Evaluation of Phenotypic Tests to Detect Extended-Spectrum β-Lactamase (ESBL)-Producing Klebsiella oxytoca Complex Strains

ABSTRACT Klebsiella oxytoca complex (KoC) species may overproduce their chromosomal class A OXY β-lactamases, conferring reduced susceptibility to piperacillin-tazobactam, expanded-spectrum cephalosporins and aztreonam. Moreover, since clavulanate maintains its ability to inhibit these enzymes, the resulting resistance phenotype may falsely resemble the production of acquired extended-spectrum β-lactamases (ESBLs). In this work, a collection of 44 KoC strains of human and animal origin was characterized with whole-genome sequencing (WGS) and broth microdilution (BMD) susceptibility testing. Comparison of ESBL producers (n = 11; including CTX-M-15 [n = 6] and CTX-M-1 [n = 5] producers) and hyperproducers of OXYs (n = 21) showed certain phenotypic differences: piperacillin-tazobactam (MIC90s: 16 versus >64 μg/mL), cefotaxime (MIC90s: 64 versus 4 μg/mL), ceftazidime (MIC90s: 32 versus 4 μg/mL), cefepime (MIC90s: 8 versus 4 μg/mL) and associated resistance to non-β-lactams (e.g., trimethoprim-sulfamethoxazole: 90.9% versus 14.3%, respectively). However, a clear phenotype-based distinction between the two groups was difficult. Therefore, we evaluated 10 different inhibitor-based confirmatory tests to allow such categorization. All tests showed a sensitivity of 100%. However, only combination disk tests (CDTs) with cefepime/cefepime-clavulanate and ceftazidime/ceftazidime-clavulanate or the double-disk synergy test (DDST) showed high specificity (100%, 95.5%, and 100%, respectively). All confirmatory tests in BMD or using the MIC gradient strip did not perform well (specificity, ≤87.5%). Of note, ceftazidime/ceftazidime-avibactam tests also exhibited low specificity (CDT, 87.5%; MIC gradient strip, 77.8%). Our results indicate that standard antimicrobial susceptibility profiles can raise some suspicion, but only the use of cefepime/cefepime-clavulanate CDT or DDST can guarantee distinction between ESBL-producing KoC strains and those hyperproducing OXY enzymes.

. Unless specified, all bioinformatics steps above were done with default parameters. Antimicrobial susceptibility tests (ASTs). Strains confirmed as KoC by using the WGS output underwent ASTs implementing the ESB1F and GNX2F broth microdilution Sensititre panels with Mueller-Hinton (MH) broth (Thermo Scientific) according to the manufacturer's instructions. ASTs were performed in duplicate leading to consistent results (therefore only one MIC value was shown in Table 2). ATCC strains Escherichia coli 25922 and Klebsiella quasipneumoniae ATCC 700603 were used as controls. MICs for antibiotics were interpreted according to the 2022 European Committee on Antimicrobial Susceptibility Testing (EUCAST) criteria (40). For minocycline and cefoxitin, the Clinical and Laboratory Standards Institute (CLSI) criteria of 2022 were used (39). We defined the strains as hOXY-KoC those with an MIC of CRO $2 mg/mL and lacking genes encoding ESBLs (bla ESBLs ) or plasmid-mediated AmpCs (bla pAmpCs ).
Phenotypic confirmatory tests for ESBL production. Based on the AST results, all KoC strains nonsusceptible (NS) to CRO (MIC $2 mg/mL) were further analyzed with several inhibitory-based confirmatory tests to detect ESBL producers. As for the ASTs, these assays were repeated two times leading again to consistent results (therefore, only one value was shown in Table 3).
The performance of two broth microdilution (BMD) tests was extrapolated from the results of the MIC ESB1F Sensititre panel: CTX/CTX-clavulanate (CTX-CL) and CAZ/CAZ-clavulanate (CAZ-CL). Moreover, MIC gradient strip tests (Liofilchem) with FEP/FEP-clavulanate (FEP-CL), CAZ alone and CAZ-avibactam (CZA) alone were assessed on MH agar plates (Oxoid). The results of these 4 MIC confirmatory assays were interpreted as ESBL-positive if the strain in the presence of the inhibitor had a $8-fold (or $3 2fold) MIC decrease compared with the MIC of the cephalosporin alone (33,39).
Finally, KoC strains were studied with the double-disk synergy test (DDST) on MH agar plates with disks of CTX (5 mg; Liofilchem), CAZ (10 mg; Liofilchem), FEP (30 mg; Liofilchem) and ATM (30 mg; Bio-Rad) placed with a distance center-to-center of 25 mm (DDST-25) and 30 mm (DDST-30) around a disk of amoxicillin-clavulanate (AMC; 20/10 mg; Bio-Rad). An ESBL-positive result was indicated when the inhibition zone around at least one of the cephalosporins or ATM disks expanded or there was a keyhole toward the AMC disk (33).
Data availability. The draft genome assemblies are deposited in GenBank under BioProject PRJNA894995.

RESULTS AND DISCUSSION
The clavulanate-based phenotypic confirmatory tests show good performance and reliable results in detecting ESBL-producing E. coli and K. pneumoniae strains. In contrast, such assays resulted in high false-positive rates when performed with hOXY-KoC strains (10,(19)(20)(21)(22)(23)(24). It should also be noted that level of identity at the amino acid and at the nucleotide levels of those OXYs may generate false-positive results with immunochromatographic or PCR-based assays designed to detect CTX-M ESBLs, respectively (29,30). In addition, although being faster, these non-phenotypic tests are more expensive, making their implementation limited to the screening of suspected carbapenemase producers (31,32).
The scope of our study was to use a well-defined collection of KoC strains to find a possible phenotypic-based strategy to ensure the identification of ESBL producers among those that are ESC-NS. From an epidemiological point of view, the correct detection of such strains can help to accurately define their prevalence. Furthermore, since bla ESBLs are transferable on mobile genetic elements, separation of ESBL-KoC and hOXY-KoC strains has important public health and infection control implications (e.g., isolation measures and consequent costs) (33). This is particularly true in countries with a relatively low prevalence of carbapenemase producers (e.g., Switzerland) that still implement such rules for ESBL producers, especially those belonging to Klebsiella spp. (32).       Synergy results are shown with "1" (synergy with clavulanate), Molecular features of KoC strains. Based on the WGS analysis, the 44 isolates were mainly identified as K. oxytoca (n = 27; 61.4%) and K. michiganensis (n = 14; 31.8%) species (Table 1).
In total, 11 ESBL-KoC strains were identified: 6 of human origin harbored bla CTX-M-15 , while 5 from animals possessed bla CTX-M-1 . Most ESBL producers possessed various plasmid-mediated ARGs against different classes of antibiotics and 4 of them also co-carried the bla OXA-1 that encodes a b-lactamase conferring resistance to PTC (Table 1) (34). Analogous data regarding the molecular characteristics of ESBL-KoC strains are scarce. Of note, most of the reported human isolates possessed the bla CTX-M-15 or bla SHV-12 ESBL encoding genes (2,35,36), while those of animal origin carried bla DHA-1 , bla CTX-M-9 , bla CTX-M-15 , or bla SHV-12 (15,16). However, in these studies, characterization of bla ESBLs/pAmpCs was obtained using only PCR-based methods and identification was generically reported as K. oxytoca. Moreover, only two surveys reported the corresponding sequence types (STs) of the ESBL-KoC strains as we have done in the current study (33,35). This lack of high-quality typing was also evident in studies evaluating the performance of phenotypic confirmatory tests for ESBL detection (see below) (10,(19)(20)(21)(22)(23)(24)37).
The remaining 33 KoC strains in our collection did not possess any bla ESBL or bla pAmpC gene. Based on the MIC of CRO, 21 of these strains were categorized as hOXY-KoC, while the last 12 isolates were defined as WT KoC (WT-KoC) strains for simplicity. Overall, both hOXY-KoC and WT-KoC strains possessed much less ARGs compared to ESBL producers. Notably, most hOXY-KoC were isolated from clinical samples of humans who were hospitalized, whereas WT-KoC strains were mainly detected in animals admitted from the community (Table 1).
Numerous OXY-types were detected in the overall collection of 44 KoC strains, including five newly reported (Table 1). Of note, strain R1057 hyperproduced OXY-2-5, a previously described variant of OXY-2 (Pro167Ser) that hydrolyzes CAZ at much higher level than the WT OXYs (7).
Overall, we emphasize that previous studies analyzing the susceptibility of KoC strains and the performance of phenotypic confirmatory tests for ESBL production did not provide an accurate molecular characterization as we did in the present work (10,(19)(20)(21)(22)(23)(24)37). Such information is essential to interpret the overall phenotypic and confirmatory test results illustrated below.
Phenotypic characteristics of KoC strains. Looking at the results of the ASTs (Table 2), we first noted that, consistently with the genotypic data, ESBL-KoC strains showed a frequency of associated resistance to non-b-lactam antibiotics higher than the hOXY-KoC isolates. This was particularly true for trimethoprim-sulfamethoxazole (SXT) and gentamicin (GEN ). Nevertheless, even this information was not useful for establishing a strategy to distinguish the two groups of ESC-NS-KoC isolates. We further emphasize that the strain producing the variant OXY-2-5 (R1057) displayed a phenotype almost identical to 8 out of 11 CTX-M-producing KoC strains (i.e., susceptible to PTC, non-susceptible to CTX and CAZ, and co-resistant to SXT) ( Table 2).
Overall, our data indicate that phenotypic results for PTC, ESCs and ATM cannot be used to distinguish between contemporary ESBL-KoC and hOXY-KoC strains. Special attention should be made to PTC and CAZ (Table 2). Three ESBL-KoC strains were in the resistant range for PTC because they coproduce the OXA-1 b-lactamase, whereas R1057 was fully susceptible (MIC #4 mg/mL) to the drug. Moreover, 4 ESBL-KoC of animal origin were only moderately resistant to CAZ (MICs of 2-4 mg/mL) because they produce the CTX-M-1 that does not significantly hydrolyze this substrate (38).
Performance of phenotypic confirmatory tests. Since a clear distinction between ESBL-KoC and hOXY-KoC strains based on the ASTs was difficult, we further evaluated the performance of 10 different inhibitor-based confirmatory tests for ESBL detection.
As shown in Table 3, none of the confirmatory tests resulted in false-negative results with the 11 ESBL-KoC strains (sensitivity, 100%). In particular, all assays provided results without any ambiguity (i.e., higher than the cutoffs used to define a strain as ESBL-positive) (33,39). Consistently, this high sensitivity was noted by numerous authors implementing various confirmatory assays and also testing strains producing non-CTX-M-type ESBLs (e.g., SHV-12 and TEM-types) (20)(21)(22)(23)(24)37). On the other hand, our study showed that CTX/CTX-CL BMD test, FEP/FEP-CL gradient strip test and DDST-25 gave a very high number of false-positive results when tested against hOXY-KoC strains (specificity of 53.8%, 56.8%, and 55.2%, respectively). Both the CAZ/CAZ-CL BMD test and the CTX/CTX-CL CDT performed better, but still showed less than ideal specificity (87.5% and 91.3%, respectively).
The low specificity that we recorded for the FEP/FEP-CL gradient strip test when implemented for ESC-NS-KoC strains was already reported by others (19)(20)(21)(22). The same authors also observed an overall low performance for other gradient strips (i.e., CAZ/CAZ-CL and CTX/CTX-CL) that were not evaluated in the present study (19)(20)(21)(22). Regarding the BMDbased confirmatory tests, two different studies used the MicroScan ESBL confirmatory panel to evaluate a total of 7 ESBL-(of which 6 producing CTX-Ms) and 9 hOXY-KoC strains. As a result, the CAZ/CAZ-CL assay showed 100% sensitivity and specificity, whereas for the CTX/CTX-CL they were 100% and 69.2%, respectively (24,37).
Our analysis indicated that the best performance in detecting ESBL-KoC strains was achieved with the FEP/FEP-CL CDT and the DDST-30 (100% specificity for both), but also the CAZ/CAZ-CL CDT showed an acceptable specificity of 95.5% (Table 3; Fig. S1).
Previous data regarding the performance of specific CDTs and DDSTs in the context of KoC is lacking. Sturn et al. tested 4 ESBL-(all TEM-types) and 17 hOXY-KoC strains with the CAZ/CAZ-CL and CTX/CTX-CL CDTs resulting in a sensitivity and specificity of 100% and 85%, respectively. However, despite the good performance, the results were a combination of the two CDTs. Interestingly, the 3 false-positives hOXY-KoC observed in that latter study produced the OXY-2-5 variant (Pro167Ser in OXY-2) (19). In another study, Wiegand et al. used a collection of 5 ESBL-(including 4 producing CTX-Ms) and 9 hOXY-KoC strains to evaluate the performance of four CDTs: CAZ/CAZ-CL, CTX/CTX-CL, cefpodoxime/cefpodoximeclavulanate, and cefpirome/cefpirome-clavulanate. Combining all CDTs, the authors reported 80% sensitivity and 88.9% specificity. In the same study, a DDST (CAZ, CTX, cefpodoxime and cefpirome disks placed at 25 to 30 mm away from AMC) showed overall sensitivity and specificity of 80% and 55.6%, respectively (20). However, in those two above-mentioned analyses, CTX (30 mg) and CAZ (30 mg) CLSI-recommended disks were implemented (39), whereas in the present work disks of CTX (5 mg) and CAZ (10 mg) have been used, as suggested by the EUCAST (33,40). Therefore, a comparison with our results does not seem meaningful.
Finally, for the very first time, we assessed the performance of gradient strips with CAZ and CZA along with a CDT with CAZ/CZA to recognize ESBL-KoC strains (Table 3). Since avibactam is a potent inhibitor of class A, C and some D b-lactamases (41), we hypothesized that CZA-based confirmatory tests could show more reliable results than those using clavulanate. However, the CDT showed a specificity of 87.5%, while the gradient strip assay resulted in 6 false-positives ESBL producers (specificity, 77.8%). We also noted that, in line with other studies (42), all ESBL-KoC and hOXY-KoC strains resulted in the EUCAST susceptible ranges for CZA (e.g., MIC 90s of 0.38 and 0.75 mg/mL, respectively; Table 3) (40).
Conclusions. Standard antimicrobial susceptibility profiles for KoC strains can raise some suspicion of ESBL production. However, a clear distinction between ESBL-KoC and hOXY-KoC strains is difficult. With our strain collection, such distinction was achieved only by implementing the FEP/FEP-CL CDT or the DDST-30, whereas all gradient stripand BMD-based confirmatory tests (regardless of the specific cephalosporin used) did not perform well. It is important to note that the FEP/FEP-CL CDT is not suggested by the CLSI (39), while the EUCAST indicates it only to detect the ESBL production among the group 2 Enterobacteriaceae (species expressing chromosomal AmpC genes) (33). Moreover, the DDST with CAZ, FEP and ATM disks is proposed by the EUCAST, but clear indications about the distance with the AMC disk and the concentration of antibiotics is not yet provided (33).
The CDT with CAZ/CAZ-CL disks may also be implemented as confirmatory tests, but production of OXY variants with potent activity against CAZ (e.g., OXY-2-5) can affect its specificity. In this context, it is worth underlining that the true prevalence of these CAZ-resistant hOXY-KoC strains that may generate phenotypic results identical to those of the CTX-M producers is not known.
In conclusion, the approach to detect contemporary ESBL-KoC strains should not consist on the use of standard gradient strip-or BMD-based confirmatory tests. In contrast, we suggest the simultaneous implementation of FEP/FEP-CL and CAZ/CAZ-CL CDTs or, alternatively, the DDST-30 including at least CAZ, FEP and ATM disks. The use of CAZ (10 mg) and CAZ-CL (10/10 mg) EUCAST-recommended disks seems to perform better than those suggested by the CLSI (30 mg and 30/10 mg, respectively) (33,39,40). However, further specific and comparative studies should address this aspect.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only. SUPPLEMENTAL FILE 1, PDF file, 0.8 MB.