Phusion DNA polymerase, restriction enzymes, and T4 ligase used in plasmid construction were purchased from Thermo Fisher Scientific (Waltham, MA, United States). DNA-spin Plasmid DNA Purification Kit and MEGA quick-spin Total DNA Purification Kit were purchased from iNtRON Biotechnology (Burlington, MA, USA). Antibiotics and inducers were purchased from Gold Biotechnology, Inc. (St. Louis, MO, USA). PpGpp standard was purchased from Jena Biosciences (Jena, German). Other chemicals were purchased from Sigma-Aldrich (St. Louis, MO, USA).

The E. coli K-12 strain NCM3722, sourced from the Coli Genetic Stock Center (Yale, USA), was employed as the wild-type strain. All strains and plasmids are listed in Supplementary Tables S1, S2, respectively. E. coli MDS42pdu strain was used for cloning. DNA Golden Gate Assembly and BglBrick (Lee et al., 2011) were used for plasmid construction. Primers were designed by Benchling (San Francisco, CA, USA) and synthesized by Integrated DNA Technologies (Coralville, IA, USA).

The ppGpp RNA-based biosensor (pET-28c S2) was purchased from Addgene (plasmid #171920) and transformed into NCM3722 ybhC: T7 RNAP strain, which was made through inserting a T7 RNA polymerase expression cassette into the WT NCM3722 genome. Specifically, T7 RNA polymerase gene, LacI, LacI operator and Int 1 (T7-RNAP) were cloned from E. coli BL21(DE3) and was inserted into the ybhC gene site (Ye et al., 2021) of NCM3722 according to the pTarget-pCas homologous recombination protocol as described previously (Jiang et al., 2015, 2017). Colony PCR was used for mutant selection and the DNA fragment sequences were further confirmed by Sanger sequencing (South Plainfield, NJ, USA).

Genes encoding Aas, PlsB, and GFP were overexpressed from plasmids, which were constructed by inserting their PCR amplified DNA sequence into a low-copy number (SC101origin of replication), chloramphenicol resistant plasmid under the control of a LlacO-1 promoter (Lee et al., 2011).

All strains were grown from single colonies and cultivated at 37 °C with 250 rpm of shaking following previous protocols (Hartline et al., 2020, 2022; Zhang et al., 2024). Luria-Bertani (LB) media (10 g NaCl, 10 g Tryptone and 5 g Yeast Extract) was used for cloning and amplifying population from single colony. M9 minimum media [75 mM MOPS at pH 7.4, 2 mM magnesium sulfate, 1 mg/L thiamine hydrochloride, 10 μM iron(II) sulfate, 100 μM calcium chloride, 3 μM ammonium heptamolybdate, 0.4 mM boric acid, 30 μM cobalt(II) chloride, 15 μM copper(II) chloride, 80 μM manganese(II) chloride, and 10 μM zinc sulfate] supplemented with different carbon sources, such as glucose (12 mM), oleic acid (4 mM), malate (18 mM), succinate (18 mM), fumarate (18 mM), or pyruvate (18 mM), was used for persister assay. All cultures were supplemented with appropriate antibiotics (ampicillin, 100 mg/L; kanamycin, 50 mg/L; streptomycin, 100 mg/L; chloramphenicol, 25 μg/mL; carbenicillin, 100 mg/L; gentamicin, 20 mg/L).

Nutrient shift protocols were developed following previous methods (Kotte et al., 2014; Hartline et al., 2022; Zhang et al., 2024). In brief, exponentially growing cells cultured in M9 glucose media were harvested by centrifugation (4,500 × g, 10 min, 4 °C) and washed three times with ice-cold carbon-free M9 media to remove pre-shift carbon completely. The cells were subsequently resuspended in M9 media, which was supplemented with oleic acid, fumarate, succinate, pyruvate, or malate, respectively, to an initial optical density (OD₆₀₀) of 0.1–0.2. The cell culture was challenged with appropriate antibiotics (ampicillin, 100 mg/L; carbenicillin, 100 mg/L; gentamicin, 20 mg/L) immediately after the nutrient shift. At designated time points post-nutrient shift, 1 mL aliquots were collected, centrifuged (4,500 × g, 5 min, 4 °C), and washed four times with ice-cold PBS (pH 7.4) to remove residual ampicillin. Cells were resuspended in PBS, serially diluted, and 5 μL aliquots were plated onto LB agar plates. After overnight incubation at 37 °C, colonies were counted to determine the colony-forming units per milliliter of culture. For strains harboring gene overexpression plasmids, 1 mM IPTG was added to the pre-shift M9 glucose culture at OD600 ≈ 0.1 and maintained throughout the assay. For NCM3722 yhbC:T7 RNAP cells carrying the ppGpp biosensor, 1 mM IPTG was added 2 h before the nutrient shift and maintained throughout the assay.

Cell cultures were pelleted by centrifugation (6,500 × g, 4 °C, 5 min), washed, and resuspended in 10 μL of M9 media containing ampicillin and the appropriate carbon source. Next, 1 μL aliquot of the resuspended cells was transferred onto a 1% M9 agarose pad supplemented with ampicillin and the same carbon source, prepared according to Wilmaerts et al. (2022). The agarose pad was then sealed with a coverslip and incubated at 37 °C in a humid chamber (Tokai Hit, Incubation Systems for Microscopes, Japan) and imaged using a Nikon Eclipse Ti microscope (Nikon Instruments Inc., USA) equipped with an EMCCD camera (Photometrics Inc., Huntington Beach, CA, USA) and a 100×, NA 1.40, oil-immersion phase-contrast objective lens. An X-Cite 120 LED was used as the light source. Time-lapse images were acquired every 5 min by an automated scanning function of the microscope with a built-in perfect focus system.

To quantify ppGpp using the biosensor, 200 μM of DFHBI-1 T was added to 1 mL of cell culture, followed by incubation in the dark at room temperature for 30 min. Cells were then pelleted by centrifugation (6,500 × g, 5 min) and resuspended in 10 μL 1x PBS. A 1 μL aliquot of the resuspended cells was used for imaging. Images were acquired using microscope as mentioned above with a FITC (λ_ex = 480 ± 30 nm, λ_em = 510 ± 10 nm, Nikon Instruments Inc., USA) filter cube for spectral separation. More than 300 single cells per sample were collected and analyzed using the Nikon NIS-Elements software package following previous methods (Han and Zhang, 2020).

ppGpp quantification was performed as described before (Radzikowski et al., 2016) with slight modifications. At least 2 × 10¹⁰ cells were harvested by centrifugation (10,000 × g, 4 °C, 1 min), and cell pellet was immediately resuspended in 3 mL of 1 M ice-cold formic acid. Samples were then incubated at 0 °C for 60 min with intermittent vertexing. Following incubation, the suspension was centrifuged (21,000 g, 4 °C, 1 min) and the supernatant was transferred to a new tube and flash-frozen in liquid nitrogen. Water and formic acid were removed by freeze-drying overnight. The dried metabolites were resuspended in 0.13 mL of 0.1 M formic acid by vertexing and ultrasonication, then filtered by centrifugation (21,000 × g, 4 °C, 1 min) through a Spin-X centrifuge filter (Corning, 0.22-μm nylon membrane, 4,000 g). Samples were analyzed using an Agilent 1,260 HPLC system equipped with a multi-wavelength detector (MWD) monitoring absorbance at 260 nm. A PL-SAX anion exchange column (10 μm, 150 × 4.6 mm, 1,000Å, Agilent) was used at a column temperature of 60 °C. A 35-minuate linear gradient of two solvents was applied, changing from Buffer A (0.01 M K₂HPO₄, pH 2.6) to Buffer B (0.5 M K₂HPO₄, pH 3.5) in 30 min, using a flow rate of 1 mL/min. Peak areas were quantified using a ppGpp standard curve prepared in 0.1 M formic acid. Intracellular concentrations were calculated based on the number of cells harvested and cell volume (Volkmer and Heinemann, 2011).

Following a nutrient shift from ¹²C-glucose to ampicillin-containing ¹³C-Oleic acid, samples were collected at 5, 9, and 24 h post shift. At least 2 × 10¹⁰ cells were added to pre-chilled 0.9% (wt/vol) NaCl solution (1:1 volume ratio) to quench the metabolism. Cells were then pelleted by centrifugation (8,000 g at 4 °C, 10 min), and the pellet was resuspended in 1.5 mL of 6 M HCl and hydrolyzed at 100 °C for 20 h. The hydrolysates were then transferred to microcentrifuge tubes, and debris was removed by centrifugation (20,000 × g, 5 min). Amino acids were derivatized the by adding 300 μL of tetrahydrofuran (THF) and TBDMS (1:1, vol/vol), followed by incubation at 70 °C for 1 h. Samples were analyzed with GC–MS after removing the debris with centrifugation and data was processed as described previously (Hollinshead et al., 2019; Sun et al., 2025).

After performing a GLU → OA + AMP shift, samples were collected at 5, 9, and 24 h post shift. Cells exponentially growing in M9 ¹³C-glucose media were collected as reference. At least 2 × 10¹⁰ cells were collected and added to pre-chilled 0.9% (wt/vol) NaCl solution (1:1, vol/vol) to quench metabolism. Cell pellet was then collected by centrifugation at 5,000 × g, 0 °C for 3 min, and then resuspended in 750 μL methanol-chloroform solution (volume ratio 1:1). Experimental samples after the nutrient shift were combined with reference sample (cells exponentially growing in M9 ¹³C-glucose) and then incubated at 4 °C with shaking at 250 rpm for 4 h. After incubation, 500 μL of ice-cold HPLC grade water was added, and the mixture was vortexed for 2 min. The mixture was then centrifuged at 600 × g for 8 min at −10 °C to allow phase separation. The aqueous phase was carefully transferred to a clean tube and filtered using a 3 kDa cut-off centrifugal filter (21,000 × g, −10 °C, 90 min). The filtrate was frozen in liquid nitrogen and freeze-dried. Dried metabolites were resuspended in 100 μL of methanol-chloroform (1:1, vol/vol) and analyzed with LC–MS as previously described (Hollinshead et al., 2016, 2019).

Statistical analysis was conducted using GraphPad Prism. The data utilized originate from at least three biological replicates, where applicable. Statistical outcomes are denoted by asterisks (*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001), indicating significant results, or by the absence of asterisks (non-significant).

Cells demonstrate remarkable metabolic plasticity, allowing them to efficiently use a wide range of metabolic substrates and adapt to changing environmental conditions. This adaptability is essential for their survival and sustained function (Cabodevilla et al., 2013). Following a GLU → OA + AMP switch, 56% of cells survived 24-h of ampicillin treatment (Hartline et al., 2022; Zhang et al., 2024). We first assessed whether this high persistence level is unique to the GLU → OA + AMP switch or can be triggered by a nutrient shift from glucose to other gluconeogenic carbon sources in the presence of ampicillin. We selected pyruvate (PYR), malate (MAL), fumarate (FUM), and succinate (SUC) for nutrient shift persister assays, as these carbons represent different entry points into glycolysis and the TCA cycle. Cells at designated time points after nutrient shift were collected, washed and plated on antibiotic-free plates to determine survivor counts and generate time-kill curves (Figure 1A). Unlike GLU → OA + AMP shift, glucose shifts to all four tested gluconeogenic carbons exhibited normal cell killing dynamics, with persistence levels ranging from 10⁻⁶ to10⁻⁷ (Figure 1B), consistent with wild-type (WT) E. coli persistence levels in exponentially growing cell cultures (Vázquez-Laslop et al., 2006). Additionally, these shifts did not induce notable transient tolerance, suggesting E. coli resumed growth quickly after switching without the extended lag phase as observed during a shift to OA (Supplementary Figure S1).

To test whether the elevated persistence level after GLU → OA + AMP switch is specific to ampicillin, we performed the nutrient shifts and time-kill experiments using an alternative beta-lactam antibiotic, carbenicillin (CRB), or an aminoglycoside, gentamicin (GEN). For both CRB and GEN, high levels of persistence (>1% persisters at 24 h post-shift) were only observed after GLU → OA + ABX shift, not with other gluconeogenic carbon shifts (Figure 1C), indicating this effect is specific to the GLU → OA + ABX condition.

The unique high persistence makes the GLU → OA + ABX shifts a compelling model to study how metabolism drives persister formation. We focused on ampicillin due to its characteristic lysis-based killing, which can be readily visualized under the microscope. Time-lapse microscopy was used to monitor cell morphological changes post nutrient shifts. After shifting from glucose to pyruvate, succinate, malate, or fumarate, cells lysed in the presence of ampicillin as indicated by sudden disruptions in cell outlines (Supplementary Videos S1–S4). Cell lysis typically occurred within 1–4 h post-shift, consistent with the time-kill data. Cell lysis in the presence of ampicillin is also consistent with the primary killing mechanism: inhibition of peptidoglycan biosynthesis weakens bacterial cell wall, making it unable to withstand the internal pressure (Kawai et al., 2023). In contrast, cells after the GLU → OA + AMP shift remained intact without lysis throughout the 24-h ampicillin treatment, indicating ampicillin failed to weaken bacterial cell walls (Figure 1D; Supplementary Video S5).

Glucose is the preferred carbon source for E. coli. Previous studies have shown that shifting cells from a preferred to a less preferred carbon source can trigger the stringent response, mediated by the signaling molecules guanosine tetra- and penta-phosphate, collectively known as (p)ppGpp (referred to here as ppGpp) (Ahn-Horst et al., 2022; Zhu and Dai, 2023). Accumulation of ppGpp downregulates multiple growth-related processes and upregulates stress response pathways. Under extreme conditions, ppGpp promotes dormancy and enhances persistence level (Amato et al., 2013; Kotte et al., 2014; Amato and Brynildsen, 2015; Radzikowski et al., 2016). Given this role, we investigated whether the high persistence observed after the GLU → OA + AMP downshift is associated with elevated ppGpp level.

We used an RNA-based ppGpp biosensor developed by Sun et al. (2021) to measure ppGpp levels in single cells following nutrient shifts. The biosensor comprises a ppGpp-specific aptamer, a transducer sequence, and a fluorogenic RNA reporter. Upon ppGpp binding, the transducer promotes hybridization that enables the RNA reporter to bind its cognate dye, DFHBI-1T, activating fluorescence. We transformed E. coli with the ppGpp biosensor and kept the RNA constitutively transcribed while cultivating the cells in either a steady state growing phase in M9 glucose media or subjected to nutrient shifts to M9 OA media supplemented with ampicillin.

Cells grown on glucose displayed weak green fluorescence, establishing the baseline ppGpp level (Figure 2A). At 30 min after the GLU → OA + AMP shift, fluorescence was slightly higher but not statistically different from the glucose control. However, by 3 h, a subpopulation of cells exhibited a substantial increase in fluorescence intensity and lasted through 6 h post-shift, indicating elevated ppGpp levels (Figures 2A,B). This increase coincided with the transient tolerance period observed in earlier persistence assays (Figure 1B) and is likely due to impaired FA utilization, potentially caused by limited FA degradation enzymes (Zhang et al., 2024). The presence of high-fluorescence cells supports a potential role for ppGpp in persistence formation.

To further confirm ppGpp accumulation dynamics, we also quantified the population-average intracellular ppGpp levels using HPLC, following published methods (Radzikowski et al., 2016). The standard curve and chromatograms are presented in Supplementary Figure S2. The population level ppGpp change closely mirrored biosensor results (Figure 2C), showing an approximately 3-fold increase at 3 h followed by a slight decline at 6 h. In contrast, no significant ppGpp elevation was observed after shifting glucose to other gluconeogenic carbon sources (Figures 2D–G). Instead, ppGpp levels steadily declined, likely due to cell death post-shift. Collectively, these findings confirm that ppGpp accumulates following the GLU → OA + AMP shift. Its accumulation kinetics and heterogeneity exhibiting a striking resemblance to the observations from time-kill survival kinetics.

We next aimed to dissect the molecular mechanisms linking ppGpp to persistence during the GLU → OA + AMP shift. Previous studies have shown that ppGpp inhibits bacterial growth-related processes in a concentration-dependent and sequential manner, with different pathways requiring distinct ppGpp levels for inhibition (Balsalobre, 2011; Steinchen et al., 2020; Czech et al., 2022). Thus, the specific ppGpp-triggered growth-inhibition mechanism may be condition-dependent.

We first hypothesized that the elevated ppGpp level was triggered by carbon starvation. When growing E. coli in glucose media, FA degradation pathway is tightly repressed by catabolic repression. After the GLU → OA + AMP shift, cells require up to 6 h to fully express FA degradation enzymes (Zhang et al., 2024), creating an extended starvation period during the lag phase. This prolonged starvation does not occur when cells are shifted from glucose to other gluconeogenic carbons (Supplementary Figure S1). We hypothesized that shortening the starvation period might reduce ppGpp-triggered persister formation. To test this, cells were pre-grown in M9 OA media to fully induce FA degradation enzymes, then transferred to M9 glucose for 1 or 2 generations, corresponding to 1/2 or 1/4 dilution of FA degradation enzyme levels. We then performed the GLU → OA + AMP nutrient shift and time-kill assay (Figure 3A). The results revealed that cells grown in glucose for 1 generation retained sufficient FA degradation capacities to immediately metabolize OA, resulting in rapid killing of 99.99% of cells (Figure 3B). In contrast, cells grown in glucose for two generations exhibited a up to 3-h transient tolerance period following the shift due to insufficient initial FA degradation enzymes. FA catabolism subsequently resumed, leading to the death of 99.99% of cells. High persistence was only observed when the lag phase was long enough to induce carbon starvation, consistent with the ppGpp accumulation kinetics (Figures 2C–G).

We next hypothesized that FA starvation drives ppGpp accumulation through holo-ACP. After the GLU → OA + AMP shift, FA degradation is activated while FA biosynthesis is repressed by FadR and FabR (My et al., 2015). Repression of FA biosynthesis depletes acyl-ACP and holo-ACP pool size relatively increases. Unacylated holo-ACP then activates the ppGpp synthase activity of SpoT, while reducing its hydrolase activity (Battesti and Bouveret, 2006; Angelini et al., 2012; Fernández-Coll and Cashel, 2020), thereby driving ppGpp accumulation. Elevated ppGpp further amplifies the effect by inhibiting FA biosynthesis enzymes, thus forming a positive feedback loop. To test this, we overexpressed the E. coli acyl-ACP synthetase (encoded by aas), which ligates free FAs to holo-ACP, regenerating acyl-ACP (Jackowski et al., 1994). We performed GLU → OA + AMP shift and time-kill assays using the aas overexpression strain (Figure 3C, aas⁺), and quantified the ppGpp concentration with HPLC. Aas⁺ mutant strain showed a trend of reduced ppGpp accumulation at 3 h and, more notably, failed to sustain the elevated level seen in the WT at later time points, consistent with acyl-ACP synthetase restoring acyl-ACP as imported FAs accumulated overtime (Figure 3D) (Jackowski et al., 1994). This reduced ppGpp allowed growth to resume, leading to rapid antibiotic killing. Consistently, the time-kill assay revealed a substantially reduced persistence observed in the aas⁺ strain, with 95% cells killed after 24 h, compared to 44% in the wildtype strain. As a negative control, overexpression of green fluorescent protein (GFP) from the same promoter showed no effect on persistence (Figure 3C, gfp⁺), ruling out the possibility of vector and protein overexpression-induced differences. We further hypothesized that ppGpp promotes persister formation in wildtype E. coli by inhibiting PlsB, the sn-glycerol-3-phosphate acyltransferase that catalyzes the first committed step of phospholipid biosynthesis (Heath et al., 1994; Yao and Rock, 2013). To test so, we constructed a plsB overexpression mutant (plsB⁺) to alleviate ppGpp inhibition through increasing the enzyme copy number. We found that the plsB⁺ strain displayed reduced persistence, with 97% cells killed after 24 h in OA + AMP (Figure 3C). HPLC results showed that ppGpp kinetics in plsB+ (Figure 3E) largely matched with that in wildtype. Thus, excess PlsB likely enabled continued phospholipid synthesis despite ppGpp, accelerating recovery and antibiotic killing. Finally, overexpressing both aas and plsB (aas-plsB⁺) did not further decrease persistence beyond single overexpression (~97.6% killed, Figure 3C). This lack of synergy is likely due to the metabolic coupling: PlsB consumes acyl-ACP, the product of acyl-ACP synthetase, regenerating holo-ACP and offsetting the effect on ppGpp (Figure 3F).

At last, we sought to understand whether persisters formed after the GLU → OA + AMP shift remains metabolic activity or enter complete dormancy. To examine amino acid and protein synthesis, we performed GLU → OA + AMP shift using unlabeled (¹²C) glucose and fully ¹³C-labeled OA. Cells were harvested at 5, 9, and 24 h post-shift, with 5 h representing the end of the transient tolerance phase and the last two time points within the persistence window. Total proteins were extracted, hydrolyzed into amino acids, and analyzed via GC-MS (Figure 4B). At 5 h post-shift, ¹³C incorporation into proteinogenic amino acids was minimal, indicating very low flux into amino acid biosynthesis, consistent with the lack of FA catabolism due to insufficient β-oxidation enzymes. By 9 h, elevated ¹³C enrichment was detected across all measured amino acids. These results demonstrates that carbon derived from β-oxidation entered central pathways for amino acid and protein biosynthesis, confirming that persisters remained metabolically active rather than fully dormant during persistence (Blattman et al., 2024; Fung et al., 2025). The ¹³C incorporation involves β-oxidation to acetyl-CoA, the TCA cycle, gluconeogenesis, and the pentose phosphate pathway (e.g., valine and histidine, Figure 4A). Additional ¹³C incorporation continued beyond 9 h, but at a slower rate, reflecting emerging metabolic bottlenecks despite substrate availability.

To further examine metabolic activity, we quantified intracellular metabolite pools using LC–MS and compared them to exponentially growing glucose culture (Figure 4C). Substantial changes were observed across key metabolites, marked by the depletion of malate, 2-phosphoglycerate (2PG), and fructose 1,6-bisphosphate (FBP), alongside the accumulation of succinate and sedoheptulose 7-phosphate (S7P). This signature reflects a defined metabolic strategy to cope with stress and indicates potential bottlenecks for growth resumption.

The depletion of FBP and 2PG confirms switch away from glucose catabolism and relief of inhibition on gluconeogenesis, allowing reverse flux from TCA intermediate up to gluconeogenesis. The sharp reduction in malate, despite abundant succinate, suggests its rapid conversion to oxaloacetate or pyruvate. Meanwhile, the significantly elevated S7P abundance indicates an active pentose phosphate pathway (PPP) activity. This metabolic shift represents a key adaptive response, as ampicillin treatment induces oxidative stress as part of lethality (Dwyer et al., 2014; Zhang et al., 2024). To survive, persister cells must detoxify reactive oxygen species (ROS) and maintain essential redox homeostasis. Channeling carbon flux through the PPP serves as a critical survival strategy to meet this demand, as it generates the NADPH required for antioxidant defense and redox balance under antibiotic pressure. The accumulation of S7P also suggests a persister state, as ppGpp shuts down the expensive DNA and RNA synthesis, which reduces the demand for nucleotides building block, ribose 5-phosphate (R5P), and as a result, leads to the accumulation of S7P. This metabolic reprogramming trend is consistent with the ppGpp-dependent transcriptional response profile observed during amino acid starvation (Traxler et al., 2008). Together, these results show that persisters after the GLU → OA + AMP shift undergo active carbon flux reprogramming rather than metabolic shutdown. Activation of gluconeogenesis and the pentose phosphate pathway provide a survival strategy to balance redox homeostasis and counter oxidative stress, thereby enabling persistence and evasion of β-lactam killing (Kawai et al., 2019).