Edwardsiella tarda LTB4 used in this study was obtained from Professor Xiaohua Zhang, Ocean University of China University. LTB4 was grown at 30°C for 24 h in 50 ml Luria-Bertani (LB) broth in 250-ml flasks.

Measurement of minimum inhibitory concentration (MIC) was performed as previously described (National Committee for Clinical Laboratory Standards, 1999). In brief, LTB4 was cultured in LB medium with twofold serially diluted ceftazidime (CAZ, Guangzhou QiYun Biological Technology) ranging from 0.01 to 160 μg/ml. Overnight bacterial cultures were diluted 1:100 in fresh LB medium and cultured at 30°C to an OD600 of 0.5. The tray contained a series of twofold dilutions of antibiotics. Ninety microliters of LB containing CAZ and 10 μl of logarithmic phase cells with 10⁷ CFU/ml of LTB4-R_CAZ or LTB4-R_CAZ were incubated in a microwell plate for 24 h at 30°C. The lowest concentration showing no visible growth was recorded as the MIC. At least three biologic replicates were performed.

Real-time quantitative PCR (qRT-PCR) was carried out as previously described (Li et al., 2016). Bacterial cells were harvested at OD600 = 1.0. The total RNA of each sample was isolated with TRIzol (Invitrogen, United States). Reverse transcription-PCR was carried out on a PrimeScript^TM RT Reagent Kit with gDNA eraser (Takara, Japan) with 1 μg of total RNA according to manufacturer’s instructions. qRT-PCR was performed in 384-well plates, and each well contained a total volume of 10 ml liquid including 5 ml 2X SYBR Premix Ex Taq^TM, 2.6 μl PCR-grade water, 2 μl cDNA template, and 0.2 μl each pair of primers (10 mM). The primers are listed in Supplementary Table 1. All the assays were performed on the LightCycler 480 system (Roche, Germany) according to the manufacturer’s instructions, and four independent samples were assayed for both the control group and the test group. The cycling parameters were listed as follows: 95°C for 30 s to activate the polymerase; 40 cycles of 95°C for 10 s; and 60°C for 30 s. Fluorescence measurements were performed at 70°C for 1 s during each cycle. Cycling was terminated at 95°C with a calefactive velocity of 5°C per second, and a melting curve was obtained. To analyze the relative expression level of the target gene, we converted the data to percentages relative to the value of no-treatment group. At least triplicate biological repeats were carried out.

Bacterial sample preparation was carried out as previously described (Zhang et al., 2019). In brief, 10 ml OD600 = 1.0 cells were quenched with cold methanol and sonicated for 5 min at 200 W. Samples were centrifuged at 12,000 rpm for 10 min. Supernatant, containing 1 μg/ml ribitol (Sigma–Aldrich) as internal analytical standard, was transferred into a new tube and dried by vacuum centrifugation device (LABCONCO). The dried extracts were then incubated with 80 μl methoxyamine hydrochloride (20 mg/ml, Sigma–Aldrich) in pyridine (Sigma–Aldrich) for 90 min at 37°C and derivatization was done with an identical volume of N-methyl-N-(trimethylsilyl)trifluoroacetamide (Sigma–Aldrich) for another 30 min. Samples were centrifuged at 12,000 rpm for 10 min, and the supernatant was transferred into new tubes. Gas chromatography-mass spectrometry (GC-MS) analysis was performed with an Agilent GC-MS instrument. Spectral deconvolution and calibration were performed using AMDIS and internal standards as previously described. A retention time (RT) correction was performed for all the samples, and then the RT was used as a reference against which the remaining spectra were queried, and a file containing the abundance information for each metabolite in all the samples was assembled. Metabolites from the GC-MS spectra were identified by searching in the National Institute of Standards and Technology (NIST11.L) Mass Spectral Library. Among the detected peaks of all chromatograms, compound peaks were considered as endogenous metabolites and the same metabolite names were merged. The resulting data matrix was normalized by the concentrations of added internal standards and the total intensity. This file was then used for subsequent statistical analyses. The abundance of a metabolite was scaled by total abundance of all metabolites in a sample as its relative abundance for further analysis. Hierarchical clustering was completed in the R platform with the package gplots¹ using the distance matrix. Multivariate statistical analysis included principal component analysis (SIMCA-P 12.0.1), which was used to discriminate sample patterns. GraphPad Prism 7 was used to draw figures.

Measurement of enzyme activity was performed as previously described with a few modifications (Zhang et al., 2019; Jiang et al., 2020c). In brief, the harvested cells were collected at OD600 = 1.0 and then re-suspended in sterile saline to OD600 = 1.0 after washing. Samples of 30 ml were collected by centrifugation at 8,000 rpm for 5 min. Pellets were re-suspended in phosphate-buffered saline (PBS) and broke down by sonication for 2 min at a 200-W power setting on ice, and then centrifuged at 12,000 rpm for 10 min to remove insoluble materials. Supernatants containing 200 μg of total proteins were transferred to a pyruvate dehydrogenase (PDH) reaction mix (0.15 mM MTT, 1 mM MgCl₂, 0.5 mM PMS, 0.2 mM TPP, 2 mM sodium pyruvate, and 50 mM PBS), an α-ketoglutarate dehydrogenase (KGDH) reaction mix (0.15 mM MTT, 1 mM MgCl₂, 0.5 mM PMS, 0.2 mM TPP, 50 mM α-ketoglutaric acid potassium salt, and 50 mM PBS), a succinate dehydrogenase (SDH) reaction mix (0.15 mM MTT, 1 mM PMS, 5 mM sodium succinate, and 50 mM PBS), and a malate dehydrogenase (MDH) reaction mix (0.15 mM MTT, 1 mM PMS, 50 mM PBS, and 50 mM malate) to a final volume of 200 μl in a 96-well plate. Subsequently, the plate was incubated at 37°C for 5 min for PDH/KGDH/SDH/MDH assay and then measured at 566 nm for colorimetric readings. The plate was protected from light during the incubation. The dehydrogenase activity was assayed in quadruplicate.

The activity of superoxide dismutase (SOD) and catalase (CAT) was determined by commercial kits (Nanjing Jiancheng, China). The process is as follows: 10 ml of bacteria with OD600 = 1.0 was collected, washed twice with PBS, and suspended in PBS. The suspension was broke down by sonication for 2 min at a 200-W power setting on ice and then centrifuged at 12,000 rpm for 10 min to remove insoluble materials. The supernatant was taken, and the protein concentration was determined. The activity of SOD was measured with 150 μg total protein. The follow-up test was carried out according to the manufacturer’s instructions. Then, SOD activity was measured at 450 nm, and CAT activity was measured at 405 nm. The plate was protected from light during the incubation.

Detection of ATP was determined by a BacTiter-Glo^TM Microbial Cell Viability Assay (Cat. G8231, Promega, Madison, WI, United States) as previously described (Cheng et al., 2019). In brief, bacterial cells were harvested in the OD600 of 1.0 by centrifugation for 10 min at 12,000 g and washed twice by centrifugation with sterile saline. Cells were resuspended with saline solution and adjusted the OD600 to 1.0. Then, 50-μl samples were added to a 96-well plate and mixed with an equal volume of the kit solution. Then, the absorbance was measured using VICTOR X5 (PerkinElmer, Turku, Finland) according to the manufacturer’s instructions. The concentration of ATP was calculated according to the standard curve of ATP.

Antibacterial assay was carried out as described previously (Peng et al., 2015). Overnight bacterial cultures were diluted 1:100 in fresh LB medium and cultured at 30°C to an OD600 of 1.0. Bacterial cells were collected by centrifugation at 8,000 rpm for 5 min. The samples were then washed with sterile saline three times; suspended in M9 minimal media containing 10 mM acetate, 1 mM MgSO₄, and 100 μM CaCl₂; and diluted to an OD600 of 0.2. Fe³⁺ or/and ceftazidime were added and incubated at 30°C and 200 rpm for 6 h. To determine bacterial count, 100 μl of cultures was obtained and then serially diluted. An aliquot of 10μl of each dilution was plated in TSB agar plates and incubated at 30°C for 22 h. The plates only with 20–200 colonies were counted, and the colony-forming unit per milliliter was calculated.

Bacterial membrane potential was measured by BacLight Bacterial Membrane Potential Kit (Invitrogen). In brief, bacteria were diluted to 10⁶ CFU/ml and stained with 10 μl of 3 mM DiOC₂, followed by incubation for 30 min. Samples were analyzed using a FACSCalibur flow cytometer (Becton Dickinson, San Jose, CA, United States). The green/red fluorescence was detected with 488–530/610 nm. The membrane potential was determined and normalized as the intensity ratio of the red fluorescence and the green fluorescence. Relative proton motive force (PMF) was determined by test samples compared with control samples.

LTB4 was cultured in LB medium with or without twofold serially diluted ceftazidime, leading to LTB4-R_CAZ and LTB4-S, respectively. The MIC of LTB4-S was 0.039 μg/ml CAZ, while that of LTB4-R_CAZ was 0.625 μg/ml CAZ. There was a 16-fold difference between the two strains (Figure 1A). Consistently, a higher viability was detected in LTB4-R_CAZ than LTB4-S (Figure 1B). To explore whether LTB4-R_CAZ had a reduced ROS production, ROS production was detected in LTB4-R_CAZ and LTB4-S. A lower ROS production was determined in LTB4-R_CAZ than LTB4-S (Figure 1C). These results indicate that ROS production is reduced in LTB4-R_CAZ compared with LTB4-S.

These above results motivated us to speculate that ROS plays an important role in ceftazidime resistance. To demonstrate this, the viability of LTB4-R_CAZ was detected in the presence or absence of ROS promoter or/and inhibitor with ceftazidime. There was a stronger resistance to H₂O₂ in LTB4-S than Escherichia coli K12 and thereby 5 mM H₂O₂ was used (Supplementary Figure 1). The promoters Fe³⁺ and H₂O₂ (positive control) potentiated ceftazidime-mediated killing. However, the inhibitor thiourea not only eliminated the potentiation caused by the promoters but also inhibited the ceftazidime-mediated killing (Figure 2A). The effect of Fe³⁺ was increased in a dose-dependent manner and was related to the concentration of ceftazidime used (Figure 2B). When the concentration of Fe³⁺ was fixed, the killing efficacy was ceftazidime dose dependent (Figure 2C). To validate the role of the ROS promoter or/and inhibitor in the ceftazidime-mediated killing, we measured the ROS production of LTB4-R_CAZ in the presence or absence of the ROS promoter or/and inhibitor. ROS production was elevated and reduced in the presence of the inhibitor thiourea and promoters Fe³⁺, H₂O₂, or ceftazidime alone, respectively. Comparatively, the elevated ROS ranked as H₂O₂ > ceftazidime > Fe³⁺. When the synergistic use of ceftazidime with one of the promoters or/and the inhibitor was performed, they promoted and inhibited ROS level, respectively (Figure 2D). These results support the conclusion that ROS is related to ceftazidime resistance and promotes ceftazidime-mediated killing.

To understand why ROS are reduced in LTB4-R_CAZ, the P cycle and SOD degradation were investigated. The P cycle is a recently illustrated cycle, which provides respiratory energy in E. tarda (Su et al., 2018), and is related to ROS biosynthesis (Figure 3A). For the investigation of the P cycle, qRT-PCR was used to measure the expression of genes in the P cycle. Among the 18 genes detected, 11, 2, and 5 were reduced, elevated, and unchanged, respectively. Specifically, the 11 reduced genes encode all enzymes detected except for citrate synthase (CS) and isocitrate dehydrogenase (ICDH). The two elevated genes encode PDH and MDH. The five unchanged genes encode CS, ICDH, SDH, fumarate reductase (FRD), and pyruvate kinase (PK) (Figures 3B,C). Consistently, a lower activity of PDH, KGDH, and SDH was detected in LTB4-R_CAZ than that in LTB4-S (Figure 3D). NADH, membrane potential, and ATP were reduced in LTB4-R_CAZ compared with LTB4-S (Figures 3E–G). These results indicate that the inactivation of the P cycle forms a characteristic feature in LTB4-R_CAZ, which is related to the reduced ROS production. For the investigation of ROS degradation, the activity of ROS degradation enzymes was measured. Among these enzymes working for the ROS degradation, SOD and CAT are widely employed to indicate the antioxidant response. Therefore, the activity of SOD and CAT was detected in LTB4-R_CAZ and LTB4-S. A similar activity of SOD and CAT was measured between LTB4-R_CAZ and LTB4-S (Figures 3H,I). These results indicate that antioxidant response is not changed in LTB4-R_CAZ. Taken together, the lower ROS in LTB4-R_CAZ is attributed to a reduction of ROS generation instead of an increase of ROS degradation.

The above results motivated us to explore whether Fe³⁺ promotes global metabolism including the P cycle to elevate ROS level and potentiate ceftazidime-mediated killing. To do this, GC-MS-based metabolomics was used to compare metabolic profiles between media with or without Fe³⁺. Four biological and two technical replicates were performed in each group, yielding eight data sets with 65 metabolites in a sample. Among the 65 metabolites, 33 (50.8%) showed differential abundance (p < 0.05), with 27 at higher abundance and 6 at lower abundance in the presence of Fe³⁺ (Figure 4A). Fourteen pathways were enriched, of which alanine, aspartate, and glutamate metabolism; the TCA cycle; and aminoacyl-tRNA biosynthesis were the top three pathways by impact (Figure 4B). Differential metabolites at abundance are listed in Figure 4C, where all detected metabolites were elevated in the TCA cycle (Figure 4C). Principal component analysis identified two principal components, where component t[1] distinguished LTB4-R_CAZ from LTB4-R_CAZ + Fe³⁺ (Figure 4D). Discriminating variables were identified on an S-plot (Figure 4E). Cutoff values were ≥ 0.05 for absolute value of the covariance p and ≥ 0.5 for correlation p (corr). Nine putative biomarkers were identified, including decreased abundance of 9-hexadecenoic acid, acetic acid, and myo-inositol and elevated abundance of citric acid, oxalic acid, putrescine, butanoic acid, threonine, aspartic acid, and palmitic acid. Among the elevated metabolites, citric acid plays a role in the P cycle. These results indicate that Fe³⁺ impacts the metabolic profile, where the P cycle is elevated.

To further demonstrate the Fe³⁺-mediated activation, gene expression and enzyme activity were measured in the P cycle. qRT-PCR analysis showed that out of 18 genes tested, 5, 3, and 10 were elevated, reduced, and unchanged, respectively. In detail, the five elevated genes encode PDH, ACO, ICDH, SCS, and FRD, and the reduced genes encode KGDH, SDH, and PK (these enzymes are encoded by two genes) (Figures 5A,B). The activity of PDH, KGDH, SDH, and MDH was elevated (Figure 5C), which was further supported by increased NADH, membrane potential, and ATP (Figures 5D–F). These results indicate that Fe³⁺ promotes the P cycle but does not affect the activity of SOD and CAT (Figure 5G).

To further validate that Fe³⁺ promotes ROS via the P cycle, inhibitors of the P cycle were used to investigate whether the inhibition affects ROS and viability of LTB4-R_CAZ in the presence of both Fe³⁺ and ceftazidime. Furfural is a non-competitive inhibitor for PDH, and malonic acid is competitive against SDH. Lower ROS were detected in the presence than the absence of furfural or malonic acid, while higher ROS were determined in the medium with than without Fe³⁺ in the presence or absence of ceftazidime and furfural or malonic acid (Figure 6A), suggesting that Fe³⁺ partly reverts the inhibition mediated by the two inhibitors. Consistently, a lower viability was detected in the medium with than without Fe³⁺ in the presence of furfural or malonic acid (Figure 6B). To understand the mechanisms by which Fe³⁺ promotes the P cycle, we supposed that Fe³⁺ activates the activity of enzymes in the P cycle. To explore this, the activity of PDH, KGDH, SDH, and MDH was measured in a medium with the indicated concentrations of Fe³⁺. The activity of the four enzymes was increased in a Fe³⁺ dose-dependent manner in vitro (Figure 6C). These results indicate that Fe³⁺ promotes the activity of PDH, KGDH, SDH, and MDH in the P cycle directly, which is related to the elevation of ROS level in the presence of Fe³⁺.

To ensure the Fe³⁺-induced potentiation is generalizable to cephalosporins and clinically isolated pathogens, three types of cephalosporins (ceftazidime, cefoperazone, and cefazolin) and 12 strains of bacterial pathogens (E. tarda, E. coli, Klebsiella pneumoniae, and Pseudomonas aeruginosa) were used. The 12 strains of pathogens are resistant to at least three classes of antibiotics and thereby belong to multidrug-resistant bacteria. However, Fe³⁺ effectively promotes the three drugs to kill all pathogens (Figure 7). These results indicate that the Fe³⁺-induced potentiation is generalizable to the drugs and bacterial pathogens.