Introduction

Within the dynamic and often hostile environment of the host, nutrient scarcity presents a pervasive challenge for invading bacteria, promoting the evolution of survival strategies and shaping pathogen persistence.^1,2 Among the macronutrients essential for bacteria, phosphorus (P) serves as a critical element for life. It is a fundamental component of key biomolecules such as nucleic acids, phospholipids, and ATP, and plays crucial roles in the essential processes of genetic inheritance, energy metabolism, membrane integrity, and signal transduction.^3,4 The primary assimilable form of phosphorus for bacteria is the orthophosphate anion (PO₄^3-), commonly known as inorganic phosphate (Pi). However, Pi is typically found at very low concentrations (<4μΜ) in diverse environments, from oceans (0.1-3μΜ),^5,6 and soil (0.0105 to 10.5μΜ, <1μΜ near plant roots),^7,8 to critically host-associated niches including macrophages,⁹ Salmonella-containing vacuoles,¹⁰ and cystic fibrosis sputum.¹¹ This localized Pi scarcity is a direct consequence of host nutritional immunity and intense microbial competition.^12,13 Evidence for widespread bacterial adaptation to Pi limitation is abundant, with metagenomic and metatranscriptomic studies consistently revealing robust transcriptional upregulation of high-affinity phosphate acquisition (Pho) regulon genes across various chronic infections and dysbiosis conditions.^11,14,15 Bacteria cope with these pervasive conditions via the conserved PhoBR two-component system, which activates genes for Pi acquisition, transport, and storage when extracellular Pi falls below 4 μM.^3,16,17

Beyond direct Pi assimilation, Pi depletion triggers broader adaptive responses crucial for pathogen survival, including enhanced virulence and, critically, antimicrobial resistance (AMR).^18–20 For example, Pi limitation upregulates virulence factors in Vibrio cholerae, EHEC, and Pseudomonas aeruginosa, and conversely, Pi supplementation can reduce infection severity.^21–24 More importantly, Pi depletion is increasingly recognized as a potent environmental cue driving AMR across diverse bacterial taxa, particularly against cationic antibiotics like polymyxins. This is evidenced by increased vancomycin resistance in Streptomyces coelicolor,²⁵ and enhanced polymyxin B/E resistance in Pseudomonas fluorescens, P. aeruginosa, and Acinetobacter baumannii under Pi-limited conditions.^11,26–28 Given that Pi limitation is a prevalent and dynamic stressor in host niches, understanding its role in promoting AMR is vital for developing effective counter-strategies.

Mechanistically, Pi depletion-induced AMR is often linked to the remodelling of bacterial membranes, where phosphorus-containing phospholipids are substituted with diverse non-phosphorous lipids. This strategy spares intracellular phosphate and often modifies membrane charge, which is critical for resistance to cationic antimicrobials. Examples include the accumulation of ornithine lipids (OLs) in Vibrio cholerae,²⁹ diacylglyceryl-N,N,N-trimethylhomoserine (DGTS) in Rhizobium meliloti,³⁰ and Cryptococcus neoformans,³¹ glycolipids in Agrobacterium tumefaciens,³² and extensive lipidome remodelling in Mycobacterium tuberculosis.³³ Specifically for polymyxin resistance, P. aeruginosa synthesizes neutral monoglucosyl diacylglycerol (MGDG) and glucuronic acid diacylglycerol (GADG),¹¹ while A. baumannii produces positively charged ornithine- and lysine-containing aminolipids (OL and LL) under Pi limitation.²⁷ These diverse lipid modifications effectively reduce the net negative charge of the bacterial membrane, decreasing polymyxin binding and improving resistance.

However, these established lipid-remodelling strategies are not universally conserved. Critically, genes for OL biosynthesis are absent in most Enterobacteriaceae, a family of paramount clinical importance, and MGDG/GADG synthesis is not a predominant mechanism for polymyxin resistance in this group.^34–36 Given the widespread prevalence of Pi limitation in host environments and the escalating threat of polymyxin resistance in Enterobacteriaceae, a fundamental question with significant clinical implications arises: what alternative, yet unidentified, mechanisms underpin Pi depletion-induced polymyxin resistance in this critical bacterial family, especially in the absence of the established OL and MGDG/GADG pathways?

In this study, we address this critical gap by reporting that Pi limitation also induces polymyxin resistance in Enterobacteriaceae, but through a distinct mechanism. Using E. coli K-12 MG1655 as a model, we demonstrate that this process involves the upregulation of conserved Ugd and ArnBCADTEF proteins, which catalyse the addition of the positively charged 4-amino-4-deoxy-L-arabinose (L-Ara4N) to lipid A. This adaptive lipid A modification is driven by a novel Pi-Mg-Fe-PmrAB regulatory circuit: loss of membrane-stabilizing magnesium (Mg²⁺) and compensatory iron (Fe³⁺) mobilization to the cell membrane activate the PmrAB two-component system, leading to upregulation of the ugd-arn operon. Crucially, interfering with this circuit, via Mg²⁺ supplementation or iron chelation, effectively reverses resistance in representative Enterobacteriaceae species, but not in non-enteric species like P. aeruginosa. Phylogenetic analyses further reveal distinct clades for E. coli PmrB (EcoPmrB) and P. aeruginosa PmrB (PaePmrB), with PaePmrB lacking the conserved “ExxE” motif required for Fe³⁺ sensing. Our findings uncover a conserved, targetable stress-responsive circuit underlying Pi depletion-induced polymyxin resistance within Enterobacteriaceae, offering new avenues for managing stress-induced antibiotic resistance.

Results

Pi depletion induces polymyxin resistance in Enterobacteriaceae

To investigate whether Pi limitation induces antibiotic resistance in Enterobacteriaceae, we assessed the susceptibility of two representative species, E. coli MG1655 and Salmonella enterica serovar Typhimurium ATCC14028. Bacteria were initially cultured in standard MOPS synthetic minimal medium containing replete Pi (1.32 mM). Subsequently, cultures were transferred to either Pi-replete (Pi+, 1.32 mM) or acute Pi-depleted (Pi-, 0 mM) conditions for 3 hours, a protocol adapted from R G Groat et. al.³⁷ Following this, bacterial survival was determined via 3-hour polymyxin B killing assays (Figure S1). Polymyxin B concentrations for these assays were established based on the minimum inhibitory concentrations (MICs) of strains in MOPS minimal medium (Table S1) and a preliminary dose-dependent killing assay (Figure S2). Results showed that Pi depletion enhanced the survival of both species against polymyxin B (Figure 1A). Similar Pi depletion-induced polymyxin resistance was also observed in E. coli O157 and four uropathogenic E. coli strains isolated from urinary tract infection patients in the Queen Mary Hospital, Hong Kong (Figure 1B). Consistent with previous reports, P. aeruginosa PAO1 subjected to Pi depletion also display enhanced resistance against polymyxin B (Figure 1B). These results suggest that Pi depletion-induced polymyxin resistance is a widespread phenomenon among Gram-negative bacteria.

Figure 1.

Pi depletion induces 4-amino-4-deoxy-L-arabinose (L-Ara-4N) modification of lipid A in Enterobacteriaceae

Since these species lack the MGDG/GADG and OL pathways described in P. aeruginosa and A. baumannii, we employed E. coli K-12 MG1655 as a model to comprehensively investigate the regulatory mechanisms underlying this stress-induced polymyxin resistance. We performed quantitative proteomic analysis on E. coli MG1655 cells subjected to the identical growth and treatment conditions as for the polymyxin B challenging assay (Figure S1). Among the 2990 proteins detected which cover 69% of the 4358 proteins annotated in the UniProt E. coli MG1655 database, abundance of 225 proteins was found to be altered following acute Pi depletion, with 142 proteins upregulated and 83 downregulated (Figure S3) (Table S2). In addition to proteins belonging to the Pho regulon, such as PhoB, PstB, PhoU, and PhnB (Figure 2A) in the COG functional group of transportation and metabolism of inorganic ions (13% of upregulated proteins) (Figure S3), the most abundant functional groups among the significantly upregulated proteins include those involved in cell wall/membrane/envelope biogenesis (13%) and energy production and conversion (10%) (Figure S3). Conversely, proteins involved in translation, ribosomal structure and biogenesis (33%) and amino acid transport and metabolism (17%) were significantly downregulated (Figure S3).

Figure 2.

Specifically, among the most upregulated proteins (log₂FC>10) induced by Pi depletion (Table S2), two proteins encoded in the arnBCADTEF operon: ArnB and ArnC, were identified (Figures 2A and 2B). Additionally, two other proteins in the Arn pathway, ArnA and Ugd, were also up-regulated with more than five-fold in response to Pi depletion treatment (Figures 2A and 2B). This finding is particularly intriguing, because Ugd and Arn proteins are known to catalyze the biosynthesis of 4-amino-4-deoxy-L-arabinose (L-Ara4N) and its modification of lipid A (Figure 2B), which confers resistance to the cationic antibiotic polymyxins due to simultaneous neutralization of a negative charge and introduction of a positive charge to lipid A.³⁸ To verify the upregulation of these proteins, we first examined the mRNA level of ugd and arnB genes by RT-qPCR and found their levels were 16- and 27-fold higher, respectively, under Pi- than under Pi+ conditions (Figure 2C). Next, we constructed ArnB-3×FLAG in MG1655 chromosome and observed a significant production of the ArnB protein under Pi- conditions, while its level was undetectable under Pi+ conditions (Figure 2D). We also constructed promoter-lacZ fusions and observed more than 10-fold higher promoter activities for both ugd and arn genes under Pi- conditions than Pi+ conditions (Figure 2E), confirming the upregulation of the Ugd-Arn system under Pi depletion.

We then extracted lipid A molecules from E. coli MG1655 cells cultured under identical conditions as in the proteomic analysis, and analyzed the extract using ESI-MS. Native lipid A, a hexacetylated glucosamine disaccharide with two phosphate groups at position 1 and 4’ (m/z 1796.98), was detected in E. coli cells cultured under Pi sufficient conditions (Figure 2F). In constant, the predominant lipid A species in E. coli cells subjected to Pi depletion were found to be modified with L-Ara4N (m/z 1928.67) (Figure 2F). Next, we deleted arnT, arnB, and ugd and found that deletion of these genes abolished Pi depletion-induced polymyxin B resistance in E. coli (Figure 2G).

The induction of L-Ara4N modified lipid A, along with decrease in the unmodified form, was also observed in E. coli O157 (Figure S4A), a representative UTI strain UTI-3 (Figure S4B), and S. Typhimurium ATCC14028 (Figure S4C), each exhibiting varying ratios. These fundings indicate that L-Ara4N modification of lipid A is widespread in Enterobacteriaceae and correlates with increased polymyxin B resistance. We also analyzed lipid A species in P. aeruginosa PAO1 and found that instead of the L-Ara4N modification, a 2-hydroxymyristate modification of lipid A, which is catalyzed by the aspartyl beta-hydroxylase LpxO, was detected in Pi-depletion treated cells (Figure S4D). Altogether, these results indicated that unlike the strategy observed in PAO1 where Pi stress induces resistance by replacing the negatively charged phospholipids with MGDG/GADG, in E. coli, resistance is mediated by lipid A modification with L-Ara4N catalyzed by Ugd and Arn proteins.

To delineate the sensitivity of the Ugd-Arn pathway to varying Pi levels, we measured the promoter activities of P_ugd-lacZ and P_arn-lacZ across a gradient of phosphate concentrations. We observed that the activities of both P_ugd-lacZ and P_arn-lacZ remained robustly activated, at levels comparable to complete Pi depletion (0 mM), even when 0.040 mM or 0.132 mM Pi was supplemented to the medium (Figure S5). When Pi supplementation reached 0.4 mM, their upregulation was attenuated by 60%. These findings collectively suggest that moderate Pi limitation (below 0.4mM) is effective in sustaining the activation of ugd-arn genes (Figure S5).

The PmrAB two component system is activated to up-regulate the arn operon during Pi depletion

To elucidate the regulatory mechanisms underlying ugd-arn upregulation during Pi limitation, we first investigated the involvement of the global phosphate regulators PhoB and PhoR. We found that deleting either PhoB or PhoR did not affect the enhanced polymyxin B resistance or the upregulation of the arn operon observed in Pi-depleted E. coli (Figures 3A and S6). This observation suggested that other regulatory systems were primarily responsible for upregulating the arn gene expression under Pi depletion.

Figure 3.

The arn operon and ugd are known targets of the PmrAB two-component system (TCS) in E. coli. ³⁹ (Note the E. coli homolog of this system is BasSR; we adopt the PmrAB nomenclature in this study to reflect its direct role in polymyxin resistance and its wider usage in the literature.)⁴⁰ To determine if this regulation also extended to Pi depletion, we constructed ΔpmrB and ΔpmrA mutants. Both the enhanced polymyxin B resistance and the upregulation of the arn operon under Pi depletion were abolished in these mutants (Figures 3A and 3B), indicating a critical role for PmrAB in arn regulation and polymyxin B resistance during Pi depletion. The promoter activity of ugd remained unaffected in ΔpmrB and ΔpmrA mutants (Figure S7), suggesting a distinct regulatory input for ugd. Given that the PhoPQ TCS can indirectly activate the PmrAB regulon in S. Typhimurium via PmrD (though this indirect regulation appears absent in E. coli), we also examined the contribution of PhoPQ.⁴¹ Deletion of phoP or phoQ did not alter the enhanced polymyxin B resistance or arn operon upregulation under Pi depletion (Figures 3A and S6). These findings suggest that the arn system is activated by PmrAB in response to an as-yet-unidentified secondary signal induced by Pi depletion, rather than by direct Pi depletion sensing or through the PhoPQ system.

Given the distinct regulation we observed for ugd promoter activity, we further investigated its response to the PhoBR system. We measured ugd promoter activity in ΔphoB, ΔphoR, ΔphoP, and ΔphoQ mutants and found P_ugd-lacZ activity decreased by approximately 30% in both ΔphoB and ΔphoR mutants, but not in ΔphoP and ΔphoQ mutants (Figure S7). This indicates that ugd is regulated by the PhoBR system, potentially alongside other pathways, during Pi depletion.

Given the central role of the PmrAB system in mediating Pi depletion-induced polymyxin resistance, we next investigated whether eptA, another PmrAB-regulated gene, contributes to this phenotype. We constructed a ΔeptA mutant and assessed bacterial survival in a polymyxin B killing assay. Deletion of eptA had no impact on the enhanced polymyxin B resistance (Figure S8), indicating that eptA is not required for the resistance phenotype induced by Pi depletion. This observation was further supported by ESI-MS analysis (Figure 2F), which did not detect phosphoethanolamine (pEtN) modification of lipid A, a hallmark of EptA activity, in Pi-depleted cells. Collectively, these results strongly suggest that the activation of the Arn system via PmrAB is the primary mechanism responsible for Pi depletion-induced polymyxin resistance.

To assess whether PmrAB activation occurs under Pi depletion, we examined the phosphorylation status of the response regulator PmrA using Phos-tag SDS-PAGE, which reliably detects phosphorylated PmrA in cells treated with known activator Fe³⁺.⁴² Results showed that phosphorylated PmrA was detected in cells subjected to Pi depletion (Figure 3C), demonstrating that PmrAB TCS is indeed activated under Pi depletion condition. To further confirm this in vivo, we performed a FRET-based whole-cell transcription factor-promoter binding assay.⁴³ We incorporated a bright fluorescent unnatural amino acid, L-(7-hydroxycoumarin-4-yl) ethylglycine (CouA)⁴⁴ at position G137 within the DNA-binding domain of PmrA through genetic encoding, allowing it to serve as the FRET donor (Figure 3D).⁴³ The nucleic acid dye SYTO9 was utilized to stain promoter DNA of PmrA regulon (Figure 3D),⁴³ acting as the FRET acceptor. The whole-cell FRET assay revealed that, similar to the treatment with excessive FeSO₄ (500μM),⁴³ a FRET signal indicative of PmrA binding to its target promoters was observed in cells subjected to Pi depletion without excessive FeSO₄ (Figure 3E). These results confirmed that Pi depletion induced activation of the PmrAB signal transduction system in E. coli.

Pi depletion drives Fe³⁺ elevation in E. coli cells to activate the PmrAB TCS

PmrAB has been previously reported to be activated by ferric ion (Fe³⁺), zinc ions (Zn²⁺), aluminum ions (Al³⁺), or acidic pH.^45,46 Since no Al³⁺ was added in the culture, and the pH remained largely unchanged (7.2-7.0) across the Pi depletion treatments (Figure S9), we focused on the two metal ions and examined cellular metal profiles including iron (Fe), zinc (Zn), magnesium (Mg), copper (Cu), and calcium (Ca). ICP-MS analysis revealed a significant, greater than 5-fold increase in cellular Fe content under Pi depletion, compared to Pi-sufficient conditions (Figure 4A). Conversely, cellular Mg levels decreased, while Cu, Zn, and Ca levels remained largely unchanged (Figure 4A). Corroborating this cellular Fe accumulation, the Fe content in the spent medium decreased after 3 hours of E. coli incubation under Pi depletion (Figure 4B), indicating active Fe mobilization from the medium into E. coli cells. These findings strongly implicated Fe³⁺ as the activating signal for PmrAB under Pi depletion.

Figure 4.

Given that S. Typhimurium PmrB (StPmrB) senses Fe³⁺ through conserved glutamic acid residues (E36, E39, E61, E64) in its periplasmic domain,⁴² we hypothesized a similar mechanism for E. coli PmrB (EcoPmrB). To test this, we generated chromosomal pmrBE36A and pmrBE39A mutations in E. coli MG1655 (Figure 4C). Consistent with their proposed role in Fe³⁺ sensing, these mutations abolished the Parn-lacZ reporter response to exogenous Fe³⁺ (100 µM) (Figure S10). Crucially, both mutations also abolished arn operon upregulation under Pi depletion (Figure 4D), demonstrating that Pi depletion-induced PmrAB activation requires the Fe³⁺ sensing capability EcoPmrB.

To confirm the necessity of Fe mobilization, we observed that arn operon upregulation was abolished when cells were grown in Fe-free, Pi-depleted medium (Figure 4E). Although Pi depletion significantly impaired bacterial growth (doubling time increased from ∼1 hr to ∼6 hrs), co-depletion of Fe did not further exacerbate this effect, confirming that the observed PmrAB activation specifically responds to Pi-related Fe mobilization rather than general growth inhibition (data not shown). Investigating the sensitivity of this response, we found that reintroduction of Fe to Pi-depleted cells restored Parn-lacZ activity in a dose-dependent manner (Figure 4F), further confirming the role Fe mobilization in activating PmrAB and arn genes during Pi depletion. Importantly, arn activation in Pi-starved cells occurs at significantly lower exogenous Fe concentrations (∼10-fold less) than in Pi-sufficient conditions (Figure 4F), suggesting an enhanced cellular capacity for Fe mobilization or sensing under Pi depletion.

Pi depletion triggers Mg²⁺ loss, leading to Fe³⁺ mobilization to membrane and PmrAB activation

To elucidate the mechanism of Fe mobilization during Pi depletion, we first investigated the role of canonical iron uptake pathways. Surprisingly, deletion of genes encoding outer membrane iron-chelator complex receptors and enterobactin synthetase (fecA, fepA, fiu, cirA, entE), individually or as a penta-deletion mutant (Δ5Fe), did not abolish Parn-lacZ induction during Pi depletion (Figure S11A). ICP-MS analysis further confirmed that cell-associated Fe still increased by over 5-fold in Δ5Fe cells during Pi depletion (Figure S11B). These results indicated that Fe-mediated PmrAB activation during Pi depletion was independent of enhanced intracellular Fe uptake and suggested an alternative mechanism, likely involving Fe association with the cell surface, consistent with the known ability of PmrAB to sense periplasmic and surface-bound Fe³⁺.⁴²

The outer leaflet of the Gram-negative outer membrane (OM) primarily consists of lipid A, whose negatively charged phosphate groups are typically neutralized by cell-associated Mg²⁺ and Ca²⁺ to maintain membrane stability.⁴⁷ Our ICP-MS analysis revealed a 44% reduction in cellular Mg content, while Ca²⁺ levels remained unchanged, in E. coli cells following 3 hours of Pi depletion treatment (Figure 4A). Monitoring the dynamics revealed that cellular Mg progressively decreased upon transferring E. coli cells to Pi-depleted medium, continuing over a 5-hour period (Figure 5A). This Mg²⁺ loss was independent of Fe, as a similar degree of loss occurred in Fe-free, Pi-depleted medium (Figure S12). Conversely, cellular Fe increased, peaking after two hours of Pi depletion before slightly declining (Figure 5A). The temporal correlation between Fe accumulation and Parn-lacZ activity (Figure 5B) suggested active Fe mobilization to the cell surface during Pi depletion, potentially to compensate for Mg²⁺ loss and stabilize the membrane. The subsequent activation of the PmrAB-Arn system that remodel lipid A by decreasing negative charges, might thus represent an adaptive response to physiological changes, such as OM instability, caused by Mg²⁺ dissociation. Supporting this, deletion of pmrA led to an even greater and sustained increase in Fe mobilization during Pi depletion (Figure 5C), which could be toxic to E. coli.⁴²

Figure 5.

To directly assess Fe association with the cell surface, we washed cell pellets with deferoxamine (DFOM)-containing buffer, a membrane-impermeable Fe³⁺-specific chelator.⁴⁸ ICP-MS analysis showed that washing with DFOM-supplemented buffer significantly reduced Fe content in cells treated with Pi depletion but had no significant effect on cells harvested from Pi-sufficient conditions (Figure 5D). Quantification revealed that Pi-starved cells exhibited a two-fold greater loss of Fe upon DFOM washing compared to Pi-sufficient conditions (Figure 5E), demonstrating that Pi depletion promoted Fe mobilization to cell surfaces, where it remained susceptible to removal by a membrane-impermeable chelator.

To uncover the driving force behind this enhanced Fe mobilization, we supplemented the culture medium with excess Mg (10 mM) to counteract Mg loss. This supplementation significantly reduced Fe mobilization to MG1655 cells during Pi depletion (Figure 6A); correspondingly, P_arn-lacZ activity declined as Mg concentration increased in the Pi-depleted medium (Figure 6B). These results strongly suggest that Fe mobilization to the cell surface during Pi depletion is dependent on Mg²⁺ loss. Further, monitoring Mg²⁺ dynamics in the spent medium revealed a continuous increase in Mg levels throughout the Pi depletion treatment (Figure 6C), confirming active Mg²⁺ dissociation from the cells. Similar Mg²⁺ release was observed even during combined Mg and Pi depletion (Figure S13). These results suggested that while Mg²⁺ is essential for bacterial growth, an imbalance with phosphorus during Pi depletion appears to trigger active Mg²⁺ dissociation from the cells. A comparable 45% Mg²⁺ loss was also observed in P. aeruginosa PAO1 under similar conditions (Figure S14). However, Fe mobilization was less pronounced (2.4-fold vs. 5.4-fold in E. coli), reflecting species-specific adaptation strategies.¹¹

Figure 6.

Given that membrane-associated Mg²⁺ primarily neutralizes the negative charges of lipid A to stabilize the OM, we hypothesized that Mg²⁺ loss during Pi depletion alters OM properties, and Fe mobilization may serve to stabilize the OM. To test this, we employed an OM-sensitive probe.⁴⁹ The probe reports membrane surface properties through two orthogonal readouts: an overall fluorescence increase, which reflects probe disassembly upon binding to the anionic bacterial surface, and the pyrene excimer-to-monomer emission ratio (I₄₈₂/I₃₇₅), which captures differences in local surface interactions. The 2D sensor output revealed distinct signals between cells grown under Pi-sufficient and Pi-depleted conditions (Figure 6D), indicating altered OM characteristics. Notably, cells grown in Fe-free, Pi-depleted medium exhibited even greater shifts (Figure 6D), suggesting that Fe mobilization and subsequent activation of the PmrAB-Arn pathway served as an adaptation strategy to maintain OM integrity during Pi stress. This strategy likely prevents uncontrolled Fe accumulation and toxicity,⁴² ultimately resulting in stress-induced polymyxin resistance. A schematic model illustrating this process is presented (Figure 7).

Figure 7.

Disrupting the Mg-Fe-PmrAB-Arn circuit counters Pi depletion-induced polymyxin resistance in Enterobacteriaceae

Our findings above identified a Mg-Fe-PmrAB-Arn circuit underpinning the polymyxin resistance induced by stress adaptation to Pi depletion. To evaluate whether interference with this pathway could mitigate the stress-induced polymyxin resistance, we supplemented Pi depletion medium with excess Mg²⁺ (10mM) or applied a membrane impermeable chelator DFOM (10μM). Both approaches effectively suppressed Pi depletion-induced polymyxin resistance in E. coli and S. Typhimurium (Figure 8A). In contrast, neither intervention reversed polymyxin resistance in P. aeruginosa despite various combinations of Mg²⁺ supplementation and Fe chelation (Figure S15). This highlights species-specific adaptative strategies and aligns with several observations from our study as well as several previous research.^11,50 Specifically, in P. aeruginosa, despite a 45% Mg loss during Pi depletion, Fe levels increased only two-fold compared to over five-fold in E. coli, consistent with its reliance on MGDG/GADG lipids synthesis regulated by the PhoBR system (Figure 8B).¹¹ Additionally, P. aeruginosa exhibits a distinct lipid A modification, 2-hydroxy myristate (Figure S4D), that does not confer polymyxin resistance.⁵¹ Phylogenetic analysis of 250 PmrB homologs revealed that EcoPmrB and PaePmrB cluster in separate clades, with PaePmrB lacking the conserved “ExxE” motif essential for Fe³⁺ sensing (Figure 8C). Together, these findings underscore the diverse stress response strategies employed by different classes of Gram-negative bacteria and highlight the importance of understanding species-specific regulatory mechanisms for developing targeted interventions against stress-induced antibiotic resistance.

Figure 8.

Discussion

Bacterial adaptation to environmental stressors is crucial for survival and pathogenesis. Phosphate (Pi) scarcity, a pervasive condition in diverse environments including host-associated niches such as macrophage phagosomes, salmonella-containing vacuoles, and cystic fibrosis sputum, is a potent cue driving antimicrobial resistance (AMR) against cationic antibiotics such as polymyxins.^9–11 Adaptive mechanisms previously characterized for Pi depletion-induced AMR often involve remodeling bacterial membranes by substituting phosphorus-containing phospholipids with Pi-free, neutral, or positively-charged lipids.^11,26–28 However, the specific mechanisms in clinically important Enterobacteriaceae, which largely lack these canonical phospholipid remodeling pathways, have remained largely unknown. In this study, we unveil a conserved, yet distinct, polymyxin B resistance mechanism in Enterobacteriaceae: Pi limitation triggers a cascade of cationic imbalance and Fe³⁺ signaling, leading to L-Ara4N modification of lipid A and outer membrane remodeling, a process distinct from phospholipid-based resistance pathways (Figure 7). This intricate Mg-Fe-PmrAB regulatory circuit represents a previously unappreciated ecophysiological vulnerability, offering a novel strategy to mitigate stress-induced polymyxin resistance.

Maintaining a balanced P/Mg ratio is paramount for bacterial cellular physiology, with ATP and rRNA serving as primary cytoplasmic reservoirs for both Pi and Mg²⁺.^52–55 Previous work by Bruna et al. demonstrated that bacteria suppress Pi uptake under Mg²⁺ limitation to preserve cellular function.⁵³ Herein, our study extends this understanding by showing that Pi scarcity itself actively triggers Mg²⁺ release from E. coli cells. This observation underscores the tightly coupled nature of Pi and Mg²⁺ homeostasis. Our comprehensive proteomic analysis further supports this, revealing that proteins involved in translation, ribosomal structure, and biogenesis were among the most significantly downregulated groups during Pi depletion (Figures 1A and S3). Concurrently, we observed a steady decline in cellular ATP levels under Pi-limited conditions (Figure S16), suggesting a profound arrest in ribosome biogenesis and protein synthesis that likely contributes to the observed Mg²⁺ release and cellular loss.

In Enterobacteriaceae, Mg²⁺ homeostasis is primarily governed by the PhoPQ two-component system, which is well-characterized in S. Typhimurium to be activated under low extracellular Mg²⁺ conditions.^56,57 Given the observed Mg²⁺ release during Pi-depletion, we investigated the involvement of this classical Mg²⁺-sensing PhoPQ system. Our results showed that despite cellular Mg²⁺ loss during Pi depletion, the transcription of key PhoPQ regulon genes, such as phoQ and mgtA, remained unchanged (Figure S17). This indicates that while cellular Mg²⁺ levels are perturbed, they do not drop sufficiently to activate the canonical PhoPQ response. Instead, our data compellingly demonstrate that under acute Pi depletion, Enterobacteriaceae preferentially engage an alternative, Fe³⁺-mediated membrane remodeling pathway. This pathway activates the arn operon and subsequent L-Ara4N modification independently of the classical Mg²⁺-sensing PhoPQ system, highlighting a distinct adaptive strategy.

This Fe³⁺-mediated lipid A modification strategy in Enterobacteriaceae offers a stark contrast to Pi depletion-induced polymyxin resistance mechanisms described in other Gram-negative bacteria, notably P. aeruginosa. In P. aeruginosa, Pi depletion primarily activates the PhoBR regulon, orchestrating the biosynthesis of phosphorus-free glycolipids (MGDG and GADG) to replace negatively charged phospholipids and maintain membrane integrity (Figure 8B).¹¹ While P. aeruginosa possesses a PmrAB homolog, its PmrB sensor kinase lacks the critical Fe³⁺-sensing “ExxE” motifs found in Enterobacteriaceae PmrB, precluding direct Fe³⁺ response. Instead, PmrAB activation in P. aeruginosa is typically coupled to the PhoPQ system, which senses low extracellular Mg²⁺. Thus, while P. aeruginosa employs PhoBR-regulated phospholipid remodeling as its primary Pi depletion response, Enterobacteriaceae, largely lacking these glycolipid pathways, have evolved to utilize a secondary, Fe³⁺-sensing mechanism. Here, the PmrAB system, via its specific Fe³⁺-sensing “ExxE” motifs in PmrB, orchestrates membrane remodeling through L-Ara4N modification of lipid A as a key physiological adaptation to Pi depletion.⁵⁰ While P. aeruginosa also exhibits lipid A modification (e.g., 2-hydroxymyristate by LpxO) during Pi depletion, this modification does not contribute to polymyxin resistance and is regulated independently of PhoBR, PhoPQ, or PmrAB, highlighting distinct physiological roles.⁵¹

The biosynthesis of L-Ara4N and its transfer to lipid A are mediated by Ugd and ArnABCDEFT proteins, encoded by the ugd gene and the arn operon, respectively^58–61. Our data reveal a sophisticated regulatory separation: during Pi depletion, arn transcription is tightly controlled by the PmrAB system, while ugd expression is not PmrAB-dependent and exhibits partial regulation by the PhoBR system (Figure S6). This bifunctional control is not unexpected, as ugd encodes UDP-glucose dehydrogenase, an enzyme with dual roles in L-Ara4N biosynthesis and in the production of exopolysaccharides like colanic acid.^62,63 Therefore, while PmrAB activation of arn is absolutely critical for L-Ara4N modification of lipid A during Pi limitation, ugd regulation reflects its broader involvement in diverse cellular metabolic and stress responses.

While preparing this manuscript, a recent study by Baijal et al. (2024) reported that the ppk gene modulates arn operon upregulation under combined nutrient limitations (transfer from rich LB to low-Pi, no-amino-acid MOPS medium) in E. coli.⁶⁴ Inspired by this, we investigated the role of ppk in polymyxin resistance specifically under Pi depletion, our primary focus. We found that ppk deletion did not affect arn or ugd promoter activity under sole Pi-depleted conditions (Figure S18), indicating its dispensability when Pi is the only limiting factor. To reconcile these findings, we replicated the Baijal et al. experimental conditions, confirming ppk-dependent arn upregulation under combined Pi and amino acid scarcity (Figure S19A). Our subsequent experiments, systematically disentangling these stresses, revealed that while both Pi and amino acid depletion induce arn upregulation, Pi depletion is the dominant driver, and ppk-dependent regulation emerges only under combined Pi and amino acid scarcity (Figure S19B). These results clarify that while ppk influences arn expression under specific multifactorial stresses, our identified PmrAB-arn pathway is independently and dominantly activated by Pi depletion alone, providing precise mechanistic insight into this fundamental environmental stressor.

In both natural habitats and host-associated niches, Pi scarcity imposes dual challenges: nutrient acquisition and defense against antimicrobial threats. The Pi-Mg-Fe-PmrAB regulatory circuit we have uncovered likely represents a sophisticated adaptive strategy for Enterobacteriaceae, facilitating their survival in highly heterogeneous environments with fluctuating Pi levels, such as during host colonization, transmission, or environmental persistence.⁶⁵ This PmrAB-mediated membrane remodeling not only confers resistance to polymyxins but may also enable bacteria to preserve membrane integrity and withstand other cationic antimicrobial peptides (CAMPs) produced by competing microbes or host defenses.

In summary, our findings elucidate a conserved yet distinct regulatory circuit in Enterobacteriaceae that intricately links Pi and Mg²⁺ homeostasis to Fe³⁺-mediated lipid A modification, driving stress-induced antibiotic resistance. This mechanism is notably distinct from those characterized in P. aeruginosa or A. baumannii. Crucially, we demonstrate that targeting this circuit effectively mitigates stress-induced resistance. Given that polymyxins remain indispensable as last-line antibiotics for treating multidrug-resistant Gram-negative bacterial infections, our study not only offers critical insights into the fundamental molecular mechanisms underlying stress-induced resistance but also highlights a promising avenue for informing novel resistance management strategies.

Materials and methods

Bacterial strains and growth conditions

Strains, plasmids, and primers used in this study are listed in the S2 and S3 Tables. Bacteria were cultured in lysogeny broth (LB) broth/plates or MOPS minimal medium. MOPS medium was freshly prepared from 1× MOPS stock by supplementing 0.1 % (w/v) glucose and 1.32 mM K₂HPO_4. Acute Pi depletion was applied by transferring bacteria culture into the non-Pi MOPS medium. 1× MOPS consists of 40 mM MOPS, 4 mM Tricine, 10 μM FeSO₄·7H₂O, 9.5 mM NH₄Cl, 0.276 mM K₂SO₄, 0.5 μM CaCl₂·2H₂O, 525 μM MgCl₂, 50 mM NaCl, 0.003 μM (NH₄)₆Mo₇O₂₄, 0.4 μM H₃BO₃, 0.03 μM CoCl₂, 0.01 μM CuSO₄, 0.08 μM MnCl₂, and 0.01 μM ZnSO₄. Antibiotics applied to select and maintain the construct were chloramphenicol (25 μg/ml), kanamycin (20 μg/ml) and ampicillin (100 μg/ml).

Lipid A extraction and ESI-MS analysis

Lipid A was extracted from designated E. coli cells using Bligh-Dyer method as described with minor modifications.⁶⁶ E. coli MG1655 cultured in MOPS medium to OD₆₀₀ of 0.5 was harvested as cells under Pi sufficient condition. The same volume of cells was washed with Pi depletion medium for 3 times and resuspended in Pi depletion medium and shake for 3 hrs at 37°C as cells subjected to Pi depletion. 200 ml of each type of cell pellets with three biological replicates were harvested by centrifugation at 4500 rpm for 30 min and washed with H₂O. The cell pellets were suspended in 76 ml of a single-phase Bligh-Dyer mixture (chloroform: methanol: water=1:2:0.8; v/v/v) and stirred at room temperature for 1 h. Insoluble debris was collected by centrifugation at 4500 rpm for 30 min, washed with 60 ml single-phase Bligh-Dyer mixture, and suspended in 27 ml of 12.5 mM sodium acetate (pH 4.5). The mixture was heated at 100°C for 30 min and cooled to room temperature. The suspension was mixed with 30 ml chloroform and 30 ml methanol. Following vigorously shaken for extraction of lipid A and centrifugation for 30 min, the lower phase containing lipid A was collected and dried with a rotary evaporator.

All the mass spectra of lipid A samples were acquired using Waters SYNAPT quadrupole time-of-flight mass spectrometer (MS) equipped with an electrospray ionization source⁶⁷. Lipid A samples were dissolved in a mixture of chloroform and methanol (4 : 1; v/v) and subjected to ESI/MS in the negative ion mode. Data acquisition and analysis were performed using MASS LYNX V4.1 software, Waters, Milford, MA, USA.

ICP-MS and DFOM treatment

Total cellular metal content was quantified using inductively coupled plasma mass spectrometry (ICP-MS) following previous description with minor modifications⁶⁸. E. coli cells equivalent to OD₆₀₀ as 5 were harvested by centrifugation at 4500 rpm for 15 mins and washed three times with ice-cold PBS buffer. Cell pellets were subjected to acid digestion with 100 μl 65.0 % HNO₃ (Fluka) at 65°C for 16hrs. Samples were diluted to a final volume of 15 ml with 1% HNO₃ and subjected to ICP-MS detection on Agilent 7900 system (Agilent Technologies, CA). Samples were further diluted if the signals exceeded the liner range of standard curve. The standard curves of tested metals were generated from multi-element calibration standard-2A for ICP-MS (Agilent). Metal contents in bacteria were calculated based on the standard curve and results were normalized to original cell number.

Deferoxamine mesylate salt (sigma) were applied to examine the degree of iron association to E. coli cell surface. E. coli cells equivalent to OD₆₀₀ as 5 were harvested by centrifugation at 4500 rpm for 15 mins. Cell pellets were resuspended in HEPES buffered saline with and without 50μM DFOM and incubated in 37°C for 30 mins with 220rpm agitation. Total cellular metal content of these treated cells was quantified using inductively coupled plasma mass spectrometry (ICP-MS) as described above.

OM feature monitoring using PI-BactD

OM feature of E. coli was monitored by an OM-sensitive probe PI-BactD following previous description with minor modifications.⁴⁹ E. coli cells grown under designated conditions were harvested by centrifugation at 13000 rpm for 2 mins and washed with PBS buffer (10 mM, pH 7.2) twice. Cell pellets were resuspended in PBS buffer to an OD₆₀₀ of 0.5. Fluorescence probe PI-BactD (gift from prof. Xu Z, Dalian) was added into 1mL cell suspension to a final concentration of 2 mM and incubated at room temperature for 1 min. Output signals of two channels: fluorescence increase (ΔS/S₀) and fluorescence ratio metric changes (I₄₈₂/I₃₇₅) were recorded to plot the 2D graph (Thermo Scientific Varioskan® Flash spectral scanning multimode reader). Six replicates were conducted for each strain and treatment.

Construction of Ppromoter-lacZ (Parn-lacZ as example) fusion in the single-copy plasmid pNN387

A DNA fragment corresponding to the -200 to -1 region relative to the ATG start codon of arnB was amplified using E. coli MG1655 genomic DNA as the template and primers specific to this fragment flanked with NotI and HindIII restriction sites at the 5’- and 3’-end, respectively. Following digestion with NotI and HindIII and purification using the MiniBEST agarose gel DNA extraction kit (TaKaRa), the fragment was ligated into the plasmid pNN387 which was treated with the same restriction enzymes using the quick ligation kit (NEB). The resulting construct was selected on LB agar plates supplemented with chloramphenicol (25 μg/ml) and was verified by DNA sequencing (BGI, Shenzhen).

Beta-galactosidase activity assay

β-gal activity assay was performed based on a previous description with modifications.⁶⁸ Cells were grown under designated conditions until OD₆₀₀ reached 0.4. 10 μg/ml tetracycline was added to stop cell growth and protein synthesis, and cells were left on ice until being assayed. 100μl of cell culture and 400 μl of Z buffer (60 mM Na₂HPO₄, 40 mM NaH₂PO₄, 10 mM KCl, 1 mM MgCl₂, 50 mM β-mercaptoethanol, pH 7.0, 4 °C) were mixed. One drop of 0.1% SDS and two drops of chloroform were added to lyse the cells. The reactions were initiated by adding 100 μl of 4 mg/ml ortho-nitrophenyl-β-galactoside into cell suspension and were incubated at 28 °C. Color development in the mixtures was monitored. Finally, 250μl of 1M Na₂CO₃ was added to stop the reaction. Color of the mixtures were measured using spectrophotometer at OD₄₂₀ and OD₅₅₀. The expression value in Miller units was calculated, and results are presented as the mean of three biological replicates.

RNA extraction and RT-qPCR

Bacterial cells grown under designated conditions was subjected to RNA extraction when the OD₆₀₀ reached 0.4. Cells were harvested by centrifugation at 4,500 rpm for 10 min at 4 °C. After removing the supernatant, the cell pellet was stored at -80 °C to aid cell lysis. Total RNA was extracted using the MiniBEST Universal RNA Extraction Kit (Takara, Japan) according to the manufacturer’s instruction. Reverse transcription (RT) was performed using PrimeScript^TM RT reagent Kit (Takara, Japan). Quantitative PCR was performed using specific primers and the TB Green® Premix Ex Taq™ (Takara, Japan) in a 20 μL reaction system. The reaction was performed in the ABI StepOnePlus real time PCR system with rrsB as reference gene to normalize the relative expression of the target genes. The results were presented as fold change expression of the target genes, and results were presented as the mean of three biological isolates.

Construction of clean gene deletion, point mutation and tag insertion in chromosome of E. coli MG1655

Gene deletion, mutation, and tag insertion in chromosomal were achieved employing a CRISPR/Cas9-mediated approach developed by Zhao et al. with modifications.⁶⁹ For gene deletion (deletion of lacZ as example), E. coli MG1655 was electrotransformated with pCAGO plasmid which contains both SpCas9 and λ-RED to obtain MG1655-pCAGO. A single colony of the resulting transformant was inoculated into LB medium supplemented with ampicillin (100 mg/ml) and grew overnight at 30°C. 100 μl of the overnight culture was inoculated into 10 ml fresh culture medium and cells were grown at 30°C with IPTG (0.1 mM) induction to OD600 as 0.6. The cells were collected and washed with cold ddH₂O for three times, generating competent cells for electroporation. An aliquot comprising 100 μl competent cells was mixed with 800 ng of the editing-cassette which encompasses 40 bp of sequence among the CDS of lacZ gene followed by the chloramphenicol resistant gene, the N20PAM sequence (GTCCATCGAACCGAAGTAAGG), 40 bp upstream-homologous arm of lacZ gene, and 40bp downstream-homologous arm of lacZ gene, in a 2 mm Gene Pulser cuvette (Bio-Rad) and was subjected to electroporation at 2.3 kV. Following recovering the transformants by cultivation in LB medium for 2 hrs at 30°C, the mixture was spread onto LB agar plate supplemented with ampicillin (100mg/ml), chloramphenicol (25 mg/ml), and 1% glucose (to avoid the background expression of the Cas9 protein). Single colonies were collected and verified by colony PCR to obtain an authentic clone in which the editing-cassette was embedded in chromosomal. The confirmed colony was inoculated into LB medium supplemented with ampicillin (100 mg/ml), IPTG (0.1 mM), and L-arabinose (0.2%) at 30°C overnight to enable expression of SpCas9 and the λ-RED recombinase. The cells were then streaked on LB agar plates supplemented with ampicillin (100mg/ml). Following colony PCR, desired colonies containing deletion of the lacZ gene were verified by DNA sequencing (BGI, Shenzhen). For point mutation, (mutation of pmrB E36 to A as example) E. coli MG1655 was conducted following the same procedures except for the editing cassette sequence which encompasses 40 bp of upstream-homologous arm of GAA (Glu36) followed by the chloramphenicol resistant gene, the N20PAM sequence, the 40 bp upstream-homologous arm of GAA (Glu36), GCA (Ala) and 40 bp downstream-homologous arm of GAA (Glu36). For tag insertion, (insertion 3xFLAG to C-terminal of arnB gene as example) E. coli MG1655 was conducted following the same procedures except for the editing cassette sequence which encompasses 40 bp of upstream-homologous arm of stop codon of arnB gene followed by the chloramphenicol resistant gene, the N20PAM sequence, the 40 bp upstream-homologous arm of stop codon of arnB gene, tag sequence and 40 bp downstream-homologous arm of stop codon of arnB gene (include stop codon). After obtaining all desired genome modifications, the editing plasmid was cured by growing the resulting cells at 42°C for 48 hrs.

Construction of pET28A-pBAD-pmrA for PmrA protein expression

A DNA fragment encoding the PmrA protein was amplified using genomic DNA of E. coli MG1655 as the template and primers specific to this fragment flanked with NcoI and XhoI restriction sites at the 5’- and 3’-end, respectively. After digestion with NcoI and XhoI, and purification using the MiniBEST agarose gel DNA extraction kit (TaKaRa), the purified fragment was ligated into the plasmid pET28a-pBAD and pET28a-pBAD-3xFLAG ⁴³ which was treated with the same restriction enzymes using the quick ligation kit (NEB). The resulting construct was selected on LB agar plates supplemented with kanamycin (20 μg/ml) and was verified by DNA sequencing (BGI, Shenzhen). All plasmids constructed are listed in the supplementary S3 Table.

Immunoblot analysis

Protein levels of ArnB and PmrA were analyzed by SDS-PAGE following Western Blot. 1ml of cells with OD₆₀₀ of 0.5 was collected and resuspended in 200 μl sample buffer (13 mM Tris, 0.2% SDS, 5% glycerol, 0.004% bromophenol blue, pH 6.8) before being subjected to cell lysis by sonication. Following centrifugation at 13000 rpm for 10 min at 4°C, 20 μl of the sample was separated on SDS-PAGE gels (TGX FastCast Acrylamide Kit, 10%, BIO-RAD). Proteins were then transferred onto a polyvinylidene difluoride (PVDF) membrane (BIO-RAD) using a Trans-Blot SD Semi-Dry Transfer Cell (BioRad) at constant voltage (25 V) for 7 min. Membranes were blocked for 1 h with 5% non-fat milk solution in TBST (20 mM Tris, 150 mM NaCl, 0.1% Tween-20) at room temperature followed by incubation with 1:10,000 primary antibody (monoclonal or polyclonal anti-His antibody) in 1% non-fat milk in TBST at room temperature with shaking for 1 hour. After washing with TBST, the membrane was incubated with the secondary antibody (Goat anti-mouse IgG HRP conjugate) in 1% non-fat milk in TBST at room temperature with shaking for 1 hour. Chemiluminescence of protein bands was detected by ECL Plus Western Blotting Detection Reagents (GE Healthcare, USA), and imaged using a X-ray film development by Alliance Q9 (UVITEC, United Kingdom).

Phos-tag SDS-PAGE

In vivo phosphorylation of PmrA was examined using Phos-Tag SDS-PAGE following our previous description with minor modifications.⁴³ E. coli MG1655ΔpmrA cell harboring pET28a-pBAD-pmrA-3×FLAG (AY7068) was inoculated into MOPS minimum medium supplemented with kanamycin (20 μg/ml) and grew overnight at 37°C. 100 μl of the overnight culture was inoculated into 5 ml fresh MOPS medium and grew at 37°C for 4 hrs. Expression of PmrA was induced with 1mM L-arabinose for 1 hr. Cells were collected by centrifugation at 13000 rpm and were washed with MOPS medium supplemented with 0.1, 0.5, or 1 mM FeSO₄ or Pi- MOPS medium twice and shake for 1 hr at 37°C in the corresponding medium. 1ml of cells with OD₆₀₀ of 0.5 was collected for the subsequent analysis. Cell pellet was washed with cold PBS twice and was resuspended in 200 μl sample buffer (13 mM Tris, 0.2% SDS, 5% glycerol, 0.004% bromophenol blue, pH 6.8) before being subjected to cell lysis by sonication. Following centrifugation at 13000 rpm for 10 min at 4°C, 20 μl of the sample was loaded onto a 10% Tris-Gly polyacrylamide gel containing 25 mM Phos-tag Acrylamide (Wako) and 50 mM MnCl₂. Samples were separated by electrophoresis for 2 hrs at 4°C, following washing with 50 ml transfer buffer containing 10 mM EDTA for 20 min to remove residual Mn²⁺. Proteins on the gel were transferred onto a PVDF membrane. The membrane was blocked with 5% non-fat milk in TBST buffer and was incubated subsequently with monoclonal ANTI-FLAG^® M2 antibody (Sigma) and goat anti-mouse IgG (H+L)-HRP conjugate (Bio-rad). To detect FLAG-tagged protein bands, the membrane was submerged in the ECL Plus Western Blotting Detection Reagents (GE Healthcare) for 2 min and then subjected to the imaging system (UVITEC Cambridge) to record signals.

Amber codon mutation of selected CouA incorporation sites in pET28A-pBAD-pmrA

Selective amino acid located in the DNA binding domain of PmrA was subjected to replacement by CouA, each of these CouA incorporation sites were mutated to a TAG amber codon by site-directed mutagenesis PCR. PCR was conducted using a pair of primers for the desired mutagenesis and pET28A-pBAD-pmrA plasmid as the template. PCR product was purified using MiniBEST agarose gel DNA extraction kit (TaKaRa). Following DpnI digestion of the template for 16 hrs, the PCR product was transformed into E. coli DH5α. Transformants were recovered on LB agar plates supplemented with kanamycin (20 μg/ml) at 37°C overnight. Colonies recovered were verified by DNA sequencing (BGI, Shenzhen).

In vivo PmrACouA - promoterSYTO 9 binding assay

Binding of PmrA_CouA and promoter_{SYTO 9} was monitored by FRET assay following our previous description with minor modifications⁴³. Plasmid pET28a-pBAD-pmrA-G137TAG was co-transformed with pEVOL-CouRS into E. coli MG1655 ΔpmrA strain, resulting in AY7065 constructs. A fresh single colony of AY7065 was inoculated into MOPS medium supplemented with kanamycin (20 μg/ml) and chloramphenicol (25 μg/ml) and grew overnight at 37°C. 75 μl of the overnight culture was inoculated to 7.5 ml fresh medium to subculture the bacteria. Following culturing at 37°C for 4 hrs, CouA incorporation and expression of PmrA-G137CouA was induced mildly by supplementing 0.9% arabinose and 1mM CouA at 37°C for 5 hrs. Approximately 3.5×10¹⁰ cells were harvested and washed twice with fresh MOPS medium. Cells were resuspended in MOPS medium as control, or MOPS medium supplemented with 0.5 mM FeSO₄, or Pi- MOPS medium and shake for 1 hr at 37°C. After centrifugation at 4500 rpm for 15 min at 4°C, cell pellet was resuspended in 600 μl of PBS. 85 μl of the suspension was added in each of 6 wells (3 wells for control and 3 wells for experimental group) on a 96-well black plate preloaded with 15 μl of 50 μM SYTO 9 and was incubated for 15 min at room temperature in the dark before being subjected to fluorescence recording (Thermo Scientific Varioskan® Flash spectral scanning multimode reader).

Comparative proteomic analysis

E. coli MG1655 cultured in MOPS medium to OD₆₀₀ of 0.5 was harvested as cells under Pi sufficient condition. The same volume of cells was washed with Pi depletion medium for 3times and resuspended in Pi depletion medium and shake for 3 hrs at 37°C as cells subjected to Pi depletion. Each type of cell pellets with three biological replicates were lysed by resuspending in 20 μl BugBuster reagent (Novagen, USA) (containing 25 mg/ml lysozyme, 1 unit DNase I) at room temperature for 30 mins following centrifugation at 13000 rpm at 4°C for 20 min. Supernatant was transferred to a clean microcentrifuge tube and mixed with 4×SDS sample buffer (40% glycerol, 8% SDS, 240 mM Tris/HCl pH 6.8, 0.04% bromophenol blue) supplemented with 5% β-mercaptoethanol. Sample were incubated at 55 °C for 25 min. 20 µL of each sample were loaded onto a polyacrylamide gel composed of a 10% resolving (lower) gel and a 6% stacking (upper) gel following staining with Coomassie blue staining and fractionated into eight fractions. Gel slices were cut into 1 mm³ cubes and destained with 50% acetonitrile (ACN) in 50 mm triethyl ammonium bicarbonate (TEAB) and dehydrated with pure CAN and dried in a vacuum concentrator. Sequencing grade modified trypsin (Promega) diluted by an ice-cold buffer containing 50 mM NH4HCO3 and 10% (v/v) ACN was added to cover the dehydrated gel pieces on ice. After 30 min, more trypsin-containing buffer was added to cover the expanded gel pieces followed by overnight incubation at 37 °C. Finally, the resulting tryptic peptides were extracted from the gel by equilibrating the samples with 50% ACN and 5% formic acid (FA) for 20 min in a 37 °C shaker. The extraction was repeated once, and all fractions were combined and vacuum dried. The protein digest was resuspended in 100 μl of 100 mm TEAB, and then 4 μl of 0.6 m sodium cyanoborohydride (NaBH₃CN) were added. Peptides from PI+ and Pi- were labeled with 4 μl of 4% formaldehyde (CH₂O) and deuterated formaldehyde (CD₂O), respectively. The reaction mixture was vortexed and incubated at room temperature for 1 h. The reaction was quenched by sequential addition of 16 μl of 1% (v/v) ammonia solution and 8 μl of formic acid. Finally, light- and heavy-labeled peptides were mixed, and vacuum dried immediately before further mass spectrometric analyses. Extracted peptides were analyzed by Nanoflow LC-MS/MS analyses using the EASY-nLC 1000 System (Thermo Scientific, USA) and LTQ Velos (Thermo Scientific, USA).

Data processing was performed following a previous description⁷⁰. For protein identification, MS/MS spectra were extracted from RAW files using MaxQuant (http://maxquant.org/, version 1.5.3.30). MASCOT (Matrix Science, version 2.3.02) was used to perform database search against an E. coli database (strain MG1655, with sequence files downloaded from UniProt in FASTA format, 4,358 entries. Average mass was selected with precursor mass tolerance of 1.5 Da and fragment mass tolerance of 0.8 Da. Methionine oxidation was set as a dynamic modification. The resulting assignments were filtered to achieve a false-discovery rate (FDR) of <1% at both the peptide and protein levels using the target-decoy method. Relative abundances of protein were represented by spectral counts, which are the number of repeated identification of peptides during the entire analysis.

To identify proteins that were differentially expressed in the sets of samples, the spectral counts data were subjected to statistic by DESeq package (Anders and Huber 2010) in R (Team 2014). Briefly, to eliminate the variance caused by sample loading and machine reading depth, the total counts data of two biological conditions was normalized to the same size. The following statistics were performed: foldChange (fold change from condition A to B), pval (p value for the statistical significance of this change), and padj (p value adjusted for multiple testing with the Benjamini-Hochberg procedure, which controls false discovery rate). Data set with padj <0.05 were considered to be significant (Anders and Huber 2010), and corresponding proteins were listed in S1 Table.

Polymyxin challenging assay

Challenging E .coli cells with designated concentration of polymyxin B (Sigma. USA) were performed as described with modifications.⁷¹ E. coli cells grown under designated conditions were harvested by centrifugation at 13000 rpm for 1min. Cell pellets were re-suspended in MOPS medium to a final concentration of 10⁸ cells/ml. The cell culture was treated with polymyxin B with designated concentration for 1-3 hours at 37 °C without agitation. Cell culture prior to and following polymyxin treatment was plated on LB plate, CFU was counted after 12-16 hours incubation at 37°C. Survival of polymyxin challenging was determined by the survival rate of E. coli cells after drug treatment. Results are presented as the mean of three biological replicates.

Phylogenetic analysis of EcoPmrB homologs

The PmrB sequences from E. coli K-12 MG1655 were used as a query against γ-proteobacteria via BLASTP searches in the UniProt database, with an E-threshold of 10 and the BLOSUM62 substitution matrix. The top 250 retrieved sequences were aligned using the MUSCLE algorithm implemented in MEGA12 software. A phylogenetic tree of EcoPmrB homologous was constructed using the Maximum Likelihood method, with branch support assessed via adaptive bootstrap analysis using 562 bootstrap replicates in MEGA12. The JTT model was applied as the evolutionary model for tree construction.

Statistical analysis

Statistical analysis was performed using GraphPad Prism. Group comparisons between two groups were performed using two-tailed Student’s t test. Group comparisons between multiple groups were performed using one-way ANOVA or two-way ANOVA with post hoc tests.

Supporting information

Figure S1.

Figure S2.

Figure S3.

Figure S4.

Figure S5.

Figure S6.

Figure S7.

Figure S8.

Figure S9.

Figure S10.

Figure S11.

Figure S12.

Figure S13.

Figure S14.

Figure S15.

Figure S16.

Figure S17.

Figure S18.

Figure S19.

Table S1.

Table S2.

Table S3.

Table S4.

Data availability

All data are available in the main text or the supplementary materials.

Acknowledgements

We thank Ms. Yining Chen in Prof. A. Yan’s lab for proofreading the manuscript. We thank Prof. Changhao Bi (Tianjin Institute of Industrial Biotechnology, Chinese Academy of Sciences) for the generous gift of pCAGO plasmid. We thank Prof. Zhaochao Xu and Prof. Lu Miao (CAS Key Laboratory of Separation Science for Analytical Chemistry, Dalian Institute of Chemical Physics, Chinese Academy of Sciences) for the generous gift of OM-sensitive probe PI-BactD. We thank Prof. Patrick C. Y. Woo (Department of Microbiology, University of Hong Kong) for the generous gift of clinical E. coli strains.

Additional information

Funding

This work was supported by the

Research Grants Council of Hong Kong, Theme-based Research Scheme T21-705/20-N (A.Y. and T.Z.)

Natural Science Foundation of China 22193062 (T.Z.)

Research Grants Council of Hong Kong, Collaborative Research Fund C7033-20G (A.Y.)

National Key Research and Development Program of China 2022YFA1304500 (X.L.)

Natural Science Foundation of China 22174003 and 21974002 (X.L.)

Author contributions

Conceptualization: Guangming Zhang, Ziqing Deng, Aixin Yan.

Investigation Guangming Zhang, Ziqing Deng, Jiezhang Jiang, Minji Wang, Xiaoyuan Wang, Xiaoyun Liu, Aixin Yan.

Formal Analysis Guangming Zhang, Ziqing Deng, Xiaoyuan Wang, Xiaoyun Liu, Aixin Yan.

Funding Acquisition: Xiaoyun Liu, Aixin Yan.

Resources: Xiaoyuan Wang, Xiaoyun Liu, Aixin Yan.

Supervision: Xiaoyuan Wang, Xiaoyun Liu, Aixin Yan.

Visualization: Guangming Zhang.

Writing – original draft: Guangming Zhang, Aixin Yan.

Writing – review & editing: Guangming Zhang, Xiaoyun Liu, Aixin Yan.

Funding

Research Grants Council, University Grants Committee (T21-705/20-N)

• Aixin Yan

MOST | National Natural Science Foundation of China (NSFC) (22193062)

• Aixin Yan

Research Grants Council, University Grants Committee (C7033-20G)

• Aixin Yan

MOST | National Key Research and Development Program of China (NKPs) (2022YFA1304500)

• Xiaoyun Liu

MOST | National Natural Science Foundation of China (NSFC) (22174003)

• Xiaoyun Liu

MOST | National Natural Science Foundation of China (NSFC) (21974002)

• Xiaoyun Liu