange antimicrobial resistance [27]. Furthermore, the concurrent carriage of multiple β-lactamase genes with overlapping hydrolytic activity was confirmed in each of our analyzed strains. The implications of this have not yet been entirely elucidated, but it might have provided an evolutionary advantage for K. pneumoniae by offering an additional reliable basis for resistance [27].
In the current evaluation, concordance of more than 96% between phenotypic and genomic predictions of antimicrobial resistance was observed. Similarly, Ruppe et al., obtained more than 96% concordance in a study that included 187 Enterobacterales strains tested phenotypically using the disc diffusion method [33]. However, despite WGS being considered a robust surveillance tool [26,32], there is still insufficient published evidence to support its use as a highly accurate instrument to predict antimicrobial susceptibility phenotypes from genomic traits, and it is necessary to establish a consensus regarding which database to interrogate for the detection of antimicrobial resistance genes [33,34].
Of note, some discrepancies were identified in this study. Two isolates, BC3_TM and BC8_BM, were categorized as resistant to gentamicin by Vitek 2 Compact and susceptible by disc diffusion, but no AME or RMT genes were detected. However, inadequate correspondence between genotypes and inference of the presence of substrates with resistance to aminoglycosides have been extensively reported based on the European Committee on Antimicrobial Susceptibility Testing (EUCAST) clinical breakpoints in association with expert rules [35]. Instead, consistent with Ruppe et al. [33], the BC9_TM strain was found to possess the aac(6 )-Ib-cr5 gene but without phenotypic expression of amikacin resistance.
Vaziri et al. underscored the fundamental role of the qnrB gene in the K. pneumoniae genome, which, in association with the aac(6 )-Ib-cr5 gene, contributes to increased resistance to aminoglycosides, fluoroquinolones, and cephalosporins [36], a conclusion that was not supported by our findings. A recent nationwide epidemiological survey conducted in Greece confirmed a divergence between the resistance phenotype and the AME genotype for aminoglycosides, probably as a consequence of various competing resistance mechanisms associated with the distinct catalytic activities of AME genes [37]. In addition, molecular detection of a resistance gene does not necessarily indicate expression and activity [38].
The catA1 acetyltransferase and the nonenzymatic cmlA5 genes, encoding resistance to chloramphenicol, were detected in most of our strains. However, discrepant results for chloramphenicol were noted in the case of four strains (BC1_TM, BC2_BM, BC4_BM, and BC10_TM), showing resistance based on Vitek 2 Compact and susceptibility based on the Kirby-Bauer method, while no resistance genes were detected by WGS. The contribution of another mechanism responsible for chloramphenicol resistance (high expression of efflux systems) should not be underestimated [32], given that all of our isolates harbor the oqxAB efflux pump.
Previous studies in Romania have reported on the worrisome emergence of the rmtC gene in NDM-1-positive isolates belonging to the order Enterobacterales [39,40]. This observation is supported by the results of the present study, in which isolate BC7_BM was found to co-carry the rmtC gene along with bla NDM-1 , bla CTX-M-15 , and bla CMY-16 resistance genes and express pan-aminoglycoside resistance. Similarly, the co-expression of NDM-1 and rmtC genes, along with at least one ESBL-encoding gene in K. pneumoniae isolates, has previously been documented in Kenya, Turkey, and Saudi Arabia [41][42][43]. Since the first description of the RMT gene aminoglycoside resistance methylase (armA) in K. pneumoniae in 2003 [44], eight other plasmid-mediated variants (rmtA to rmtH) and N1-A1408 methyltransferase (MTase) (npmA) have emerged in Gram-negative pathogens in various parts of the world [45]. In contrast to our findings, a study performed in Switzerland between 2017 and 2020, which analyzed 103 carbapenem-and aminoglycoside-resistant Enterobacterales strains, did not find any K. pneumoniae isolates harboring the rmtC gene; the most frequently identified was the armA gene [46]. Among the RMT genes, armA, rmtB, and rmtC genes are distributed worldwide and confer high-level, broad-range aminoglycoside resistance, including against plazomicin, a newly approved aminoglycoside compound that can evade virtually all clinically relevant AMEs, including acetyltransferase (aac), phosphotransferase (aph), and adenylyltransferase (aad or ant) [45]. The association between RMT genes and bla NDM , bla KPC , and mcr genes is a multifaceted topic because of the possible involvement in the expansive dissemination of extensively pandrug-resistant organisms [45].
Among our three bla KPC-2 -carrying isolates resistant to meropenem-vaborbactam (BC3_TM, BC5_TM, and BC8_BM), detailed molecular analysis verified their affiliation with the international high-risk lineages ST258 and ST101, with evidence of either GD134-135 duplication in the ompK36 gene or insertion of ISKpn26 (IS5) upstream of ompK36 in association with SNPs and substitutions in the ompK37 porin gene. Furthermore, even though BC6_BM showed susceptibility to meropenem-vaborbactam, an FS mutation at amino acid 166 in ompK35 was recorded. These mutations are responsible for alterations in permeability that contribute to reduced susceptibility to meropenem-vaborbactam, as previously reported [47][48][49]. Meropenem-vaborbactam penetrates the outer membrane of K. pneumoniae mainly through ompK35 and ompK36 porins, but vaborbactam prefers the latter, which has a narrower inner channel [50]. However, the single presence of a nonfunctional ompK35 porin, without mutation in ompK36, is associated with MIC values of ≤0.06 mg/L meropenem-vaborbactam [47], and a higher degree of meropenemvaborbactam inactivation was demonstrated only upon ompK36 porin loss of function, either in isolation or concurrently with ompK35 [50].
In agreement with other reports from diverse geographical regions [30][31][32], chromosomal mutations associated with colistin resistance were revealed in all examined strains in the current study, and no plasmid-borne mcr genes were identified.
MgrB, a small negative regulatory transmembrane protein, represses phoPQ signaling, but the expression of this TCRS regulator is upregulated in the presence of a modified mgrB gene, and it remodels LPS through the addition of cationic L-Ara4N, which prevents colistin molecules attaching to the LPS membrane [17,18,51]. Mutation of mgrB gene was the basis for resistance most frequently encountered in our study (11 out of the total strains) and was induced by either ISKpn25 (n = 6), ISKpn26 (n = 1), or a missense mutation M27K (n = 4), predicted as deleterious by bioinformatics tools. Several recent studies have highlighted the pivotal role of ISs in contributing to the emergence of colistin resistance by disrupting the mgrB gene [19][20][21][30][31][32]. Our findings provide evidence that plasmid IncFIB(pQil) encodes ISkpn25, while the ISKpn26 element is associated with IncFIA(HI1) and IncR plasmids. Fordham et al. demonstrated similar connections between these IS elements and their companion plasmid families, with the exception of the relationship between ISKpn26 and IncFII(pHN7A8) plasmids [19]. The M27K mutation in the mgrB gene was also reported to mediate colistin resistance in a previous study [52]. In our analysis, only strain BC5_TM presented the M27K mutation as the sole potential marker of resistance to colistin. In contrast, Liu et al. concluded that the expression of MgrB protein was not affected by the M27K mutation, even though the strain exhibited an MIC of 32 mg/L colistin [53]. Inactivation of mgrB has also been shown to promote virulence in K. pneumoniae isolates by suppressing the initiation of host defense reactions and limiting the action of multiple antimicrobial peptides [54].
In comparison with prior studies, our analysis indicates the occurrence of five previously unreported point mutations in phoP and phoQ genes, which potentially mediate colistin resistance and heteroresistance. We observed an alteration in the phoP gene induced by a novel deleterious L4F substitution in the BC2_BM isolate, while several new intolerant SNPs (L26Q, L224Q, Q426L, and Q317K), as anticipated by bioinformatics tools, were detected in phoQ in both CR and ChR strains. Meanwhile, other researchers have mentioned the deleterious L26Q substitution in the phoP gene [55]. Furthermore, some mutations closely related to those above were detected in the phoQ gene (such as L30Q, L26P, L96R, and L257P substitutions) [31,[56][57][58] and the phoP gene (V3F) [58] and have previously been described as mediating colistin resistance.
Within the complex chromosomal cascade mechanism that confers colistin resistance, phoP can also stimulate the production of PmrA protein, which belongs to the pmrAB TCRS, either directly or indirectly through the adaptor PmrD protein, leading to the addition of cationic pEtN and L-Ara4N moieties to LPS [18,51,59]. In our investigation, only pmrB, as part of the pmrAB TCRS, was subject to diverse mutations (R256G, T157P, P95L, and the newly reported V352E). The R256G substitution has been widely reported to contribute to colistin resistance [30][31][32]53], but this deleterious SNP has also been found in polymyxinsusceptible strains [30,58], suggesting that this alteration alone might not be sufficient to increase MIC values for colistin [58]. Interestingly, all seven isolates detected with this substitution presented at least an additional colistin-resistant element and phenotypically expressed resistance, except the two parental ChR strains, with MIC ≤0.5 mg/L colistin by BMD. In addition, T157P substitution has been reported to disrupt the α-helix secondary structure of mutated PmrB protein, with consecutive activation of PmrA [60], and this SNP was mentioned by several authors [32,52,56,60]. In agreement with the analysis of KPC-3-producing Colombia strains by Jajol et al. [60], our KPC-2-producing BC8_BM strain, with the same T157P mutation, belongs to the same ST258 clone. The two PmrB mutations, R256G and T157P, are among the most frequently reported as being associated with colistin resistance [61], and the rare P95L alteration in PmrB was confirmed to confer resistance by complementation assay [62].
Notably, our BC2_BM and BC3_TM isolates, with colistin-susceptible results by Vitek 2 Compact AST N222 cards but an MIC value of ≥8 mg/L colistin by BMD, did not present any mgrB gene disruption by ISs. Consequently, in the first case, a combination of mutations in mgrB (M27K), phoP (L4F), and phoQ (Q426L) genes was noted, while in the second case, an L26Q substitution in the phoQ gene was associated with R256G mutation in the pmrB gene. However, several reports have mentioned that Vitek 2 Compact is inappropriate for colistin susceptibility testing [63][64][65], especially for strains showing an MIC > 1 mg/L by BMD [63]. The manufacturer of Vitek 2 Compact has also recommended using an alternative method prior to reporting colistin results obtained with several types of testing cards in this automated system [66], at least until this issue can be addressed through technology upgrades to be implemented in the future.
Moreover, the crrAB TCRS can activate Pmr A [51,59], and gain-of-function mutations in crrB alone can activate the expression of genes leading to colistin resistance without any contribution from the pmrAB TCRS [59]. Interestingly, our findings revealed a rare P151S substitution in the crrB gene only in the resistant mutant BC14_TM_C_m isolate, which, in contrast to its susceptible parental strain BC13_TM_C_hR, expressed an MIC value of >64 mg/L colistin. The P151S substitution detected in the putative histidine kinase domain has been validated to induce elevated resistance to colistin [67]. Similarly, Jajol et al. reported a mutation with a subtle difference, P151L in crrB, as conferring colistin resistance [52], while Pitt et al. confirmed a P158R substitution in the same gene [30].
Remarkably, another significant aspect of our study is the molecular basis of the two pairs of ChR (parental and the corresponding resistant mutants) isolates, which are reported for the first time in Romania [65]. These two pairs of strains co-harbored an mgrB alteration caused by an ISKpn25 element insertion at the same position (with the deletion of nucleotides 1-5) and an R256G substitution in pmrB. An identical molecular profile was observed with the BC6_BM strain, which, unlike the two parental ChR strains, had an MIC value of 32 mg/L colistin indicated by BMD. In particular, the two pairs of isolates were supplementarily accompanied by a new L224Q substitution in phoQ, and the two mutants presented either an additional amino acid change P151S in crrB or Q317K in phoQ.
Despite the presence of a disrupted mgrB gene in the two parental ChR strains, a possible explanation for the results of colistin susceptibility according to BMD might be the potential suppressor effect of the L224Q mutation in phoQ. In addition, BMD has been considered an unreliable method for the detection of colistin heteroresistance [25,68,69]. Pitt el al., reported that K. pneumoniae isolates exclusively carrying an ISKpn26-like element exhibit an MIC of ≥64 mg/L colistin, while the introduction of a mutated phoP (P47L or A95S) or phoQ (N253T or V446G) gene into an IS mgrB-altered strain led to a decrease in the MIC [30]. These mutations have been shown to disturb the pathways involved in phoQ, phoP, and pmrD expression [30]. However, the complex and challenging mechanisms underlying colistin resistance have not been entirely decoded, and accurate detection accompanied by functional analysis is required in order to clarify their influence on resistance [17,30,51,59]. Moreover, there are still insufficient data on the genetic basis of colistin heteroresistance, although there is evidence that mutations in phoPQ, pmrAB, and mgrB regulatory systems or in lpxM and yciM alleles are involved [61,[69][70][71][72] and that the mcr-1 gene is not connected with this phenomenon based on PCR results [68,71].
The strains included in our collection were concentrated in either established (ST258, ST101, and ST147) or emerging (ST307) international clonal lineages [11,27,73,74]. The circulation of bacterial clones ST258, ST101, and ST307 in Romania has been previously documented in both clinical and wastewater specimens [39,40,75]. The present analysis reaffirms the prominent relationship between ST258 and bla KPC genes [27], whereas our strains assigned to ST101 carried bla OXA-48 , bla KPC-2 , and bla NDM-1 genes, consistent with Palmieri et al. [76]. K. pneumoniae genotype ST101 has been associated with resistance to carbapenem, colistin, aminoglycosides, fluoroquinolones, and fosfomycin, with the potential to become a "perfect storm" clone, considering the similarity with the genetic profiles of hypervirulent strains [77]. In addition, a recent Romanian study demonstrated that it is able to persist in wastewater samples after chlorine treatment [78]. All of our ST101 isolates expressed capsular KL17 and somatic O1v1 antigens, and these findings are also supported by other studies [77,79]. K. pneumoniae ST147 and ST307 have been associated with pandrug resistance and have been reported from endemic regions as well as in global nosocomial outbreaks, with proven links to KPC, NDM, OXA-48-like, and Verona integron-encoded MBL (VIM) [73]. In agreement with Peirano et al. [73], our six strains, assigned to ST147 and ST307, were found to carry the ESBL bla CTX-M-15 gene along with gyrA S83I and parC S80I mutations. Contrary to a recent report on an outbreak in northeast Germany induced by K. pneumoniae ST307 co-harboring bla NDM-1 and bla OXA-48 genes and colistin resistance [74], the bla OXA-48 gene was not detected in our ST307 isolate, nor were any virulence genes or elements conferring susceptibility to chloramphenicol detected.
It is noteworthy that the two parental strains, BC11_TM_B_hR and BC13_TM_C_hR, showed the same phenotypic and genomic profiles belonged to the successful ST147 clone, carried the same mgrB, phoQ, pmrB, and crrB alterations, and were recovered from the same intensive care unit (ICU) approximately 3 weeks apart from different patients, all of which are indicative of silent clonal expansion and intraward dissemination. There is also concern regarding the close relationship of these two ChR isolates from Târgu Mures , with the BC6_BM strain isolated in Baia Mare, suggesting the interregional propagation of ST147. Furthermore, the OXA-48-producer cluster ST101 revealed both interward and interregional spread, while the KPC-2-producer cluster ST258 showed interregional transmission. The constituent isolates of each cluster possessed similar or almost similar plasmids, implying plasmid-mediated propagation of bla OXA-48 and bla KPC-2 genes.
The interdependence between KPC-2-producing isolates with FII(K) and IncX3 plasmids and between OXA-48 producers with IncL and IncR plasmids concurs with previous observations made by Becker et al. in Germany [80]. In contrast to a prior study conducted at the same medical institution in Târgu Mures , on strains collected between 2012 and 2013 that demonstrated the presence of the IncR plasmid replicon in five K. pneumoniae NDM-1-positive isolates [81], our two NDM-1 producers did not harbor this replicon type. Moreover, a significant match between the BC7_BM strain and the reference plasmids suggests the persistence and evolution of a bla NDM plasmid in this geographical area, which was previously named pKOX_NDM1-like by Phan et al. [82]. Additionally, the concurrent carriage of IncFIB(K) and IncFIB(pQil) replicons belonging to the same incompatibility class is in agreement with the results of Villa et al., demonstrating the great versatility of IncF plasmids [83]. These extrachromosomal DNA molecules often exhibit a multireplicon status [12,83], which enables the acquisition of plasmids harboring incompatible replicons when replication is promoted by a compatible replicon [83]. However, the secondgeneration raw reads used in our research did not offer the same possibility to construct the entire sequence of a plasmid as in the case of long-read sequencing [32,38].
The present investigation has some limitations. The retrospective nature of the study did not allow us to provide a reliable picture of all significant clinical and therapeutic aspects or previous hospitalizations, which was beyond the scope of this manuscript. However, the data on previous colistin therapy in patients diagnosed with ChR strains may be found elsewhere [65]. BMD was performed only for colistin as the research was focused on detecting the molecular basis of resistance and heteroresistance to this antimicrobial agent. Complementation experiments with wild-type alleles to validate novel mutations potentially conferring colistin resistance and heteroresistance were not performed because of logistical constraints and should be included in future investigations.

Future Directions
Future studies should be oriented toward the reliable detection of both existing and novel mutations, combined with functional analysis, to establish their real contribution to colistin resistance and heteroresistance. Additional research should be conducted to expand our knowledge of the complex mechanisms underlying colistin resistance and heteroresistance; for the last phenomenon to establish clinical relevance, we should formulate a harmonized international definition and develop a standardized detection methodology. Furthermore, investigations using long-read sequencing technology will offer an adequate resolution for exploring plasmid structures.

Bacterial Strains, Setting, and Design of the Study
A total of 10 unique clinical CR K. pneumoniae CPE strains were analyzed: 5 selected from patients admitted to Dr. Constantin Opris , County Emergency Hospital, Baia Mare, Romania, between January 2017 and April 2021 and 5 from patients at Târgu Mures , County Emergency Clinical Hospital, Romania, between January 2017 and April 2019. Additionally, 2 ChR K. pneumoniae strains obtained from Târgu Mures , in March 2019 and their corresponding colistin-resistant mutants were included in the study. The collection comprising the 10 CR and 2 ChR isolates has been phenotypically characterized in an earlier study [65].
The first medical center is a public 920-bed general acute care nonteaching hospital in the northwest region of Romania, and the second is a large 1089-bed teaching hospital located in Transylvania, in the central region; they are located approximately 200 km apart.
CR K. pneumoniae CPE isolates were randomly selected based on (i) MIC values greater than 2 mg/L colistin (categorized as resistance) [84], as determined using the BMD method; (ii) diverse specimens collected from various hospital wards; (iii) carbapenemase types identified phenotypically (a set of n = 2 KPC, n = 2 OXA-48-like, and n = 1 MBL for each medical institution); and (iv) date of collection.
Pathogens were mostly isolated from patients in the ICU (n = 6) and were obtained from different anatomical sites, as summarized in Table 8.

Demographic Data of Patients
Data from electronic medical records available in the 2 laboratories are presented in Table 8.

Phenotypic Bacterial Identification and Antimicrobial Susceptibility Testing
All strains were identified at the species level using standard techniques and a Vitek 2 Compact system (BioMérieux SA, Marcy l'Etoile France). AST-XN05 and AST-N233 pair testing cards, starting from the same inoculum, were used with the Vitek 2 Compact system, and AST-N222 cards were added for particular strains. Vitek 2 Compact version 9.02 software was used. The Kirby-Bauer disc diffusion method was used in all cases. Meropenem-vaborbactam (30 µg), imipenem-relebactam (35 µg), ceftazidime-avibactam (14 µg), and doripenem (10 µg) were tested exclusively by disc diffusion.
The EUCAST breakpoint [84] was applied for the interpretation of all antibiotic susceptibility test results, except for tigecycline, for which US Food and Drug Administration (FDA) criteria were adopted [85].
Concordance between phenotypic and genotypic susceptibility test results was calculated individually for Vitek 2 Compact (14 isolates × 22 antimicrobial agents = 308 combinations) and the disc diffusion method (14 isolates × 24 agents = 336 combinations) as the proportion of concordant results in the total combinations tested by each method. In this study, results of susceptibility or susceptibility to increased exposure were considered to be discrepant if a resistance molecular marker was noted or resistance results were not linked with genetic elements.
Routine and extended quality control were performed according to EUCAST [84] with the following reference strains: Escherichia coli ATCC 25922, E. coli NCTC 13846, E. coli ATCC 35218, K. pneumoniae ATCC 700603, and K. pneumoniae BAA 2814. Other details regarding quality control for all methods used can be found elsewhere [65].
The isolates were frozen at −70 • C and subcultured twice on solid medium before additional testing.

Sequencing and Assembly of Draft Genomes
The 14 genomes of CR, ChR, and corresponding resistant mutants of CPE strains were sequenced using the Ion Torrent PGM platform (Thermo Fisher Scientific, Waltham, MA, USA) according to the 400 bp protocol for library preparation, which includes enzymatic shearing, Ion OneTouch2 emulsion PCR, enrichment, and Hi-Q View sequencing kits (Thermo Fisher Scientific). The sequences obtained were subjected to de novo assembly into contigs using the Assembler SPAdes plugin version 5.12 (21,33,55,77, and 99 k-mers) [88] installed on the Ion Torrent Server. For epidemiological interconnections and worldwide distribution, all the raw reads in the study were deposited in the publicly available European Nucleotide Archive (ENA) database under project PRJEB53146 (ERR9860230-ERR9860243) (https://www. ebi.ac.uk/ena/browser/home, accessed on 11 August 2022).
WGS data were analyzed using both commercial and free online bioinformatics tools.
Plasmid-mediated mcr genes, the presence and integrity of TCRS genes (phoPQ, pmrAB, and crrAB), and the regulatory transmembrane protein-coding mgrB gene were investigated to identify colistin resistance determinants. ISfinder was used to depict the insertion sequence (IS) types that led to gene disruption [93]. In order to predict whether amino acid substitutions identified in PhoPQ, PmrAB, and CcrAB had an impact on the biological function of the proteins, 2 bioinformatics tools were applied: the Protein Variation Effect Analyzer (PROVEAN) version 1.1 [94], with a default score threshold set at -2.5 for binary classification, and the sorting intolerant from tolerant (SIFT) algorithm [95], with default parameters. The corresponding genes from colistin-susceptible K. pneumoniae NC_009648.1 were used as references.

Molecular Typing of Isolates and Plasmid Profiling
In order to depict the allelic profile of the genomes, MLST, ST [96], and cgMLST complex type (CT) genes (2358 target genes) were used with Ridom SeqSphere+ software [97]. For cgMLST, a default threshold of 15 allele differences was selected, and a minimum spanning tree with the option pairwise ignoring missing values was generated.

Conclusions
This study highlights the significant challenges involved in the phenotypic and molecular diagnoses of colistin resistance and heteroresistance. Each of our isolates presented at least one mutation in the mgrB, phoPQ, pmrAB, or crrAB gene, predicted to confer colistin resistance. We report on a novel and rare potential suppressor mutation in phoQ, L224Q, with possible involvement in heteroresistance to colistin, which in the presence of an altered mgrB, modified by IS, leads to low MIC values for colistin, as determined using BMD and other phenotypic methods aside from the Rapid Polymyxin NP test and PAP assays. Evidence for the silent intrahospital dissemination of heteroresistant mutant ST147 clones is thus provided.
This research underlines the importance of WGS in identifying a catalog of molecular markers of various classes of antibiotics coupled with descriptions of circulating plasmids and clones, data that can be useful when investigating the dynamics of the dispersion of K. pneumoniae MDR isolates, in addition to being an innovative and promising strategy for continuous surveillance of antimicrobial resistance in order to limit worldwide transmission.