Context-dependent siderophore exploitability shapes microbial community structure

Abstract

Siderophores are classically viewed as shared iron-scavenging public goods, yet their ecological roles in multispecies communities remain incompletely understood. Here, we establish a synthetic microbial community to dissect how different siderophores, their uptake compatibility and spatial structure shape iron competition. Using Corynebacterium glutamicum as a model, we show that this siderophore non-producer accesses diverse xenosiderophores, including enterobactin secreted by Escherichia coli. However, exploitation was constrained and co-cultures converged to stable compositions. Dose–response experiments combined with mathematical modelling indicated that the producer retains more effective access to enterobactin than the exploiter. The presence of Pseudomonas putida altered this interaction, as it exploited enterobactin while producing pyoverdine, a siderophore inaccessible to the other community members that restricted their iron access. Across different cultivation scales, community dynamics was strongly influenced by spatial organization and initial composition. These findings identify siderophores as context-dependent iron-allocation agents that can promote microbial coexistence or exclusion.

Graphical Abstract

Graphical Abstract.

Introduction

Iron is essential for almost all living organisms because it serves as a cofactor in fundamental cellular processes including respiration, central carbon metabolism, and DNA synthesis [1, 2]. Yet, despite its abundance in many environments, biologically accessible iron is often scarce under aerobic conditions, where iron occurs predominantly as poorly soluble ferric iron [3]. To overcome this limitation, many bacteria, fungi, and plants produce siderophores that are small high-affinity metabolites that chelate ferric iron and deliver it to cells through dedicated uptake systems [4]. Because siderophore biosynthesis imposes a substantial metabolic burden, these compounds are generally produced and secreted only under iron limiting conditions [5, 6]. To date, more than 500 siderophores have been identified, encompassing a wide range of structural classes, including catecholate, carboxylate, hydroxamate, and mixed-type siderophores [7, 8]. Enterobactin, produced by Escherichia coli and Klebsiella pneumoniae, is considered the archetypal siderophore and exhibits the highest reported affinity for iron [9].

Siderophores are not only iron-scavenging molecules but also mediators of microbial interactions [2]. In bacterial communities, siderophore-bound iron can be accessed only by organisms carrying compatible uptake systems. As a result, siderophores may function as shared resources, privatized goods, or competitive agents [2, 10]. Gram-negative bacteria recognize specific iron-loaded siderophores via matching outer membrane receptors, like FepA of Escherichia coli for enterobactin [11], or FpvA in Pseudomonas putida for pyoverdine uptake [12]. In contrast, Gram-positive bacteria scavenge them by specific membrane-anchored siderophore-binding lipoproteins and ABC transporters [13]. Many microbes encode receptors and transporters for siderophores that they do not synthesize themselves, enabling them to exploit xenosiderophores produced by neighbors [14–17]. For example, E. coli synthesizes and utilizes enterobactin, but also has the receptor FhuA for ferrichrome, which is a siderophore produced by many fungal species like Penicillium spp. or Ustilago spp. [18, 19]. Moreover, Pseudomonas aeruginosa upregulates the siderophore transport system FemIRA when facing mycobactin produced by Mycobacterium spp. [20]. Xenosiderophore uptake and utilization reduces the metabolic cost of iron acquisition, but it can also deprive competing organisms of accessible iron [2, 21, 22]. Conversely, siderophore producers can restrict iron availability to organisms lacking matching receptors [23, 24]. These dynamics make siderophores a powerful determinant of microbial competition and coexistence in iron-limited habitats such as soil, the gut, and further host-associated microbiomes [25]. In these environments, xenosiderophore exploitation is emerging as an important ecological mechanism and a potential target for intervention. For example, the soil bacterium Pseudomonas fluorescens pirates desferrioxamines and coelichelin from Streptomyces ambofaciens [26], whereas in the nasal microbiome, corynebacterial species consume siderophores produced by Staphylococcus aureus, thereby reducing S. aureus growth [27]. Likewise, several Pseudomonas spp. have been shown to produce pyoverdines with significant antimicrobial properties against several human pathogens due to provoking iron limitation [28]. To date, the molecular basis of siderophore production, regulation, and chemical diversity has been extensively studied, while in recent years their ecological roles in diverse microbial communities are an emerging field of research [2, 29]. Much previous work has focused on individual siderophore systems [30, 31], pairwise producer–cheater interactions [24, 32, 33], or the effects of siderophores on microbial co-cultures [34–36]. However, natural communities often contain multiple phylogenetically diverse species that differ in siderophore production, uptake compatibility, and spatial organization. Under these conditions, the ecological role of a siderophore may not be fixed: the same molecule could promote growth of one community member while excluding another, depending on receptor compatibility, relative abundance, and the spatial structure of the habitat. A tractable synthetic system comprising model bacteria with well-defined traits is therefore valuable for elucidating how sharing, exploitation, and exclusion emerge and interact.

Here, we established a synthetic community to examine context-dependent siderophore interactions among phylogenetically distant species and the benefits and detriments of heterologous siderophores. We chose Corynebacterium glutamicum as a model xenosiderophore exploiter because corynebacteria were recently described as common exploiters in natural microbiomes [27]. C. glutamicum is a well-established, genetically tractable model organism that does not produce siderophores on its own [37], but exhibits a broad repertoire of putative xenosiderophore transport proteins for iron acquisition that was not yet studied in detail (Table S1) [38, 39]. We first identified diverse bacterial and fungal siderophores, including enterobactin, that support growth of iron-starved C. glutamicum. We then built a tripartite community consisting of the well-established model species C. glutamicum as an exploiter, E. coli as an enterobactin-producing public-good provider, and Pseudomonas putida as both an enterobactin exploiter and a pyoverdine (private good) producer. By analysing community development and composition across scales from single cells to structured communities on plates, we show that siderophore-mediated interactions are strongly context dependent. Enterobactin acts as a public good that supports the growth of each community member, whereas pyoverdine acts as a private good—a non-exploitable good that restricts iron access from E. coli and C. glutamicum even in the presence of exploitable enterobactin. A mathematical model parameterized with experimental data suggests that exploiters can utilize the siderophore less efficiently than producers, e.g. due to a substantially lower receptor affinity, constraining exploitation despite access to enterobactin. Spatial structure further modulated these interactions by promoting niche formation for exploiters while also facilitating enterobactin privatization by the producer. Together, these findings highlight the context dependency of siderophore-mediated interactions and underscore the central role of spatial organization in shaping microbial community dynamics.

Material and methods

Microbial strains and cultivation conditions

All microbial strains used in this study are listed in Table S3. Microbial strains were streaked from frozen 20% glycerol stocks on 1.5% (w/v) agar medium containing complex medium. For C. glutamicum, brain heart infusion (BHI) (Difco, BD, Heidelberg, Germany) (37 g/L BHI) medium was used as complete medium. All other microbial strains were cultivated in lysogeny broth (LB) (10 g/L tryptone, 5 g/L yeast extract, 10 g/L NaCl) medium, while Ustilago maydis was cultivated in LB or yeast extract-peptone-sucrose (YEPS) (10 g/L yeast extract, 10 g/L peptone, 10 g/L sucrose) medium. Strains were cultivated exclusively at 30°C, with exception for E. coli in first pre-cultures at 37°C. For liquid cultivations, single colonies from agar plates were picked and incubated in 5 ml complex medium in reaction tubes at 170 rpm or in 1 ml medium in deep-well plates (VWR International, Radnor, PA, United States) at 900 rpm. Upon cultivating for ~8 h in BHI or LB medium, resulting cell pellets were used to inoculate a second preculture in CGXII minimal medium (20 g/L (NH₄)₂SO₄, 1 g/L K₂HPO₄, 1 g/L KH₂PO₄, 5 g/L urea, 42 g/L 3-morpholinopropane-1-sulfonic acid (MOPS), 13.3 mg/L CaCl₂*2H₂O, 0.25 g/L MgSO₄*7H₂O, 0.2 mg/L biotin, 0, 1 or 10 mg/L FeSO₄*7H₂O, 10 mg/L MnSO₄*H₂O, 0.1 mg/L ZnSO₄*7H₂O, 0.02 mg/L CuSO₄*5H₂O, 0.002 mg/L NiCl₂*6H₂O; pH was adjusted with KOH to pH of 7.0.) [40] containing 2% (w/v) glucose as carbon source. To starve cells of the second pre-culture from iron prior to experiments, CGXII minimal medium without FeSO₄ was used (referred to “iron-free CGXII”). Protocatechuate (PCA) (3,4-dihydroxybenzoic acid), a ferric chelator, is usually added to the CGXII minimal medium to facilitate iron uptake and was therefore left out in our cultivations [40, 41]. For E. coli, 1 μM FeSO₄ was added to the second pre-culture. Second pre-cultures were performed in 20–25 ml minimal medium in shaking flasks at 120 rpm or in 1 ml medium in 96-deep-well plates (VWR International, Radnor, PA, United States) at 900 rpm.

Main cultures were inoculated at an optical density at 600 nm (OD₆₀₀) of 1, if not stated otherwise, and either performed in shaking flasks in respective volumes or for online monitoring of growth in 750 μl in 48-well microtiter Flower Plates in a BioLector I microtiter cultivation system (Beckman Coulter GmbH, Aachen, Germany) [42]. In the microtiter cultivation system, backscatter (a.u.) was measured in 30 min intervals as scattered light intensity at λ: 620 nm (signal gain: 20). In this context, backscatter is the intensity of light scattered by the cells in the culture, serving as a proxy for biomass. P. putida cultures were inoculated at an OD₆₀₀ of 0.5 in a volume of 1 ml. As standard conditions, CGXII minimal media was used as described above. For analysis of growth in presence of LB-based spent media, cells were inoculated in two-fold concentrated iron-free CGXII medium supplemented with 2*2% glucose and 2*150 μM 2,2′-dipyridyl (hereafter referred to as DIP). For CGXII-based spent media, cells were inoculated in two-fold concentrated CGXII medium with 2*2% glucose, 3.6 μM FeSO₄ and no DIP. Each cell-media suspension was mixed with 50% of respective spent medium. The addition of DIP is a well-established strategy to mimic non-accessible iron pools in siderophore studies [43–46]. Siderophores have a higher affinity for iron than this chelator, thereby enabling iron uptake by those cells, which can use the specific siderophore. When enterobactin was added as pure compound (Sigma–Aldrich, dissolved in DMSO), 3.6 μM FeSO₄ and 150 μM DIP was added in liquid cultivations.

Bacterial strains containing plasmids mediating a kanamycin resistance were cultivated in the presence of appropriate amounts of the antibiotic (C. glutamicum: 25 μg/L; E. coli: 50 μg/L) to ensure plasmid maintenance. If needed, LacI-P_tac-derived gene expression was induced by the addition of isopropyl-β-D-1-thiogalactopyranoside (IPTG).

Production and harvesting of siderophore containing spent medium

Spent medium refers to the siderophore-enriched growth medium remaining after cultivating of a respective siderophore-producing organism, but cells removed.

For screening purposes, LB-based spent medium was used: Cells of siderophore-producing strains from an overnight culture were transferred at an OD₆₀₀ of 0.45 to 50 ml LB medium supplemented with 200 μM DIP, while U. maydis cells were cultivated in YEPS medium containing 300 μM DIP.

For more specific analyses, CGXII-based spent medium was used: E. coli or P. putida strains from an overnight culture were transferred at an OD₆₀₀ of 0.45 to 50 mL CGXII media as described above, but containing 0.36 μM FeSO₄ and 360 μM DIP.

After 2 days of cultivation in shake flasks, the culture supernatant was harvested and sterile filtered (0.2 μm). Harvested spent medium was stored at −20°C until further use.

Recombinant DNA work

Standard molecular methods, such as DNA restriction, PCR, and Gibson assembly, were performed according to the manufacturer’s instructions or to previously described standard protocols [47, 48]. All plasmids and primers used in this study are listed in Tables S4 and S5. Oligonucleotide synthesis and Sanger DNA sequencing was performed by the company Eurofins Genomics (Ebersberg, Germany).

In this work, all plasmids were constructed using the enzymatic Gibson assembly approach. Plasmids were transferred into competent C. glutamicum cells by electroporation as described previously [49]. For genomic deletions and integrations, derivates of the suicide vector pK19-mobsacB were used [50]. Two step homologous recombination and selection was done as described previously [51].

Plasmids were transferred into competent E. coli cells by heat shock or TSS transformation [52, 53]. Genomic deletion of the entC gene in strain E. coli PC 0886 ΔentC was achieved with the Lamda red recombinase-based deletion method described previously [54]. The disruption cassette consisted of the neoR resistance gene, encoding the neomycin phosphotransferase providing resistance against neomycin and kanamycin. This gene was flanked by FLP recombinase recognition target (frt) sites, which are needed for later elimination of the neoR gene. The up- and downstream sequences of entC (~200 nt) formed the ends of the disruption cassette long sequences, allowing for replacement of the entC gene by the disruption cassette by homologous recombination.

For genomic integrations, the deletion protocol was modified. An integration cassette was used instead of the disruption cassette. The design was comparable to the disruption cassette, but the sequence that should be integrated was inserted between the neoR–frt sequence and the right homologous flank.

For the genomic integration of the constitutive mTagBFP (for fluorescent labeling) and the IronSenseR–mCherry, tetra parental patch-mating was performed on 1% LB agar plates overnight at 30°C. Freshly streaked strains of P. putida KT2440 WT or P. putida ΔpvdD were plated as acceptor strains. Helper strain E. coli HB101 pRK2013 and E. coli DH5αλpir pTNS1 either helped with pili formation or delivered necessary transposase genes respectively. Lastly a E. coli Pir2 carrying the miniTn7 plasmid with the IronSenseR-mCherry or constitutively expressed fluorescent marker gene (mTagBFP) was used as a donor strain. After incubation, cells were streaked on LB plates containing gentamycin (25 μg ml⁻¹) and irgansan (25 μg ml⁻¹) and cultivated over night to select for P. putida KT2440 WT or P. putida ΔpvdD integration cassette. Then single clones were streaked on LB containing kanamycin (25 μg ml⁻¹) and irgansan (25 μg ml⁻¹) as well as LB containing gentamycin (25 μg ml⁻¹) and irgansan (25 μg ml⁻¹). Clones only growing on LB containing gentamycin and irgasan were likely carrying the Tn7 integration, which was then confirmed via colony PCR.

Siderophore sharing in structural environments

Double layer agar plate assays were performed to study siderophore sharing in structural environments. LB or CGXII plates consisted of a solid 1.5% agar bottom layer and a soft 0.4% agar top layer to facilitate diffusion. Plates were prepared both with and without DIP to compare the effects of “free iron” availability versus a condition where iron could only be acquired via siderophores. For LB-based agar plates, 200 μM DIP was added, but no additional iron. For CGXII-based agar plates, 3.6 μM FeSO₄ was added and 100 μM DIP was utilized, while for “full-iron conditions” 36 μM FeSO₄ were used without any addition of DIP.

Xenosiderophore utilization screen (LB-based)

Cells from LB overnight cultures of different putative siderophore producers (Table S3) were transferred into 0.9% NaCl at cell density of 0.24*10¹⁰ cells/ml. 40 μL cell suspension were streaked on freshly prepared plates as thick lane. After overnight cultivation of producer strains at 30°C, iron-starved C. glutamicum cells were transferred into 0.9% NaCl at cell densities of 10¹⁰, 10⁹, 10⁸, 10⁷, 10⁶, 10⁵ and 10⁴ cells/ml. 3 μL of these cell suspensions were dropped at different distances to the lane of siderophore producers. Images were taken after 48 h cultivation at 30°C.

Siderophore-mediated co-cultivations (CGXII-based)

Cells were cultivated in iron-free CGXII as second pre-culture. Subsequently, cells corresponding to an OD₆₀₀ of 1 were washed and resuspended in 0.9% NaCl. Each 2 μL of cell suspension was dropped on the CGXII-soft agar layer either as monocultures, or in duo or trio combinations dropped ~5 mm apart. Plates were incubated at 30°C for 3 days and images were taken under a steromicroscope Zeiss Axio Zoom V16 with respective filter units (GFP: λ_Ex: 489 nm, λ_Em: 509 nm (gfp); HE mPlum: λ_Ex: 587 nm, λ_Em: 649 nm (E2-crimson); DAPI: λ_Ex: 358 nm, λ_Em: 463 nm (bfp)) once after 48 h and 120 h. Analysis of percentage area was performed using the open source program ImageJ/Fiji [55, 56]. Area per fluorescent channel image was extracted using individual background correction thresholds.

In vivo measurement of the biosensor for iron in C. glutamicum

For the in vivo online measuring iron levels in C. glutamicum, the fluorescent protein LSSmApple of the original IronSenseR [57] was replaced by mCherry, constructing IronSenseR-mCherry plasmid. For C. glutamicum, an N-terminal deca-histidine tag followed by a TEV cleavage site was directly fused to the biosensor encoding gene cassette. Electrocompetent C. glutamicum were transformed with the plasmid pPREx210xHis-TEV-MDtxR_G149GC. Single colonies were cultivated in 800 μl ml BHI supplemented with kanamycin in a deep-well plate at 30°C and 900 rpm for 8 h. The second pre-culture was prepared in iron-free CGXII medium containing 2% (w/v) glucose supplemented with 100 μM IPTG. Cultures were incubated overnight (16 h) at 30°C and 900 rpm in deep-well plates. For the main culture, cells were inoculated to an initial OD₆₀₀ of 0.5 in 10 ml CGXII minimal medium supplemented with 2% (w/v) glucose, 3.6 μM FeSO₄ as iron source, no PCA and 100 μM IPTG. Where indicated, enterobactin was added to a final concentration of 10 μM. Cultures were incubated in 100 ml shaking flasks at 30°C for 2 h on a rotary shaker. Cells corresponding to an OD₆₀₀ of 0.5 in 1 ml culture volume were harvested by centrifugation (5000 rpm, 4°C, 5 min), washed, and resuspended with ice-cold MOPS buffer (20 mM, pH 7.0). A volume of 100 μL was transferred to a microplate and analysed using a multimode microplate reader (Infinite M1000Pro, Tecan). Emission spectra were recorded in 5 nm increments from 490 to 750 nm for GFP upon excitation at 453 nm, and from 580 to 750 nm for mCherry upon excitation at 550 nm. Fluorescence values were corrected by subtracting the signal obtained from buffer-only controls. To obtain the ratios, the peak of GFP fluorescence (λ_em = 510 nm) was divided by the peak value of mCherry (λ_em = 610 nm). For P. putida the strain carrying the integration of the IPTG inducible IronSenseR (P. putida ΔpvdD::attTn7 IronSenseR-mCherry), was inoculated in LB at 30°C and 1200 rpm overnight. The second preculture was inoculated at an OD₆₀₀ of 0.5 and incubated overnight in LB at 30°C and 1200 rpm with the addition of 500 μM IPTG. For the main culture, cells were inoculated to an initial OD₆₀₀ of 0.1 in 1 ml LB 500 μM IPTG. Where indicated, 10 μM enterobactin and 250 μM DIP was added. Cultures were incubated for 6 h at at 30°C and 1200 rpm. 100 μL of the culture was harvested washed and resuspended in 100 μL PBS, transferred to a microtiter plate and analysed using a multimode microplate reader (Infinite M1000Pro, Tecan). Two distinct emission spectra of GFP (λ_ex = 480 nm, λ_em = 500–600 nm) and mCherry (λ_ex = 587 nm, λ_em = 600–700 nm) were recorded. To obtain the ratios, the peak of GFP fluorescence (λ_em = 508 nm) was divided by the peak of mCherry (λ_em = 608 nm).

MS-measurements of enterobactin and organic acids

For MS measurements, four replicates of C. glutamicum were cultivated as described above in an iron-free CGXII pre-culture. In the main culture, C. glutamicum was cultivated at a starting OD₆₀₀ of 1 in 50 ml CGXII supplemented with 2% glucose and either (i) 3.6 μM FeSO₄, 150 μM DIP, no PCA (iron limited) (ii) 3.6 μM FeSO₄, 10 μM enterobactin, 150 μM DIP, no PCA or (iii) 36 μM FeSO₄. Samples were taken after 5, 10, 30, 120, 300, and 1440 min.

For measurements of extracellular enterobactin, cells were centrifuged and supernatant filtered (0.22 μm). For intracellular measurements of enterobactin and organic acids, cells corresponding to two OD units were washed in 10 ml 0.9% NaCl and cell culture was rapidly filtered through a nylon membrane with a pore size of 0.45 μm using a vacuum filtration system (Merck Millipore). The filter was washed twice with 5 ml ice-cold 0.9% NaCl. The membrane was immediately placed into 2 ml tubes preloaded with steel (2 mm) and glass beads (Retsch) and snap-frozen in liquid nitrogen. The entire sampling and quenching procedures were completed within 45 s.

Enterobactin and organic acids were extracted from the cells using a method adapted from [58]. Details can be found in the supplementary material.

Flow cytometry of trio cultures

P. putida, and E. coli were precultured in LB, while C. glutamicum was precultured in BHIS overnight at 30°C and 1200 rpm. The second preculture was inoculated at an OD₆₀₀ in CGXII medium with 2% glucose and 3.6 μM FeSO₄. For the main culture, the three different strains were mixed at similar ratios to a total OD₆₀₀ of 1, in CGXII medium with 2% glucose containing either 0.36 μM FeSO₄, 3.6 μM FeSO₄, 0.36 μM FeSO₄+ 250 μM DIP or 3.6 μM FeSO₄+ 250 μM DIP and cultivated for 24 h at 30°C and 1200 rpm and then reinoculated at an OD₆₀₀ of one in fresh media inoculating further 24 h. During cultivation and at the end 20 μL samples were taken washed and diluted to an OD₆₀₀ of 0.05 in 200 μL of PBS. Afterwards samples were analysed in the flowcytometer (Amnis CellStreamTM System, Merck, now Cytek Biosciences, California, USA). Cell populations were gated for size and complexity via front scatter and side scatter. Afterwards, populations were further gated for mCherry, GFP, or mTagBFP fluorescence. For analysis of the Flowcytometry data, the CellStream Analysis Software (Merck, now Cytek Biosciences, California, USA) was used.

Pyoverdine measurements

Precultures were performed in LB at 30°C and 1200 rpm. Main cultures were inoculated at an OD₆₀₀ of 0.05 in LB, with 250 μM DIP, 250 μM DIP and 10 μM enterobactin or no addition. To measure pyoverdine production, after cultivation 20 μl of culture were transferred to 96-well clear bottom microtiter plates and diluted with 80 μL of PBS. Using a plate reader, Infinite M1000 PRO (Tecan Austria GmbH) absorption was measured at 600 nm and pyoverdine fluorescence (λ_ex = 398 nm, λ_em = 455 nm) was measured in the same device.

Single-cell cultivation in microfluidic chips

Microfluidic chips were fabricated by soft lithography using a two-layer SU-8 master on a four-inch silicon wafer. The Clewin5 design comprised cultivation chambers (60 × 100 × 0.9 μm) connected via 9.8 μm deep channels to two opposing 10 μm wide inlets. Each inlet supplied eight rows of thirty chambers. PDMS (Sylgard 184; 10 wt% curing agent) was poured onto the SU-8 master, then cured for 2 h at 80°C. It was then cut and punched (Ø 0.5 mm). The PDMS plates were bonded to glass substrates (D263® Bio, 39.5 × 34.5 × 0.175 mm) using O₂ plasma activation (25 s, 510 W, Femto Plasma Cleaner, Diener Electronics, Germany).

All experiments were performed in PCA-free CGXII minimal medium (1% w/v glucose). C. glutamicum was pre-cultured at 30°C in three steps: 8 h BHI, followed by 16 h and 4 h iron-limited CGXII. E. coli was grown in LB (37°C), then 16 h iron-limited CGXII, and finally 4 h CGXII with 3.6 μM FeSO₄. All cultures were initiated from single colonies on agar plates. No DIP was required to induce iron-restriction in microfluidic conditions.

For microfluidic experiments, the chips were inoculated with monocultures or co- cultures adjusted to an OD₆₀₀ of 0.5. Time-lapse imaging was performed every 30 min using a Nikon Eclipse Ti2 microscope (100× oil objective, DS-Qi2 camera, Perfect Focus System) The fluorescence of the genomically encoded E2-Crimson, GFP and mTagBFP were excited using a Sola LED illumination system with the appropriate filter cubes (E2-Crimson: λ_Ex = 562/40 nm, λ_Em = 641/84 nm; GFP: λ_Ex = 520/40 nm, λ_Em = 540/40 nm; mTagBFP: λ_Ex = 438/24 nm, λ_Em = 483/32 nm).

Initial preprocessing of the raw microscopy data was performed in Fiji (Version 1.54p) by cropping to ~100 × 60 μm. Time-lapse sequences were exported as TIFF files and uploaded to an open microscopy environment (OMERO) server.

Subsequent image analysis was implemented in custom Jupyter workflows based on Acia, using Omnipose for segmentation [59]. Fluorescence signals enabled cell differentiation, and segmentation artifacts were excluded by intensity, size, and position-based filtering Cell areas were extracted from segmentation masks and aggregated per frame.

Subsequent analyses focused on the temporal progression of colony area, defined as the cumulative area of all cells per species within a chamber. Growth rates were calculated as the difference quotient between consecutive time points. Relative population dynamics were quantified as the fraction of species-specific area relative to the total cell area per chamber.

Model and simulation of microfluidic co-culture experiments

We formulated a deterministic, unstructured population model to describe the interaction between E. coli and C. glutamicum in microfluidic chambers. The system is represented by coupled ordinary differential equation (ODE) based on Monod-type kinetics. We assumed: well-mixed chamber, with no spatial gradients, no explicit glucose limitation, growth-associated metabolite exchange only, and no cell death or maintenance term. The production of enterobactin and its binding to iron is simplified into the production of an abstract product . The set of governing reactions read:

(1)

(2)

(3)

where is the minimal growth rate of E. coli in absence of , is the maximum growth rate of E. coli, is the half saturation constant towards of E. coli, is the half saturation constant towards of C. glutamicum, and are the concentrations of both organisms, and are the change of E. coli and C. glutamicum concentration and is the change of the concentration of

In this scenario E. coli produces a product which itself need to take up again to stimulate growth. The production and the consumption of by both organisms is assumed to be growth dependent.

(4)

(5)

(6)

These proposed rates have been adjusted to fit qualitatively to our observed data (Table S6 and S7).

Parametrization

Maximum specific growth rates of both organisms were experimentally determined or obtained from literature (Table S6). The half-saturation constant for E. coli was taken from literature, whereas the corresponding parameter for C. glutamicum was varied to test the hypothesized interaction mechanism.

Production and uptake coefficients were defined relative to maximum growth rates and adjusted to reproduce qualitative experimental trends. All parameter values and units are summarized in Supplementary Tables S6 and S7.

Varying starting conditions in microfluidic experiments

To demonstrate the effects of the varying starting conditions given in microfluidic cultivation chambers, we performed two parameter sweeps. In the first step we varied the starting concentrations of total cell dry-weight, which corresponds to different numbers of starting cells which are inside microfluidic chambers. The variation of cell dry-weight corresponds with different start cell areas of all cells in a chamber between 10 μm² and 100 μm².

In the second step the simulated starting cell dry weight concentration is simulated as 19 different compositions of C. glutamicum and E. coli, between 5% and 95% of the entire cell dry weight concentration. The starting concentration of was set, so the calculated growth rate for E. coli is 95% of its maximum growth rate.

Varying affinity relationships of co-culture participants

To investigate the impact of potential variants of affinity towards we performed another two-dimensional parameter sweep. We varied the affinity of C. glutamicum towards the limiting product by scaling its Monod constant relative to the producers Monod constant from 0.01 to 100 To evaluate the system’s performance during growth, the mean specific growth rates for both species were calculated exclusively for the growth phase up to a total biomass of 76.15 g/L, corresponding to an observed projected cell area of 2500 μm².

Numerical integration

The model was solved numerically using Python (Version 3.8.15) and the scipy.integrate library (Version 1.9.1). The LSODA integrator was employed via the solve_ivp function. Specific growth rates were recalculated post-simulation based on the resulting concentration profiles and the governing Monod kinetics. Solutions were evaluated at 0.1 h intervals over a total simulation time of 30 h.

Results

Corynebacterium glutamicum exploits chemically diverse xenosiderophores under iron limitation

Given the predicted capacity of the non-siderophore-producing C. glutamicum to exploit xenosiderophores (Table S1), we first determined which siderophores support its growth under iron limitation. To this end, we systematically assessed the growth of iron-starved C. glutamicum in proximity to microorganisms producing chemically distinct siderophores, including catecholates, hydroxamates, and mixed-type siderophores (Fig. 1A). As expected, C. glutamicum failed to grow on solid iron-restricted media in monoculture (Fig. S1). However, robust growth was observed when growing next to enterobactin producing E. coli and Serratia odorifera, bacillibactin secreting Bacillus subtilis and Bacillus amyloliquefaciens, as well as the ferrichrome producing fungus Ustilago maydis (Fig. 1A). In contrast, no growth was observed next to Pseudomonas putida and Pseudomonas taiwanensis, both secreting pyoverdine variants. This xenosiderophore utilization pattern is further supported when analysing growth in liquid cultures supplemented with 50% LB-based spent medium of selected siderophore producing organisms grown under iron depletion (Fig. 1B).

Figure 1.

C. glutamicum is able to exploit several chemically diverse xenosiderophores. (A) The potential of seven known siderophore producers to promote growth of neighboring iron-starved C. Glutamicum on agar-solidified LB medium containing 250 μM DIP (2,2′-dipyridyl) was investigated. C. glutamicum cell suspensions were spotted at different distances to the pre-grown lane of siderophore producer. Moieties involved in iron chelation that determine siderophore classification are shown on the right (adapted from [104]). One representative replicate is shown. (B) Liquid cultivation of C. glutamicum in 50% two-fold CGXII minimal media without added iron, and 50% spent medium harvested from selected organisms as in (A) grown in LB supplemented with 200 μM DIP. Graph colors represent siderophore types, while LB control is shown in black. n = 3 biological replicates.

Together, these results demonstrate that C. glutamicum can exploit a diverse set of xenosiderophores, spanning multiple iron-chelation chemistries and originating from both Gram-positive and Gram-negative bacteria, and even from fungi, to overcome iron limitation.

C. glutamicum is an exploiter of enterobactin produced by E. coli

Having demonstrated that C. glutamicum can exploit different xenosiderophores, we next focused on its interaction with enterobactin and its producer E. coli. Based on the dose-dependent growth stimulation of C. glutamicum observed upon supplementation of iron-limited medium with LB-based E. coli spent medium (SM), a 50% SM dilution supplemented with DIP was used in subsequent experiments (Fig. S2A–B).

To test whether this growth-promoting effect specifically depends on enterobactin, we generated an enterobactin deficient E. coli strain (ΔentC), lacking the gene encoding for the isochorismate synthase EntC that catalyzes the conversion of chorismate to isochorismate in the first step of enterobactin synthesis [60]. In comparison to E. coli WT SM, SM of the ΔentC mutant did not support growth of C. glutamicum (Fig. 2A). This effect was also observed for CGXII-based SM, i.e. minimal media (Fig. S2C). Using a plasmid-based LacI-P_tac-controlled entC expression, the growth-promoting effect could be restored (Fig. S2D). Supplementation of pure enterobactin also enhanced growth of iron-starved C. glutamicum (Fig. 2B).

Figure 2.

C. glutamicum exploits enterobactin produced by E. coli. (A) Comparison of the growth of C. glutamicum in 50% LB-based spent medium of E. coli WT with that in spent medium of non-enterobactin-producing E. coli ΔentC. (B) C. glutamicum growing in 50% LB supplemented with 200 μM DIP and either 10 μM pure enterobactin (red) or 1 μL/mL DMSO as negative control. n = 3 biological replicates. (C) MS-measurement of intra- and extracellular enterobactin content in C. glutamicum cultivated with 10 μM enterobactin in the presence of 150 μM DIP. n = 4 biological replicates. (D) Schematic depiction of ratiometric fluorescent-based modified biosensor “IronSenseR-mCherry” [57]. (E) Comparing iron content by utilizing plasmid-based IronSenseR-mCherry in C. glutamicum cultivated in CGXII with or without 10 μM enterobactin. (F) Schematic depiction of C. glutamicum and E. coli making use of a non-accessible iron pool via E. coli-produced enterobactin. (G) Flow cytometry measurements of liquid co-cultured E. coli-gfp (yellow) and C. glutamicum-E2-crimson (red) under iron-restricted (250 μM DIP) conditions at a ratio of 1:1. Culture was transferred into fresh media after 24 (indicated by ‘). Samples taken at 6 h and 24 h after cultivation start and each 6 h and 24 h after the fresh transfer. Statistical significance was assessed using a one-sample Student’s t-test against the null hypothesis μ = 50%. P values were adjusted for multiple comparisons using the Holm method (P ≤ .05 (*), P ≤ .01 (**)). n = 3 biological replicates. (H) Representative merged stereomicroscopy images of mixed drop (macrocolony) assay on iron-limited CGXII plates with C. glutamicum-E2-crimson (red) and E. coli-gfp (yellow) as indicated in varying ratios and two different seeding densities (10⁻² and 10⁻³). (I) Quantified area of C. glutamicum (red) and E. coli (yellow) deduced from the microscopy pictures in (H), set in relation to 100%. Statistical significance was assessed using a one-sample Student’s t-test against the null hypothesis μ = 50%. P values were adjusted for multiple comparisons using the Holm method (P ≤ .05 (*), P ≤ .01 (**), P ≤ .001 (***), P ≤ .0001 (****)). “ns” thereby implies that the outcome balances towards 50%. Horizontal dashed line represents null hypothesis (50%), while diagonal dashed line represents the initial inoculation. n = 5 biological replicates. Ec = E. coli, Cg = C. glutamicum. (J) Fitness of E. coli in relation to C. glutamicum (selection coefficient) vs. starting frequency of E. coli (based on area (%)) at two seeding densities 10⁻² (dark yellow, circles) and 10⁻³ (light yellow, triangles). p = proportion, f = final, 0 = start.

We next asked whether enterobactin is directly taken up by C. glutamicum and thereby relieves intracellular iron limitation. Using liquid chromatography-mass spectrometry (LC–MS) measurements intracellular enterobactin was readily detectable immediately after addition, while declined over time, indicating rapid uptake under iron limitation (Fig. 2C). However, this fraction may also contain cell-associated enterobactin and does by itself not confirm the uptake of the molecule. Relative to iron-replete conditions, intracellular citrate levels showed a reduced log₂-fold change under enterobactin supplementation compared with the no-iron control, suggesting that enterobactin availability alters citrate-associated iron acquisition in C. glutamicum [61] (Fig. S3). This is also reflected in further TCA cycle intermediates like aconitate and isocitric acid. These results additionally indicate relieved intracellular iron limitation in the presence of the siderophore. Extracellular enterobactin levels decreased over time, but significant amounts were still detectable after 24 h, showing that enterobactin is not fully degraded by C. glutamicum under the tested conditions (Fig. 2C).

To further assess enterobactin-mediated intracellular iron availability, the recently described DtxR-based, ratiometric fluorescent iron biosensor (IronSenseR) [57], was modified for in vivo measurements by exchanging the Large-stokes shift reference domain (LSSmApple) by a more rapidly maturating fluorescent protein (mCherry) (Fig. 2D) and was first validated in vitro (Fig. S4). The modified IronSenseR variant indicated binding to ferrous iron in low micromolar range as it was observed for the original IronSenseR previously [57] thus indicating that the modification did not affect the biosensor function nor its specificity for ferrous iron (Fig. S4). Using this biosensor variant (IronSenseR-mCherry), we observed increased intracellular iron levels upon addition of pure enterobactin, as indicated by an increased GFP/mCherry ratio (Fig. 2E). Together, these results demonstrate that enterobactin is directly exploited by C. glutamicum and relieves intracellular iron limitation (Fig. 2F).

Enterobactin exploitation supports stable pairwise coexistence

As we could demonstrate that C. glutamicum is a versatile xenosiderophore exploiter, we next asked how E. coli-secreted enterobactin exploitation by C. glutamicum unfolds in in vivo co-cultivations. To address this, C. glutamicum and E. coli strains were labelled with constitutively expressed fluorescence reporters E2-Crimson and GFP, respectively, and first analysed in liquid microtiter cultivations using flow cytometry (Fig. 2G). Under iron-restricted conditions and a starting ratio of 1:1, the organisms likewise remain at a balanced coexistence after 48 h, with a slight, but significant overrepresentation of C. glutamicum after 24 h. Within liquid cultivations, enterobactin availability is equally distributed and sustained cell-to-cell contacts are limited due to shaking. We therefore extended the analysis to spatially structured co-cultures, using macrocolonies in which access to siderophores is likely shaped by cell density, spatial distance, and uptake efficiency. To this end, plate-spot assays were performed under both iron-restricted and iron-replete conditions (Fig. S5).

As controls, E. coli WT grew in monoculture under both iron-restricted (3.6 μM FeSO₄ + 100 μM DIP) and iron-replete (36 μM FeSO₄) conditions. In contrast, neither C. glutamicum nor E. coli ΔentC showed growth under iron limitation (Fig. S5A–B). We then co-cultivated C. glutamicum with either E. coli WT or E. coli ΔentC on iron-restricted CGXII agar plates at different starting ratios and two seeding densities (10⁻² and 10⁻³) (Fig. 2H) as initial cell densities could influence exploitation and therefore final community composition [62]. As expected, there was no growth when C. glutamicum and E. coli ΔentC were co-cultivated under iron limitation (Fig. S5C). In co-culture with E. coli WT, prominent growth of C. glutamicum was observed across all inoculation ratios, confirming that C. glutamicum depends on producer-derived enterobactin also in spatially structured environments. The two species grew in intricate spatial arrangements within the macrocolonies, leading to the formation of distinct red or yellow fluorescent patches irrespective of the ratio and seeding density. Furthermore, quantification of species-specific colony areas revealed that enterobactin exploitation remained markedly constrained: At the higher seeding density (10⁻²), which corresponds to closer spatial proximity, co-cultures converged to a balanced (1:1) outcome across most starting ratios, with only a slight shift toward E. coli when it was initially overrepresented (one-sample Student’s t-test against null hypothesis μ = 50%) (Fig. 2I, S5D). These outcomes indicate a stable coexistence maintained by negative-frequency dependent selection [63, 64], as shown by the selection coefficient crossing zero near an initial E. coli frequency of 50% (Fig. 2J). Thus, E. coli has a relative fitness advantage when initially rare, but this advantage decreases as its starting frequency increases.

At lower seeding densities (10⁻³), this balance was less robust. When C. glutamicum was inoculated at equal or lower initial abundance E. coli became dominant, consistent with greater spatial segregation and reduced access to diffusible enterobactin. Even a 5- to 20-fold excess of C. glutamicum did not result in its dominance, but instead the community stabilized near a 1:1 ratio (Fig. 2I).

These results suggest that enterobactin is not equally accessible to both partners in macrocolonies like it is the case for the liquid cultivation. Instead, C. glutamicum appears to benefit only from the fraction of enterobactin that exceeds recapture by the producer, implying that exploitation depends on the presence of an exploitable surplus. This interpretation is supported by the comparison to E. coli WT/E. coli ΔentC co-cultures, in which both strains share the same uptake system and grew largely in proportion to their initial inoculation ratio, independent of seeding density (Fig. S5E). However, when the ΔentC mutant is present in excess, a clear difference from E. coli WT/C. glutamicum macrocolonies could be observed. While the proportion of C. glutamicum never exceeded a level of ~50%, the excess of the enterobactin-deficient E. coli strain retained the initial ratio of the siderophore producer (E. coli WT). This observation suggests that the efficiency of enterobactin utilization and/or uptake is significantly lower in C. glutamicum than in the E. coli ΔentC strain.

Single-cell analysis and modelling reveal siderophore-receptor affinity-constrained exploitation in pairwise co-culture

To resolve the dynamics underlying constrained enterobactin exploitation at the single cell level with higher spatiotemporal resolution, we next analysed co-cultivation of fluorescently labelled E. coli and C. glutamicum as single-layer microcolonies in microfluidic chambers under iron-restricted conditions (Fig. 3A, Table S2, Video S1–S2). As confirmed previously, C. glutamicum grew only in the presence of enterobactin-producing E. coli (Fig. 3B), but neither in monoculture nor in co-culture with E. coli ΔentC (Fig. 3C). These differences were also reflected in the temporal development of total single-cell area obtained by image-based cell segmentation (Fig. 3D).

Figure 3.

Single-cell analysis and modelling reveal affinity-constrained exploitation in pairwise co-culture of C. glutamicum and E. coli. (A) Schematic layout of the microfluidic device used for co-cultivation. (B) Representative segmented frames of the co-culture of C. glutamicum WT and E. coli WT at 0, 7.5 and 15 h in CGXII medium under iron limitation (3.6 μM FeSO₄) at 30°C. (C) Representative segmented frames of the co-culture of C. glutamicum WT and E. coli ΔentC at 0, 7.5 and 15 h under identical conditions. (D) Mean total single-cell area (μm²) over time. Left, C. glutamicum WT (red) and E. coli WT with N = 24; right, C. glutamicum WT and E. coli ΔentC with N = 30. (E) Temporal development of cell area shown separately for each organism. Each line represents an individual chamber. Left, C. glutamicum WT and E. coli WT with N = 24; right, C. glutamicum WT and E. coli ΔentC with N = 30. (F) Experimental and model-derived growth-rate dependencies on colony area, fitted with species-specific parameters across colony sizes and exploiter fractions (minimal, 0–0.33; moderate, 0.33–0.66; high, >0.66; color-coded). Dashed lines indicate simulated data, solid lines experimental data. (G) Model-derived growth landscapes illustrating the effects of initial fraction of E. coli within the community and relative affinity ratios of exploiter and producer (K_s,x2 for C. glutamicum /K_s,x1 for E. coli) for the shared resource. Growth rates represent averages until a total cell area of 2500 μm² is reached. The white line indicates the baseline model assumption.

Tracking species-specific cell area over time revealed that also at this scale E. coli and C. glutamicum converged to a stable community composition, independent of the initial inoculation ratio (Fig. 3E). In the presence of WT E. coli, C. glutamicum accounted for ~70% of the final population area (N = 24 chambers), whereas its fraction remained minimal when enterobactin was absent and E. coli ΔentC dominated through residual growth (N = 30). Although this equilibrium differed from the approximately balanced outcome observed in liquid culture and on plates (Fig. 2G–I), all systems consistently showed stable coexistence at a defined community composition rather than unrestricted exploitation.

To determine whether the population size of C. glutamicum does affect E. coli growth at the microcolony scale due to enterobactin exploitation, growth rates for each organism were calculated between image frames (0.5 h frame rate) across microcolonies of different sizes and compositions (Fig. 3F). Whereas E. coli growth was largely insensitive to colony composition, C. glutamicum growth strongly depended on the relative abundance of exploiter cells and declined sharply once C. glutamicum exceeded ~70% of the colony. This pattern is consistent with the idea that enterobactin remains more readily available to the producer, whereas exploitation by C. glutamicum is less effective.

To conceptually and quantitatively explore the social behavior observed in macrocolonies, we implemented a simplified ODE model that was directly fitted and compared to the experimentally derived single-cell data. To this end, diverging affinities, based on Monod kinetics (kS-values) were assumed to test the possibility of different enterobactin-receptor affinities for both organisms (Fig. 3F and G). Although this model does not directly prove the underlying mechanism, it successfully reproduced the experimentally observed interdependence of growth rates when the exploiter was assumed to have 10-fold lower affinity for the shared resource than the producer and consumption and uptake rates were fitted to the given microfluidic data. The model further predicted that increasing enterobactin-receptor affinities of the exploiter promotes collapse of the producer population and can lead to a “tragedy of the commons” [65], whereas coexistence is stabilized when the producer retains superior access to the shared siderophore (Fig. 3G).The model predicts that very high exploiter affinity is also detrimental to the exploiter itself.

To experimentally assess the postulated difference in affinity, we calculated EC₅₀ (half-maximal effective concentration [66]) values from growth-rate responses to increasing enterobactin concentrations. This analysis indicated a higher apparent affinity for enterobactin in E. coli ΔentC than in C. glutamicum (EC_50-ΔentC = 0.296 ± 0.039 μM; EC_50-Cg = 0.744 ± 0.087 μM; Fig. S5F and G), thus supporting the model-based hypothesis. These values do not directly reflect receptor-siderophore binding affinity, but rather the overall efficiency with which enterobactin exploitation is converted into iron uptake and growth. Together, these findings support our hypothesis in which E. coli captures enterobactin more efficiently, likely via its dedicated high-affinity receptor FepA, while C. glutamicum accesses only a shared and spatially available fraction of the siderophore. This limits exploitation and stabilizes pairwise coexistence, with community dynamics shaped by the interplay of initial composition, siderophore availability, and species-specific affinity.

Pyoverdine acts as a private good in a three species community by preventing C. glutamicum enterobactin exploitation

Siderophore-based microbial interactions in natural contexts are more complex than expressed in one-to-one relationships. To address this in our model system, we extended our public good producer/exploiter system toward a multispecies setting. To this end we chose Pseudomonas putida as a third interacting organism, which produces the mixed-type siderophore pyoverdine. As expected, pyoverdine could not be utilized by C. glutamicum (Fig. 4A, Fig. S6B, D–E). Likewise, E. coli was also unable to utilize pyoverdine, as inferred from co-cultivation assays with the ΔentC mutant (Fig. S6C–E). However, P. putida efficiently exploited enterobactin produced by E. coli—as shown by using the pyoverdine-deficient P. putida ΔpvdD mutant (Figs S6 and S7). The ΔpvdD strain carries a deletion of the gene of the non-ribosomal peptide synthetase PvdD, which is essential for the assembly of the peptide backbone of pyoverdine [67].

Figure 4.

Siderophore-mediated interaction in the trio-consortium of C. glutamicum, E. coli, and P. putida reveals siderophore characteristics as public and private goods. (A) Growth of C. glutamicum with either 50% CGXII-based SM from E. coli WT or P. putida WT. n = 3 biological replicates. (B) Bar plot of fluorescence measured at λ_ex = 398 nm and λ_em = 455 nm from P. putida cultivation with and without enterobactin (EB) and/or DIP supply, representing pyoverdine levels. (C) Comparing iron content by utilizing genomically integrated IronSenseR-mCherry in P. putida ΔpvdD cultivated with or without 10 μM enterobactin in the presence of 250 μM DIP. (D) Bar plot of backscatter (a.u.) at 24 h of C. glutamicum on CGXII-based SM mixes of either 30% E. coli WT with 20% P. putida WT, 30% E. coli WT with 20% P. putida ΔpvdD, 30% E. coli ΔentC with 20% P. putida WT, 30% E. coli ΔentC with 20% P. putida ΔpvdD or 50% P. putida supplemented with 3.6 μM FeSO₄. n = 3 biological replicates. (E) Schematic depiction of the siderophore-mediated interactions within a trio consortium consisting of C. glutamicum (public good exploiter), E. coli (public good provider), and P. putida (private good producer and public good exploiter). (F) Flow cytometry measurements of co-cultured E. coli-gfp, C. glutamicum-E2-crimson, and P. putida-bfp (left) or ΔpvdD (right) under iron limited conditions. n = 3 biological replicates. (G) Representative stereomicroscopy pictures of mixed drop assay on iron-restricted plates with C. glutamicum-E2-crimson (red), E. coli-gfp, and P. putida-bfp, as well as siderophore-deficient mutants, as indicated in two different seeding densities in a 1:1:1 ratio and 16× vs. 80× zoom. (H) Schematic overview on ideal start conditions as inoculated for plate assays. (I) Quantified area of C. glutamicum, E. coli, and P. putida deduced from the microscopy pictures as exemplarily shown in (G), set in relation to 100%. n = 9 biological replicates. (J) E. coli outcome analysis with a ratio of E. coli outcome over initial area in percentage for two seeding densities (10⁻² and 10⁻³) in all duo and trio co-cultivations at 1:1:1 ratio. Statistical significance by Student’s t-test is highlighted (P ≤ .05 (*), P ≤ .01 (**), P ≤ .001 (***), P ≤ .0001 (****)).

Enterobactin uptake by P. putida was further supported by its physiological response. In the presence of pure enterobactin, and likewise in E. coli WT spent medium, P. putida produced less pyoverdine (Fig. 4B, Fig. S7C and D), consistent with reduced intracellular iron limitation. Furthermore, by using the IronSenseR-mCherry variant in P. putida, we observed increased intracellular iron levels in the ΔpvdD strain upon addition of enterobactin (Fig. 4C), confirming that P. putida can acquire iron from enterobactin. Previous studies already showed homology between the enterobactin receptors FepA of E. coli and PfeA of P. aeruginosa [68]. Taken together, the described properties establish P. putida as a new consortium member capable of acting as a second enterobactin exploiter producing the non-exploitable (private) siderophore pyoverdine.

We next tested how simultaneous exposure to enterobactin and pyoverdine affects C. glutamicum growth. To mimic a three-species siderophore environment in monoculture, we combined spent media from E. coli WT or ΔentC with spent media from P. putida WT or ΔpvdD (30% and 20%, respectively; Fig. 4D). E. coli WT spent medium supported growth under iron limitation, but addition of P. putida WT spent medium abolished growth despite the presence of enterobactin. Growth was restored with pyoverdine-deficient ΔpvdD spent medium as well as by iron supplementation. Thus, pyoverdine can act as a private iron-scavenging good that restricts C. glutamicum access to iron, even when exploitable enterobactin is present.

Spatial context determines community composition in the three-species consortium

Having shown that pyoverdine can override enterobactin-mediated growth promotion under iron limitation, we next examined how this interaction shapes a three-species consortium across liquid and spatially structured environments.

In liquid microtiter cultivations under iron restriction, using a 1:1:1 inoculation ratio, WT P. putida became strongly enriched after 24 h (Fig. 4F). P. putida initially declined during the first hours, indicating that its dominance was not immediate but emerged over time. This outcome is consistent with P. putida combining exploitation of enterobactin with pyoverdine production, which limits iron access for the other community members. Accordingly, when pyoverdine production was abolished, the community remained close to the inoculated ratio after 24 h. Conversely, in the absence of enterobactin, WT P. putida prevailed, whereas ΔpvdD-containing combinations showed little to no growth (Fig. S8C and D). Under iron-replete conditions without chelator, P. putida did not become overrepresented (Fig. S8A and B), confirming that its dominance depends on iron limitation.

We next asked whether spatial structure modifies this interaction by restricting siderophore diffusion and access. On iron-limited CGXII plates, WT P. putida again displaced both partners at high seeding density, while C. glutamicum and E. coli persisted at lower abundance than in liquid culture (10⁻²; Fig. 4G, Fig. S9). However, spatial structure also enabled local niche formation. In the ΔpvdD background, E. coli and C. glutamicum grew in close proximity, whereas P. putida ΔpvdD was largely confined to peripheral regions. As expected, loss of enterobactin production in E. coli enabled P. putida to dominate the community. Thus, when diffusion is limited, the balance between public enterobactin and private pyoverdine becomes locally structured rather than uniformly shared across the consortium.

Varying inoculation ratio and seeding density further revealed strong context dependency (Fig. 4G–I, Fig. S10). Under 1:1:1 WT conditions, P. putida became overabundant at the high inoculation density, whereas the lower density and higher initial E. coli abundance improved its persistence, consistent with increased spatial separation and reduced siderophore piracy. At 10⁻³ seeding density, C. glutamicum grew in distinct patches surrounded by E. coli, separating it from P. putida. Without pyoverdine, E. coli and C. glutamicum benefited from high seeding density, whereas lower density maintained compositions closer to the inoculated ratios. Without enterobactin, P. putida remained dominant across conditions, whereas under iron-replete conditions C. glutamicum outcompeted both partners (Fig. S10).

We finally asked the question of what is the impact of exploitation on the producer. We therefore quantified final-to-initial area ratios and demonstrated that P. putida significantly impaired E. coli growth despite the higher affinity of enterobactin according to literature values [9] (Fig. 4J). This effect was strongly reduced in the ΔpvdD background, although its magnitude varied with initial conditions. Overall, P. putida restricts E. coli growth while consistently benefiting from co-cultivation with E. coli, particularly when C. glutamicum is also present (Fig. S10F). By contrast, C. glutamicum exploitation did not affect E. coli growth, consistent with our previous results (Fig. 3F).

To further resolve how initial community composition influences community dynamics, we performed analogous experiments in microfluidic chambers and followed microcolony development, composition and architecture by time-lapse microscopy (Table S2, Fig. 5). Ternary plots revealed that high initial abundances of either C. glutamicum or P. putida resulted in a pronounced shift towards P. putida dominance, often reaching 70%–100% of the final community. By contrast, when E. coli initially represented more than 50% of the community, it persisted at high abundance, ~80%–90%, while also supporting C. glutamicum growth (Fig. 5A, Videos S3–S5). This pattern may reflect spatial segregation that enables E. coli to limit its exposure to P. putida. Replacing WT P. putida with the ΔpvdD mutant reduced P. putida overabundance, with its final proportion stabilizing at ~50% (Fig. 5B, Videos S6–S8). C. glutamicum also grew in some chambers in the presence of WT P. putida, but only when the initial spatial arrangement allowed C. glutamicum patches to form within surrounding E. coli cells, thereby locally separating it from P. putida (Video S3).

Figure 5.

Microfluidic single-cell analyses of community composition dynamics in three-species consortia. (A) Ternary plots showing the relative abundance of C. glutamicum WT, E. coli WT, and P. putida WT. Each point represents a single chamber, positioned according to the fractional cell area of each species. The plots show the initial composition at 0.5 h (top) and the composition after 14.5 h (bottom). Chambers with strong dominance of one species are highlighted by colored outlines and corresponding representative chambers are shown to the right, including their temporal development and segmented images at t = 0.5 h and 14.5 h. (B) As in (A), but with P. Putida ΔpvdD instead of WT. All experiments were performed in CGXII medium with 3.6 μM FeSO₄ at 30°C. All scale bars 10 μm.

Collectively, these findings show that siderophore-mediated interactions in multispecies communities are highly context dependent. In this three-species community, enterobactin produced by E. coli functions as a public good for both C. glutamicum and P. putida, whereas pyoverdine acts as a private good by reducing iron availability for competitors lacking compatible uptake systems. Spatial organization and initial community composition strongly modulate these interactions by determining the extent to which diffusible siderophores are shared, recaptured, or rendered inaccessible.

Discussion

Over the past two decades, many studies of siderophore-mediated social interactions have used relatively simple systems, including interactions among strains of the same species and pairwise interactions between species. These studies have revealed that bacteria can adopt distinct ecological roles as producers, cheaters, or exploiters. Producers synthesize and secrete siderophores, cheaters benefit from siderophores produced by clonemates, and exploiters access xenosiderophores released by other species [2, 23, 69–71]. However, natural microbial communities are not built from isolated pairwise interactions. They contain multiple, phylogenetically diverse species with different siderophore repertoires, uptake systems, affinities, and spatial arrangements. Consequently, the ecological function of a siderophore is not fixed, but emerges from the community context in which it is produced and used [23, 24, 28, 72]. In this study, we deciphered the impact of heterologous siderophores in a microbial community. We demonstrate that enterobactin from E. coli acts as a public good that can support the growth of the exploiters C. glutamicum and P. putida under iron limitation. In contrast, pyoverdine produced by P. putida acts as a private good by inhibiting growth of the competing community members. These results support a broader view of siderophores as context-dependent iron-allocation agents that can either promote coexistence or enforce exclusion, depending on the surrounding community and physical environment.

Corynebacteria are increasingly recognized as common siderophore exploiters in natural microbiomes, yet the extent and ecological consequences of this strategy have remained difficult to dissect mechanistically [73]. By using C. glutamicum, we demonstrate that a non-producing bacterium can exploit chemically diverse xenosiderophores from bacteria and fungi. Such diversity reflects adaptation to iron-limited, siderophore-rich environments and is consistent with the Black Queen hypothesis, whereby organisms can benefit from losing costly functions, if these are provided by other members of the community [74, 75]. Related siderophore-mediated interactions have previously been studied in Pseudomonas, including interactions among isolates differing in siderophore production and exploitation [10, 24, 76]. These studies established important principles of siderophore-mediated competition and coexistence within closely related taxa. Here, we extend this framework to a synthetic community of phylogenetically distant species with distinct siderophore production and uptake repertoires, allowing us to examine how heterologous siderophores simultaneously mediate sharing, exploitation, and exclusion.

Siderophore systems have long served as model systems for studying microbial cooperation [77], as also reflected in the interaction between E. coli and C. glutamicum observed here. Across structured plates and microfluidic chambers, exploitation never resulted in a collapse of the producer population. Instead, C. glutamicum-E. coli co-cultivation reproducibly converged to stable community compositions, suggesting that enterobactin exploitation by C. glutamicum is constrained. This dependency resulted in negative frequency-dependent selection, as typical for siderophore mediated interactions, and thereby promoted stable coexistence between producer and exploiter [2, 78].

Modelling and EC₅₀ analyses further suggest that this stability may be related to asymmetric access to enterobactin. The producer E. coli, carrying the cognate FepA receptor for uptake, may capture enterobactin more efficiently than the exploiter C. glutamicum. In this scenario, the exploiter accesses only the fraction of enterobactin that escapes producer recapture. This is consistent with the established ecological principle that stable exploitation of a diffusible public good can be promoted by asymmetric access, such that the producer retains privileged uptake [79, 80]. With limited dispersal and dependent on distances, public goods tend to stay in the vicinity of the producer, allowing a benefit from own production [79]. Previous models have described a “public good dilemma”: for a successful cooperation the overall sum of the benefit for the producer plus the benefit received by the exploiter must be bigger than the overall cost of production [80]. This interpretation is consistent with previous work, showing that enterobactin can exhibit density-dependent privatization, remaining partially cell-associated at low cell densities and becoming increasingly shared as extracellular enterobactin accumulates [81]. If exploiter public good access becomes too efficient, the shared resource may no longer stabilize coexistence but instead drives a “tragedy of the commons” scenario as is often observed during interactions between producers and cheaters. In this scenario, cheaters can still take up the corresponding siderophore-iron complexes via the same receptors and thus with the same efficiency as the producers [2, 65, 72]. Although this is not observed during interaction between C. glutamicum and E. coli, a related stabilizing principle has been proposed for cross-feeding mutualisms. When exploiter and producer have similar uptake affinities, stable coexistence may still be maintained if the exploiter reciprocates by providing a public good or another trait of comparable ecological value [82–84]. Exemplarily, this was recently observed in Burkholderia contaminans making use of the siderophore bacillibactin produced by Bacillus velezensis, while providing branched-chain amino acids [36].

Finally, the three-species community revealed an additional layer of ecological complexity beyond the stable coexistence. Although P. putida can exploit enterobactin, its own pyoverdine was inaccessible to both C. glutamicum and E. coli. As a result, pyoverdine abolished the growth benefit that enterobactin otherwise provided to the non-producing exploiter. This is especially interesting considering enterobactin has been reported to possess a higher binding affinity to iron (pFe ≈ 34.3) [85, 86] than pyoverdine (pFe ≈ 27) [87]. Thus, our observation underlines that the ecological outcome of multispecies communities cannot be predicted solely from siderophore binding affinities. Instead, additional factors must be considered, such as siderophore production rates, since the equilibrium of iron binding can shift toward a siderophore with lower binding affinity if produced at sufficiently high quantities [88]. It further depends on how efficient the iron-siderophore complex can be accessed by individual species. This mechanism is certainly relevant beyond the here described model consortium, particularly in communities containing pyoverdine-producing pseudomonads and taxa with only partially overlapping siderophore networks [24, 89]. In addition, differences in the fate of siderophores after uptake may influence community dynamics: In E. coli, ferric enterobactin is transported into the cytoplasm of the cell and hydrolysed via an esterase to facilitate iron release. In contrast, pyoverdine remains in the periplasm after uptake and can again be secreted after iron delivery [90, 91]. However, whether and how enterobactin is degraded, modified, or otherwise processed after exploitation by C. glutamicum and P. putida remains unknown and will require further investigation. Overall, our observations in the tripartite community show that the same siderophore network can produce different ecological outcomes depending on spatial arrangement and starting community composition [92]. The molecules remain the same, but the accessibility landscape changes. Well-mixed environments may favor broader sharing of diffusible siderophores, whereas structured habitats can promote local producer recapture, exploiter niche formation, or exclusion by non-exploitable siderophores such as pyoverdine [79, 93, 94].

Beyond that, siderophore-mediated interactions are only one part of the underlying community structure. Identity effects, i.e. strain-specific traits like baseline growth differences, resource-use efficiency or even general iron demand, further shape the functioning of the community [95]. Although recent work investigated that strain identity effects account for even more variation in microbial communities than interaction effects do, it is also claimed that this gap reduces rapidly in heterogeneous communities including public goods like siderophores [96].

Elucidating interspecific siderophore-mediated interactions in complex microbial communities is highly relevant, as previous studies already highlighted competition for iron as a driver for eco-evolutionary dynamics in natural environments [94, 97, 98]. For example, studies of plant root-associated microbiomes in tomato rhizospheres demonstrated that siderophores can alter microbiome–pathogen interactions resulting in either infection facilitation or pathogen suppression depending on the siderophore types [99]. Moreover, it was shown that higher diversity communities yield in favored siderophore production [100].

Although the three species of this study were not co-isolated from one habitat, they represent ecologically relevant siderophore interaction types that may occur in soil-, rhizosphere-, or host-influenced microbial communities. We therefore use them as a tractable synthetic consortium, rather than as a direct reconstruction of a natural community, to facilitate deciphering distinct modes of siderophore-mediated interactions in spatially structured environments in a bottom-up approach. More broadly, our findings show that siderophore chemistry, uptake compatibility, relative affinity, and spatial structure together determine, which species benefit from (heterologous) iron-siderophore complexes and which are excluded. This framework provides a basis for understanding iron competition in natural microbiomes and for designing siderophore-shaped synthetic communities in biotechnology, agriculture [101, 102], and host-associated systems [103].

Declaration of generative AI and AI-assisted technologies in the writing

During the preparation of this work the authors used ChatGPT in order to improve language of the manuscript. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the manuscript.

Conflicts of interest

We declare no conflict of interest.

Funding

The research was supported by the German Research Foundation (DFG) through the Collaborative Research Center 1535 Microbial Networking (project ID 45809666). The Center for Structural Studies is part of StrukturaLINK Rhein-Ruhr which is funded by the Deutsche Forschungsgemeinschaft (DFG Grant number 573727698) and INST 208/761–1 FUGG. The CMML is supported by the Cluster of Excellence on Plant Sciences (CEPLAS) under Germany’s Excellence Strategy EXC-2048/1 under project ID 390686111. The research visit of A.K. in Leiden was supported by an EMBO Scientific Exchange Grant (12048). Á.T.K. and the position of R.A. were supported by the European Research Council under the European Union’s Horizon 2020 research and innovation program (ERC Synergy, MicroClock, 101166968). Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Research Council Executive Agency. Neither the European Union nor the granting authority can be held responsible for them.

Data availability

The datasets generated and/or analysed during the current study are included in this published article and its supplementary information files. Further, full raw data sets can be found in the ARC Data Hub https://archive.nfdi4plants.org/records/b6rne-tnd39. It is available under the DOI: 10.60534/yt353-ycp92. The code for modelling has been deposited in a public repository (https://jugit.fz-juelich.de/m.pesch/race-for-iron-model-code.git), ensuring full reproducibility of all results presented in this study. Additionally, the model has been deployed in a browser and can be interactively interrogated under computational-biology-aachen.github.io/mxl-web/dynamic-entrobactin.