Evaluation of Piperacillin-Tazobactam ETEST for the Detection of OXA-1 Resistance Mechanism among Escherichia coli and Klebsiella pneumoniae

ABSTRACT

Globally, piperacillin-tazobactam resistance among Escherichia coli and Klebsiella pneumoniae is driven by OXA-1 beta-lactamases. Expression of bla_OXA-1 yields piperacillin-tazobactam MICs of 8 to 16 μg/mL, which straddle the susceptible/susceptible-dose dependent breakpoint set by the Clinical and Laboratory Standards Institute in 2022. Variability of the reference broth microdilution method (BMD) was evaluated by manufacturing BMD panels using 2 brands of piperacillin, 2 brands of tazobactam and 2 brands of cation-adjusted Mueller-Hinton broth. In addition, ETEST, which harbors an intermediate dilution of 12 μg/mL was evaluated for the ability to differentiate isolates with and without bla_OXA-1. A collection of 200 E. coli and K. pneumoniae, of which 82 harbored a bla_OXA-1 gene, were tested. BMD variability was on average 1.3-fold, within the accepted 2-fold variability of MICs. However, categorical agreement (CA) between BMD reads was 74.0% for all isolates and 63.4% for those with a bla_OXA-1 gene and 81.3% for those without bla_OXA-1 detected (P = 0.004, Pearson’s Chi Square). ETEST overall CA with the BMD mode was 68.0% and essential agreement (EA) was 80.5%. For isolates with bla_OXA-1, CA was 50.0% and EA was 69.5%, versus 80.5% and 88.1%, respectively, for isolates without bla_OXA-1 (P < 0.0001 for both comparisons). All ETEST errors were major errors (false resistance) compared to BMD mode. However, the negative predictive value of the ETEST for the presence of bla_OXA-1 was 94.1%, compared to only 74.2% negative predictive value for BMD. Clinicians and microbiologists should be aware of the challenges associated with testing piperacillin-tazobactam in regions where bla_OXA-1 is prevalent.

INTRODUCTION

In 2018, Harris et al. published the results of the MERINO trial, failing to demonstrate non-inferiority of piperacillin-tazobactam versus meropenem for the treatment of patients with bloodstream infections caused by ceftriaxone not-susceptible Escherichia coli or Klebsiella pneumoniae (1). These findings led to significant concern regarding the continued use of piperacillin-tazobactam for the treatment of serious infections caused by ceftriaxone-resistant Enterobacterales (2, 3). In MERINO, no association between piperacillin-tazobactam MIC, measured by ETEST, and outcome could be found (1). However, post hoc analyses of the MERINO isolates revealed an association between mortality and piperacillin-tazobactam MICs ≥32 μg/mL, when measured by reference broth microdilution (BMD) (4). This finding is supported by a multitude of publications that demonstrate poor probability of achieving pharmacokinetic/pharmacodynamic (PK/PD) therapeutic targets (i.e., greater than 50% time above MIC) for piperacillin-tazobactam MICs of ≥16 μg/mL (5). Based in part on these data, the Clinical and Laboratory Standards Institute (CLSI) revised the piperacillin-tazobactam breakpoints for the Enterobacterales to ≤8 μg/mL (susceptible), 16 μg/mL (susceptible dose-dependent, SDD), and ≥32 μg/mL (resistant) in 2022 (5). The SDD breakpoint was chosen to enable the opportunity for extended-infusion piperacillin-tazobactam strategies, i.e., 4.5 g q6h as a 3-h infusion or 4.5 g q8h as a 4-h infusion, where a greater time above the MIC is predicted (5). SDD also serves as a buffer zone due to noted challenges with obtaining accurate piperacillin-tazobactam MICs, even by reference methods (5). The U.S. Food and Drug Administration (FDA) has not yet accepted the CLSI breakpoints, but the rationale for these is under review by FDA.

During the deliberations at CLSI over the piperacillin-tazobactam breakpoints, concern regarding testing inaccuracies for piperacillin-tazobactam were frequently cited by committee members and attendees. Unpublished anecdotes from participants at the meetings suggested significant challenges associated with obtaining a reproducible MIC result for piperacillin-tazobactam by the reference BMD, particularly when different lots of cation-adjusted Mueller-Hinton broth or piperacillin and tazobactam powders were used. Furthermore, post hoc analysis of MERINO trial isolates demonstrated high rates of very major errors (VME, false susceptibility) by commercial test methods used during the course of the study, when these were compared to BMD. In these analyses, BMD was performed on the isolates using a different inoculum, on a different day than the commercial methods, and BMD MICs were evaluated once per isolate, leaving some uncertainty regarding how to best interpret these findings (4). The preferred method for comparing two test systems is to use the same inoculum for both tests, as inoculum variability, even when adjusted to within the acceptable CLSI ranges, can significantly impact MIC results (6, 7). In addition, in our experience, testing errors are equally attributable to the initial reference BMD result, as they are to commercial test inaccuracies (8, 9). Since the publication of MERINO, bioMérieux has reformulated the piperacillin-tazobactam ETEST and shown, in a well-controlled trial, excellent performance of this new formulation compared to BMD. However, this study was conducted using FDA, not current CLSI breakpoints (10).

Globally, piperacillin-tazobactam resistance among ceftriaxone not-susceptible isolates of E. coli and K. pneumoniae is largely driven by narrow-spectrum, tazobactam-resistant OXA-1 beta-lactamases (4, 11). Challengingly, the presence of these enzymes result in piperacillin-tazobactam MICs of 8 or 16 μg/mL (4, 11), which straddle the susceptible/SDD breakpoint set by CLSI, as well as overlap the wild-type epidemiological cutoff for piperacillin-tazobactam, which is ≤8 μg/mL (5). The presence of a bla_OXA-1 resistance mechanism is challenging to detect by currently available phenotypic methods. The BMD reference method is performed using 2-fold dilutions of antimicrobial, which limit the ability of this test to detect nuanced variations in MIC values (9); an MIC that lies between 8 and 16 μg/mL may intermittently test at 8 or 16 μg/mL. ETEST offers a unique opportunity to evaluate for this resistance mechanism, as an intermediate dilution of 12 μg/mL is present on the test strip. In addition, ETEST may yield MICs slightly higher than reference BMD (as has been shown for other antimicrobials [12]), which could provide better differentiation of an isolate with bla_OXA-1 mechanisms from those that do not.

In this study, we evaluated the reproducibility of BMD MICs for piperacillin-tazobactam, through the use of different lots of piperacillin, tazobactam and cation-adjusted Mueller-Hinton broth. In addition, we sought to evaluate whether piperacillin-tazobactam ETEST was better able to predict the presence of bla_OXA-1 genes, in a large collection of well-characterized clinical isolates of E. coli and K. pneumoniae.

MATERIALS AND METHODS

Isolate collection.

Bacterial isolates were obtained from JMI laboratories, with the request of an even distribution of E. coli and K. pneumoniae (n = 100 each) displaying diverse piperacillin-tazobactam MICs, with and without the bla_OXA-1 gene detected. Isolate characterization by JMI was performed as described elsewhere (12). In this collection, 42 E. coli and 40 K. pneumoniae harbored a bla_OXA-1 gene. Isolate MIC distribution and beta-lactamase genes are shown in Table S1, from JMI. One hundred and four isolates (52%) had an MIC of 8 μg/mL, 16 μg/mL or 32 μg/mL (i.e., within a doubling dilution from the 2022 breakpoints) by the original JMI BMD MIC, as shown in the Fig. S1. Isolates were shipped frozen to the laboratory and subcultured twice on TSA supplemented with 5% sheep’s blood (BBL, BD, Sparks MD) prior to testing with BMD or ETEST (bioMérieux, Durham, NC).

MIC testing.

Broth microdilution (BMD) was performed according to standards described by CLSI (13). Briefly, panels were prepared in-house using an HTF plate dispenser (MDZ Automation, Chicago, IL) into 96-well polystyrene plates. Antimicrobials on each panel included piperacillin ranging from 512 μg/mL to 0.5 μg/mL in a 2-fold dilution series in cation-adjusted Mueller-Hinton broth (CA-MHB). An additional intermediate dilution of 12 μg/mL of piperacillin was included, to replicate the concentrations available by ETEST. Tazobactam was held at 4 μg/mL in each well. Two brands of piperacillin (USP, Rockville, MD; and Sigma-Aldrich, St. Louis, MO), two brands of tazobactam (USP and Sigma) and two brands of CA-MHB (BD and Difco, Sparks MD) were used to generate 6 unique dilution series of piperacillin-tazobactam on each panel. A meropenem (Sigma-Aldrich) control was included, which ranged from 4 μg/mL to 0.004 μg/mL in a 2-fold dilution series in each brand of CA-MHB (i.e., two dilution series of meropenem). Panels were frozen at −70°C prior to use and a single panel lot was used for the entire study. Quality control was performed using E. coli ATCC 25922, E. coli ATCC 35218, and Pseudomonas aeruginosa ATCC 27853 on each day of testing.

Isolates were tested by making a suspension equivalent to a 0.5 McFarland standard in water. Isolates were inoculated to BMD and Mueller-Hinton Agar (MHA, BD) for ETEST using the same inoculum. Plates were incubated at 35°C in ambient air for 16 to 18 h and read by two readers. ETEST results were interpreted according to the manufacturer’s instructions.

Data analysis.

Given all 6 piperacillin-tazobactam BMD results were considered a ‘gold standard’, the modal MIC from the 6 replicates was used as the reference MIC in analyses. JMI MICs were not used in the analysis, but rather only to select isolates for study. Essential agreement (EA, i.e., MIC agreement ±1 log₂ dilution) and categorical agreement (CA, i.e., agreement of category interpretations of susceptible, susceptible-dose dependent and resistant) for individual BMD MICs or for ETEST to BMD, was evaluated by using the modal MIC value for each isolate as reference. Error rates were defined as very major (number false susceptible), major (number of false resistant) and minor (SDD by one method but either susceptible or resistant by the other), compared to the modal MIC interpretation. VME rates were scored using the number of reference (modal) MICs that were resistant as the denominator, and ME rates were scored using the number of reference (modal) MIC results that were susceptible was the denominator. Error rate bounded method of analysis was performed as described by CLSI in the M23 guideline (14). Clinical breakpoints applied in this study (Table S2), included the current (2022) CLSI breakpoints of ≤8 μg/mL (susceptible), 16 μg/mL (SDD) and ≥32 μg/mL (resistant) (15), and the current (September 2022) FDA breakpoints of ≤16 μg/mL (susceptible), 32 to 64 μg/mL (intermediate) and ≥128 μg/mL (resistant) (https://www.fda.gov/drugs/development-resources/fda-recognized-antimicrobial-susceptibility-test-interpretive-criteria).

MIC values were transformed to log₂ values prior to other statistical analyses, which were performed using GraphPad Prism, as described previously (16). Variability was assessed by taking the standard deviation from the mean of log₂ transformed MIC across six BMD reads (16).

RESULTS

Variability of piperacillin-tazobactam BMD MICs.

Piperacillin-tazobactam BMD was performed in replicates of 6, using the same inoculum but different lots of piperacillin, tazobactam, and CA-MHB. The number of unique BMD TZP MICs for each isolate was evaluated. Sixty-six isolates had one MIC, 111 had 2, 22 had 3, and one isolate had 4 piperacillin-tazobactam MIC values measured. No association between the number of MICs observed and the presence or absence of bla_OXA-1 was found (not shown). Variability from the mode ranged from 4 dilutions below to 2 dilutions above the modal piperacillin-tazobactam MIC (Table 1), and no trend by species or presence/absence of bla_OXA-1 was observed. On average, MIC variability was 1.3-fold across piperacillin-tazobactam conditions (i.e., the average of standard deviations of log₂ transformed MIC values) (16). This was the same for E. coli, K. pneumoniae, and isolates with and without bla_OXA-1 and is less than the 2-fold dilution expected for MIC testing by BMD (14). Only 12 isolates displayed greater than 2-fold variability (ranging from 2.05 to 4.18-fold variability), one of which harbored bla_OXA-1. Among these 12 isolates, two displayed categorical disagreement between MICs. The first was a K. pneumoniae isolate with BMD piperacillin-tazobactam MICs of 4 μg/mL (susceptible, n = 4 readings) and 16 μg/mL (SDD, n = 2 readings) and that harbored a bla_LEN-2 gene as the sole detected beta-lactamase. The second was a K. pneumoniae isolate with piperacillin-tazobactam MICs of 2 μg/mL (n = 1), 4 μg/mL (n = 1), 8 μg/mL (susceptible, n = 2), and 16 μg/mL (SDD, n = 2) and that harbored a bla_SHV-11 gene as the sole detected beta-lactamase.

TABLE 1.

Number of readings for BMD piperacillin-tazobactam MICs^a

	Dilutions from the mode
Isolate group	−4	−3	−2	−1	0	1	2
All isolates	8	6	16	140	951	75	4
E. coli	4	0	7	67	481	39	2
OXA-1	0	0	5	34	191	20	2
No OXA-1	4	0	2	33	290	19	0
K. pneumoniae	4	6	9	73	470	36	2
OXA-1	0	0	1	32	184	22	1
No OXA-1	4	6	8	41	286	14	1

^aReadings ±1 dilution from the mode were considered in essential agreement (EA).

Overall, EA with the piperacillin-tazobactam mode was 186/200 (93.0%), and CA by CLSI breakpoints was 148/200 (74.0%). As expected based on the analysis above, EA did not statistically differ between isolates with and without bla_OXA-1 (P = 0.12, Pearson’s Chi Square test). However, a significant difference was noted in CA (P = 0.004, Pearson’s Chi Square test) between isolates with (52/82, 63.4%) and without (96/118, 81.3%) bla_OXA-1 (not shown). Overall, 52 minor errors were observed across BMD reads, compared to the mode. Thirty of 52 (57.7%) minor errors were for isolates with bla_OXA-1 (13 E. coli and 17 K. pneumoniae), and 22 (12 K. pneumoniae, 10 E. coli) were for isolates without a bla_OXA-1 gene. Among the minor errors, 35 occurred due to one or more MIC reading higher than the mode, including 22 from a susceptible mode to SDD and 13 from SDD to R. For the 17 isolates with one or more MIC lower than the mode that resulted in minor errors, 10 were from a SDD mode to susceptible and 7 had a resistant mode with one or more readings at SDD. Four isolates with minor errors were associated with MICs that were not in EA from the mode, 1 of which harbored bla_OXA-1 and 1 of which was E. coli with a bla_TEM-1 gene detected.

Meropenem was used as a control and tested in duplicate using one lot of meropenem and two brands of CA-MHB. All results where within EA and CA of each other, and 88% (176/200) isolates demonstrated the exact same MIC by both dilutions. All QC results for both piperacillin-tazobactam and meropenem were within CLSI ranges, with no clear skewing of results on either end of the QC range across the different testing parameters.

Performance of ETEST compared to BMD.

We first evaluated ETEST using the BMD mode as the reference (Table 2). EA was 80.5% (161/200) and CA was 68% (136/200), including 15 major errors (14.8%) and 49 (24.5%) minor errors (Table 2), when assessed using CLSI breakpoints. Eight of the 15 ME were for isolates with variable BMD MICs, all of which had 1 to 2 MIC readings of 16 μg/mL (SDD), which would have been classified as a minor error. Among the minor errors, 45 (91.8%) were due to a higher MIC by ETEST than the BMD mode. Twenty-nine (59.2%) minor errors were for isolates that also displayed minor errors by BMD against itself, 25 of which had at least one BMD read in CA with the ETEST result. Overall, EA was 69.5% (57/82) and CA was 50% (41/82) for isolates with bla_OXA-1, whereas it was 88.1% and 80.5%, respectively, for isolates without bla_OXA-1 (P < 0.0001 for both EA and CA differences by Chi-square test).

TABLE 2.

Summary of categorical agreement for isolates using CLSI breakpoints, based on presence of bla_OXA-1 gene

Resistance gene	N	% EA	% CA	VME	%	ME	%	MIN	%
bla OXA-1	82	69.5	50.0	0	0	13	52.0	28	34.1
No blaOXA-1	118	88.1	80.5	0	0	2	2.6	21	17.7
All	200	80.5	68.0	0	0	15	14.9	49	24.5

As this isolate collection was enriched for isolates near the breakpoint (i.e., n = 94 isolates had MICs of 8, 16, or 32 μg/mL by the modal MIC, including 64 with bla_OXA-1), an error-rate bounded method of evaluation was applied (Table 3). By this method, an increased error rate is accepted when the isolates are within EA with the reference BMD MIC, with up to 40% minor errors accepted for isolates within a dilution of the SDD breakpoint (i.e., MIC 8 to 32 μg/mL) and up to 5% minor errors for those outside this MIC range (≤4 μg/mL and ≥64 μg/mL) (14). Minor errors for both isolates with reference MICs 8to 32 μg/mL and those with ≤4 μg/mL exceeded these limits (Table 3). Similarly, the major error rate for isolates with an MIC within a dilution of the susceptible breakpoint, and those with MICs ≤4 μg/mL exceeded acceptable rates of <10% and <2%, respectively (Table 3) (14). As ETEST is an FDA-cleared product, we evaluated performance using FDA breakpoints, again using the error-rate bounded method (Table 3). Major error rates were within acceptance limits, but minor error rates fell outside accepted limits, at 25.8% for isolates with MICs ≤8 μg/mL (acceptable is <5%) and 28.5% for isolates with MICs ≥256 μg/mL (acceptable is <5%), albeit only 20 isolates harbored MICs in this bin.

TABLE 3.

Piperacillin-tazobactam ETEST performance versus BMD modal MIC, applying CLSI and FDA breakpoints for 200 isolates of E. coli and K. pneumoniae^a

MIC range by reference (μg/mL)	N	VME	%	ME	%	MIN	%
CLSI breakpoints
≤4	62			2	3.3	5	8.1
8–32	94	0	0	13	13.8	42	44.6
≥64	44	0	0	0	0	0	0
All	200	0	0	15	14.8	49	24.5
FDA breakpoints
≤8	101			1	1	26	25.8
16–128	84	0	0	2	2.4	30	35.7
≥256	15	0	0			1	6.7
All	200	0	0	4	2.9	57	28.5

^aCLSI, Clinical and Laboratory Standards Institute; N, number; VME, very major error; ME, major error; MIN, minor error; SDD, susceptible-dose dependent MIC; I, intermediate MICs. Bolded values exceed acceptance limits set in M23 (14).

Given the variability of the reference method, performance was recalculated by excluding isolates that did not demonstrate EA or CA across the BMD dilutions. Among 186 isolates with all BMD results within EA of the mode, 153 (82.3%) were in EA with ETEST and 125 (67.2%) were in CA (by CLSI breakpoints) with ETEST. If isolates with no CA across the BMD reads were excluded (n = 52), EA between BMD mode and ETEST was 83.1% (123/148) and CA was 81.8% (121/148). In the latter calculation, 7 isolates displayed major errors (9.0% of isolates with susceptible modes) and 20 isolates displayed minor errors (13.6% of all isolates tested). Among the major errors, 6 (85.7%) were for E. coli and all 7 (100%) were for isolates with bla_OXA-1 (Table 4). Among the 20 minor errors, 14 (70%) had an ETEST MIC that was in EA with the BMD mode. Fourteen (70%) were for E. coli and 10 (50%) were for isolates with bla_OXA-1 (Table 4), all of which yielded a more resistant result by ETEST. Again, the association between better CA by ETEST compared to BMD mode for isolates without bla_OXA-1 (86/96, 89.6%) versus those with bla_OXA-1 (35/52, 67.3%) remained (P < 0.001, Pearson’s Chi Square). Similarly, EA was better between ETEST and BMD mode for isolates without (85/96, 88.5%) than with (38/52, 73.1%), bla_OXA-1 for this subset of isolates (P = 0.01, Pearson’s Chi Square).

TABLE 4.

Summary of isolates with errors between piperacillin-tazobactam BMD mode and ETEST, excluding isolates without categorical agreement across BMD reads

		TZP BMD MIC (μg/mL)							BMD mode interpretation	ETEST MIC (μg/mL)	ETEST interpretation	OXA
Isolate	Species	TZP A	TZP B	TZP C	TZP D	TZP E	TZP F	Mode
Major errors
935537	E. coli	8	8	8	8	8	8	8	S	64	R	yes
1010382	E. coli	8	8	8	8	4	8	8	S	24	R	yes
1047153	E. coli	8	8	4	4	4	8	8	S	>256	R	yes
1089754	E. coli	8	4	4	4	4	8	4	S	32	R	yes
1096213	E. coli	8	8	8	8	8	8	8	S	48	R	yes
1103942	E. coli	4	4	4	4	2	4	4	S	32	R	yes
1085556	K. pneumoniae	8	8	8	8	8	8	8	S	32	R	yes
Minor errors
854173	E. coli	8	8	8	8	8	8	8	S	12	SDD	no
875258	E. coli	12	8	12	12	12	8	12	SDD	24	R	no
878462	E. coli	16	12	12	16	12	16	16	SDD	32	R	no
881178	E. coli	8	8	4	8	8	8	8	S	12	SDD	no
919308	E. coli	12	12	12	12	12	12	12	SDD	24	R	no
932991	E. coli	4	4	4	8	8	8	4	S	16	SDD	no
952507	E. coli	12	12	12	12	12	16	12	SDD	24	R	no
939687	E. coli	8	4	8	8	4	4	8	S	12	SDD	yes
941524	E. coli	16	12	12	12	12	12	12	SDD	32	R	yes
984652	E. coli	2	4	4	4	4	4	4	S	12	SDD	yes
990076	E. coli	8	8	8	8	8	8	8	S	16	SDD	yes
1089395	E. coli	8	8	8	8	8	8	8	S	12	SDD	yes
1166103	E. coli	8	8	4	4	4	4	4	S	16	SDD	yes
1166699	E. coli	16	12	12	12	12	12	12	SDD	48	R	yes
1027048	K. pneumoniae	2	2	4	2	2	2	2	S	12	SDD	no
1162424	K. pneumoniae	8	4	4	4	4	4	4	S	12	SDD	no
1168943	K. pneumoniae	32	32	32	32	32	32	32	R	16	SDD	no
952367	K. pneumoniae	16	12	12	12	12	12	12	SDD	24	R	yes
982114	K. pneumoniae	16	12	12	12	12	12	12	SDD	24	R	yes
1152213	K. pneumoniae	12	12	12	12	12	16	12	SDD	24	R	yes

To attempt to understand if performance was driven by a given brand of piperacillin powder, tazobactam powder, or CA-MHB, ETEST performance was scored against individual BMD reads. In this analysis, CA ranged from a low of 68.5% (using USP piperacillin, Sigma tazobactam and Difco CA-MHB) to a high of 83% (obtained both using USP piperacillin, USP tazobactam and Difco CA-MHB, and Sigma piperacillin, USP tazobactam and Difco CA-MHB). No clear association between performance and lot of powder or media was observed (Table 5). CA differences between BMD conditions were driven by differences in minor errors, which ranged from 24% (USP piperacillin, Sigma tazobactam and Difco CA-MHB) to 10.5% (USP piperacillin, USP tazobactam and Difco CA-MHB). Major errors were similar between BMD results, ranging from 12 (12.9%) to 16 (15.4%), Table 5.

TABLE 5.

Evaluation of the performance of ETEST compared to different BMD conditions^a

	BMD condition
Error type	USP piperacillin USP Tazobactam BD CA-MHB	Sigma piperacillin USP Tazobactam BD CA-MHB	USP piperacillin Sigma Tazobactam BD CA-MHB	USP piperacillin USP Tazobactam Difco CA-MHB	Sigma piperacillin USP Tazobactam Difco CA-MHB	USP piperacillin Sigma Tazobactam Difco CA-MHB
CA	69.5%	80.5%	78.5%	83.0%	83.0%	68.5%
Very major errors	0 (0%)	0 (0%)	0 (0%)	0 (0%)	0 (0%)	0 (0%)
Major errors	14 (13.9%)	15 (15.3%)	16 (15.4%)	13 (13.7%)	12 (12.9%)	15 (15.3%)
Minor errors	47 (23.5%)	24 (12.0%)	27 (13.5%)	21 (10.5%)	22 (11.0%)	48 (24.0%)

^aBMD, broth microdilution; CA-MHB, cation-adjusted Mueller Hinton Broth; CA, categorical agreement.

Evaluation of BMD and ETEST compared to presence of OXA-1.

Overall, ETEST provided a better differentiation between isolates with and without the bla_OXA-1 gene than did BMD modal MICs (Fig. 1, Table 6). The addition of a 12/4 μg/mL dilution in the BMD panels did not improve differentiation. Only 4 isolates that harbored bla_OXA-1 were classified as susceptible by ETEST, versus 25 by BMD. The sensitivity of a not -susceptible (SDD or resistant) interpretation for the presence of bla_OXA-1 was 69.5% (95% CI, 58.2 to 78.9%) by BMD versus 95.1% (87.3 to 98.4%) by ETEST. Specificity was lower for both tests, at 64.4% (55 to 72.8%) and 54.2% (44.8 to 63.3%) for BMD and ETEST, respectively. Importantly, the negative predictive value of a susceptible ETEST MIC was 94.1% for bla_OXA-1 versus 74.2% for BMD.

FIG 1.

Piperacillin-tazobactam MICs by BMD mode (A) or ETEST (B), stratified by the presence (black bars) or absence (white bars) of bla_OXA-1.

TABLE 6.

Evaluation of BMD and ETEST versus presence of bla_OXA-1 gene, using CLSI breakpoints^a

	BMD		ETEST
Parameter	%	95% CI	%	95% CI
Sensitivity	69.5	(58.2−78.9)	95.1	(87.3−98.4)
Specificity	64.4	(55−72.8)	54.2	(44.8−63.3)
Positive predictive value	57.5	(47.2−67.3)	59.1	(50.2−67.4)
Negative predictive value	75.2	(65.5−83.1)	94.1	(84.9−98.1)

^aBMD, broth microdilution; CI, confidence interval.

We evaluated the ability of different reference BMD conditions, the sensitivity for the presence of bla_OXA-1 of a not susceptible BMD MIC ≥16 μg/mL ranged from 65.8 – 71.9%, still much lower than ETEST (not shown).

DISCUSSION

The prevalence of narrow-spectrum OXA-1 penicillinases is not well appreciated by many infectious disease physicians or clinical microbiologists but is estimated to be present in roughly 30% of ceftriaxone not-susceptible K. pneumoniae and E. coli in the United States (JMI data on file). No commercial tests are currently available detect bla_OXA-1 by molecular methods, leaving clinicians to rely on the piperacillin-tazobactam MIC as the sole mechanism available to detect this resistance mechanism. However, expression of bla_OXA-1 leads to only modestly elevated piperacillin-tazobactam MICs of 8 to 16 μg/mL, which overlap the ECV and the revised 2022 CLSI susceptible breakpoint of ≤8 μg/mL. In this study, we demonstrated reference BMD is poor at differentiating isolates with and without this enzyme, even when a nonstandard dilution of 12 μg/mL was included in the panel (Table 6). While not well studied on other commercial systems, an increased rate of very major errors by Vitek-2 (bioMérieux) has been shown for isolates with bla_OXA-1, when Vitek-2 MICs are compared reference BMD (4). Generating confidence among clinical microbiologists, pharmacists and infectious diseases physicians on the accuracy of piperacillin-tazobactam results is paramount, given the alternative is to transition all patients to a carbapenem, which would foster further antimicrobial resistance (17). For critical cases, use of ETEST “confirmation” of susceptible results may provide added reassurance of susceptibility, much like the use of a vancomycin MIC of 1.5 μg/mL determined by ETEST has come to inform use of alternatives to vancomycin for the treatment of S. aureus bacteremia (18). In this study, we showed a negative predictive value of an ETEST MIC ≤8 μg/mL of 94% for the presence of bla_OXA-1. These data, coupled with the updated 2022 CLSI breakpoints, provide substantial reassurance that an isolate is treatable with piperacillin-tazobactam (5). We did not evaluate the performance of disk diffusion in this study, as prior data suggest the formulation of piperacillin-tazobactam disk may need to be adjusted (15). However, it is possible the disk diffusion test may also serve as a better differentiator of isolates with and without bla_OXA-1 than is BMD. Of course, continued use of piperacillin-tazobactam in these scenarios must be taken into consideration with the clinical context of the patient’s infection, response to treatment and comorbidities, and not the piperacillin-tazobactam MIC alone, regardless of the method used to measure its value (19). This being said, the presence of bla_OXA-1 was associated with increased risk of 30-day mortality for patients randomized to piperacillin-tazobactam in MERINO (4). While no studies have assessed the efficacy of piperacillin-tazobactam for ceftriaxone not-susceptible infections in regions of low incidence of bla_OXA-1, preliminary data suggest efficacy (20).

This study highlights several factors important with regard to reference BMD, which is currently used by FDA, CLSI and European agencies to assess the performance of commercial antimicrobial susceptibility testing (AST) products. First, we reaffirm there is no such thing as a single ‘reference’ MIC. By adjusting parameters within the reference, i.e., use of different reference powders and CA-MHB, CA for BMD to itself fell below what is currently accepted for commercial devices (i.e., CA was <90%). While we demonstrated substantial variability of categorical interpretations by the BMD method, the method performed as expected for 95.5% of isolates, with only 12 displaying variability beyond the expected 2-fold variation across MICs. We also were unable to demonstrate that the presence or absence of bla_OXA-1 led to increased variability of MIC results (Table 1). We did not evaluate the impact of inoculum, nor variability of readers, as the BMD tests were set up using the same inoculum and read by the same individuals. Inoculum has been repeatedly shown to be one of the most significant variables that leads to MIC differences by BMD (21, 22), even when inocula within the CLSI acceptable range of 2 × 10⁵ to 8 × 10⁵ CFU/mL are used (7). We would anticipate the variability observed herein may be inflated if individual inocula were used for each BMD condition. Importantly, we did find that CA was worse for isolates with bla_OXA-1, which would be expected, given the fact that expression of this enzyme leads to MICs that are bisected by the current CLSI breakpoints. While FDA breakpoints are not associated with this dilemma (not shown), they are not predictive of treatment success (5) and 56.1% of isolates with bla_OXA-1 in this study were interpreted as susceptible by the FDA breakpoints.

Performance of ETEST, when scored by an individual BMD condition set, ranged substantially, with CA as low as 68.5% and as high as 80.3% across BMD parameters (Table 5). Importantly, subtle changes such as formulations of reagents used herein, may lead a test system to move from “acceptable” to “unacceptable” correlation with the reference. No one condition we evaluated is considered superior over others, as the reference method itself does not stipulate which brands of powders or CA-MHB to use (13) and both antimicrobial powders and CA-MHB conformed with CLSI standards (13), per the manufacturer. We were unable to document any particular set of reagents that skewed BMD performance (Table 5), and it may be that slight variance of the stock concentrations prepared and small pipetting errors could have driven the variability of the method, not necessarily the stock powders themselves, although this supposition is worth further investigation. Any beta-lactam combination agent is subject to further such variability as two, not one, antimicrobials must be controlled. All quality control results were within range and did not display skewing toward one end of the range or the other (not shown). This fact is one limitation of the currently used QC strains and CLSI-published QC ranges, which, by design, yield acceptable performance in the face of variability in media lots, inocula and interlaboratory testing (14). Importantly, we did not evaluate the impact of media brand on ETEST variability, and it is entirely possible similar fluctuation of MIC results would be observed for this test method, as were observed by BMD.

The categorical errors observed in this study, for both BMD against itself, and for ETEST, are without doubt inflated by the challenging set of isolates evaluated (i.e., ~45% of isolates with MICs 8 to 32 μg/mL). The accepted variability of BMD is ±1 doubling dilution. When a majority of isolates evaluated harbor an MIC near a breakpoint, it is inevitable that this accepted variability of the BMD test will result in categorical errors (21). These data should be used to inform selection of isolates for both manufacturers and clinical laboratories when evaluating AST devices. Performance will inevitably be poor if isolates are enriched for those with MICs near a breakpoint. This is a particular problem for piperacillin-tazobactam, where the breakpoints abut the ECV (5). In contrast, the modal meropenem MIC for this isolate collection was 0.032 μg/mL, which is six doubling dilutions away from the susceptible breakpoint of ≤1 μg/mL (15). Even if meropenem MICs varied substantially across “reference” BMD methods, it would not be anticipated to impact CA rates. In contrast, the EA for piperacillin-tazobactam for BMD was 93%, but still resulted in unacceptable minor error rate of 26.5%, simply due to the location of piperacillin-tazobactam MICs in the studied population to the breakpoint.

In summary, we demonstrated that the piperacillin-tazobactam ETEST susceptibility (MIC ≤8 μg/mL) provided good sensitivity to rule out the presence of bla_OXA-1 in a collection of ceftriaxone not-susceptible E. coli and K. pneumoniae. BMD reproducibility for piperacillin-tazobactam was within expected ranges, but due to the location of the breakpoint to the wild-type piperacillin-tazobactam MICs, CA for both the reference to itself, and ETEST to the reference, was low, with unacceptably high rates of minor errors, particularly for isolates with the bla_OXA-1 resistance mechanism. Laboratories should be aware of this challenge associated with piperacillin-tazobactam MIC testing, when undertaking validation studies prior to implementing the CLSI breakpoints. Evaluation of EA should be taken into primary consideration when assessing the results of such validation studies, with higher than typically allowable minor errors if isolates are within EA, as recommended by the International Organization for Standardization in the ISO 20776-2 standard (23). Clinicians should be aware of the enhanced rate of variability for piperacillin-tazobactam, understanding that the CLSI susceptible breakpoint of ≤8 μg/mL was chosen over a susceptible breakpoint of ≤16 μg/mL (which was supported by PK/PD data and the MERINO trial) in large part due to this MIC variability issue. Finally, we note that bla_OXA-1 is not the only piperacillin-tazobactam resistance mechanism. Others, such as hyper-expression of bla_TEM-1 may similarly impact piperacillin-tazobactam MIC performance (24). The transition of susceptibility testing from phenotypic alone assessments to the inclusion of some genotyping, such as the evaluation for tazobactam-resistant beta-lactamases, will likely help further improve judicious use of piperacillin-tazobactam, and other antimicrobials, in the future.

Funding Statement

Funded by an Investigator-initiated grant to RMH