Heterogeneity in the response to the metabolic environment drives fosfomycin heteroresistance

Abstract

Heteroresistance is a form of antibiotic resistance in which a phenotypically unstable subpopulation of resistant cells coexists with a majority susceptible population. Many mechanisms of heteroresistance are unclear and the impact of the extracellular environment on this phenomenon is unknown. Studying fosfomycin heteroresistance in Enterobacter cloacae complex, we observed that glucose availability greatly increased the frequency of the resistant subpopulation. Glucose downregulated the glycerol and fosfomycin importer GlpT via catabolite repression. Heteroresistance was generated by heterogeneity in single-cell expression of GlpT, in addition to expression of fosA, which acted as a resistance enhancer. Due to the impact of environmental glucose on glpT expression, we observed that the resistant subpopulation was more frequent during infection in murine models of hyperglycemia/diabetes. Thus, we reveal a paradigm where heterogeneity in response to the environment tunes resistant subpopulation frequency, which suggests that the metabolic context of infection in specific disease states may be important to consider in treatment decisions.

Choby et al. show an antibiotic-resistant subpopulation is generated by heterogeneity in expression of an antibiotic importer repressed by environmental glucose, suggesting diabetes-related infection environments may influence antibiotic resistance.

Introduction

The rise and spread of antibiotic resistance among bacterial pathogens is a growing challenge to preventing and treating infections¹. Heteroresistance (HR) is a unique form of antibiotic resistance which defies conventional classification, as isolates harbor a minor subpopulation of resistant cells in a large majority susceptible population². When this resistant subpopulation is extremely rare (<1 in 10,000 cells), it is usually undetected by traditional antimicrobial susceptibility testing. The resistant cells grow in the presence of an antibiotic and are transiently enriched, highlighting that they are not persisters, which are a different type of subpopulation that cannot replicate in the presence of drug³. When the antibiotic is removed, the frequency of the phenotypically resistant subpopulation returns to baseline levels. Heteroresistance has recently been shown to be a cause of unexpected antibiotic treatment failure in human patients^4,5.

Fosfomycin was discovered over 50 years ago and is now increasingly used as oral therapy for urinary tract infections⁶. There is also renewed interest in its utility as an intravenous treatment for multi-drug-resistant infections, given the dearth of drugs available to treat highly resistant infections and the fact that it is from a unique antibiotic class. Fosfomycin inhibits peptidoglycan synthesis by covalently binding and inactivating the UDP N-acetylglucosamine enolpyruval transferase, MurA⁷. Fosfomycin HR is widespread in carbapenem-resistant Enterobacterales (CRE); we previously screened CRE clinical isolates through the Georgia Emerging Infections Program’s Multi-site Gram-negative Surveillance Initiative and observed that fosfomycin HR was detected at a greater rate (72% of isolates) than HR to any other antibiotic tested⁸.

Conventional resistance to fosfomycin is known to occur primarily through mutations to the fosfomycin target MurA, fosfomycin importers, or acquisition of a fosfomycin resistance determinant⁹. Fosfomycin enters the cell through the glycerol-3-phosphate transporter (GlpT) or the hexose phosphate transporter (UhpT); therefore, mutations that inactivate their function can confer resistance by blocking fosfomycin entry^10–12. Acquisition of the glutathione-S-transferase FosA also confers resistance; FosA conjugates glutathione to fosfomycin to inactivate the drug¹³. Various chromosomal and plasmid-borne fosA variants exist and are common among Enterobacter and Klebsiella clinical isolates but have not extensively spread into Escherichia coli¹⁴. While much is understood about routes to fosfomycin resistance, what generates population-wide heterogeneity in fosfomycin resistance (heteroresistance) is unclear.

The Enterobacter cloacae complex comprises a group of species with intrinsic and frequently acquired antibiotic resistance and the capacity to cause various infections. Generally, these species exist in a variety of environmental habitats and can colonize the mammalian gut. However, they can be opportunistic, nosocomial pathogens of patients who are immunocompromised or otherwise more susceptible to infection. Relatively little is known mechanistically about the virulence factors and pathogenesis of these organisms¹⁵. Globally, Enterobacter spp. were the 7th most frequent cause of death by bacterial pathogens, generally from lower respiratory infections, bloodstream infections, and peritoneal and intra-abdominal infections¹⁶. Treatment options for multidrug resistant Enterobacter infections are limited, contributing to the Centers for Disease Control’s designation of extended-spectrum β-lactamase (ESBL)-producing Enterobacterales as a serious threat and carbapenem-resistant Enterobacterales an urgent threat¹⁷. Thus, a greater understanding of antibiotic resistance and heteroresistance is critical to improving the capacity to treat infections caused by antibiotic-resistant Enterobacter.

In this work, we investigate fosfomycin HR in a model CRE clinical isolate, Enterobacter strain Mu208. We find that the minority resistant subpopulation is supported by expression of FosA and synthesis of the FosA cofactor, glutathione, confirming their role in fosfomycin resistance. Surprisingly, we show that the population heterogeneity is not linked to FosA, but rather heterogeneity in the activity of the glpT promoter, controlling expression of the fosfomycin importer. Cells with low glpT expression constitute the resistant subpopulation. A transcriptional repressor, GlpR, and transcriptional activator, CRP, both play important roles in regulating glpT expression and responding to environmental carbon sources through catabolite repression. Notably, glucose import represses glpT expression via CRP and thereby decreases population-wide susceptibility to fosfomycin. This is recapitulated in murine models of hyperglycemia/diabetes; the frequency of the fosfomycin resistant subpopulation is increased in the presence of excess glucose during infection. Together, our data reveal a paradigm where heterogeneity in expression of an antibiotic importer, combined with expression of a resistance enhancer (FosA), drives heteroresistance. Because expression of the importer is sensitive to the metabolic environment, the metabolic context of infection in specific disease states may impact antibiotic treatment outcome.

Results

Mu208 is fosfomycin HR and causes treatment failure in mice

We assessed the features of fosfomycin heteroresistance (HR) in a clinical isolate of the Enterobacter cloacae complex, E. kobei strain Mu208. We used population analysis profile (PAP; Supplementary Fig. 1) to measure HR, which is performed by growing a bacterial strain in broth culture without antibiotic, then plating a range of dilutions of the culture on agar containing a range of concentrations of fosfomycin. Cells that are resistant to the antibiotic generate colonies on fosfomycin plates and are enumerated as colony forming units (CFU) at each concentration so that a proportion of the total population which survives at a given concentration can be calculated. Thus, PAP assesses resistant cells in the population prior to antibiotic exposure. The breakpoint concentration differentiates resistant versus susceptible isolates in clinical testing for each antibiotic; the Clinical & Laboratory Standards Institute (CLSI) fosfomycin breakpoint is 256 µg/mL¹⁸. Mu208 exhibited fosfomycin HR with a subpopulation of 1–10% of cells resistant to 256 µg/mL fosfomycin. This contrasts with the susceptible Enterobacter isolate Mu819 in which all the cells were uniformly killed at a concentration of fosfomycin below the clinical breakpoint (Fig. 1a and Supplementary Fig. 2a). In a time-kill assay, there was a period of initial killing by fosfomycin but subsequent rapid growth of Mu208, with the replicating population comprised of fosfomycin resistant cells (Fig. 1b and Supplementary Fig. 2b). The frequency of the resistant subpopulation which grew in the time-kill assay was unstable: after enrichment of the resistant cells in broth containing fosfomycin, the resistant population returned to the baseline frequency following subsequent growth in fosfomycin-free media (Fig. 1c). Together, these data establish that strain Mu208 exhibited unstable fosfomycin heteroresistance.

Fig. 1. Enterobacter cloacae complex isolate Mu208 is fosfomycin heteroresistant and causes treatment failure in mice.

a Population analysis profile (PAP) of Mu208 and E. cloacae complex isolate Mu819 plated on Mueller–Hinton agar (MHA) containing fosfomycin; the proportion of surviving colonies is quantified relative to MHA containing no fosfomycin, from three independent experiments with two biological replicates each. Data are presented as mean values ± SD. b Time kill of Mu208 and Mu819: survival in media containing 512 μg/mL fosfomycin over time and quantification of surviving colony forming units on MHA (Mu208 total and Mu819) and MHA containing 512 μg/mL fosfomycin (Mu208 resistant) from two independent experiments with five biological replicates each for Mu208 and three replicates each for Mu819. Data are presented as mean values ± SD. c Quantification of the fosfomycin resistant subpopulation of Mu208 after growth in media alone (baseline), subcultured into 256 μg/mL fosfomycin and grown for 20 h, and subsequently subcultured into fresh media without fosfomycin. At the end of each growth, an aliquot was diluted and plated onto MHA containing 256 μg/mL fosfomycin to quantify the resistant subpopulation, from three independent experiments with three biological replicates each. ** indicates p = 0.0044 and *** indicates p = 0.0004 by RM one-way ANOVA with Sidak’s multiple comparisons test, F (1.847, 14.78) = 20.57. (d–e) Survival of female mice infected with d Mu819 or e Mu208, with and without fosfomycin treatment. Mice (n = 10 per group, except n = 9 for Mu819+fos) were infected intraperitoneally (IP), and fosfomycin was administered at 200 mg/kg, IP. ** indicates p = 0.0028 by Log-rank (Mantel–Cox) test. f Schematic of glutathione biosynthesis and inactivation of fosfomycin, genes indicated in red were transposon hits. g PAP of Mu208, the ΔfosA deletion mutant, and its complement, from two independent experiments with two biological replicates each. Data are presented as mean values ± SD. h PAP of Mu208, a gshA deletion mutant, and its complement, from two independent experiments with three biological replicates each. Data are presented as mean values ± SD. (i–j) Survival of female mice (n = 10 per group) infected with i Mu208 or j Mu208 ΔgshA, with and without fosfomycin treatment. Mice were infected intraperitoneally (IP), and fosfomycin was administered at 200 mg/kg, IP. ** indicates p = 0.0058 by Log-rank (Mantel–Cox) test. Source data are provided as a Source Data file.

Next, we tested whether Mu208 was able to cause treatment failure in a murine model of peritonitis. As expected, fosfomycin treatment rescued mice from a lethal dose of the susceptible Mu819 isolate (Fig. 1d). However, fosfomycin treatment did not rescue mice infected with Mu208, demonstrating that a fosfomycin HR isolate can cause in vivo treatment failure (Fig. 1e).

FosA and glutathione are required for the resistant subpopulation

We sought to understand the mechanistic basis for the phenotypic heterogeneity that results in fosfomycin heteroresistance. An increase in copy number of antibiotic resistance genes (gene amplification) has been found to generate an antibiotic-resistant subpopulation in a variety of species to various antibiotics^2,19–21. We therefore assessed whether the fosfomycin resistant subpopulation in Mu208 was a result of gene amplification. However, genome sequencing of resistant subpopulations did not identify areas of the genome with elevated gene copy number >2 except at the chromosome and plasmid origins (Supplementary Data 1). Further, a ΔrecA strain, which is unable to perform RecA-dependent homologous recombination important for some gene amplification events, remained fosfomycin HR (Supplementary Fig. 2c). Taken together, these data indicate that fosfomycin HR in Mu208 is independent of gene amplification events.

Therefore, we turned to a transposon screen to identify genetic factors responsible for the fosfomycin resistant subpopulation in Mu208. We screened ~80,000 colonies following transposon mutagenesis for loss of the ability to form a colony on agar containing fosfomycin (Supplementary Fig. 3 and Supplementary Data 2). Multiple transposon mutants with a disruption in a chromosomal fosA2/fosA^EC family variant¹⁴ (hereafter referred to as fosA), exhibited reduced fosfomycin resistance, validating the screen. fosA encodes a glutathione-S-transferase which conjugates glutathione to the epoxide ring of fosfomycin, inactivating the drug. Similarly, some transposon mutants with reduced fosfomycin resistance were associated with synthesis of the FosA cofactor glutathione (Fig. 1f). We generated a ΔfosA strain which was susceptible to fosfomycin, demonstrating that FosA was required for the highly resistant subpopulation (Fig. 1g and Supplementary Fig. 4a). Nevertheless, heterogeneity remained in the ΔfosA strain, with subpopulations that had a greater relative resistance than the majority population, but did not survive to the breakpoint concentration. This appears to be a common feature of Enterobacterales lacking fosA²², and also suggests that the generation of heterogeneity is FosA-independent. Interestingly, complementation in trans with a multi-copy plasmid resulted in homogeneous resistance (100% of cells are resistant) to even high concentrations of fosfomycin (Fig. 1g).

Fig. 3. Transcriptional regulation of GlpT determines resistant subpopulation frequency.

a Schematic of regulation of glpT and its promoter by (1) the GlpR repressor, binding of which is inhibited by the presence of glycerol-3-phosphate, and (2) CRP bound to cAMP, synthesized by CyaA. b PAP of Mu208 and mutants of cyclic AMP receptor protein (Δcrp) and glycerol phosphate repressor (ΔglpR), from two independent experiments with three biological replicates each. Data are presented as mean values ± SD. c glpT expression (P_glpTmCherry) in the ΔglpR mutant and corresponding survival upon exposure to 128 µg/mL fosfomycin, from three independent experiments with one biological replicate each; data were determined from 700 to 1000 cells for each experiment. Data are presented as mean values ± SD. d Proportion of fosfomycin resistant colonies in the absence or presence of exogenous cyclic AMP in M9 broth with 0.8% glycerol, from two independent experiments with four biological replicates each. *** indicates p = 0.0006 by two-tailed paired t-test, t = 5.867, df = 7. e Transcript abundance of glpT measured by qPCR in the absence or presence of exogenous cyclic AMP in M9 broth with 0.8% glycerol from D from two independent experiments with four biological replicates each; ** indicates p = 0.0019 by two-tailed paired t-test, t = 4.816, df = 7. f glpT expression (P_glpTmCherry) in the Δcrp mutant and corresponding survival upon exposure to 128 µg/mL fosfomycin, from three independent experiments; data were determined from 700 to 1000 cells for each experiment. Data are presented as mean values ± SD. g crp expression (P_crpmCherry) in Mu208 mutant and corresponding survival upon exposure to 128 µg/mL fosfomycin, from three independent experiments with one biological replicate each; data were determined from 700 to 1000 cells for each experiment. Data are presented as mean values ± SD. Source data are provided as a Source Data file.

Similarly, inactivation of glutathione synthesis in a generated ΔgshA strain also reduced the frequency of the resistant subpopulation (Fig. 1h and Supplementary Fig. 4b). Exogenous glutathione increased the resistant subpopulation frequency, even in the absence of endogenous glutathione synthesis (ΔgshA), but the frequency remained much lower than in the wildtype strain (Supplementary Fig. 4c–e). Exogenous addition of the glutathione precursor L-glutamate also increased the resistant subpopulation frequency but was dependent on endogenous glutathione synthesis (Supplementary Fig. 4c). Therefore, the fosfomycin resistant subpopulation requires FosA and its cofactor glutathione.

We were intrigued by the contribution of endogenous glutathione synthesis to fosfomycin HR and the availability of glutathione during infection. Glutathione is an important mammalian cofactor that can achieve millimolar intracellular and micromolar extracellular concentrations²³. To test whether bacterial-derived glutathione is required even in the presence of mammalian-derived glutathione, we infected mice with a lethal dose of the ΔgshA glutathione synthase mutant. While fosfomycin treatment failed to rescue mice from infection with wildtype Mu208, the ΔgshA strain did not cause treatment failure (Fig. 1i, j). Thus, synthesis of the FosA cofactor glutathione is critical to the fosfomycin resistant subpopulation and the ability of strain Mu208 to cause fosfomycin treatment failure in vivo.

glpT inactivation confers fosfomycin resistance

Having established that Mu208 encodes the fosfomycin resistance determinant FosA and can synthesize the FosA cofactor glutathione, it was unclear why the majority of the population was susceptible to the breakpoint concentration of fosfomycin. Fosfomycin can enter the cell through the glycerol-3-phosphate importer GlpT, and we observed a fosfomycin susceptible mutant in our transposon screen in glycerol phosphate metabolism, glpD (Supplementary Data 2). glpD encodes an aerobic glycerol-3-phosphate dehydrogenase, and a glpD deletion mutant exhibited a reduced frequency of the resistant subpopulation relative to wildtype Mu208 (Fig. 2a and Supplementary Fig. 5a). GlpD is at the intersection of lipid and carbon metabolism; it converts glycerol-3-phosphate to glycerone phosphate²⁴. Thus, inactivation of glpD may increase glycerol-3-phosphate levels and affect glycerol-3-phosphate metabolism, so we evaluated the role of glpT on fosfomycin HR.

Fig. 2. Expression of the fosfomycin and glycerol phosphate importer GlpT correlates with survival in fosfomycin.

a PAP of Mu208 and a glpD deletion mutant, from two independent experiments with three biological replicates each. Data are presented as mean values ± SD. b PAP of Mu208, a glpT deletion mutant, and the glpT mutant complemented in the chromosome under control of its native promoter or a constitutive promoter, from two independent experiments with three biological replicates each. Data are presented as mean values ± SD. c Representative images of Mu208 att:: P_glpTmCherry on agar pads in the absence of fosfomycin, with a low mCherry cell and high mCherry cell in insets, followed by images of the time-lapse following fosfomycin addition. d, e P_glpTmCherry expression intensity prior to fosfomycin exposure among cells. The fate of the cells was tracked after fosfomycin addition and classified as those that replicated (“Resistant”) or did not replicate (“Susceptible”) in three independent experiments with one biological replicate each; data were collected from 700–1000 cells for each experiment. Data are presented as mean values ± SD. d Shows P_glpTmCherry and fate in wildtype Mu208 following 128 µg/mL fosfomycin, e shows P_glpTmCherry and fate in ΔfosA following 8 µg/mL fosfomycin. Source data are provided as a Source Data file.

Fosfomycin can be imported by GlpT and/or the hexose phosphate importer UhpT. Mu208 encodes both GlpT and UhpT, but uhpT has a frame-shift mutation that may ablate expression or generate a protein lacking the first 21 amino acids (mutations in uhpT are prevalent among Enterobacter isolates^25,26). We tested the role of each transporter in fosfomycin susceptibility. The ΔglpT strain was impervious to even 2048 µg/mL fosfomycin (Fig. 2b and Supplementary Fig. 5b), suggesting it is the primary fosfomycin importer in this strain under the conditions tested. On the other hand, the ΔuhpT strain demonstrated no change in fosfomycin HR compared to wild-type, consistent with GlpT acting as the primary fosfomycin importer in Mu208 (Supplementary Fig. 6a). Compared to the ΔglpT strain, which was not killed, wildtype Mu208 exhibited ~5 logs of killing at high concentrations of fosfomycin, suggesting glpT is expressed under these conditions. We complemented the glpT deletion in cis at the attB site under the control of a constitutive promoter (P_tet) or the native promoter (P_glpT). When glpT was constitutively expressed, the majority of the population was killed at much lower fosfomycin concentrations compared to the strain complemented with the native promoter, suggesting transcriptional control of glpT has a substantial impact on fosfomycin resistance (Fig. 2b and Supplementary Fig. 5b).

Our data support a hypothesis in which GlpT is the primary fosfomycin importer in Mu208. Genetic inactivation of glpT conferred homogenous resistance, presumably because fosfomycin entry into the cell is severely diminished. Mutations which ablate the expression or function of GlpT could represent a route to fosfomycin resistance in our assays. However, we hypothesized that most of the fosfomycin resistant population were not glpT mutants, because the frequency of the resistant subpopulation was much greater than the expected mutation rate (Fig. 1a), the resistant subpopulation rapidly returned to baseline frequency when fosfomycin was removed suggesting this is a flexible phenotypic resistance trait rather than being caused by a more stable form of resistance such as mutation (Fig. 1c), and because the resistant subpopulation, if generated entirely by random inactivating mutations to glpT, would exist at the same frequency across fosfomycin concentrations.

Nevertheless, we evaluated the resistant subpopulation in Mu208 for changes in glpT function by sequencing or phenotypic testing. We sequenced colonies collected from agar containing fosfomycin and performed variant analysis. We only found a mutation in glpT or its promoter in 1/7 biological replicates (Supplementary Fig. 6b). Phenotypically, mutations in glpT render bacteria unable to grow on glycerol phosphate as the sole carbon source, as it cannot be transported intracellularly. Therefore, we screened fosfomycin resistant colonies by patching on M9 agar with glycerol phosphate as the sole carbon source. From resistant colonies collected on 256 µg/mL fosfomycin, glpT null colonies appeared at a rate of 1 per 100,000 total CFU, much lower than the frequency of the total resistant subpopulation at 1 per 300 total CFU (Supplementary Fig. 6c). While phenotypically null glpT mutants exist in cultures of Mu208, they comprised only a small fraction of the resistant subpopulation across most concentrations. However, when a strain only survived at a frequency of 1 per 10⁵ total cells, because of genetic mutation or substantial killing at high fosfomycin concentration, the majority of those surviving colonies were phenotypically glpT null (Supplementary Fig. 6d). Thus, there was a rare subpopulation of resistant cells with null glpT function which exists at approximately 1 in 10⁵ that is highly fosfomycin resistant. However, the vast majority of the resistant subpopulation has intact GlpT.

Expression of GlpT is heterogeneous and predicts resistance

We hypothesized that population-level variation in glpT expression could potentially explain the phenotypic heterogeneity in fosfomycin resistance. We generated a strain encoding P_glpTmCherry in the chromosome to monitor activation of the glpT promoter. Mu208 att::P_glpTmCherry was imaged on agar pads, in the absence of fosfomycin, and fluorescence was measured at the single cell level. Prior to fosfomycin exposure, the population of cells exhibited heterogeneity in P_glpT activity. Subsequently, the fate of each cell was tracked individually following fosfomycin addition, and we observed that cells with lower P_glpT were those that survived (Fig. 2c, d, Supplementary Video 1, and Supplementary Fig. 7a). In the absence of fosfomycin, all cells replicated uniformly regardless of initial P_glpTmcherry activity (Supplementary Fig. 7b). To confirm this finding was specific to P_glpT and rule out the possibility that the resistant cells were globally less transcriptionally active, we similarly measured P_gshAmCherry and found that gshA promoter activity was similar between resistant and susceptible cells (Supplementary Fig. 8a).

Importantly, the expression pattern and heterogeneity of P_glpTmCherry remained in the ΔfosA strain (Fig. 2e). The ΔfosA mutant is susceptible (Fig. 1g), with a subpopulation only evident at very low fosfomycin concentrations, because FosA-mediated resistance is absent. At 8 μg/mL fosfomycin, we observed approximately 2% survival of the ΔfosA strain, and those surviving cells had generally lower P_glpTmcherry expression than the rest of the population (Fig. 2e), suggesting that population heterogeneity is fosA-independent but rather driven by levels of glpT expression.

GlpR and CRP control glpT expression and fosfomycin resistance

In E. coli, glpT expression is regulated by the transcriptional repressor GlpR²⁷ and the carbon catabolite repression activator CRP²⁸ (cyclic-AMP receptor protein). We observed high amino acid homology between CRP of E. coli MG1655 and Mu208 (100% identity) as well as between GlpR from each strain (90% identity). We also identified putative binding sites in the glpT promoter for CRP and in the glpT coding sequence for GlpR based on high similarity to E. coli²⁷, suggesting a similar regulatory cascade in Mu208 (Fig. 3a). We generated deletion mutants of each regulator and assessed fosfomycin resistance by PAP. The ΔglpR strain was more susceptible to fosfomycin compared to wildtype Mu208, consistent with GlpR repressing P_glpT and thus glpT expression being higher in ΔglpR, leading to greater fosfomycin uptake. In contrast, the Δcrp strain was homogeneously resistant, consistent with CRP activating P_glpT and therefore glpT expression being lower in Δcrp, causing less fosfomycin uptake (Fig. 3b and Supplementary Fig. 9a). These phenotypes were complemented in trans (Supplementary Fig. 9b, c). We measured P_glpTmCherry in the ΔglpR background and found that all cells in the population had high P_glpT activity and were killed by 128 μg/mL fosfomycin (Fig. 3c). At lower fosfomycin concentrations, the ΔglpR strain exhibited resistant cells (Fig. 3b). Therefore, we repeated the imaging at 32 μg/mL fosfomycin, where some cells survived, and again determined that surviving cells had lower P_glpTmcherry activity as compared to susceptible cells (Supplementary Fig. 9d). Thus, we conclude that GlpR repression of glpT expression in a population is critical for the generation of the fosfomycin resistant subpopulation.

CRP activation requires cyclic adenosine monophosphate (cAMP), which can be synthesized by adenylate cyclase (cyaA). CRP binds cAMP, and the CRP-cAMP complex binds the glpT promoter and induces expression, among its other regulated genes (Fig. 3a). To further investigate the role of CRP in fosfomycin HR, we added exogenous cAMP to Mu208 grown in minimal media and observed that it increased glpT expression (Fig. 3e) and reduced the frequency of the resistant subpopulation (Fig. 3d and Supplementary Fig. 9e, f). Further, we determined that inactivation of cyaA phenocopied the crp deletion, leading to homogenous fosfomycin resistance (Supplementary Fig. 9g). We next tested whether the constitutive expression of glpT could overcome the resistance conferred by deletion of crp. Constitutive expression of glpT rendered even the crp deletion strain susceptible, consistent with our model that the effects of CRP on fosfomycin resistance occur via activation of the glpT promoter (Supplementary Fig. 9h). Thus, we conclude that CRP binds cAMP synthesized by CyaA, activating P_glpT, enhancing fosfomycin entry, which thereby reduces fosfomycin resistance.

Based on these results, we hypothesized that the resistant subpopulation would exhibit lower CRP activity. First, we confirmed that CRP activated the glpT promoter; we observed that all the cells in the Δcrp strain had low P_glpT activity and were resistant (Fig. 3f). We then generated a crp reporter strain (P_crpmCherry) and observed heterogeneity amongst the population, with low P_crp cells surviving fosfomycin treatment (Fig. 3g). This further indicates that the heterogeneity in fosfomycin HR in Mu208 is due to heterogeneous glpT expression, which is linked with heterogeneous  CRP activity. Importantly, all reporters phenocopied their respective parental strain when assayed by fosfomycin PAP (Supplementary Fig. 8b–d). In sum, the carbon catabolite repression pathway exhibits heterogeneity, and there is a minor population of low CRP cells. These low CRP cells correlate with low glpT expression, as CRP is a transcriptional activator of P_glpT. Low P_glpT cells survive fosfomycin treatment because GlpT is the primary fosfomycin importer in strain Mu208, and less fosfomycin enters cells when GlpT expression is low.

Glucose represses glpT via CRP and enhances the resistant subpopulation

The important role played by GlpR and CRP in regulating glpT expression and fosfomycin resistance suggested that carbon source availability is a key regulator of fosfomycin HR. GlpR is regulated by environmental glycerol-3-phosphate, while CRP activity is modulated by adenylate cyclase, whose activity is inhibited when environmental glucose is imported by the phosphotransferase (PTS) system. In further support of this hypothesis, ptsG was identified in our transposon screen for loss of fosfomycin resistance (Supplementary Data 2). ptsG encodes the glucose-specific EIICB component of the sugar PTS system; PtsG imports and phosphorylates glucose to generate intracellular glucose-6-phosphate. Therefore, we assessed whether exogenous glucose would impact the frequency of the fosfomycin resistant subpopulation at the breakpoint concentration. Addition of glucose led to a 20-fold increase in the frequency of the fosfomycin resistant subpopulation in Mu208 (Fig. 4a). Further, at baseline in the absence of additional glucose, there was a ~2 log reduction in the resistant subpopulation frequency in ΔptsG compared to Mu208, and the ΔptsG strain did not respond to exogenous glucose, highlighting that PtsG is the critical glucose importer in this strain (Fig. 4a and Supplementary Fig. 10a, b). Import and phosphorylation of glucose by PtsG is known to reduce cAMP synthesis by CyaA, thereby reducing CRP activity²⁹ and subsequent expression of glpT. We measured glpT transcript abundance by qRT-PCR and found that excess glucose resulted in ~100-fold reduction in relative glpT transcript abundance (Fig. 4b), consistent with glucose increasing fosfomycin resistance via glpT repression.

Fig. 4. Glucose enhances the frequency of the resistant subpopulation by repressing glpT.

a Proportion of colonies resistant to fosfomycin in Mu208, the ΔptsG deletion mutant, and its complement, in Mueller–Hinton broth with and without 1% exogenous glucose, from two independent experiments with three biological replicates each. ns indicates p = 0.2102, ** indicates p = 0.0027, and **** indicates p < 0.0001 by mixed-effects analysis (two-sided) with Geisser-Greenhouse correction and Šídák’s multiple comparisons test, F (2.471, 11.86) = 581.5. b glpT abundance measured by qPCR following growth in MHB + 2% glucose relative to MHB alone, from two independent experiments with five biological replicates total. **** indicates p < 0.0001 by two-tailed paired t-test, t = 66.62, df = 4. c Proportion of colonies resistant to fosfomycin in Mu208 grown in M9 media with 0.4% glycerol and increasing concentration of glucose (doubling from 0.015625 to 0.25%), from two independent experiments with three biological replicates each. d Proportion of colonies resistant to fosfomycin in Mu208 grown in M9 media with 0.8% glycerol, M9 with 0.6125% glucose (+) or M9 with 2.5% glucose (++), from two independent experiments with three biological replicates each. ns indicates p = 0.2332 (+glucose), p = 0.0731 (++glucose), ** indicates p = 0.0014 (+glucose), p = 0.0011 (++glucose), *** indicates p = 0.0008, and **** indicates p < 0.0001 by RM one-way ANOVA with Šídák’s multiple comparisons test, F (2.402, 12.01) = 242.1. Population analysis profiles from strains Mu208, ΔglpT, ΔgshA, and ΔfosA, e following growth in M9 + 0.8%glycerol as the sole carbon source or f following growth in M9 + 0.4% glucose as the sole carbon source, from two independent experiments with three biological replicates each. Data are presented as mean values ± SD. glpT expression (P_glpTmCherry) in Mu208 and corresponding survival upon exposure to 128 µg/mL fosfomycin in g M9 + 0.8% glycerol as the sole carbon source or h M9 + 0.4% glucose as the sole carbon source. From three independent experiments with one biological replicate each, data were determined from 700 to 1000 cells for each experiment. Data are presented as mean values ± SD. Source data are provided as a Source Data file.

Because of the key role of carbon source availability on fosfomycin HR, we turned to M9 minimal media to control the carbon source present. When Mu208 was grown with glycerol as the sole carbon source, the frequency of the subpopulation resistant to 256 µg/mL fosfomycin was below 1%, but low concentrations of exogenous glucose resulted in a dose-dependent increase in the frequency of the resistant subpopulation (Fig. 4c and Supplementary Fig. 10c, d). The effect of glucose was not the result of enhanced total growth of the cultures, as bacterial growth yield was equivalent in each condition. We hypothesized that the 2-log increase in resistant subpopulation frequency following growth in glucose versus glycerol as the primary carbon source was mediated largely by a reduction in glpT promoter activation. To assess this, we measured the resistant population frequency of the strain encoding glpT under the P_tet constitutive promoter. We found that glucose did not enhance the frequency of the resistant population in this strain with non-natural control of glpT (Fig. 4d), compared to the control strain with glpT under its native promoter, consistent with this hypothesis.

To further study the impact of exogenous glucose, we performed PAP of Mu208, ΔglpT, ΔgshA, and ΔfosA strains following growth in M9 broth with glycerol or glucose. The entire population of wildtype Mu208 survived up to 64 µg/mL in M9 glycerol but up to 256 µg/mL in M9 glucose, with similar shifts for the ΔgshA and ΔfosA strains (Fig. 4e, f, and Supplementary Fig. 10e, f), consistent with glucose repressing glpT expression and thus mediating reduced fosfomycin entry. We performed time-lapse microscopy with P_glpTmCherry following growth in the same conditions and observed a shift in the peak fluorescence of the population from intensity of ~250 in glycerol to <100 in glucose; the resistant subpopulation frequency shifted from 8% in glycerol to ~95% in glucose (Fig. 4g, h). These data are consistent with glucose enhancing the frequency of the fosfomycin resistant population primarily through glpT repression and thereby limiting fosfomycin import.

Glucose enhances the resistant subpopulation in Enterobacterales

We next sought to test whether glucose was able to enhance the frequency of the resistant subpopulation among carbapenem-resistant Enterobacter clinical isolates. We grew four additional isolates collected by the Georgia Emerging Infections Program (GaEIP) in Mueller–Hinton broth with or without additional glucose (Supplementary Fig. 11a) or in M9 with either glycerol or glucose as the sole carbon source (Supplementary Fig. 11b). In both, glucose resulted in increased frequency of the population resistant to 256 μg/mL fosfomycin. Next, we performed PAP of 8 GaEIP isolates each of carbapenem-resistant Escherichia and Klebsiella following growth in M9 media with glycerol or glucose as the sole carbon source. Escherichia isolates were susceptible by PAP, consistent with infrequent carriage of the fosA resistance gene¹⁴. However, six of the eight E. coli isolates demonstrated a significant increase in the proportion of the population resistant to 64 μg/mL fosfomycin when grown in glucose as compared to glycerol (Supplementary Fig. 12a, d). While Klebsiella isolates were either HR or resistant to fosfomycin, five of eight showed a significant increase in the proportion of the population resistant to 256 μg/mL fosfomycin when grown in glucose (Supplementary Fig. 12b, e). In contrast, when we performed PAP of Staphylococcus aureus isolates (model strains Newman and JE2, and three previously described clinical isolates³⁰) following growth in Mueller–Hinton broth with or without additional 2% glucose added, we observed that glucose did not enhance the proportion of the population resistant to 32 μg/mL fosfomycin (Supplementary Fig. 12c, f). Together, our data support the hypothesis that glucose can generally increase relative fosfomycin resistance in Enterobacter, Escherichia, and Klebsiella.

The fosfomycin resistant subpopulation is enriched in diabetic mice

Changing the carbon source resulted in substantial changes to the frequency of the fosfomycin resistant subpopulation (Fig. 4). However, the mammalian infectious niche is not well-represented by growth conditions used for clinical susceptibility testing, or by M9 broth with a sole carbon source. Glucose levels in patients suffering from bacterial infection can vary based on co-morbidities, and diabetic patients are particularly vulnerable to bacterial infections³¹. Therefore, we sought to evaluate whether excess glucose in vivo altered the frequency of the fosfomycin resistant subpopulation during infection. We used two models of hyperglycemia/diabetes: streptozotocin (STZ)-treated mice to represent type I diabetes and the Lep^ob/Lep^ob obese mouse model of type II diabetes. We infected STZ- and control-treated mice in the peritoneum with Mu208. Following 24 h of infection, a peritoneal wash and the spleen and liver (sites of dissemination) were collected. Samples were plated on agar containing zero or the breakpoint concentration of fosfomycin to enumerate the frequency of the resistant subpopulation. The frequency of the population resistant to 256 µg/mL fosfomycin increased in the peritoneum and spleen of STZ-treated mice compared to controls, without differences in the overall bacterial burden (Fig. 5a and Supplementary Fig. 13a). When we similarly infected Lep^ob/Lep^ob mice and the heterozygous control mice, we found the fosfomycin resistant subpopulation increased in frequency in the peritoneum and liver of Lep^ob/Lep^ob mice compared to controls (Fig. 5b and Supplementary Fig. 13b). Because fosfomycin failed to treat wildtype mice infected with Mu208 (Fig. 1e, i), we were unable to compare the outcomes of infection and fosfomycin treatment of STZ-treated or Lep^ob/Lep^ob mice relative to wildtype mice. These data suggest that during infection, Mu208 responds to host glucose and is more resistant to fosfomycin. More broadly, these results show how the host metabolic environment encountered by pathogens could impact antibiotic resistance.

Fig. 5. Glucose availability increases the frequency of the resistant subpopulation in murine models of diabetes/hyperglycemia.

a, b Proportion of colonies resistant to fosfomycin in Mu208 plated directly from indicated tissues following infection of male mice. Data are presented as geometric mean values ± geometric SD. Mice were infected with ~2 × 10⁸ colony forming units for ~24 h. a Prior to infection, mice were treated with control or streptozotocin (STZ), from three independent experiments with 13 mice total for control and 12 mice total for STZ treated. * indicates p = 0.0308 (peritoneum), 0.0390 (spleen), and ns indicates p = 0.5076 by one-way ANOVA with Šídák’s multiple comparisons test, F (5, 69) = 6.762. b B6.Cg-Lep^ob/J heterozygous or homozygous (Lep^ob/Lep^ob) mice were infected, from two independent experiments, with 8 mice total for WT/Lep^ob and 10 mice total (one absent data point for peritoneum) for Lep^ob/Lep^ob. *** indicates p = 0.0002, ns indicates p = 0.6988, and * indicates p = 0.0229 by one-way ANOVA with Šídák’s multiple comparisons test, F (5, 47) = 6.756. Source data are provided as a Source Data file.

Discussion

We investigated the mechanism of fosfomycin heteroresistance in Enterobacter cloacae complex strain Mu208, which caused fosfomycin treatment failure in a murine model of peritonitis (Fig. 1e). 10% of the total cells of this isolate are resistant to the breakpoint concentration of 256 μg/mL (Fig. 1a), which is relatively high considering that isolates with resistant subpopulations as low as 0.001% of the population caused treatment failure in mice^8,32. The ΔgshA mutant (lacking a glutathione synthetic gene and thus resulting in lower glutathione availability for the FosA resistance enzyme to inactivate fosfomycin) did not cause treatment failure; this isolate does have a population that survives at the breakpoint and higher, but at a much lower frequency than the wildtype strain (Fig. 1h), demonstrating that subpopulation frequency is likely a contributing factor to causing treatment failure. Whether a high frequency of cells surviving at the breakpoint or the existence of a subpopulation that can resist high concentrations of antibiotic (Mu208 has subpopulations that survive beyond 2048 μg/mL fosfomycin) is more critical to causing treatment failure is unclear. Our conclusions regarding the ability of Mu208 to cause treatment failure are limited because of its high frequency resistant subpopulation, making it challenging to separate the effects of heteroresistance versus overall population resistance. However, we can conclude that a heteroresistant strain can cause fosfomycin treatment failure.

We observed that the ΔgshA mutant, defective in glutathione (GSH) synthesis, had a reduced frequency of the fosfomycin resistant subpopulation (Fig. 1h). Adding GSH to the media at very high concentrations, up to 20 mM, only modestly increased resistant subpopulation frequency in the ΔgshA mutant (Supplementary Fig. 4c–e). This is consistent with data indicating that cytoplasmic GSH levels are challenging to complement exogenously^33,34 (FosA is cytoplasmic and thus GSH would need to reach the cytosol for use by this resistance enzyme). This may in part be due to the hydrolysis of periplasmic GSH by gamma-glutamyltranspeptidase GGT before it is imported to the cytosol³⁵. In mammals, glutathione is generally measured at low millimolar concentrations intracellularly (1–8 mM^23,36) but significantly lower levels extracellularly (2–3 μM^37,38 in plasma). Some extracellular niches have higher GSH concentrations: >400 μM in lung epithelial lining fluid³⁸ and 1–6 mM in bile^39,40. Thus, we conclude that glutathione synthesis is critical for Mu208 to cause treatment failure, but whether this is because this strain was unable to completely import the available GSH and provide it to FosA, or whether host sources of GSH were insufficient in quantity, remains unresolved.

Despite encoding FosA, the majority of Mu208 cells were killed by the breakpoint concentration of fosfomycin. We observed that most of the cells were susceptible at this concentration because they expressed sufficient GlpT, a glycerol importer that also imports fosfomycin. Mu208 was amenable to single-cell microscopy because the resistant subpopulation frequency was relatively high at the breakpoint concentration, but it is worth noting that imaging assays, with 700–1000 cells quantified, have a much higher limit of detection than the population analysis profile, which detects killing or survival of ~10⁷ cells. Mutant strains lacking genes involved in fosfomycin resistance were assessed by single-cell microscopy at lower drug concentrations where the frequency of cells resilient to fosfomycin was high enough to overcome the high limit of detection (ΔfosA in Fig. 2e, ΔglpR in Fig. 3c and Supplementary Fig. 9d). Despite these limitations, examining glpT expression and cell fate with single-cell resolution provided critical insights into the means of survival of the resistant subpopulation.

Additionally, we did not expect to measure the fluorescence and survival of any glpT null mutants, as it is unlikely a cell which exists at 1 in 100,000 (Supplementary Fig. 6c) would be visualized in an assay measuring 700–1000 cells. However, data presented here establishes that in Mu208, random, low frequency glpT inactivating mutants play a minor role in the resistant subpopulation. Following enrichment of the resistant subpopulation in broth, the population returned to baseline frequency when fosfomycin was removed, which confirmed that there was not a strong overall selection of stable resistant mutants (Fig. 1c). Increasing expression of glpT by deleting glpR, which would not affect glpT null mutants, exerted a robust effect on the resistant subpopulation, consistent with glpT transcriptional regulation being a key driver of the resistant subpopulation (Fig. 3b). glpT null mutants only became dominant among resistant cells when sufficiently high concentrations of fosfomycin eliminated the vast majority of resistant cells with intact glpT. For wildtype Mu208, the resistant cell frequency dropped to 1 in 10⁵ at 2048 μg/mL fosfomycin, and those surviving cells were mostly glpT null (Supplementary Fig. 6d). This is consistent with the differing mechanisms of resistance; the vast majority of cells which survived fosfomycin had intact glpT but reduced expression. This population survived to 1024 μg/mL, and less so at 2048 μg/mL. glpT null cells likely had a higher single cell MIC due to complete glpT inactivation, and survived to at least 2048 μg/mL, but their very low frequency resulted in only a minor contribution to the composition of the total resistant subpopulation across the population.

The regulation of glpT is complex. Altering the carbon source environment affected glpT expression, resulting in shifts in fosfomycin entry and cellular resistance. When glucose is present, it is imported by PtsG and converted to glucose-6-phosphate, reducing the phosphorylation of the PtsG partner protein Crr. This state of Crr results in dampened cAMP synthesis by CyaA, such that CRP is not active and unable to activate glpT transcription. Further, when glucose is abundant, GlpR represses glpT, resulting in poor fosfomycin entry into the cell and fosfomycin resistance (Fig. 6a). In the absence of glucose, phosphorylated Crr would activate CyaA, activating CRP and thus glpT expression. If glycerol or glycerol-phosphate are present, GlpR would be de-repressed, leading to elevated levels of glpT and fosfomycin import, yielding low fosfomycin resistance (Fig. 6b). These changes can occur globally and likely explain the overall shift in resistance profiles between, for instance, strain Mu208 grown in M9 glycerol versus M9 glucose (Fig. 4e, f).

Fig. 6. Proposed model of the effects of glucose availability on fosfomycin heteroresistance in Enterobacter Mu208.

a, b Repression or activation of glpT transcription by the presence of glycerol or glucose carbon source is mediated through GlpR and CRP. c, d Population wide distribution of glpT expression across resistant subpopulations. The effect of glucose shifted the entire population’s glpT expression lower, but heterogeneity in glpT expression and relative fosfomycin resistance remained.

On the other hand, for any one growth condition, heterogeneity of resistance existed in wildtype Mu208 (Fig. 6c, d). We attribute this heterogeneity partially to a broad distribution in P_glpT activity (Fig. 2b), likely mediated by heterogeneity in CRP activity (Fig. 3f). We observed susceptible and resistant cells, and the resistant subpopulation demonstrated a range of MICs; some cells survived at the breakpoint but not greater, while other resistant cells survived beyond 2048 μg/mL. One might expect that the most resistant cells were those that had the absolute lowest level of glpT expression in the single-cell assays. However, the relative resistance within the subpopulation is likely the product of low glpT expression as well as FosA expression and activity, and gshA/glutathione synthesis. We observed that resistant cells demonstrated a broad range of gshA promoter activity, suggesting differences in glutathione abundance across the population (Supplementary Fig. 8a). Thus, P_glpT activity is predictive of susceptibility, in the sense that if the cell demonstrates mCherry intensity >225, it will be susceptible (Fig. 2d), but this cannot describe how resistant a single resistant cell is because of the important contributions of FosA and glutathione.

We conclude that the ability of environmental glucose to enhance the frequency of the resistant subpopulation (Fig. 4) is largely due to repression of glpT. Glucose has global effects on cellular metabolism and could also enhance resistance by increasing glutathione synthesis, altering cell wall metabolism, or through other mechanisms. However, addition of glucose had no effect on the frequency of the resistant subpopulation in the strain engineered to have constitutive, non-regulated P_glpT activity (Fig. 4d), demonstrating that any effects of glucose outside of affecting glpT transcription are minor. We observed the effects of glucose in vitro and in vivo, and note that the concentrations used in M9 broth correlate well with levels of mammalian blood glucose. Non-diabetic blood glucose is below 100 mg/dL (0.1% glucose w/v) when fasted or below 200 mg/dL (0.2% w/v) non-fasted, and diabetic blood glucose concentrations up to and exceeding 240 mg/dL (0.24% w/v) are observed in humans⁴¹ and 350 mg/dL (0.35%) in the mice used in this study^42,43. We observed more than a 1-log increase in the fosfomycin resistant subpopulation frequency with the addition of as little as 0.0625% glucose (Fig. 4c). Together, these data indicate that host-derived glucose could have meaningful effects on fosfomycin heteroresistance at concentrations relevant to non-diabetic and diabetic physiology.

Despite glucose’s effect on glpT expression, glpT heterogeneity only generated heteroresistance in conjunction with resistance imparted by FosA. fosA deletion strains were completely killed below the breakpoint concentration of fosfomycin (Figs. 1g and 4e), demonstrating FosA is a resistance enhancer. The heterogeneity in glpT expression (Fig. 2e) and killing by low concentrations of fosfomycin (Figs. 1g and 4e) show heterogeneity is intact in the absence of fosA, and that heterogeneity and resistance are distinct traits.

Fosfomycin HR presents challenges clinically. Depending on the frequency of the resistant subpopulation at the breakpoint concentration, HR isolates may be designated resistant or susceptible by conventional antimicrobial susceptibility testing. We showed that Mu208 can cause treatment failure in mice when treated with fosfomycin, and treatment of HR isolates can occur in humans when an isolate is incorrectly designated susceptible by susceptibility tests. Unfortunately, clinical fosfomycin susceptibility testing is fraught with confusion. Disk diffusion or Etest are often used. Resistant colonies are sometimes visualized within the zone of clearing on these fosfomycin susceptibility tests. However, there is disagreement about their relevance and how to interpret these colonies. In part because glpT mutants exhibit negative fitness costs^10,44, European (EUCAST) guidelines are to ignore these resistant colonies and classify such isolates as fosfomycin susceptible⁴⁵. It is critical that the underlying biology of fosfomycin resistance be clearly understood to accurately guide effective clinical therapy.

Most surprisingly, our data suggest that the host metabolic environment or co-morbidities may affect the outcome of antibiotic treatment. We observed that excess host glucose in models of hyperglycemia/diabetes increased the frequency of the resistant subpopulation in strain Mu208 during infection. This is notable because an estimated 15% of adults in the United States have diabetes⁴⁶, and of those, around half have elevated hemoglobin A1C levels consistent with hyperglycemia^46,47. Our data highlight the importance of understanding how the host metabolic environment affects bacterial metabolism, antibiotic resistance in vivo, and treatment outcome.

Methods

Ethics statement

All murine experiments were conducted in compliance with approved protocols and guidelines of the Emory University Institutional Animal Care and Use Committee (IACUC), protocol #201700106.

Isolate information

Enterobacter kobei Mu208 is a clinical isolate previously described⁸, collected as part of the Georgia Emerging Infections Program (GaEIP) in Georgia, USA. Additional GaEIP isolates of Enterobacter, Klebsiella, and Escherichia were also used. Strains and plasmids in this paper are detailed in Supplementary Table 1.

Reagents

Mueller–Hinton agar (MHA; BD Difco) and Mueller–Hinton broth (MHB; BD Difco) were used throughout as indicated. M9 broth was prepared with 1× M9 Salts (BD Difco), 100 µM CaCl₂, 2 mM MgSO₄, and carbon source as indicated, either glycerol (Promega #H5433) or anhydrous dextrose (VWR #BDH9230). For growth of E. cloacae before plating on MHA containing fosfomycin, bacteria were grown at 37 °C with shaking, 1.5 mL volume in 12 × 75 mm polypropylene aeration tubes (Globe Scientific #110438). When MHB was used, 25 µg/mL glucose-6-phosphate was included per CLSI guidelines⁴⁸. YESCA was made by autoclaving 1 g/L yeast extract (Fisher BP1422) and 10 g/L casamino acids (Acros #61204) in water. YESCA agar included 20 g/L agar (Sigma A1296).

Screening for GlpT function used M9 agar made by 2× M9 broth containing 2% glycerol phosphate disodium salt hydrate (Sigma # G6501) combined with 2× agar (Sigma A1296).

Antibiotics used were: fosfomycin disodium salt (TCI #F0889), kanamycin sulfate (Teknova #K2150; 35 µg/mL for strain Mu208), tetracycline (Alfa Aesar #J61714; 20 µg/mL for Mu208), chloramphenicol (Fisher #BP904, 50 µg/mL for Mu208) and spectinomycin dihydrochloride pentahydrate (Sigma #S4014; 50 µg/mL for E. coli, 125 µg/mL for Mu208).

Reduced L-glutathione (Sigma #G4251), L-glutamic acid (Sigma #49449), and adenosine 3′,5′-cyclic monophosphate sodium salt monohydrate (Sigma A6885) were added at concentrations indicated.

Medium gel pads used for cell imaging under microscope: UltraPure^TM agarose (Invitrogen #16500100) and MHB was used to prepare MHB 1.5% agarose gel pad. M9 sole carbon source 10% polyacrylamide (PA) gel pad was made by mixing acrylamide (Bio-Rad #1610140, 10% (w/v)), N, N-methylene bisacrylamide (Bio-Rad #1610142, 0.15% (w/v)), ammonium persulfate (Bio-Rad #1610700, 0.75% (w/v)) and TEMED (Bio-Rad #1610800, 0.25% (w/v)), with M9 + 0.8% glycerol or M9 + 0.4% glucose. Polymerization of PA gel pad was performed under room temperature for 20 min⁴⁹.

Whole genome sequencing

Mu208 was plated on MHA containing 0 or 256 µg/mL fosfomycin Biomass was collected and subject to whole genome sequencing following extraction with the Wizard Genomic DNA Purification Kit (Promega). The genome of Mu208 grown without fosfomycin (mostly susceptible) was subjected to Illumina (650 Mbp) and Nanopore (300 Mbp ONT) sequencing to create a reference genome and quantify gene copy variation. Porechop⁵⁰ was used to trim residual adapter sequence from the Oxford Nanopore Technology (ONT) reads that may have been missed during basecalling and demultiplexing. De novo genome assemblies were generated from the ONT read data with Flye⁵¹ under the nano-hq (ONT high-quality reads) model. Additional Flye options initiate the assembly by first using reads longer than an estimated N50 based on a genome size of 6 Mbp. Subsequent polishing used the Illumina read data with Pilon⁵² under default parameters. To reduce erroneous assembly artifacts caused by low quality nanopore reads, long read contigs with an average short read coverage of 15× or less were removed from the assembly. Assembled contigs were evaluated for circularization via circlator⁵³ using the ONT long reads. Assembly annotation was then performed with Bakta⁵⁴ using the Bakta v5 database. The genome of the resistant population from growth on agar with fosfomycin was subject to Illumina sequencing and mapped to the respective reference genome with CNV analysis. Quality control and adapter trimming was performed with bcl2fastq⁵⁵. Reads were mapped to their respective references via bwa mem⁵⁶. PCR and optical duplicates were marked and excluded from the analysis using PicardTools’ ‘MarkDuplicates’⁵⁷ functionality. Aligned read counts were imported into R’s CNOGpro⁵⁸. package. CNV events were called via CNOGpro using a bootstrapping method, which calculates an average gene number event giving a possible upper and lower bound. Variant calling was performed with breseq⁵⁹. Genome sequencing and analysis were performed by SeqCenter (seqcenter.com).

Population analysis profile

Population analysis profile (PAP) was performed as described previously⁸ and indicated in Supplementary Fig. 1. A given Enterobacter strain was grown overnight from a single colony streaked to MHA from −80 °C glycerol stocks. After approximately 16–20 h of growth, the culture was diluted (1:250 for MHB, except 1:100 for strains with a growth defect [ΔgshA, Δcrp, ΔcyaA], and 1:100 for M9 broth) into fresh media and grown for 6 h in MHB or 7 h in M9. The cultures were serially diluted in PBS in a 96-well plate (Falcon), and 7.5 or 10 µl of each dilution was plated on MHA containing fosfomycin as indicated. Colonies were enumerated after 24–48 h of growth. For many experiments shown, the same protocol was followed, but the cultures were plated only on MHA containing 0 or a single fosfomycin concentration indicated. PAP of Escherichia or Klebsiella was performed following overnight growth in M9 media as indicated, following inoculation with a single colony. S. aureus PAP was performed following overnight growth in Mueller–Hinton broth with or without 2% glucose added. We used the CLSI discontinued breakpoint of 32 μg/mL fosfomycin for S. aureus⁶⁰.

The surviving colonies are enumerated, and the isolate is classified as resistant if at least 50% of the total colonies grow at 1 or 2× breakpoint. An isolate is considered susceptible if less than 0.0001% (−6 logs) of the cells grow at any concentration shown. An isolate is considered heteroresistant if there is greater than 0.0001% survival at 1× breakpoint and an ≥8-fold difference between the resistance of the subpopulation and resistance of the main population. We include the breakpoint in our definition of HR because we are interested in the clinical relevance of this phenomenon beyond the biological questions posed by the existence of resistant subpopulations. It is worth noting that breakpoints can change based on clinical and regulatory changes, and occasionally breakpoints differ between regulatory agencies. Nevertheless, the inclusion of breakpoint in our definition defines subpopulations that are resistant to clinically relevant antibiotic concentrations. Defining the 8-fold resistance difference in Mu208, if <50% of the population survives at 256 µg/mL, the resistance of the main population is 128 µg/mL. The resistance of the subpopulation that survives on 1024 µg/mL is greater than 1024 µg/mL, making the fold change ≥8. The limit of detection is approximately −6.5 to 7.5 logs but varies based on the density of the culture, in the main figures the y-axis is set at −6 logs of survival. See Supplementary Fig. 1 for graphical representation.

In defining the features of heteroresistance in the isolates of this work, based on guidelines set forth by the field^2,8,61:

1. Clonality: these isolates demonstrate monoclonal heteroresistance, they are purified isolates, and single colonies are used throughout experiments.

2. Level of resistance: the MIC of the resistant subpopulation in wildtype Mu208 is ≥8× the MIC of the main population, when comparing the amount of killing by the lowest concentration of fosfomycin used throughout, and the growth of the resistant subpopulation at ≥2048 µg/mL.

3. Frequency of the resistant subpopulation: we consider the frequency of the subpopulation at 2× the CLSI breakpoint for the strains used throughout, the frequency is ~0.01% for wildtype Mu208 following growth in MHB.

4. Instability in the frequency of the resistant subpopulation: wildtype Mu208 demonstrated unstable heteroresistance. As shown in the passage experiments after selection, there is a significant reduction in the resistant population frequency within a single passage (~10 generations per passage) in antibiotic free media.

Time kill

Mu208 and Mu819 were streaked from −80 glycerol stocks to MHA for isolation, and a single colony was used to start overnight cultures in MHB grown for 18 h. Subsequently, 1.5 µl of this overnight culture was added to 1.5 mL MHB containing 512 µg/mL fosfomycin. At the timepoints indicated, 100 µl of culture was removed, serially diluted in PBS, and plated on MHA containing 0 or 512 µg/mL fosfomycin, grown for 24 h, and surviving colonies were enumerated.

Resistance stability assays

Mu208 was streaked from −80 °C glycerol stocks to MHA for isolation, and a single colony was used to start overnight cultures in 1.5 mL of M9 + 0.8% glycerol phosphate and grown 15–20 h. Fifteen microliters of the cultures were diluted into 1.5 mL of M9 + 0.8% glycerol phosphate of grown for 6 h. These cultures were diluted and plated MHA containing 0 or 256 µg/mL fosfomycin, the “baseline.” One point five microliters of the cultures were added to fresh M9 + 0.8% glycerol phosphate containing 256 µg/mL fosfomycin and grown for 16 h. Following growth, the cultures were diluted and plated MHA containing 0 or 256 µg/mL fosfomycin, the “+fosfomycin.” One point five microliters of the cultures were added to fresh M9 + 0.8% glycerol phosphate and grown for 20 h. Following growth, the cultures were diluted and plated MHA containing 0 or 256 µg/mL fosfomycin, the “drug-free media.”

Murine treatment failure

Approximately 8 week old female C57BL/6J Mus musculus (Jackson Laboratory) were injected intraperitoneally with 100 µl of Mu819, Mu208, or Mu208 ΔgshA suspended in sterile PBS at a density of ~2 × 10⁹ CFU/mL. Following infection, mice were treated at 200 mg/kg using a sterile stock of 40 mg/mL fosfomycin disodium salt via intraperitoneal injection at 2, 8, 14, 20, 26, and 32 h post-infection. Mice were closely monitored during the 96 h experiment, with mouse weights being taken every 6 h and mice were euthanized by carbon dioxide asphyxiation if moribund or below 80% starting weight, or at the conclusion of the experiment. All experiments were conducted in compliance with approved protocols and guidelines of the Emory University Institutional Animal Care and Use Committee (IACUC). Mice were housed under specific-pathogen-free conditions in filter-top cages at Emory National Primate Center, Emory University, and provided food and water ad libitum. The vivarium was maintained at 72 °F at 30–70% (optimally 40–50%) relative humidity under 12 h × 12 h light/dark cycle.

Transposon screen

Seven hundred and fifty microliters of a culture of a spontaneous streptomycin resistant variant of Mu208 was combined with 750 µl E. coli BW19851 carrying pFD1, collected, pelleted, and resuspended in 50 µl YESCA. The suspension was plated on a 0.45 µM nitrocellulose membrane placed on top of a YESCA agar plate and incubated at 37 °C for 2.5 h. The nitrocellulose was washed in a 15 ml tube with YESCA containing 1 mM IPTG and 100 µg/mL streptomycin, grown for 3 h at 37 °C, diluted, and plated on MHA containing 100 µg/mL streptomycin and 35 µg/mL kanamycin. After 24 h of growth at 37 °C, colonies were replica-plated using sterile velvet onto MHA containing kanamycin or kanamycin and 256 µg/mL fosfomycin. Colonies that appeared on kanamycin but not kanamycin+ fosfomycin were restreaked for isolation. To identify the transposon insertion site, colony PCR was performed using primers 71/72 to yield the DNA flanking the transposase. Two microliters of the PCR product from JCP71/72 was the template for subsequent nested PCR with JCP73/74. PCR products were Sanger sequenced (Azenta) using JCP74 and aligned to the Mu208 genome.

Strain construction: gene deletions

Lambda-red based allelic exchange was used to replace the coding sequence of genes with a kanamycin resistance gene, and then the gene was removed with Flp recombinase^62,63 to create unmarked in-frame deletions. The kanamycin resistance gene from pEXR6K_kanFRT was cloned using Promega GoTaq 2× master mix with Flp recognition sequence and homology to regions flanking the gene to be replaced using primers DHP175/181 (recA), DHP65/66 (fosA), DHP109/110 (gshA), DHP3/4 (glpT), DHP5/6 (uhpT), DHP49/50 (crp), JCP495/496 (glpR), DHP51/52 (cyaA), and DHP127/128 (ptsG), detailed in Supplementary Data 3. The purified PCR product was electroporated into competent Mu208 (or mutant) carrying pKD46-tet, and transformants were selected on kanamycin. Transformants were re-streaked to kanamycin for isolation and subject to PCR for successful allelic exchange as indicated by a product size change using primers flanking the gene to be replaced: primers DHP179/182 (recA), DHP73/74 (fosA), DHP117/118 (gshA), DHP9/10 (glpT), DHP11/12 (uhpT), DHP57/58 (crp), JCP497/498 (glpR), DHP59/60 (cyaA), and DHP135/136 (ptsG), detailed in Supplementary Data 3. The mutants were then made electrocompetent and electroporated with pCP20⁶³. Transformants were selected at 30 °C on chloramphenicol, patched to chloramphenicol at 30 °C and grown for 24 h, then patched to MHA and MHA+ kanamycin. Kanamycin-sensitive mutants were streaked for isolation, and PCR was used with the same flanking primers to screen for the loss of the kanamycin resistance gene.

Strain construction: plasmid-borne complementation

The coding sequences of the following Mu208 genes were cloned into pBAV-T5 at BamHI and NcoI sites using primers DHP290/291 (fosA), DHP286/287 (gshA), DHP284/285 (crp). Using Hi-fi assembly (NEB), pBAV-T5 amplified by PCR using JCP280/281 was assembled with glpR (JCP499/500), cyaA (TO131/132), and ptsG (TO142/143). Plasmids were propagated in E. coli NEB5α, Sanger sequenced (Azenta) using primers JCP339/340, and electroporated into appropriate strain with kanamycin selection.

Strain construction: chromosomally integrated plasmids

pCD13PSK was digested with BamHI and SacI. P_glpTglpT was amplified from the chromosome with primers (JCP477/478) that included homology to digested pCD13SPK and combined via NEB Hifi assembly. Similarly, glpT and its 5′ UTR were amplified with primers (JCP476/477) which included homology to pCD13PSK, and the forward primer (JCP476) included P_tet to generate pCD13PSK P_tetglpT. JCP511/512 amplified a linear product of pCD13PSK mCherry (Emily Crispell, Weiss laboratory) for Hifi assembly with Mu208 promoters to generate pCD13PSK P_genemCherry: primers amplified the promoter with homology to pCD13PSK mCherry: P_glpT (JCP513/514), P_gshA (JCP515/516), and P_crp (JCP521/522).

Plasmids were propagated in E. coli PIR2 with spectinomycin, and Sanger sequenced (Azenta) using primers JCP49/585. Mu208 wildtype or indicated mutants were made electrocompetent, transformed with pINT-ts-tet, and selected for with tetracycline. Strains carrying pINT-ts-tet were made electrocompetent and transformed with appropriate pCD13PSK plasmid. Of note, following 1 h of recovery at 37 °C, the culture was moved to 42 °C for 30 min before plating on spectinomycin. Integration of pCD13PSK at the attB site was confirmed by colony PCR using JCP474/53. To additionally confirm, the region of the chromosome encoding the cloned genes and pCD13PSK with amplified with JCP474/539 and Sanger sequenced (Azenta) using primers JCP49/585. For experiments, spectinomycin was not used because the integration is stable⁶⁴.

Imaging cells and mCherry expression

Mu208 wildtype carrying att:: P_glpTmCherry, att:: P_gshAmCherry, and att:: P_crpmCherry, along with the indicated Mu208 mutants carrying P_glpTmCherry were used to visualize population-level variation in promoter activation prior to fosfomycin exposure and to assess viability following treatment.

Imaging of cells and mCherry expression was performed as previously described^65,66. Once a cell culture reached OD₆₀₀ ~ 0.1, cells were placed on a 35 mm glass-bottom Petri dish (Cellvis) and covered with an MHB 1.5% agarose gel pad, or 10% polyacrylamide gel pad containing M9+glycerol or M9+glucose, as needed. Cells were imaged using an inverted microscope (Olympus IX83) with a 60× oil immersion phase-contrast objective lens, housed within a pre-warmed (37 °C) incubator chamber (InVivo Scientific). mCherry fluorescence was detected using a TRITC filter cube.

Snapshot imaging of cells was performed before exposure to fosfomycin. Time-lapse images of cells were captured every 30 min after adding fosfomycin to the agarose pad covering cells.

Image analysis

Fluorescence intensity of mCherry was measured using a plug-in for Fiji ImageJ software, named MicrobeJ⁶⁷ (version 5.13 I (22)). Reported intensities are the results after subtracting extracellular background fluorescence as well as autofluorescence of wildtype Mu208 cells from intracellular fluorescence signals.

Viability of cells were tracked from time-lapse images after fosfomycin exposure. Continuous cellular division and colony formation represented survival. Killing of cells by fosfomycin was defined as visible lysis or permanent arrest of cell growth.

qPCR

For cAMP addition, Mu208 was grown overnight in M9 + 0.8% glycerol, and 15 µl was added to 1.5 mL fresh M9 + 0.8% glycerol with and without 5 mM cAMP. After 7 h of growth, 1.3 mL was added to 2.6 mL RNAprotect (Qiagen #76104). RNA was subsequently isolated using RNeasy Mini Kit (Qiagen #74104), then treated with DNAse (BioLab #M0303S) according to manufacturer’s instructions. Reverse-transcriptase quantitative PCR was performed with Power SYBR Green RNA to C_T 1-Step Kit (Applied Biosystems #4389986). rplB was detected with TO168/169 and was used as the housekeeping for normalization; glpT with DHP264/265. The fold change for each biological replicate was compared to the same replicate from no cAMP addition, resulting in a value of 1 for each no cAMP replicate. Fold change was calculated using the 2^−ΔΔCT method⁶⁸.

For glucose addition, Mu208 was grown overnight in MHB, and 2.5 µl was added to 2.5 mL MHB with or without 2% glucose added and grown until OD reached ~0.4–0.5; 1 mL was added to 2 mL RNAprotect, RNA was isolated as above, and qPCR was performed as above.

GlpT function phenotypic screening

Colonies from the strains and agar containing fosfomycin concentrations indicated in Supplementary Fig. 6d were patched to M9 agar containing 2% glycerol phosphate. Mu208 wildtype and ΔglpT were patched as quality controls, wildtype grew, but ΔglpT was unable to grow. Mu208 forms small and large resistant colonies on fosfomycin, and from our screening we identified that only the large colonies exhibited the GlpT-null phenotype. By patching the more frequent small resistant colonies and less frequent large resistant colonies, we calculated the frequency of the GlpT null phenotype as reported in Supplementary Fig. 6d.

Murine diabetic models

Streptozocin treated and control treated male Mus musculus C57BL/6J mice were sourced from Jackson Laboratory, Stock # 033853. Jackson Laboratory verifies a non-fasted blood glucose exceeding 250 mg/dL prior to shipment, but most mice have glucose exceeding 350 mg/dL⁴². Male mice are the standard for STZ induced diabetes, and only Jackson Laboratory data from male mice were available⁴². These mice were infected intraperitoneally at ~10 weeks of age with 100 µl of Mu208 suspended in sterile PBS at a density of 1.7–2.4 × 10⁹ CFU/mL.

Mus musculus B6.Cg-Lep^ob/J heterozygous and homozygous male mice were sourced from Jackson Laboratory (Strain #000632). At 8 weeks male Lep^ob/Lep^ob mice have a mean non-fasted blood glucose of ~350 mg/dL and WT/Lep^ob mice have a mean of ~200 mg/dL⁴³. Male mice were selected because they more reliably experience hyperglycemia⁴³. These mice were infected intraperitoneally at ~8 weeks of age with 100 µl of Mu208 suspended in sterile PBS at a density of ~2 × 10⁹ CFU/mL.

For both series of experiments, after 24 h, mice were euthanized by carbon dioxide asphyxiation. The peritoneal cavity was washed with 3 mL sterile PBS using a syringe and 26-gauge needle, and the liver and spleen were weighed and collected in 1 mL PBS. Organs were homogenized using a BioSpec Tissue Tearor (#98570-395). The suspensions were diluted in PBS, spread, and spot-plated on MHA containing 0 or 256 µg/mL fosfomycin for enumeration. All experiments were conducted in compliance with approved protocols and guidelines of the Emory University Institutional Animal Care and Use Committee (IACUC). Mice were housed under specific-pathogen-free conditions in filter-top cages at Emory National Primate Center, Emory University, and provided food and water ad libitum. The vivarium was maintained at 72 °F at 30–70% (optimally 40–50%) relative humidity under 12 h × 12 h light/dark cycle.

Statistics and software

The number of experiments and the number of biological replicates are detailed in each figure legend, along with statistical test as appropriate. For our in vitro experiments, biological replicates are derived from independent stationary phase cultures from independent colonies, often with additional independent subcultures as described. Unless otherwise specified, the mean is shown, and error bars indicate standard deviation. Microsoft Excel Version 2311 was used to collect data. Data were analyzed using GraphPad Prism 10.1.1, 10.1.2, or 10.5.0. Affinity Designer v2.6 was used for illustrations in Fig. 6 and Supplementary Fig. 1.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

J.E.C.: conceptualization, investigation, visualization, and writing-original draft. D.A.H.: conceptualization, investigation, writing-review & editing. M.M.: investigation, visualization, writing-review & editing. T.O.: investigation, validation, writing-review & editing. V.I.B.: investigation, writing-review & editing. M.K.: supervision, resources, writing-review & editing. D.S.W.: conceptualization, investigation, writing-review & editing, supervision, resources, funding acquisition, and project administration.

Peer review

Peer review information

Nature Communications thanks Yingshun Zhou and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

D.S.W. discloses support for the research of this work from the National Institutes of Health [AI158080, AI141883, and AI197193] and a Burroughs Wellcome Fund Investigators in the Pathogenesis of Infectious Disease award. J.E.C. discloses support for the research of this work from the Cystic Fibrosis Foundation [CHOBY19F0] and the National Institutes of Health [T32 DK108735]. This work was supported by the National Institutes of Health’s Office of the Director, Office of Research Infrastructure Programs, P51 OD011132. The Multi-Site Gram-negative Surveillance Initiative is funded by the CDC. The content is solely the responsibility of the authors and does not necessarily reflect the official views of the National Institutes of Health.

Data availability

The sequencing data generated in this study from Mu208 grown in fosfomycin have been deposited in NCBI SRA: SAMN49513011 [https://www.ncbi.nlm.nih.gov/sra?LinkName=biosample_sra&from_uid=49513011]. Mu208 reference genome sequences are available at NCBI: GCF_048016555.1 [https://www.ncbi.nlm.nih.gov/datasets/genome/GCF_048016555.1/]. The copy number variation data associated with sequencing are provided in Supplementary Data 1. Materials are available upon request to D.S.W. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-77232-6.