Metagenomic Profiling and Genome‐Centric Analysis Reveal Iron Acquisition Systems in Cheese‐Associated Bacteria and Fungi

ABSTRACT

Cheese microbial communities are composed of diverse interacting microorganisms, including both inoculated and non‐inoculated strains. One limiting factor for microbial growth on cheese surfaces is iron availability. To better understand the role of iron acquisition in cheese microbial ecology, we investigated the diversity and distribution of iron uptake systems across a wide range of cheeses. We analysed 136 metagenomes and 1400 genomes and Metagenome‐Assembled Genomes (MAGs) from 44 French Protected Designation of Origin (PDO) cheeses. Using an updated set of Hidden Markov Models targeting iron acquisition genes, we identified a wide diversity of iron uptake systems. Siderophore biosynthesis and import systems were more prevalent in surface‐associated species than in those from the cheese core. About 20 different siderophore biosynthesis pathways were detected, with desferrioxamine and enterobactin‐type being the most prevalent. Genomic analyses revealed the main bacterial and fungal producers, including Glutamicibacter, Corynebacterium, Staphylococcus, and Penicillium. While siderophore biosynthesis pathways were found in a minority of MAGs, iron/siderophore import systems were widespread, suggesting the potential for cross‐feeding interactions involving siderophores. These findings enhance our understanding of microbial interactions in cheese and open perspectives for improving ripening cultures by considering iron acquisition traits.

This study describes the diversity of iron acquisition systems in cheese microorganisms and provides information on the distribution of siderophore biosynthesis pathways in both bacteria and fungi, including poorly characterised species.

1. Introduction

Cheese contains bacteria, yeasts, and moulds that contribute to the development of the typical organoleptic properties of the final product while also limiting the growth of pathogens and other undesirable microorganisms. Cheese microbial communities are very diverse, especially on the surface (Wolfe et al. 2014; Irlinger et al. 2015; Irlinger et al. 2024), where the microbial density is often higher than in the core of the cheese because of oxygen availability. Surface microorganisms play an essential role in the ripening of surface‐ripened cheeses due to their ability to consume lactic acid produced by lactic acid bacteria and to degrade caseins and fats from milk. In cheese, microorganisms come from deliberately inoculated cultures (commercial cultures, whey cultures), as well as from milk, brine, and the cheese‐making environment (Mounier et al. 2006). In some cases, uninoculated microorganisms are even able to outcompete those that are deliberately inoculated (Feurer et al. 2004; Goerges et al. 2008; Sun and D'Amico 2021). Understanding how these microorganisms interact with each other and with the cheese matrix to form microbial communities of varying complexity and whose structure changes during the manufacturing process remains a challenge.

Selection is one of the processes described in the literature that provides a better understanding of microbial community assembly phenomena in fermented foods. It represents the influence of the abiotic environment (the properties of the environment) and the biotic environment (microbial interactions) on the fitness of cells in a microbial community (Louw et al. 2023). In cheese, iron acquisition by microorganisms is the result of both biotic and abiotic selection processes (Mekuli et al. 2025). Cheese is a highly iron‐restricted habitat because of its low iron concentration (usually a few mg/kg), the fact that it is present as the poorly soluble Fe³⁺ form, and the presence of iron‐binding molecules originating from milk, such as lactoferrin (Mekuli et al. 2025). The growth of ripening bacteria of the genera Glutamicibacter, Corynebacterium, and Brevibacterium on cheese depends on the availability of iron (Monnet et al. 2010; Monnet et al. 2012). In a model cheese, iron supplementation stimulated the growth of particular species within a reconstituted microbial community composed of nine strains (Shoukat et al. 2025). In addition, biotic interactions involving iron acquisition have been identified in cheese models involving reconstituted communities composed of Hafnia alvei , Brevibacterium aurantiacum , and Debaryomyces hansenii (Pham et al. 2019), Staphylococcus and Scopulariopsis (Kastman et al. 2016), and Penicillium and Staphylococcus (Ye et al. 2023).

In order to survive in environments with limited iron availability, such as on the surface of cheese, microorganisms use various strategies for iron uptake, including the reduction of ferric iron and the uptake of ferrous iron, the acquisition of ferric citrate, and the import of iron via siderophores (Andrews et al. 2003). Siderophores are low‐molecular‐weight compounds that are synthesised by microorganisms to chelate iron with high affinity (Andrews et al. 2003; Krewulak and Vogel 2008). Based on their iron‐chelating groups, they can be categorised as hydroxamate, catecholate, carboxylate, or mixed families (Andrews et al. 2003; Timofeeva et al. 2022; Mekuli et al. 2025). Once secreted, they chelate iron, and the iron/siderophore complexes are then imported into microbial cells via specialised transport systems. In bacteria, this typically involves an ABC transporter system comprising a substrate‐binding protein, a permease, and an ATP‐binding cassette (Andrews et al. 2003; Saha et al. 2013; Mekuli et al. 2025). In Gram‐negative bacteria, TonB‐dependent receptors mediate substrate‐specific transport across the outer membrane. In fungi, iron can be imported via reductive pathways involving iron reduction or via non‐reductive systems utilising specific iron/siderophore complex transporters (Philpott 2006).

The mechanisms of iron acquisition have been widely studied in microorganisms of clinical interest such as Salmonella species, Shigella species, Mycobacterium tuberculosis, Staphylococcus aureus, Vibrio cholerae , and Pseudomonas aeruginosa (Wyckoff et al. 2007; Chao et al. 2019; Mey et al. 2021; Ghssein and Ezzeddine 2022; Schalk and Perraud 2023). However, there is a lack of in‐depth knowledge about the iron acquisition capacities of the microbial species present in food, including cheese. Such knowledge would be helpful, firstly, to gain a better understanding of the biotic and abiotic selection phenomena involved in the assembly of microbial communities in cheeses and, secondly, to facilitate the development of ripening cultures with a high colonisation capacity of the cheese matrix. A recent analysis characterised the bacterial and fungal communities of more than 2000 microbiota from French Protected Designation of Origin (PDO) cheeses (Irlinger et al. 2024). These cheeses contained many microbial species that had not been deliberately inoculated. Some of these samples were subjected to metagenomic sequencing, which allowed the reconstruction of 1119 bacterial MAGs belonging to seven phyla (Gardon et al. 2025). Several studies have demonstrated the interest of metagenomic analyses for the study of iron metabolism in aquatic environments (Garber et al. 2020; Garber et al. 2021; Zhang et al. 2023; Thoppil et al. 2025). In the present study, we aimed to gain a better understanding of the iron acquisition capacities in cheese microbial communities by analysing the metagenomic data available for French PDO cheeses (Gardon et al. 2025). For that purpose, the functions of siderophore biosynthesis, transport of iron/siderophore complexes, and other iron import systems were investigated using a Hidden Markov Model‐based approach on metagenomics contigs, metagenome‐assembled genomes (MAGs), and genomes from bacterial isolates. The last two approaches are genome‐centric approaches and allow us to identify both the bacterial and fungal taxa involved in iron transport and biosynthesis in cheese.

2. Materials and Methods

2.1. Metagenomic and Genomic Dataset

A total of 136 cheese metagenomes were obtained from the MetaPDOCheese study (Gardon et al. 2025, BioProject PRJEB64630). The dataset includes metagenomes from 44 different French PDO cheeses, along with associated metadata. The samples were categorised according to different cheese components, namely origin (rind (n = 115) and core (n = 21)), milk‐producing species (Dairy Species: cow (n = 105), goat (n = 27), and ewe (n = 4)), and cheese technological families (Hard Cooked cheese (HC; n = 12), Lactic Bloomy rind (LB; n = 28), Lactic Washed rind (LW; n = 9), Soft Bloomy rind (SB; n = 5), Soft Washed rind (SW; n = 26), Internal Blue mould (IB; n = 27), and Uncooked Pressed cheese/Semihard cheese (UP; n = 29)) (Table S1). One metagenome (AOP44_E1_core) was excluded from the dataset due to an abnormally low number of sequencing reads.

Additionally, 1105 bacterial Metagenome‐Assembled Genomes (MAGs) reconstructed from these metagenomes using the SnakeMAGs workflow (Tadrent et al. 2023), and 280 genome sequences from bacterial isolates of the 44 different French PDO samples were also obtained from the MetaPDOCheese study (Gardon et al. 2025, BioProject PRJEB64630 and PRJNA1308522). The MAGs were further categorised according to sampling location (Origin: rind (n = 1016) and core (n = 89)), milk‐producing species (Dairy Species: cow (n = 952), goat (n = 116), and ewe (n = 37)), and cheese technological families (HC (n = 234), LB (n = 100), LW (n = 63), SB (n = 18), SW (n = 246), IB (n = 170), and UP (n = 274)) (Table S2). Genomes from cheese isolates were categorised according to milk‐producing species (Dairy Species: cow (n = 225), goat (n = 51), and ewe (n = 4)), and cheese technological families (HC (n = 23), LB (n = 43), LW (n = 28), SB (n = 17), SW (n = 49), IB (n = 41), and UP (n = 79)) (Table S2). All these strains were isolated from rind samples.

2.2. Fungal MAG Recovery and Classification

In the MetaPDOCheese study (Gardon et al. 2025), 1275 MAGs were eliminated because they did not pass the prokaryotic contamination and completeness filters of CheckM (Parks et al. 2015). We recovered these MAGs and checked them for fungal markers (n = 255) using BUSCO (v5.3.2) (Manni et al. 2021) with the following options: ‐‐auto‐lineage‐euk ‐m genome, resulting in the selection of 158 fungal MAGs with a quality score of at least 50. Taxonomic assignment of these MAGs was then performed using a phylogenetic approach. For this purpose, we used UFCG v1.0.6 (Kim et al. 2023) to search and extract 62 fungal core genes in the selected MAGs and in reference genomes. For each marker, sequences from the MAGs, reference genomes, and from the UFCG database (n = 1587) were aligned with MAFFT ‐‐auto (v7.487) (Katoh and Standley 2013) and trimmed with trimal (v1.4.1) (Capella‐Gutiérrez et al. 2009) with the ‐gt 0.1 option. The concatenation of each marker's alignment was used to infer a maximum likelihood tree using IQ‐TREE (v2.2.0.3) (Minh et al. 2020) with the following options: ‐‐seqtype AA ‐m TEST ‐bb 1000 ‐alrt 1000. ModelFinder was used to find the best‐fitting model of evolution (Kalyaanamoorthy et al. 2017). Following the same method, another tree was produced using only the MAGs and the UFCG reference, Rhizoclosmatium globosum (GCA_002104985.1), to root the tree; the R. globosum reference was subsequently deleted in the final tree. The 158 fungal MAGs were further categorised according to sampling location (Origin: rind (n = 156) and core (n = 2)), milk‐producing species (Dairy Species: cow (n = 123), goat (n = 33), and ewe (n = 2)), and cheese technological families (HC (n = 0), LB (n = 38), LW (n = 15), SB (n = 6), SW (n = 20), IB (n = 54), and UP (n = 25)). The Fungal MAGs are available under NCBI BioProject PRJEB64630, BioSample: SAMN50110840‐SAMN50110997.

2.3. Compilation of Public Hidden Markov Models Related to Iron Acquisition

To identify genes associated with iron acquisition systems, 148 Hidden Markov Models (HMMs) were collected from the KOfam database (Aramaki et al. 2020) (version 2022‐03‐01), FeGenie (Garber et al. 2020), and Protein Family Models from the NCBI database (Li et al. 2021) (Table S3). Two HMM categories were defined: “Siderophore biosynthesis” and “Import.” The latter category corresponds to importers of iron or iron/siderophore complexes. However, ATPase and permease components of ABC‐type iron/siderophore transport systems, even if present in metagenomes and genomes, were not considered in the present study as they are less specific for iron transport systems than the substrate‐binding protein components. Only the substrate‐binding proteins and the receptors were considered for the detection of siderophore transport pathways.

Due to an abnormally high number of hits, four HMMs (PvdDIJL‐VabF‐VibF‐EntF‐DhbF‐MbtBEF‐PchE‐AngR‐PchF, PvsE‐AcsE_vibrioferrin_biosynthesis‐rep and Sid_RhbA_Rhizobactin_biosynthesis_Rhizobium_meliloti_Q9Z3R2 from FeGenie, and K02364 from KOfam) had to be recalibrated in order to improve their specificity. This was done by querying the four HMMs against the coding DNA sequences (CDSs) predicted from the 136 cheese metagenomes using hmmscan (HMMER v3.3.2) (Eddy 2011). The results were manually inspected by checking the hits with BLASTp analysis (Camacho et al. 2009). The bitscore thresholds were then recalibrated in order to remove false positives.

2.4. Creation of New HMMs Related to Iron Acquisition

Reference sequences, primarily composed of proteins with experimentally characterised functions, were gathered from UniProtKB and NCBI. To enhance the diversity of each protein set, these sequences were used as queries in a BLASTp search against the NCBI RefSeq and the UniProtKB/SwissProt databases. Sequences with a minimum amino acid identity of 45% over at least 70% of the query length were retained. The retrieved sequences were then de‐replicated using MMseqs2 (v14.7e284) (Steinegger and Söding 2017) with a 99% amino acid identity threshold to avoid over‐representation of the same protein sequence. Multiple sequence alignments were performed using MAFFT (v7.487) with the ‐‐auto option, followed by manual inspection and curation. In order to identify and remove false positives, phylogenetic trees with reference sequences were constructed using IQ‐TREE (v2.2.0.3), with the following options: ‐‐seqtype AA ‐m TEST ‐bb 1000 ‐alrt 1000. The curated alignments were subsequently used to generate HMMs with hmmbuild (v3.3.2). To determine optimal bitscore cutoff values for each HMM, the models were queried against a set of reference sequences and the 136 cheese metagenomes CDSs using hmmscan (v3.3.2). By manually inspecting the results, we established bitscore thresholds that best distinguish true from false positives for each HMM (see Table S3).

2.5. HMMs Search in Genomes, Metagenomes and MAGs

The set of 185 hidden Markov models (HMMs) was queried against the predicted CDSs from 136 metagenomes, 1105 microbial genome assemblies (MAGs), and 280 strain genomes using hmmscan (v3.3.2) with the e‐value (−E) filter at 1e‐5. These CDSs were then filtered by predefined bitscore cutoffs for each HMM (see Table S3), as well as by the length of the CDS recognised by the HMM (over 70%). Similarly, a subset of 29 HMMs was queried against the CDSs predicted from the 158 fungal MAGs. To determine the coverage of selected CDSs in metagenomes, reads were mapped to the CDSs using Bowtie2 (v2.4.4) (Langmead and Salzberg 2012), and the resulting SAM files were then converted into sorted and indexed BAM files using SAMtools (v1.14) (Li et al. 2009). To obtain a single normalised abundance value per HMM in each metagenome, a weighted mean of the HMM length (HMM_length) was calculated, taking into account the coverage of each CDS and its length. The RPKM_HMM (Reads Per Kilobase Million) value was then calculated using the following formula:

$$\mathrm{RPK}{\mathrm{M}}_{\mathrm{HMM}}=\frac{\sum \text{reads\hspace{0.17em}of\hspace{0.17em}CDSs}}{\frac{\text{total\hspace{0.17em}reads}}{1000000}\times \frac{\mathrm{HM}{\mathrm{M}}_{\text{length}}}{1000}}$$

Each HMM was also assigned to categories to distinguish genes involved in siderophore biosynthesis or import (iron‐siderophore complexes or other iron acquisition systems) (see columns “Iron acquisition system” and “Category” in Table S3). Cumulated RPKM values, designated RPKM_groups, were calculated by summing the RPKMs of HMMs from the same functional category (e.g., pyoverdine synthesis). HMMs recognising multiple siderophores were grouped into the Multi‐siderophore category. Due to shared HMMs recognising enterobactin, vanchrobactin, bacillibactin, and vibriobactin genes, all HMMs associated with these siderophores were grouped under the same category, designated “Enterobactin‐Type”. All the HMMs are available on GitHub (https://github.com/STabuteau/iron_cheese_metagenomes).

mOTUs (v3.0.3) (Ruscheweyh et al. 2022) and EukDetect (v1.3) (Lind and Pollard 2021) were used to quantify bacterial and fungal abundances in the metagenomes. RPKM_HMM and RPKM_groups abundances were correlated with the prokaryotic and fungal abundances by Pearson's correlation coefficient (r) analysis. Cumulated RPKM values, designated RPKM_global, were also determined by summing the RPKM values of HMMs from higher functional levels (siderophore biosynthesis, siderophore import, and iron import).

Assessment of the presence or absence of a siderophore biosynthesis pathway in a given metagenome, MAG, or genome was based on the detection of the genes involved in the biosynthesis pathway. A completeness ratio for each biosynthesis pathway expressed in percentage was calculated by dividing the number of genes detected by HMM analysis by the number of genes known to be involved in the biosynthesis pathway (Table S3). Only biosynthesis pathways with ≥ 55% completeness were estimated to be present in metagenomes, MAGs, and genomes.

2.6. Statistical Analysis and Data Visualisation

Statistical analysis to compare the RPKM_global values of core and rind metagenomic samples was done with the ggpubr (v0.6.0) package using the Wilcoxon test. Kruskal–Wallis tests followed by Bonferroni post hoc tests were performed to compare the counts of import systems between phyla using the emmeans (v1.11.0) and multcomp (v1.4‐28) packages.

Visualisation of data was performed using R (v4.4.1) with the tidyverse collection package (v2.0.0) (Wickham et al. 2019). The heatmaps were generated using the ComplexHeatmap package (v2.22.0) (Gu 2022) with Pearson distance as the distance function. Metagenomes metadata were added to the heatmaps, as well as taxa composition of metagenomes determined by Kaiju analysis (v1.7.3) (Menzel et al. 2016) with the “nr database (+eukaryotes)” (v2020/05) as a reference database, and default parameters.

All the trees were visualised on iTOL (v7) (Letunic and Bork 2024). For the fungal tree, metadata were added to the tree. Fungal MAG completeness data were obtained with BUSCO (v5.3.2). Relative abundances of fungal MAGs were obtained with CoverM (v0.6.1) (Aroney et al. 2025). Information on iron acquisition genes was obtained with hmmscan (v3.3.2) used on fungal MAG CDSs predicted with the easy‐predict mode of MetaEuk (v6.a5d39d9) (Levy Karin et al. 2020) using the UniRef100 database (v2024_04) and the following options: ‐‐metaeuk‐eval 0.0001 ‐‐metaeuk‐tcov 0.6 ‐‐min‐length 20.

Figures were edited with Inkscape software v1.3.2 (https://inkscape.org/).

3. Results

3.1. Overall Survey of the Iron Acquisition Systems in the Cheese Metagenomes

We investigated iron acquisition systems in a subset of metagenomes produced from a large‐scale analysis of over 2000 microbiota from French cheeses, registered under the Protected Designation of Origin (PDO) cheeses, produced and ripened in a defined geographical area, and covering most of the cheese families made in the world (Irlinger et al. 2024; Gardon et al. 2025). The dataset consisted of 136 cheese metagenomes from 44 different PDOs, including 115 rind samples and 21 core samples. To identify iron acquisition systems, we searched for genes involved in direct ferrous and ferric iron import, as well as genes associated with siderophore biosynthesis and import. We identified genes by homology, using 185 Hidden Markov Models (HMMs). Among them, we collected 148 HMMs from various sources. In some cases, we recalibrated them in order to improve their specificity and sensitivity (Table S3). Since the available HMMs did not cover some well‐known iron acquisition systems, we constructed and calibrated 37 additional HMMs, mainly for the biosynthesis of fungal siderophores (ferricrocin, fusarinine C, coprogen, ferrichrome) and their import systems into bacteria.

We detected a total of 147 out of the 185 HMMs related to iron acquisition mechanisms in the cheese metagenomes (Figure S1), representing 0.53% of all the CDSs of the dataset. The abundance profiles for groups of HMMs corresponding to siderophore biosynthesis and iron/siderophore, ferrous and ferric iron, or ferric citrate import systems show heterogeneity between samples (Figure 1). Some siderophore biosynthesis systems, such as enterobactin‐type biosynthesis (comprising enterobactin, vanchrobactin, vibriobactin, and bacillibactin) and desferrioxamine biosynthesis (corresponding to three HMM groups in Figure 1), are present in the majority of samples. In contrast, other siderophore HMMs, such as proteobactin or 7‐hydroxytropolone, are present in only a few samples. The latter is present mainly in goat cheese samples. Regarding import systems, ferrous and ferric iron importers, as well as multi‐siderophore import systems, are present in the majority of samples. Other import systems are more specific, such as carboxylate family siderophore importers, as well as salmochelin, pyochelin, and pyoverdine transporters, which are limited to a subset of samples.

FIGURE 1.

Heatmap of the abundance of CDSs mapping to groups of iron acquisition HMMs in the cheese metagenomes. Abundances are expressed as RPKM (reads per kilobase per million mapped reads). Siderophore biosynthesis systems are shown in the upper part of the heatmap, whereas siderophore import systems and other iron acquisition systems are in the lower part. The enterobactin‐type group includes enterobactin, vanchrobactin, vibriobactin, and bacillibactin. Phylogenetic composition data result from the taxonomic classification of the metagenomic reads using Kaiju. The samples (columns) were clustered with the Pearson method. Cheese technological families: Hard Cooked cheese (HC), Internal Blue mould (IB), Lactic Bloomy rind (LB), Lactic Washed rind (LW), Soft Bloomy rind (SB), Soft Washed rind (SW), Uncooked Pressed cheese/Semihard cheese (UP).

The clustering of the metagenomic samples reveals differences between groups of samples based on siderophore biosynthesis and iron/siderophore, ferric citrate, ferric, and ferrous iron import systems. This clustering seems more consistent with the phylogenetic composition (taxa) and sample origin (core/rind) than with the milk‐producing species or the cheese technological family (Figure 1). Rind samples exhibit higher diversity and abundance than core samples, both for siderophore biosynthesis and import systems. Core samples are dominated by Bacillota, which, in cheeses, mainly correspond to lactic acid bacteria. In about half of the core samples, there is also a high abundance of the ferrous/ferric iron import category, which corresponds to the Efe transport system, active on both ferrous and ferric iron (Miethke et al. 2013). There is a strong correlation (Pearson's correlation coefficient = 0.78) between the Bacillota phylum and the ferrous iron import category (Figure S2).

By aggregating the HMM abundances in the three categories, “siderophore biosynthesis”, “siderophore import” and “direct iron import” (ferrous and ferric import), we found that the latter category is more abundant in core samples than in rinds. In contrast, we observed the opposite for the two siderophore categories (Figure S3).

Samples dominated by the Ascomycota fungi display greater diversity in hydroxamate‐family siderophore biosynthesis categories than samples dominated by Pseudomonadota and Actinomycetota (Figure 1). We observed the opposite trend for the mixed and catecholate siderophore families. This is supported by correlation analyses, which associate Ascomycota with the biosynthesis of some specific hydroxamate siderophores (pulcherrimin, fusarinine C, coprogen‐related siderophores, and ferricrocin), while Actinomycetota and Pseudomonadota are associated with some catecholate and mixed‐family siderophore biosynthesis (Figure S2). In addition, Pseudomonadota abundance is correlated with the biosynthesis of the hydroxamate‐family siderophore desferrioxamine (corresponding to three HMM groups in the figure). Concerning iron/siderophore importers, samples dominated by Pseudomonadota and Actinomycetota displayed a higher diversity and abundance compared to those dominated by Ascomycota (Figures 1 and S2). Similarly, correlation between phylum abundance and individual HMM abundances (Figure S4), as well as correlations between microbial families or species with the iron acquisition categories or the individual HMM abundances, reveal putative associations that will be further investigated at the genome level (Figures [Link], [Link]).

3.2. Analysis of Siderophore Biosynthesis Pathways in the Cheese Samples

In order to better assess the siderophore biosynthesis capacities in the cheese samples, we searched for complete siderophore biosynthesis pathways among the CDSs identified by the HMM searches. This analysis only includes siderophores for which the biosynthesis pathways and corresponding genes have been experimentally determined (listed in the sheet HMM_list from Table S3). It was conducted on metagenomes as well as on bacterial and fungal MAGs (1105 bacteria and 158 fungi). Considering that gene clusters in metagenomes and MAGs can be fragmented and that some genes may be missing (e.g., incomplete MAGs), we took biosynthesis pathways with at least 55% of the detected genes into account (Tables S1 and S2).

Siderophore biosynthesis pathways are present in nearly all rind metagenomes (112/115) (Figure 2). The three metagenomes without detected biosynthesis pathways are from the Lactic Bloomy rind (LB) technological family. The most frequent siderophores are from the catecholate and the hydroxamate families, followed by the carboxylate and the mixed families, which are present in about half of the metagenomes. The most commonly detected siderophores biosynthesis pathways are enterobactin‐type and desferrioxamine, identified in 86% and 75% of the metagenomes, respectively. Some other siderophores such as 7‐hydroxytropolone and proteobactin are less frequent in metagenomes. There are some variations in siderophore distribution across technological families. Carboxylate family siderophores are less prevalent in the Bloomy rind (Lactic Bloomy rind and Soft Bloomy rind) and Lactic Washed rind technological families. Pathways for fungal hydroxamate family siderophores (coprogen, ferricocin, fusarinine C and pulcherimin) are absent from Hard Cooked and Soft Washed rind families and are only occasionally present in the Lactic Washed rind technological family. The prevalence of pathways for siderophores from the mixed family considerably varies depending on the cheese technology. These siderophores are absent in the 25 Internal Blue mould metagenomes and present in only one of the four Soft Bloomy metagenomes.

FIGURE 2.

Siderophore biosynthesis pathways detected in the 115 metagenomes of rind samples and in the 1172 bacterial and fungal MAGs obtained from the corresponding metagenomic sequences. Typical fungal siderophores are indicated in red. The enterobactin‐type group includes enterobactin, vanchrobactin, vibriobactin and bacillibactin. The numbers of metagenomes or MAGs containing at least one siderophore pathway from a given family are represented by the global siderophore family lines. Cheese technological families: Hard Cooked cheese (HC), Internal Blue mould (IB), Lactic Bloomy rind (LB), Lactic Washed rind (LW), Soft Bloomy rind (SB), Soft Washed rind (SW), Uncooked Pressed cheese/Semihard cheese (UP).

Among the 1172 MAGs, we detected at least one siderophore biosynthesis pathway in 311 MAGs (27%). In most cases, within a given technological family, the biosynthesis pathways that were detected in metagenomes were also detected in MAGs, except for 7‐hydroxytropolone, proteobactin and vibrioferrin, which were not or only poorly detected in MAGs. We detected fungal siderophores biosynthesis pathways in some of the fungal MAGs (indicated in red in Figure 2). We detected the rhizoferrin biosynthesis gene in MAGs but not in metagenomes. Overall, there is a quite similar pattern of siderophore distribution across technological families in the MAGs and the metagenomes, indicating that our MAGs cover most of the microbial diversity of the samples. In some cases, for example, for enterobactin‐type siderophores, the number of MAGs that encode the corresponding pathway is larger than the number of metagenomes in which the pathway is detected, which suggests that in some samples, there are several different producers of the same siderophore.

Siderophore biosynthesis data from MAGs aggregated according to metagenomic samples (Figure S9) show that 71% of the cheese samples contain several siderophore pathways (11% with one siderophore pathway and 24% with none). We found the highest number of pathways in the MAGs from cheese AOP29_B2_surf, which totalise nine biosynthesis pathways (mostly catecholate siderophores). The proportion of samples containing multiple siderophore families is 54%. In most cases, a particular siderophore pathway that is detected in a metagenome is only encoded by a single MAG (Figure S10). The pathways encoded by several MAGs mainly correspond to the enterobactin‐type category, which includes enterobactin, bacillibactin, vanchrobactin and vibriobactin, which were difficult to distinguish due to shared HMMs.

In cheese core samples, siderophore biosynthesis pathways are present in half of the metagenomes, but only in about 6% of the MAGs (5/91 MAGs) (Figure S11). The siderophore diversity is much lower than in rinds, and pathways are detected in only four technological families (Internal Blue mould, Lactic Bloomy rind, Soft Washed rind, and Uncooked Pressed cheese). Unlike rind samples, most siderophore biosynthesis pathways present in metagenomes were not detected in MAGs.

3.3. Distribution of Iron Acquisition Systems Among Cheese Bacteria

For this analysis, in addition to the 1105 bacterial MAGs, we considered the 280 bacterial genomes (Table S2) of strains isolated from cheeses of the MetaPDOCheese study (Irlinger et al. 2024; Gardon et al. 2025), representing a total of 102 genera. Among these, we identified siderophore biosynthesis pathways in at least one MAG or genome from 35 genera spanning three phyla: Actinomycetota, Pseudomonadota, and Bacillota (Figures 3 and S12).

FIGURE 3.

Siderophore biosynthesis pathways detected in the bacterial genera of the MAGs and cheese isolates. The enterobactin‐type group includes enterobactin, vanchrobactin, vibriobactin, and bacillibactin. The prevalence corresponds to the proportion of the MAGs/genomes within the genus in which the biosynthesis pathway was detected.

No mixed‐family siderophore pathways were detected in Actinomycetota. However, we identified pathways from the three other families: catecholate (enterobactin‐type siderophores), hydroxamate (desferrioxamine) and carboxylate (staphylopine). Enterobactin‐type siderophores (including enterobactin, vanchrobactin, vibriobactin, and bacillibactin) were the most abundant biosynthetic pathways detected in metagenomes, this is partly due to Actinomycetota MAGs classified as Corynebacterium, Glutamicibacter and Microbacterium. At the species level, enterobactin‐type biosynthesis pathways are highly prevalent in Corynebacterium variabile (49/50), Microbacterium gubbenense (36/38), Glutamicibacter sp002975405 (5/6) and, to a lesser extent, G. arilaitensis (19/33). Desferrioxamine biosynthesis pathway is present in only four genera of Actinomycetota, with the highest prevalence in Brachybacterium, mainly B. alimentarium (12/18). Staphylopine biosynthesis pathway was primarily detected in Glutamicibacter, especially G. arilaitensis (32/33) and G. sp002975405 (6/6).

Among the 30 Bacillota genera of the MAGs and genomes, siderophore biosynthesis pathways are present in only two: Staphylococcus and Facklamia. Both genera exclusively carry carboxylate‐family siderophores. Staphyloferrin A biosynthesis pathway is highly abundant in Staphylococcus (33/37), while staphyloferrin B pathway is only detected in Staphylococcus equorum MAGs/genomes (10/21) and one Facklamia tabacinasalis genome (1/8). We detected staphylopine biosynthesis pathway only in Staphylococcus xylosus (3/3).

The phylum with the greatest siderophore diversity is Pseudomonadota, with 10 siderophores pathways from three siderophore families. It contains no carboxylate and is the only of the three phyla that has mixed‐family siderophores. In this family, we found petrobactin, yersiniabactin and yersiniabactin‐like biosynthesis pathways in a few MAGs/genomes, while pyoverdine and aerobactin pathways are more abundant. Pyoverdine biosynthesis pathway is highly prevalent in Pseudomonas (36/39), and aerobactin pathway is prevalent in Serratia (9/10) and Vibrio (9/17). In the catecholate family, enterobactin‐type and pyochelin biosynthesis pathways are the most abundant siderophores. Enterobactin‐type biosynthesis pathways, which were the most abundant pathways in metagenomes, were detected in several Pseudomonadota genera, including Psychrobacter (58/141), Serratia (10/10) and Alcaligenes (6/6). Among the species with a large number of MAGs/genomes, enterobactin‐type pathways are highly prevalent in Psychrobacter celer (33/35). We detected pyochelin biosynthesis pathways mainly in Pseudomonas (34/39), especially in P. helleri (30/31) and P. lundensis (3/4). In addition, we detected desferrioxamine (hydroxamate family) biosynthesis pathway in a few genera, including Hafnia, Pseudoalteromonas and Psychrobacter. It is particularly prevalent in Hafnia alvei (6/7), Pseudoalteromonas sp. (9/12), Psychrobacter alimentarius (9/10), and Psychrobacter aquimaris (5/5).

Within these three phyla, we identified several species capable of producing multiple siderophores (Figure S12). In Actinomycetota, multiple siderophore biosynthesis pathways occur in Glutamicibacter arilaitensis and G. ardleyensis, both carrying genes for enterobactin‐type and staphylopine biosynthesis. In Bacillota, some Staphylococcus equorum genomes have both staphyloferrin A and B pathways, while staphyloferrin A and staphylopine pathways are found in S. xylosus . In Pseudomonadota, Pseudomonas helleri has two siderophore pathways: for pyoverdine and for pyochelin. Psychrobacter alimentarius and P. aquimaris genomes both encode enterobactin‐type and desferrioxamine pathways. Several Serratia species have both aerobactin and enterobactin‐type biosynthesis pathways.

In the previous analyses, we assigned CDSs to known siderophore biosynthesis pathways. However, a portion of the CDSs could not be assigned to any of the reference pathways considered in this study. In the 1385 MAGs and genomes analysed, we identified an average of two CDSs per MAGs/genome that could not be assigned to a pathway (see Table S2 sheet Bacterial_MAGs_Genomes_SiderophoreSynthesis, and Figure S13). The number of these unassigned CDSs varies across species and MAGs/genomes, with notably high values in Serratia and Pseudomonas MAGs/genomes. For MAGs and genomes that contain multiple unassigned CDSs, these may correspond to partial biosynthesis pathways, or to pathways that are presently uncharacterised and absent from our current reference pathways (see Table S3, sheet Bacterial_MAGs_Genomes_SiderophoreSynthesis).

We assessed the potential for iron/siderophore import in MAGs and genomes by analysing the presence of two specific components TonB‐dependent receptors for iron/siderophore translocation across the outer membrane (Gram‐negative species) and substrate‐binding components of ABC‐type iron/siderophore transporters (Figures S14 and S15). Compared to biosynthesis, we detected siderophore import components in a larger number of genera (97 among 102 genera). They are present in all genomes of cheese isolates and the majority of MAGs (942/1105). TonB‐dependent receptors are present in the Gram‐negative phyla, mainly represented in this study by Pseudomonadota and Bacteroidota, whereas substrate‐binding proteins are found in all phyla except Patescibacteria, which is represented by only one MAG. When specifically considering the substrate‐binding components of ABC‐type iron/siderophore transporters, Actinomycetota and Pseudomonadota have more transport systems than Bacillota and Bacteroidota (Figure S16). Most Pseudomonadota and Actinomycetota species have more than three substrate‐binding proteins, and several species have more than 10 (Figure S17). In particular, Corynebacterium variabile , Halomonas alkalophila and Pseudomonas helleri have median counts of 12, 14 and 10, respectively.

We also investigated the presence of ferrous import or ferric reductive iron uptake in the MAGs and genomes. We detected substrate‐binding components of ABC‐type transporters involved in ferric and ferrous iron import (e.g., EfeO, YfeA, SitA, FutA), the permease FeoB, which is associated with ferrous iron import, and the NRAMP family of iron transporters (Figure S18). We found these import systems in the majority of MAGs and genomes (1343/1385) and in all phyla, except Patescibacteria. In contrast to iron/siderophore import, these iron import systems are significantly more abundant in Bacillota than in Actinomycetota and Pseudomonadota (Figures [Link], [Link]).

3.4. Distribution of Iron Acquisition Systems Among Cheese Fungi

We also analysed the 158 fungal MAGs recovered from the cheese metagenomes. Taxonomic assignment of these MAGs was inferred using robust genome phylogeny with 1723 reference fungal genomes (Figure S21). The MAGs encompass abundant and rare taxa, with relative abundance varying from > 1% to 63% (average: 15%). We identified 22 species, including Penicillium spp. (n = 18 MAGs), Geotrichum candidum (n = 43) and Debaryomyces hansenii (n = 47) (Figure 4). The taxonomic assignment of some MAGs was more challenging due to the limited availability of reference genomes or the absence of specific marker genes within the MAGs. Among these, the two MAGs of Sporendonema sp. may correspond to Sporendonema equinum, a species commonly found in cheese environments. This assignment was supported by a BLASTn alignment of the beta‐tubulin gene identified in AOP33_8_surf_bin.7, which showed 99.57% identity and 19% coverage with the partial beta‐tubulin gene of Sphaerosporium equinum strain MUCL 38540 (GenBank: ON075085.1).

FIGURE 4.

Phylogenomic tree representing the relationship between the 158 fungal MAGs reconstructed from cheese metagenomes and presence/absence of genes involved in iron acquisition. ModelFinder was used to find the best fitting model of evolution (VT + F + I + G4) and to assess bootstrap values. Circles on internal nodes indicate bootstrap support (bootstraps over 70% are indicated by an unfilled circle and over 90% are indicated by a filled circle). The completeness level for each MAG is represented by a red pie chart. Presence of iron acquisition genes is indicated by a filled box. Import genes are separated into the RIA (Reductive Iron Assimilation) category, involved in ferrous and ferric import, and the nRIA (non‐Reductive Iron Assimilation) category, involved in iron/siderophore import. sidA is a gene involved in the first step of biosynthesis of multiple fungal siderophores (coprogen, ferrichrome, fusarinine C, ferricrocin). See Table S2 for the full list of genes. Due to the aspecificity of the KOFAM HMM K21476, pul2 detection was refined using a phylogenetic approach (see Figure S23) with reference sequences. The relative abundance of MAGs in their respective metagenomes is represented by the green bar plot. Cheese technological families: Hard Cooked cheese (HC), Internal Blue mould (IB), Lactic Bloomy rind (LB), Lactic Washed rind (LW), Soft Bloomy rind (SB), Soft Washed rind (SW), Uncooked Pressed cheese/Semihard cheese (UP).

We investigated CDSs involved in fungal siderophore biosynthesis, iron/siderophore uptake and reductive iron uptake. We focused on the biosynthesis of several fungal siderophores: coprogen‐related siderophores, ferricrocin, ferrirhodin, ferrichrome‐related siderophores, fusarinine C, triacetylfusarinine C, rhizoferrin and pulcherrimin (a pigment capable of chelating iron). We did not detect genes responsible for ferrichrome, ferrirhodin and triacetylfusarinine C in any of the MAGs. Complete or partial biosynthesis pathways for fusarinine C and ferricrocin are present in all Penicillium MAGs. We detected the coprogen pathway in nearly all P. camemberti (7/8) and more than half of the P. roqueforti MAGs (4/7). For the latter species, absence of coprogen genes detection in some MAGs may be explained by their lower completeness level (53% to 98%). Apart from the Penicillium genus, complete or partial pathways for fusarinine C and ferricrocin are also present in the Scopulariopsis sp. MAG and the two Sporendonema sp. MAGs, and the coprogen pathway is present in the most complete Sporendonema sp. MAG. We detected the rfs gene, responsible for the biosynthesis of rhizoferrin, only in the two Mucor MAGs (M. circinelloides and M. lanceolatus ). In contrast to the filamentous fungi, we did not identify complete or partial siderophore biosynthesis pathways in the yeast MAGs, except for Kluyveromyces lactis and Wickerhamiella pararugosa. For Kluyveromyces lactis, the pulcherrimin biosynthesis pathway is complete in all MAGs, except in the MAG with the lowest completeness level. Surprisingly, we detected a complete pathway of Enterobactin‐type siderophore in the Wickerhamiella pararugosa MAG AOP43_C2_surf_bin15 (Table S2 sheet Fungal_MAGs_HMMs).

Concerning import systems, we detected Fth1 and Fet3‐5, two components involved in reductive iron assimilation (RIA), and the two ferrous iron uptake (FIA) systems, Smf and Fet4. RIA and FIA import systems are present in most fungal MAGs. Importers of the SIT family (Philpott 2006), which are responsible for the import of iron/siderophore (ferrichrome‐related siderophores, ferricrocin, triacetylfusarinine C and enterobactin) are also present in most of the MAGs. No siderophore importer was detected in the two Mucor MAGs, even though they encode a rhizoferrin biosynthesis pathway. The pulcherrimin importer Pul3 is present in almost all Kluyveromyces lactis MAGs and in the Saccharomyces cerevisiae MAG.

4. Discussion

In this study, we aimed to improve our understanding of iron acquisition strategies in cheese microbial communities. Specifically, we focused on identifying genes involved in siderophore biosynthesis, iron/siderophore complex transport, and other iron import systems. To do this, we used a Hidden Markov Model (HMM)‐based approach on a large dataset comprising 136 metagenomes from 44 different PDO cheeses, 1263 metagenome‐assembled genomes (MAGs), and 280 bacterial genomes isolated in the MetaPDOCheese study (Gardon et al. 2025). This combination enabled us to obtain a community‐level overview of iron acquisition genes from metagenomes, as well as identifying the species carrying these capacities based on their genomes. This provides a complementary perspective on iron acquisition strategies in cheese microbiota. We did not use AntiSMASH and FeGenie (Garber et al. 2020), two commonly used tools for identifying iron acquisition systems in genomic or metagenomic studies. AntiSMASH relies on databases that lack several references of siderophore clusters, and occasionally struggles to accurately annotate them. In addition, this tool groups siderophore biosynthetic gene clusters into three types: Non‐ribosomal peptide synthetase (NRPS), NRPS‐metallophores, and NRPS‐independent siderophores (NIS). The first two are not specific to siderophores, which complicates siderophore identification. FeGenie uses an HMM‐based approach, but its reference database lacks fungal‐specific models that are essential for studying cheese environments. To overcome these limitations, we curated a comprehensive HMM collection by integrating models from FeGenie, KOfamScan (Aramaki et al. 2020) and NCBI (Li et al. 2021), and enriched it with custom‐built HMMs mainly targeting fungal siderophore biosynthesis genes. Although reconstructing bacterial MAGs is a well‐established approach, only few metagenomic studies have also generated fungal MAGs (Peng et al. 2021; Singh et al. 2023; Carlino et al. 2024; Tagirdzhanova et al. 2024). In the present study, we proposed a strategy to assign fungal MAGs using the UFCG tool and database (Kim et al. 2023). We used this tool to extract marker genes from fungal MAGs and to perform phylogenetic classifications based on comparisons with the marker gene database. Concerning gene annotation of the fungal MAGs, we used the MetaEuk eukaryote‐specific gene prediction tool. This approach improved the detection of genes involved in iron acquisition, such as the rhizoferrin gene. Indeed, this gene was detected in fungal MAGs, but not in the corresponding metagenomes, which underlines the importance of using domain‐specific tools in complex communities.

In our dataset, siderophore biosynthesis potential was mainly observed only in a subset of MAGs (25%). However, most rind‐associated metagenomes contained multiple MAGs encoding siderophore biosynthesis pathways, often from distinct siderophore families. The most prevalent siderophores biosynthesis pathways detected in the cheese samples were of the enterobactin‐type (catecholate) and desferrioxamine (hydroxamate). Enterobactin‐type pathways were widely distributed across several genera, though precise discrimination between enterobactin, bacillibactin, vibriobactin and vanchrobactin was not feasible with our current HMM set. At the genus level, we observed marked differences in siderophore biosynthesis potential among bacterial and fungal taxa commonly found in cheese. Enterobactin‐type biosynthesis clusters were prevalent in Corynebacterium species, especially in C. variabile , while they were rare in C. casei , highlighting species‐level differences. This is consistent with earlier reports describing enterobactin‐like gene clusters in C. variabile (Schröder et al. 2011). In Glutamicibacter, we identified biosynthesis genes corresponding to both staphylopine and enterobactin‐type. This corroborates a previous study on G. arilaitensis in which two clusters were detected: one for a catecholate or mixed catecholate/hydroxamate siderophore, and one for a NIS‐type siderophore (Monnet et al. 2010). We also detected enterobactin‐type pathways in Microbacterium, notably in M. gubbeenense . This finding is consistent with genomic data describing enterobactin‐like clusters in this species (Corretto et al. 2020). Within the Bacillota phylum, siderophores were almost exclusively detected in the Staphylococcus species. Specifically, we found pathways for staphyloferrin A and B, as well as staphylopine, in agreement with previous findings (Song et al. 2018; Souza et al. 2019; Reydams et al. 2024). In Pseudomonas, nearly all MAGs and genomes encoded both pyoverdine and pyochelin biosynthesis genes, which is coherent with known metabolic capacities of this genus (Cornelis 2010). Interestingly, most Pseudomonas MAGs detected in our dataset were assigned to P. helleri, a species for which no literature concerning siderophores is available. Literature on siderophore production in Psychrobacter is scarce (Lasa and Romalde 2017; Borker et al. 2024). Our results revealed a large diversity of siderophore biosynthesis pathways within the Psychrobacter genus. For instance, P. celer carried enterobactin‐type pathways, P. alimentarius harboured desferrioxamine biosynthesis genes, and P. sp014861395 possessed both enterobactin‐type and vibrioferrin pathways, albeit at low prevalence.

Among the fungal MAGs, we detected siderophore biosynthesis genes that are consistent with the known production capacities of cheese‐associated moulds (Emri et al. 2013; Lebreton et al. 2020; Luu et al. 2023). Penicillium species encoded pathways related to coprogen, fusarinine C and ferricrocin synthesis, while Mucor species carried the rhizoferrin biosynthesis gene. Consistent with the literature, we detected a pulcherrimin biosynthetic pathway in Kluyveromyces lactis (Krause et al. 2018). We also detected an enterobactin‐type pathway in Wickerhamiella pararugosa, which is in line with a known siderophore biosynthesis horizontal gene transfer (HGT) event that occurred in the Wickerhamiella/Starmerella clade (Kominek et al. 2019). Although ferrichrome‐type siderophores were previously detected in an Oregon blue cheese sample (Scott 1981), we did not detect any ferrichrome biosynthesis pathway in our samples. This absence may be due to the HMMs used to detect these pathways, which were constructed from sequences of strains distantly related to cheese fungal species since ferrichrome synthesis genes have not yet been identified in cheese strains. We also identified species carrying two siderophore biosynthesis pathways within the same genome, including Glutamicibacter arilaitensis, Psychrobacter alimentarius , P. aquimaris and Pseudomonas helleri. According to the literature, the ability to produce multiple siderophores can confer a competitive advantage by enhancing iron scavenging efficiency and ensuring that at least one siderophore remains inaccessible to competitors or host defences (Takase et al. 2000; Niehus et al. 2017; McRose et al. 2018; Kramer et al. 2020). Additionally, this trait may support cooperative interactions or broaden the range of targeted metals (McRose et al. 2018). Siderophores may also be involved in intracellular iron storage, such as ferricrocin (McRose et al. 2018), which was detected in Penicillium MAGs, which have three siderophore pathways. In the present study, we identified known siderophore biosynthesis pathways in cheese metagenomes and in cheese‐associated species. However, some CDSs involved in siderophore biosynthesis could not be assigned to any characterised pathway. These sequences may correspond to siderophores with uncharacterised biosynthesis pathways, or potentially represent novel siderophores. Investigating these genes could provide valuable insights to discover new siderophores (Reitz and Medema 2022) and to understand their roles in food microbial communities (Mekuli et al. 2025).

We observed that reductive iron import systems are largely present in cheese species, notably in Bacillota, which includes Lactic Acid Bacteria (LAB) that dominate the cheese core. In contrast, iron acquisition via siderophores is particularly abundant in the Actinomycetota and Pseudomonadota phyla, which are mainly encountered at the cheese surface. The reduced number of iron acquisition systems in LAB is consistent with their well‐documented low or absence of iron requirements (Bruyneel et al. 1989). Although predicting the exact type and specificity of iron/siderophore import components based only on sequence homology remains challenging, we were able to reveal differences in their diversity and abundance at different taxonomic levels. Concerning fungal MAGs, we detected iron/siderophore importer genes in most of them, including in yeasts, despite the fact that yeasts generally do not produce siderophores themselves, except for Kluveromyces spp. and members of the Wickerhamiella/Starmerella clade (Choi et al. 2024; Krause et al. 2018; Sun et al. 2024). In the bacterial MAGs, iron/siderophore import components were more widespread than siderophore biosynthesis pathways, with the highest prevalence observed in Actinomycetota and Pseudomonadota. These two phyla also harboured species with the greatest numbers of iron/siderophore transport components. This extensive repertoire of import systems at the genomic prediction level is consistent with the idea that siderophores act as “public goods”, meaning that non‐producing strains benefit from the metabolic investment of those that produce them (Smith and Schuster 2019). The degree to which these non‐producers can benefit is dependent on the diversity and specificity of their transport systems. For example, pyoverdine transporters in Pseudomonas species exhibit high specificity (Meyer 2000), whereas other transporters can accommodate a broad range of siderophores and even facilitate the exchange of siderophores between kingdoms (Clarke et al. 2001; Philpott 2006; Sheldon et al. 2016; Hannauer et al. 2010). Such differences in specificity between transporters likely contribute to the complexity of microbial interactions, highlighting the potential for cross‐kingdom interactions mediated by siderophores. This includes the possibility that bacteria may exploit fungal siderophores in cheeses.

Horizontal gene transfer (HGT) of iron acquisition genes has been documented in various environments, including marine ecosystems and human pathogens (Richards et al. 2009; Gyles and Boerlin 2014). In cheese‐associated Brevibacterium species, some genes involved in siderophore biosynthesis were detected in genomic islands resulting from HGT (Pham et al. 2017). In addition, several HGT genomic regions containing siderophore acquisition genes are prevalent in cheese‐associated bacteria, suggesting that iron is a driving force in the adaptation of bacteria/fungi to the cheese habitat (Bonham et al. 2017). One of these regions, referred to as actinoRUSTI, was detected in 15 species belonging to five genera of Actinomycetota and one Bacillota. In our dataset, we observed patterns consistent with possible HGT events in MAGs and genomes. Regarding siderophore biosynthesis, enterobactin‐type pathways were detected at a very low prevalence in Brachybacterium alimentarium (1/18). Similarly, desferrioxamine and staphylopine biosynthesis clusters were present in one B. tyrofermentans genome (1/26) and one Corynebacterium faecigalinarum genome (1/12), respectively. Among Bacillota, the staphyloferrin B pathway was found in only one Facklamia tabacinasalis genome (1/8). A putative novel Halomonas species showed low prevalence of pyochelin (1/34) and enterobactin‐type (2/34) pathways. The sporadic presence of these siderophore biosynthesis clusters within species suggests HGT events, emphasising the potential role of HGT in shaping the iron acquisition strategies of cheese‐associated microorganisms. Further genomic investigations are required in order to confirm HGT events suspected in the genomes and MAGs of this study.

Some of the identified siderophore‐producing species, are already used in commercial ripening cultures (e.g., Glutamicibacter arilaitensis, Staphylococcus equorum or Corynebacterium variabile ) (Bourdichon et al. 2018), while others are part of microbial communities originating from the natural environment where the cheese is manufactured. Several studies have shown that inoculated ripening cultures sometimes struggle to establish themselves on cheese surfaces, and are outcompeted by resident environmental strains (Feurer et al. 2004; Goerges et al. 2008; Irlinger and Mounier 2009; Sun and D'Amico 2021). The competitive advantage of environmental microorganisms may be partly due to their ability to produce or import siderophores. Indeed, iron availability is a limiting factor for the growth of typical cheese surface bacteria (Monnet et al. 2012) and siderophore production is an effective strategy for microorganisms to overcome iron scarcity (Boiteau et al. 2016). This mechanism probably contributes to the successful establishment of environmental microorganisms on the rind. Interestingly, in our study, we observed that environmental bacterial species encoding siderophore biosynthesis pathways exhibited significantly higher relative abundances compared to those lacking those pathways (Figure S22). To our knowledge, the capacity to produce siderophores is not currently used as a selection criterion for designing cheese ripening cultures. We believe that this trait could be an important lever for improving culture performance by accelerating ripening or limiting the growth of undesirable microorganisms. Our comprehensive inventory of siderophore biosynthesis pathways and their associated cheese species represents a valuable contribution to the rational design of such ripening cultures.

5. Conclusion

Our study provides a comprehensive overview of iron acquisition systems of cheese microbial communities. This was achieved by integrating metagenomic analyses and genome‐centric approaches applied to both bacterial and fungal genomes. This strategy allowed us to characterise the diversity of iron acquisition systems in cheeses and provides information on the distribution of siderophore biosynthesis pathways in cheese microorganisms, including underexplored or poorly characterised species. This study also supports the hypothesis that siderophore production provides a selective advantage in the iron‐restricted cheese surface environment and may explain why non‐inoculated strains sometimes have a better fitness than strains from ripening cultures. The present study, which is based on genomic and metagenomic analyses, reveals the potential for interactions mediated by siderophores in cheese microbial communities. However, further studies involving gene expression and siderophore production data are required to provide more direct evidence for these interactions. Overall, our results improve our understanding of the ecology of cheese microbial communities and suggest that iron acquisition mechanisms may be taken into account to design improved ripening cultures.

Author Contributions

Sibylle Tabuteau: investigation, writing – original draft, methodology, visualization, writing review and editing, software, formal analysis, data curation. Vincent Hervé: writing – original draft, writing – review and editing, software, formal analysis, investigation, methodology, data curation, resources, conceptualization. Françoise Irlinger: conceptualization, funding acquisition, writing – review and editing, resources. Christophe Monnet: conceptualization, funding acquisition, writing – original draft, writing – review and editing, supervision, project administration, investigation, visualization, resources.

Funding

This work was supported by ABIES Doctoral School and the MICA Department of INRAE.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Heatmap of the abundance of CDSs mapping to individual iron acquisition HMMs in the cheese metagenomes. Abundances are expressed as RPKM (reads per kilobase per million mapped reads). Phylogenetic composition data result from the taxonomic classification of the sequencing reads using the Kaiju tool. The samples (columns) and the HMMs (rows) were clustered with the Pearson method. Cheese technological families: Hard Cooked cheese (HC), Internal Blue mould (IB), Lactic Bloomy rind (LB), Lactic Washed rind (LW), Soft Bloomy rind (SB), Soft Washed rind (SW), Uncooked Pressed cheese/Semihard cheese (UP).

Figure S2: Heatmap of correlation between the phylogenetic composition at the phylum level and the abundance of CDSs mapping to groups of iron acquisition HMMs. The enterobactin‐type biosynthesis group corresponds to enterobactin, vanchrobactin, vibriobactin and bacillibactin. The samples (columns) and the grouped HMMs (rows) were clustered with the Pearson method.

Figure S3: Abundances of iron acquisition categories, expressed as RPKM (reads per kilobase per million mapped reads) of the CDSs mapping to the corresponding HMM groups. Each point represents one cheese metagenome. Differences between core and rind metagenomes for each category were evaluated by a Wilcoxon test.

Figure S4: Heatmap of correlation between the phylogenetic composition at the phylum level and the abundance of CDSs mapping to individual iron acquisition HMMs. The samples (columns) and the HMMs (rows) were clustered with the Pearson method.

Figure S5: Heatmap of correlation between the phylogenetic composition at the family level and the abundance of CDSs mapping to groups of iron acquisition HMMs. The enterobactin‐type group includes enterobactin, vanchrobactin, vibriobactin and bacillibactin. The samples (columns) and the grouped HMMs (rows) were clustered with the Pearson method.

Figure S6: Heatmap of correlation between the phylogenetic composition at the species level and the abundance of CDSs mapping to groups of iron acquisition HMMs. The enterobactin‐type group includes enterobactin, vanchrobactin, vibriobactin and bacillibactin. The samples (columns) and the grouped HMMs (rows) were clustered with the Pearson method. Only species with at least one Pearson's correlation coefficient > 0.4 or < −0.4 were included in the heatmap.

Figure S7: Heatmap of correlation between the phylogenetic composition at the species level and the abundance of CDSs mapping to individual iron acquisition HMMs. The samples (columns) and the HMMs (rows) were clustered with the Pearson method. Only species with at least one Pearson's correlation coefficient > 0.4 or < −0.4 were included in the heatmap.

Figure S8: Heatmap of correlation between the phylogenetic composition at the family level and the abundance of CDSs mapping to individual iron acquisition HMMs. The samples (columns) and the HMMs (rows) were clustered with the Pearson method.

Figure S9: Siderophore biosynthesis pathways detected in the 1263 bacterial and fungal MAGs obtained from the metagenomes of rind and core samples. Each column aggregates the MAGs produced from one metagenomic sample, and the heatmap shows the cumulated numbers of siderophores within each siderophore family. The samples (columns) were clustered with the Pearson method. Cheese technological families: Hard Cooked cheese (HC), Internal Blue mould (IB), Lactic Bloomy rind (LB), Lactic Washed rind (LW), Soft Bloomy rind (SB), Soft Washed rind (SW), Uncooked Pressed cheese/Semihard cheese (UP).

Figure S10: Siderophore biosynthesis pathways detected in the 1263 bacterial and fungal MAGs obtained from the metagenomes of rind and core samples. Each column aggregates the MAGs produced from one metagenomic sample. The enterobactin‐type group includes enterobactin, vanchrobactin, vibriobactin and bacillibactin. The samples (columns) were clustered with the Pearson method. Cheese technological families: Hard Cooked cheese (HC), Internal Blue mould (IB), Lactic Bloomy rind (LB), Lactic Washed rind (LW), Soft Bloomy rind (SB), Soft Washed rind (SW), Uncooked Pressed cheese/Semihard cheese (UP).

Figure S11: Siderophore biosynthesis pathways detected in the 21 metagenomes of cheese cores and in the 91 bacterial and fungal MAGs obtained from the corresponding metagenomic sequences. The enterobactin‐type group includes enterobactin, vanchrobactin, vibriobactin and bacillibactin. Typical fungal siderophores are indicated in red. The numbers of metagenomes or MAGs containing at least one siderophore pathway from a given family are represented by the global siderophore family lines. Cheese technological families: Hard Cooked cheese (HC), Internal Blue mould (IB), Lactic Bloomy rind (LB), Lactic Washed rind (LW), Soft Bloomy rind (SB), Soft Washed rind (SW), Uncooked Pressed cheese/Semihard cheese (UP).

Figure S12: Siderophore biosynthesis pathways detected in the bacterial species of the MAGs and cheese isolates. The enterobactin‐type group includes enterobactin, vanchrobactin, vibriobactin and bacillibactin. The prevalence corresponds to the proportion of the MAGs/genomes within the species in which the biosynthesis pathway was detected.

Figure S13: Number of CDSs from bacterial MAGs and genomes from cheese isolates that were detected as siderophore biosynthesis genes, but that could not be assigned to a known biosynthesis pathway. Species for which we found siderophore biosynthesis pathways in at least half of the MAGs/genomes are indicated in blue.

Figure S14: Siderophore transport CDSs detected in the genera from the bacterial MAGs and bacterial genomes of cheese isolates. TonB‐dependent receptors for iron/siderophore translocation across the outer membrane (Gram‐negative species) are in the upper part of the figure, whereas substrate‐binding components of ABC‐type iron/siderophore transporters are in the lower part. Genera for which we found siderophore biosynthesis pathways in at least half of the MAGs/genomes are indicated in blue. The prevalence corresponds to the proportion of the MAGs/genomes within the genus in which each transport CDS is detected.

Figure S15: Siderophore transport CDSs detected in the species from the bacterial MAGs and bacterial genomes of cheese isolates. TonB‐dependent receptors for iron/siderophore translocation across the outer membrane (Gram‐negative species) are in the upper part of the figure, whereas substrate‐binding components of ABC‐type iron/siderophore transporters are in the lower part. Species for which we found siderophore biosynthesis pathways in at least half of the MAGs/genomes are indicated in blue. The prevalence corresponds to the proportion of the MAGs/genomes within the species in which each transport CDS is detected.

Figure S16: Number of substrate‐binding components of ABC‐type iron/siderophore transporters detected in the bacterial phyla from the MAGs and cheese isolates. Differences between phyla were evaluated by a Kruskal–Wallis test followed by post hoc tests with the Bonferroni correction (p < 0.05).

Figure S17: Substrate‐binding components of ABC‐type iron/siderophore transporters detected in the bacterial species from the MAGs and cheese isolates. Species for which we found siderophore biosynthesis pathways in at least half of the MAGs/genomes are indicated in blue.

Figure S18: Ferrous and ferric iron transport CDSs detected in the bacterial species from the MAGs and cheese isolates. The prevalence corresponds to the proportion of the MAGs/genomes within the species in which each transport CDS is detected.

Figure S19: Ferrous and ferric iron transport CDSs detected in the bacterial phyla from the MAGs and cheese isolates. Differences between phyla were evaluated by a Kruskal–Wallis test followed by post hoc tests with the Bonferroni correction (p < 0.05).

Figure S20: Ferrous and ferric iron transport HMMs detected in the bacterial species from the MAGs and cheese isolates.

Figure S21: Phylogenomic tree representing the relationship between the 158 fungal MAGs reconstructed from cheese metagenomes and the 1723 references from the UFCG and NCBI databases. The tree was rooted with the UFCG reference, Rhizoclosmatium globosum. MAGs considered in this study are indicated in red. Phylum, family and genus of the references are indicated. ModelFinder was used to find the best fitting model of evolution (Q.yeast+F + I + G4). Bootstrap values are indicated by a black circle.

Figure S22: Relative abundance of the MAGs as a function of the presence or absence of siderophore biosynthesis pathways and the origin of the MAGs (ferment/environment). We considered that species not listed in the inventory of ferments (Bourdichon et al. 2018) corresponded to strains that were not deliberately inoculated (environmental strains). Differences between the presence or absence of siderophore biosynthesis pathways for ferment and environment were evaluated by a Wilcoxon test.

Figure S23: Phylogenomic tree representing the relationship between the 42 fungal CDS identified by KOFAM K21476 HMM and Uniprot references. Taxonomic affiliation is indicated for CDS MAGs. The tree was rooted with the Uniprot reference B8NJG8 (Cytochrome P450 monooxygenase lepD). The pul2 cluster is highlighted in green. ModelFinder was used to find the best fitting model of evolution (Q.pfam+I + G4). Bootstrap values are indicated by a black circle.

Table S1: HMMs and siderophore biosynthesis pathways detected in the metagenomes, and associated metadata from the MetaPDOcheese study.

Table S2: HMMs detected in the genomes of bacterial isolates, and bacterial and fungal MAGs considered in the present study, and associated metadata from the MetaPDOcheese study.

Table S3: Hidden Markov Models (HMMs) related to iron acquisition systems used in the present study.

Data Availability Statement

The fungal MAGs used in this study are available under NCBI BioProject PRJEB64630, BioSample: SAMN50110840‐SAMN50110997. The HMMs and the scripts used in this study are available on GitHub (https://github.com/STabuteau/iron_cheese_metagenomes).