Ethical approval was obtained from the Stellenbosch University Health Research Ethics Committee (N14/06/069, U22/05/180). A consent waiver was received since the isolates used were anonymised, the patients would neither be harmed nor benefit from the study directly, and no patient information was included in this study. The research data were stored electronically on a password-protected device accessible by the principal investigator only.

Forty-six each of E. coli and K. pneumoniae previously isolated from patient blood cultures (BacT/ALERT^® Culture Media, bioMérieux, Marcy-l’Étoile, France) at a tertiary hospital in South Africa between 01 June 2017 and 31 January 2018 as part of routine patient management were collected. Species identification was done as part of routine diagnostic procedures using the Vitek^®2 automated system (bioMérieux, Marcy-l’Étoile, France).

Agar dilution was performed by spotting 1 μL of a 1:4 dilution of a 0.5 MacFarland suspension of each bacterial isolate on Mueller-Hinton agar plates (Sigma Aldrich, St. Louis, Missouri, United States) containing glucose-6-phosphate and fosfomycin (Fosfomycin sodium VETRANAL^®, Sigma Aldrich, St. Louis, Missouri, United States) at a range of concentrations (1 μg/mL – 256 μg/mL), in two-fold increments.¹⁷ Agar dilution was performed in triplicate, and the MIC was determined as the lowest antibiotic concentration at which bacterial growth was not observed in at least two replicates following incubation at 37 °C for 16 h in ambient air.¹⁹

Gradient diffusion was performed using fosfomycin MIC strips (0.064 μg/mL – 1024 μg/mL) (Liofilchem, Roseto degli Abruzzi, Italy) on Mueller-Hinton agar plates (Green Point Media, Cape Town, South Africa) inoculated with a 0.5 McFarland bacterial suspension and incubated at 37 °C for 16 h in ambient air. The MIC was interpreted at the concentration where the zone of the ellipse intercepted with the strip.²⁰

Kirby Bauer disk diffusion was performed using fosfomycin/glucose-6-phosphate discs (200 μg) (Mast Group, Bootle, United Kingdom) on Mueller-Hinton agar plates inoculated with a 0.5 McFarland bacterial suspension. The zone of inhibition (ZOI) was measured following incubation at 37 °C for 16 h in ambient air.

Quality control was performed using the fosfomycin susceptible E. coli American Type Culture Collection 25922 strain (ATCC, Manassas, Virginia, United States), with an MIC range of 0.5 μg/mL – 2 μg/mL and a ZOI range of 22 mm – 30 mm, as specified by CLSI.¹⁷

The antibiotic susceptibility testing results were analysed using the interpretive categories and breakpoints established by the CLSI (2020) and EUCAST (2020).^17,18 For E. coli, CLSI interprets ZOI ≥ 16 mm/MIC ≤ 64 μg/mL as susceptible, ZOI = 13 mm/MIC – 15 mm/MIC = 125 μg/mL as intermediate, and ZOI ≤ 1 mm/MIC ≥ 256 μg/mL as resistant; due to the absence of breakpoints for K. pneumoniae, E. coli breakpoints were used for interpretation. For E. coli, EUCAST interprets ZOI ≥ 24 mm/MIC ≤ 32 μg/mL, as susceptible and ZOI < 24 mm/MIC > 32 μg/mL as resistant, but for K. pneumoniae, there are only agar dilution breakpoints; therefore E. coli disk diffusion breakpoints were used for interpretation. For disk diffusion, EUCAST recommends disregarding isolated colonies that are within the ZOI, while CLSI advises taking them into account.^17,18 For gradient diffusion, the manufacturer’s guidelines suggest different approaches for E. coli and K. pneumoniae: for E. coli, single colonies in the zone of the ellipse should be disregarded, while for K. pneumoniae, colonies within 3 mm of the strip should be considered for both CLSI and EUCAST interpretations.²⁰

The performance of disk diffusion and gradient diffusion relative to the gold standard method, agar dilution, was described using categorical agreement, essential agreement (for gradient diffusion), and minor, major, and very major error rates. Categorical or essential agreement of greater than 90% and minor, major, and very major error rates of less than 3% were considered acceptable.²⁴ The combinations of mutations and genes detected by whole genome sequencing analysis were compared to gold standard agar dilution MICs. Data were stored, and statistical calculations were done, in Microsoft^® Excel^® for Microsoft 365, 2022 (Microsoft^®, Redmond, Washington, United States).

According to the gold standard method, agar dilution, 98% (45/46) of the E. coli isolates were susceptible to fosfomycin, according to both CLSI and EUCAST guidelines. Of the K. pneumoniae isolates, 70% (32/46) were susceptible, 22% (10/46) resistant, and 9% (4/46) intermediate, according to CLSI breakpoints, while 63% (29/46) of the K. pneumoniae isolates were susceptible and 37% (17/46) were resistant according to EUCAST breakpoints.

For E. coli, gradient diffusion showed 100% categorical agreement with agar dilution using EUCAST breakpoints, with one major error (2%) when interpreting the MICs according to CLSI breakpoints (Table 1). The essential agreement between gradient diffusion and agar dilution for E. coli was only 11%. Gradient diffusion did not perform well against agar dilution for K. pneumoniae, with a categorical agreement of around 75% depending on the guidelines used and an essential agreement of only 65%. Disk diffusion showed a categorical agreement of 100% against agar dilution using both CLSI and EUCAST breakpoints for E. coli. Disk diffusion did not perform well for K. pneumoniae, with a categorical agreement of 46% using EUCAST and 82% using CLSI breakpoints.

Various fosfomycin resistance mechanisms were detected in E. coli, with all the isolates carrying the mdtG gene and 93% (43/46) having the E448K mutation in the glpT gene (Table 2). Mutations in uhpT and the regulator genes, cyaA and ptsI, were also detected. A fosA gene was only detected in one E. coli isolate; this isolate had the highest fosfomycin MIC at greater than 256 μg/mL and was the only E. coli isolate classified as fosfomycin resistant. In the absence of fosA, MICs ranged from 1 μg/mL to 16 μg/mL regardless of the presence of different combinations of glpT, uhpT, and cyaA mutations, and the presence or absence of mdtG genes (Table 3).

All of the K. pneumoniae isolates had a fosA gene, either fosA5 (6/46, 13%) or fosA6 (40/46, 87%), and the E350Q mutation in uhpT (Table 2). Both combinations were present in susceptible and resistant isolates with MICs that ranged from 1 μg/mL to 256 μg/mL (Table 3).

Although both gradient diffusion and disk diffusion, interpreted using CLSI and EUCAST guidelines, produced an acceptable categorical agreement for E. coli, a more extensive sample set of isolates containing more fosfomycin-resistant isolates, and with MICs around the breakpoints should be analysed to confirm these findings.

The aim of the study was to correlate the fosfomycin resistance mechanisms with the phenotypic susceptibility results. However, due to the very low prevalence of fosfomycin resistance amongst E. coli isolates, with only one resistant isolate, we were unable to test the significance of any association. Similarly, it was clear that there was no association between fosfomycin MIC and resistance mechanisms in K. pneumoniae.

Not all fosfomycin resistance genes or mutations may be included in the Comprehensive Antibiotic Resistance Database²², and therefore, mutations or genes that confer resistance could have been missed. Thus, whole genome sequence data should be manually searched to identify potential novel fosfomycin resistance mutations, which could be further validated to improve our understanding of fosfomycin susceptibility.

The expression of the resistance mechanisms, specifically the fosA genes, was not evaluated in this study. Differential expression may account for the variable MICs and susceptibility results despite the presence of these genes in all K. pneumoniae isolates.

Fosfomycin molecular resistance mechanisms were not associated with phenotypic susceptibility to fosfomycin in E. coli or K. pneumoniae, except fosA3 in one resistant E. coli isolate. Disk diffusion and gradient diffusion perform well for fosfomycin susceptibility testing for E. coli but not for K. pneumoniae. These findings support the guidelines given by CLSI and EUCAST in that disk diffusion and gradient diffusion susceptibility testing should not be performed for K. pneumoniae and that the E. coli breakpoints should not be used for interpreting K. pneumoniae susceptibility.