Candida auris is an emerging organism that has caused nosocomial infections and outbreaks in hospitals and other health care facilities globally. It is almost ubiquitously fluconazole resistant and can acquire echinocandin resistance within a few weeks of echinocandin therapy [1,2]. Reported amphotericin B resistance rates for C. auris vary considerably. Reviewing 42 publications from 2009 to 2019, Chaabane et al. [3] reported amphotericin B resistance rates ranging from 0% to 100% [3]. Since then, Lockhart et al. [4] reported 35% resistance among 54 isolates from Pakistan, India, South Africa, and Venezuela during 2012–2015, Helleberg et al. reported 0% resistance among 122 isolates from India [5], Ben Abid et al. [6] reported 0% resistance among 122 isolates from India, they also reported 79% amphotericin B resistance among 76 isolates in Qatar 2020–2021, whereas amphotericin B resistance rates varied between 33%, 59%, and 0–22.6% for Colombian isolates across studies and methods used for antifungal susceptibility testing [7–9].

These differences may reflect either clinically relevant differences in amphotericin B susceptibility potentially driven by clonality among outbreak strains, technical issues with testing or adoption of a clinical breakpoint (BP) that bisects the wild-type population. Indeed, Siopi et al. [10] recently found that the use of Sensititre YeastOne led to systematically higher MICs than Clinical and Laboratory Standards Institute (CLSI), with modal (range) MICs of 2 (1–8) versus 1 (0.25–1) mg/L, respectively. This led to amphotericin B resistance rates of 0 and 89% using the susceptibility breakpoint of 1 mg/L. Similarly, Kathuria et al. [11] found notable differences in MICs across CLSI, Vitek 2, and Etest results for 90 C. auris isolates, resulting in amphotericin B resistance of 16%, 100%, and 1%, respectively, depending on the method used.

We compared MIC results with two reference methods (European Committee on Antimicrobial Susceptibility Testing (EUCAST) and CLSI) and two brands of gradient diffusion strips (Etest and MTS strips) using a shared C. auris strain collection that includes at least three clades and is derived from multiple states in the United States.

Forty C. auris strains from nine U.S. states were included. Clade information was not available for six isolates. Of the 34 isolates that were typed, 20 were Clade I, 13 were Clade III, and one was Clade IV. The isolates were chosen in a way that only a single isolate came from any perceived geographic clone. These isolates were a fair representation of what can be expected across country-wide surveillance. Amphotericin B susceptibility was investigated using two reference methods (EUCAST E.Def 7.4 [12,13] and CLSI M27 Ed4 [14]) and two gradient diffusion strip tests (Etest [bioMérieux] and MTS [Liofilchem]). EUCAST MICs were determined three times on three separate days. Flat bottom plates (Nunc MicroWell, Nunclon Delta-treated MicroWell plates, catalogue no. 167008; Thermo Fisher Scientific) were prepared using “RPMI 1640 with MOPS and glucose pH 7” (SSI Diagnostica, catalogue no. 60984, SSI Diagnostica) and the International Organization for Standardization (ISO) method for drug dilution (two sets of plates) and serial dilution (one set of plates) [15]. MICs were read spectrophotometrically (490 nm wavelength and 90% growth inhibition) after 1 day of incubation.

CLSI MICs were determined twice on separate days using two internal lots of RPMI-1640 medium prepared on separate days from one purchased lot of medium (Sigma Aldrich; Catalogue No. R6504; Lot No. 2482886) and round bottom plates (Corning Costar Cell Culture-Treated plates, catalogue no. 3799, Fisher Scientific). Plates were prepared using the ISO method for drug dilution, as also published in the CLSI M27 Ed4 document. MICs were read visually using the endpoint of complete inhibition of visual growth after 1 day of incubation.

Etest MICs were determined in two laboratories: (a) consecutively as part of the ongoing routine testing, and (b) batched in another laboratory. In routine testing (at the CDC) a single reader read each individual Etest, but there were three people in total that read the MICs. All three were trained to read Etest by the same person and there was no discernible difference in the average MIC values across the readers (data not shown). In both laboratories, the standard procedure for Etest MIC determination was to inoculate agar once using a swab dipped once in the inoculum suspension, incubate at 35–37 °C, and read the MIC at 1 day (unless growth was insufficient). A third testing was performed strictly following Etest manufacturer instructions with double inoculation of the agar with a swab dipped twice in the inoculum suspension and plates read after 24 hours and 48 hours by three independent mycology laboratory staff members skilled in antifungal susceptibility testing.

MTS (Liofilchem) strip MICs were determined in batch testing. Suspensions of each C. auris strain (0.5 McFarland corresponding to ~1 × 10⁶ CFU/mL) were prepared. Sterile swabs were dipped once into the suspensions and used to inoculate the RPMI plates. The plates were briefly allowed to dry and then incubated at 35 °C for 1 day before reading the MICs.

For all MIC test methods, Candidia. krusei ATCC 6258 and Candidia. parapsilosis ATCC 220900 were included for quality control (QC). A summary of the retrieved QC results is presented in Table S1. Unfortunately, QC data for the consecutive testing was not retrievable, but QC strains were included every time a clinical isolate was tested, and the result was only accepted if both QC strains were within the QC range.

EUCAST and CLSI amphotericin B MIC testing were robust across the included method variables (Table 1). The modal MIC was 1 mg/L for both methods, distributions unimodal and narrow, and GM-MICs similar (range 0.74–1.07). However, applying a BP of 1 mg/L led to differential susceptibility classification (0% for EUCAST and 8–28% resistance for CLSI). For EUCAST specifically, the MICs obtained using plates prepared with serial dilution (with pipette tip changes after columns 4 and 7) and ISO dilution for plate preparation were compared [15]. A significant correlation was found between the MICs for the three experiments (Fig. 1) and no pattern was observed between MIC and position of the isolate on the plate (inner or outer rows, Tables S2–S4).

Gradient diffusion strip testing with one swab inoculation resulted in slightly lower GM-MICs but wider MIC distributions and high resistance rates (25–30%) compared with the results for the reference methods. The MICs increased when the manufacturer’s recommendation to inoculate twice was followed, resulting in modal MIC of 2–4 mg/L and resistance rates of 45–58% and 60–63% when read after 24 hours and 48 hours, respectively. Seven of the eight Etest datasets were bimodal, with the first peak around 0.125–0.5 mg/L and the second larger peak around 1–4 mg/L, the exception being the double inoculation plates read by reader one after 24 hours incubation.

A significant and either moderate (Spearman r: 0.5– < 0.7) or strong (≥0.7) correlation was found for all combinations of the 14 MIC data set obtained with EUCAST, CLSI, and gradient diffusion strip testing, except for the CLSI BMD2 dataset, for which 36/40 MICs fell at 1 mg/L. (Fig. 1). Moreover, MICs from the EUCAST, CLSI, and gradient diffusion strips all correlated to the optical density of drug-free control in EUCAST wells, suggesting that some isolates grew better than others and that this was associated with MIC (Table 2 and Fig. 1).

The EUCAST and CLSI broth microdilution MIC results were in close agreement with each other across the potential sources of variation included in the study. Thus, CLSI MIC testing was robust across two different batches of broth medium; EUCAST MIC testing was not influenced by choice of plate preparation method (using the ISO scheme for drug dilution, or serial dilution in the plate with two tip changes) nor by the position on the plate, in contrast to what can be observed for isolates from other species that grow less well in the middle rows of the plate [15]. The gradient diffusion strips generated wider and bimodal distributions with reader to reader and centre to centre variation. This was particularly notable for the consecutive Etest MIC testing data set, both compared with the batched Etest and MTS strip testing and to the reference methods. Greater variation is expected during consecutive testing; however, in this case MICs covered nine dilutions compared with four for batched testing, which may compromise reproducibility and correct susceptibility classification. Of note, all QC MICs fell within the target (EUCAST) and within the range (remaining tests), meaning that variation was not captured by QC strains (Table S1). Therefore, our results add to the concern on amphotericin B susceptibility testing using commercial methods [10,11].

A correlation was found between MICs for the isolates by the different methods. This may suggest a true differential susceptibility, or it may relate to differential growth rates of the strains. MICs in general depend on growth, the classical example being the higher MICs obtained for plates read after two rather than 1 day of incubation. This is particularly true for diffusion tests where the endpoint is not determined as a fixed proportion of the growth for the growth control for that particular isolate and incubation time. Of note, the mean optical density of positive control wells in EUCAST plates correlated to the MIC not only for EUCAST results but also for CLSI, Etest, and MTS results, suggesting that some isolates grew better than others in all three labs and that this was associated with the MIC. This raises the question of whether this MIC variation, that in reference testing is discrete, correlated to clinical efficacy or is simply a reflection of growth in vitro.

Few studies have investigated underlying mechanisms in C. auris isolates classified as amphotericin B resistant. The biosynthesis of ergosterol in Candida cells involves a number of enzymatic steps encoded by erg6, erg11, erg24, erg25, erg26, erg27, erg2, erg3, erg5, and erg4 in sequential order [17]. Amphotericin B resistance in Candida spp. has been associated with mutations in erg2 [18,19] and erg6 [20] and with combined mutations in erg11 and in erg3 or erg5 [18–22]. In C. auris, Carolus et al. [23] found that simultaneous emergence of nonsense mutations in erg3 and erg11 in experimentally evolved multidrug resistant C. auris led to decreased amphotericin B susceptibility (MIC 2 mg/L) and that a mutation in mec3, a gene involved in DNA damage homeostasis, further increased polyene MIC (4 mg/L) as determined by the CLSI method. Ben Abid et al. [6] reported a unique premature stop codon in erg3 in an isolate amphotericin B resistant (MIC 2 mg/L) with Vitek2 testing; however, nine isolates with higher amphotericin B MICs (4–≥16 mg/L) did not harbour this or other known amphotericin B associated mutations [6]. Escandón et al. [24] reported a significant association between four nonsynonymous mutations and elevated amphotericin B Etest MIC. One of these was located in a gene predicted to encode for a transcription factor similar to FLO8, which in C. albicans has been associated with biofilm formation and amphotericin B resistance [25], whereas another mutation was located in a gene predicted to encode for a membrane transporter of unknown importance for amphotericin B susceptibility. Jacobs et al. whole genome sequenced 19 consecutive isolates, all of which were classified as amphotericin B resistant by colorimetric susceptibility testing using the Sensititre YeastOne assay and four of which were resistant by Etest (MIC 2 mg/L) [26]. But no known genetic causes for resistance were found.

One of the limitations of this study is that resistance to amphotericin B may be epigenetic. If this is the case, passage of the isolates in the laboratory with no drug pressure may cause the MIC to decrease. This could also account for some of the discrepancies between laboratories. We were also unable to determine whether any of the patients from whom these isolates were collected had been treated with amphotericin B.

Breakpoints are typically set statistically to encompass 97.5–99% of the modelled wild-type MIC distribution, resulting in a breakpoint that is 1–2 two-fold dilutions above the modal MIC. In all 14 datasets in this study, the modal MIC was too close to a susceptibility BP of ≤1 mg/L. This will inevitably lead to variable MIC classification, as found here, with resistance rates varying from 0% to 63% for the same strain collection despite the modal MIC being similar. Consequently, from a technical point of view, either C. auris should be regarded as susceptible (and the BP set at 2 mg/L), or the entire population should be deemed resistant to amphotericin B to avoid random classification. A BP of 2 mg/L (one dilution higher than the modal MIC for EUCAST and CLSI would for this dataset result in the following numbers of “resistant isolates”: Etest consecutive: 2/40 (5%), Etest batched: 0/40 (0%), and MTS batched: 2/40 (5%) provided single swab inoculation is adopted, and 0% for all five EUCAST and CLSI runs. Our proposal in the absence of solid data linking the majority of elevated MICs to molecular resistance mechanisms would be to regard isolates with MIC ≤2 mg/L as wild-type isolates for which amphotericin B is a valid option but not the first treatment choice unless the organism is echinocandin resistant.