Genomic insights into specialized and primary metabolism gene clusters in methicillin-resistant Staphylococcus aureus from Egyptian community settings

Abstract

Community-acquired methicillin-resistant Staphylococcus aureus (CA-MRSA) remains a major public health concern. Beyond canonical virulence determinants, biosynthetic gene clusters (BGCs) for specialized metabolites (SMs) and primary metabolic adaptations may shape fitness in community settings. In this study, we analyzed 123 CA-MRSA isolates from Egypt (2020–2021) alongside three reference genomes using a combined genomics workflow: antiSMASH and PRISM for homology-based specialized BGCs and predicted structures, GECCO for de novo BGC detection, BiG-SCAPE for gene-cluster family (GCF) networks and gutSMASH for primary-metabolism clusters detection. Assemblies were generated from Illumina MiSeq data and annotated with Prokka; sequence types and clonal complexes were assigned with FastMLST. antiSMASH identified a total of 934 BGCs (median 8 per genome), dominated by the following BGC types: NRPS-independent siderophores (NIS) (n = 312; IucA/IucC-like siderophores), non-ribosomal peptide synthetase (NRPS) (n = 132), and opine-like metallophores (n = 123). GECCO detected 415 BGCs, including 287 unclassified and 101 NRPS-associated clusters. GutSMASH recovered 582 primary metabolism gene clusters (GCs), with fumarate-to-succinate (n = 96), arginine-to-bicarbonate (n = 89; ADI) and nitrate-reductase (n = 84) energy-capturing modules being the most frequent. NRPS and NIS content varied widely among isolates, exceeding reference genome diversity, and BiG-SCAPE networks suggested multiple isolate-specific GCFs lacking MIBiG matches. Egyptian CA-MRSA displays a rich repertoire of iron-acquisition, peptide, and metallophore BGCs alongside anaerobic/energy-efficiency primary metabolism GCs that may support survival and competitiveness in community niches. While experimental validation is needed, these pathways nominate tractable targets for anti-virulence or metabolism-directed interventions and provide a genomic baseline for CA-MRSA in the North African region.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12864-026-13121-y.

Introduction

Methicillin-resistant Staphylococcus aureus (MRSA) is a major antimicrobial resistance threat and has been designated a superbug due to its ability to withstand multiple antibiotics and cause diverse, often severe infections [1–3]. Resistance originates from the acquisition of the Staphylococcal cassette chromosome mec (SCCmec), which encodes the altered penicillin-binding protein PBP2a and associated regulatory elements, conferring resistance to nearly all β-lactams [1, 2]. MRSA initially emerged in hospitals (HA-MRSA) but has now spread widely in community settings. Community-acquired MRSA (CA-MRSA) is typically associated with SCCmec types IV/V and, in many lineages, the Panton–Valentine leukocidin (PVL) toxin [4, 5]. These strains often infect otherwise healthy individuals and have contributed to the global expansion of invasive MRSA disease.

The success of MRSA reflects the interplay of classical virulence factors, biofilm formation, and horizontal gene transfer, combined with lineage-specific adaptations that shape regional epidemiology [6–9]. Genome-wide association studies have highlighted the role of proteases and other extracellular factors in S. aureus virulence [1], while comparative clinical genomics studies show that both methicillin-resistant and methicillin-sensitive hyper-virulent lineages contribute to poor outcomes [4]. Beyond healthcare-associated contexts, recombination-driven host adaptation has been reported in avian S. aureus [3], illustrating the broad adaptive potential of this species.

CA-MRSA lineages frequently carry mobile genetic elements (MGEs) encoding toxins, adhesins, and enzymes that facilitate host colonization, tissue invasion, and immune evasion [7]. For example, PVL and other cytotoxins disrupt host cells, while proteases and nucleases aid dissemination. Protein A and biofilm matrix components help MRSA resist immune clearance and antibiotic treatment. These traits enhance persistence and transmission, but their distribution varies among clonal complexes, influencing patterns of disease [4].

Beyond these well-characterized virulence determinants, S. aureus also encodes biosynthetic gene clusters (BGCs) that direct the production of specialized metabolites (SMs). These include non-ribosomal peptide synthetase (NRPS) products [10, 11], ribosomally synthesized and post-translationally modified peptides (RiPPs) [12–15], siderophores [16–19], and metallophores [18]. Within the last few years, increasing numbers of bacteriocins have been isolated and identified from Gram-positive microorganisms, underscoring renewed interest in RiPP-associated BGCs and their ecological and antimicrobial roles [20]. Accordingly, SM BGCs identified in S. aureus and MRSA genomes include those responsible for staphylobactin and staphyloferrin A biosynthesis, the NRPS-derived aureusimine pathway, as well as bacteriocin-encoding clusters [20–22].Such metabolites can promote iron and metal acquisition, shape interbacterial competition and support niche adaptation by modulating host–pathogen interactions [23, 24]. At the same time, clusters involved in primary metabolism, such as fumarate-to-succinate [13], arginine deiminase (arginine-to-bicarbonate) [14], nitrate reductase [15] and thiamine pyrophosphate (TPP)-dependent amino acid metabolism [16] may optimize energy capture and redox balance under hypoxia, nutrient limitation, or host-imposed stress [17]. Together, these pathways extend the metabolic repertoire of MRSA and may underpin its ecological versatility.

While the role of individual toxins and biofilm determinants in MRSA pathogenicity is well-studied, large-scale mapping of BGC repertoires in CA-MRSA remains limited, particularly in regions such as North Africa where genomic data are scarce. Recent work has shown the emergence of novel multidrug-resistant S. aureus sub-lineages in South Asia [5], highlighting the importance of global surveillance. Comprehensive surveys of both specialized and primary metabolic gene clusters (GCs) could reveal lineage-specific adaptations that contribute to survival, virulence, and transmission in community settings.

Genome-mining is advancing quickly which helped in developing multiple existing bioinformatics tools which identify the different types of BGCs. Some of the tools are rule-based, which are capable of identifying BGCs using known motifs. Other tools use machine learning approaches to identify or predict novel clusters. Finally, some tools work on comparative and phylogenetic tools to link the clusters between species. An example of a rule-based tool is antibiotics and Secondary Metabolite Analysis SHell (antiSMASH) [12, 25]. AntiSMASH is the gold standard for genome-mining for BGC detection, it comprises known BGCs, and is comprehensive of several BGC types [26]. BGC prediction with Conditional Random Fields (GECCO) is another tool that uses machine learning for the identification and prediction of novel BGCs [17]. GECCO can predict novel BGCs that share similar patterns with known BGCs. Specialized Metabolic Analysis of Specialized gene clusters from the Human gut microbiome (gutSMASH) is a more specialized tool, which detected primary metabolism GCs, and was mainly built using human gut-associated bacterial with special focus on common anaerobic pathways [15]. Biosynthetic Gene Similarity Clustering and Prospecting Engine (BiG-SCAPE) is a clustering engine that groups the outputs from the former tools in gene cluster families (GCFs) [27]. Clinker visualizes the alignments between samples [23]. Prediction of genetically encoded natural product structures based on microbial genomes (PRISM) is a specialized tool for the prediction of chemical structures of the metabolites produced by NRPS and PKS gene clusters. It uses multiple search approaches; library-based approach using Hidden Markov Models (HMMs) to identify enzyme substrates, combinatorial chemical prediction to produce a library of possible molecules and structure scoring to predict molecular weight and chemical formula [16]. Other specialized tools like are also developed; namely BActeriocin GEnome mining tool (BAGEL) that specializes in mining of RiPPs and Bacteriocins [28] and DeepBGC that uses deep learning to identify novel BGCs [29].

Egyptian CA-MRSA isolates are interesting to study, as it is a unique epidemiological sample and clinically relevant. Community-acquired MRSA (CA-MRSA) has emerged as a significant public health threat distinct from its healthcare-associated counterpart (HA-MRSA), capable of infecting immunocompetent individuals outside clinical settings. In Egypt, CA-MRSA strains have been increasingly reported [30], yet their genomic characteristics, virulence potential, and metabolic gene cluster repertoire remain insufficiently characterized compared to globally dominant lineages such as USA300.

Here, we analyse 123 CA-MRSA isolate genomes from Egypt using an integrated genome-mining pipeline (antiSMASH, GECCO, gutSMASH, BiG-SCAPE, Clinker, and PRISM). Our aims were to (i) catalogue the diversity of specialized and primary BGCs, (ii) compare their distribution across clonal complexes and (iii) highlight candidate pathways for experimental validation and potential therapeutic targeting. We hypothesize that studying the BGCs of the CA-MRSA isolates in Egypt would possibly lead to a unique BGC profile that would in the future pave the way for better understanding of this pathogen and hence better targeting of it.

Materials and methods

Bacterial cultures

A total of 123 representative non-duplicate Staphylococcus aureus was previously obtained from a clinical diagnostic center in Alexandria, Egypt, which serves multiple screening laboratories. The collection period spanned from August 2020 to April 2021. The isolates were collected prior to the study period as part of historical sampling; therefore, ethics approval and patient consent were not required. To ensure diversity and minimize the inclusion of genetically related isolates, a random sampling approach was employed, considering the month of isolation and isolate source. The historical isolates were derived from various clinical specimens, such as aspirate, wound, pus and mini-bronchoalveolar lavage (mini-BAL).

Classification as CA-MRSA followed the standard epidemiological criteria established by the US Centers for Disease Control and Prevention (CDC), which defines CA-MRSA infections as those occurring within 48 h of hospital admission in individuals with no prior history of hospitalization, surgery, or positive cultures. MRSA confirmation was conducted using established morphological, biochemical, and molecular methods. Methicillin resistance was initially determined phenotypically based on resistance to cefoxitin according to CLSI guidelines (CLSI M100). Molecular detection of mecA was used to further characterize methicillin resistance mechanisms but was not the sole criterion for CA-MRSA designation [7–9, 31]. All isolates were preserved at –80 °C in brain heart broth supplemented with 15% (v/v) glycerol for future analysis. The whole study workflow is depicted in Fig. 1.

Fig. 1.

Study pipeline.

DNA extraction and whole genome sequencing

Genomic DNA was extracted from bacterial isolates using the QIAamp DNA Mini Kit (Qiagen, UK) by following the manufacturer’s protocol, with modifications to optimize the yield. The optimization included increasing the initial bacterial suspension volume to 2 mL and pre-warming the elution buffer to 60 °C. To enhance cell lysis, 200 μL of 20 mg/mL lysostaphin (Sigma-Aldrich, USA) was added to the lysis buffer. DNA was eluted in 100 μL of elution buffer and then quantified using a NanoDrop spectrophotometer (Thermo Fisher Scientific, USA). Library preparation was carried out using NexteraXT kits (Illumina, San Diego, CA) as per the manufacturer's guidelines, with barcoding applied to batches of 48 isolate gDNA samples. Sequencing was performed on Illumina's MiSeq platform using MiSeq run kits (v3) to generate 2 × 300 base paired-end reads. Quality control, adapter trimming, and quality filtering of the resulting FASTQ data were managed using the Fastp command-line tool, utilizing the error correction parameter (C) for paired-end overlap analysis [32]. De novo genome assembly was performed using SPAdes (version 3.13.0) with the careful command [33]. Assembly statistics were calculated on the filtered assembly with QUAST 4.4 [34], and the resulting assemblies underwent five rounds of polishing with Pilon to [35]. The average size of the polished assemblies was 2.71 ± 0.4 Mbp (Supplementary table S1-A). The completeness of the assemblies was subsequently evaluated using BUSCO (v5.2.2) [36] against the bacteria_odb10 lineage dataset (n = 124 conserved markers), to assess the proportion of complete, fragmented, and missing single-copy orthologs and confirm assembly integrity prior to downstream annotation (Supplementary table S1-B).

Data availability

Short-read data are available in the National Center for Biotechnology Information (NCBI) Sequence Read Archive, associated with BioProject PRJNA1135495 (https://www.ncbi.nlm.nih.gov/bioproject/?term=PRJNA1135495). Assembled genomes data and individual accession numbers are detailed in Supplementary Table (3).

Detection of homology-based specialized metabolism GCs

After filtration and quality checking for the 123 genomic files, their FASTA files were uploaded to antiSMASH [12] to detect the homology-based specialized metabolism GCs. The used algorithms were KnownClusterBlast, SubClusterBlast, ActiveSiteFinder, RREFinder, and TFBSanalysis. The detection strictness was adjusted to relaxed. Three reference genomes were chosen from the assembled genomes for Staphylococcus aureus (MRSA) on NCBI genome database. These reference genomes were; the main reference genome on NCBI genome database (https://www.ncbi.nlm.nih.gov/datasets/genome/?taxon=1280) for Staphylococcus aureus ASM1342v1 (NCTC 8325), in addition to two other assemblies ASM609491v1 (ATCC 12600 strain) which is widely used in antibiotic resistance, biofilm formation, and metabolic characterization research [37–39], and ASM289538v1 (MRSA107) which represented MRSA in clinical and environmental research in regional studies [40–42]. The three reference genomes were onwards referred to as Ref-1, Ref-2 and Ref-3, respectively. These three references had single scaffolds and had their genomes completely sequenced.

To visualize the distribution of antiSMASH-detected GCs across clonal complexes (CCs) and across samples in community-acquired Staphylococcus aureus, we used Rstudio (Version 4.5.1). Using tidyverse package, antiSMASH GC data were imported from an Excel file and converted from wide to long format to summarize the total presence of each specialized metabolism GCs within each CC. We then generated a heatmap using ComplexHeatmap [13, 14] with clonal complexes on the x-axis and specialized metabolism GC types on the y-axis.

Detection of homology-based primary metabolism GCs, SM structures & de novo specialized metabolism GCs

The genomic fasta files were run using gutSMASH [15] for the detection of homology-based primary metabolism GCs. KnownClusterBlast and Annotation for Functional Gene Categories were the additional analysis used with the server. Then PRISM 4 [16] was used to predict the chemical structures of BGCs. The command line version of GECCO software version 0.9.10 [17] was used to examine BGCs with conditional random fields (CRFs) model on the fasta files from both the isolates and the reference genomes.

To visualize the distribution of the primary metabolism GCs from gutSMASH across clonal complexes (CCs) and across samples in CA-MRSA, the same methodology of drawing the heatmap was used, as aforementioned.

Annotation and typing

Prokka, a rapid prokaryotic genome annotation tool, was utilized to identify coding sequences (CDSs), rRNAs, tRNAs, and other genomic features. Prokka was run with default parameters to ensure consistency across all genomes, generating standardized GenBank-formatted output files for downstream analysis [18]. FastMLST, a multilocus sequence typing tool that uses BLASTn to perform PubMLST searches [19], was used to determine the sequence type (ST), allelic profile and clonal complex (CC) for the draft assemblies [19].

Alignment of the detected NRPS GCs

The NRPS GC GenBank files generated by antiSMASH were then uploaded to the online server of the Clinker software [23]. Clusters were labeled according to gene function and links were made between similar clusters. The GenBank files for all the isolates and references were analyzed together.

Detection of sequence similarity networks of GCs and grouping into GCFs

Using Galaxy server with Galaxy Version 1.1.9 + galaxy2 [43], Biosynthetic Gene Similarity Clustering and Prospecting Engine (BiG-SCAPE) software was utilized to construct networks of sequence similarity between GCs in isolates and in reference genomes [27]. BiG-SCAPE grouped the isolates and the reference genomes into gene cluster families (GCFs).

Multiple alignment and phylogenetic tree for NRPS and NI-siderophore GCs

Sequences for NRPS and NI-Siderophore were extracted according to the coordinates from antiSMASH output for both isolates and references using codes stored in a GitHub repository (https://github.com/AhmedAboushanab/Genomic-Insights-into-Specialized-and-Primary-Metabolism-Gene-Clusters). The extracted sequences were used in building an NRPS-based phylogenetic tree and an NI-Siderophore-based phylogenetic tree using multiple alignment program for amino acid or nucleotide sequences (MAFFT) online server version 7 (https://mafft.cbrc.jp/alignment/server/) [44, 45].

Several tools (PRISM, GECCO, Clinker, and MAFFT) were used with their default options to allow for reproducibility, decrease the risk of bias, accelerate and benchmark the pipeline.

Statistical analysis and BGC enrichment

Fisher’s exact test was used to calculate BGC enrichment in each CC using Rstudio (Version 4.5.1). In addition, it was also used to Jaccard co-occurrence matrix for top specialized metabolites, and top 19 primary metabolites. Finally, Logistic for presence and absence of NIS and metallophore in CC after normalization by genome size.

Results

Siderophore, NRPS and metallophore GCs were the most abundantly detected specialized metabolism GCs

Compared to the reference genomes, the screened isolates revealed higher diversity of specialized metabolism GCs with fifteen distinctly detected GC types (Figs. 2A and 3A). The NI-siderophore GC exhibited the highest prevalence, particularly in CC1 and CC121, with a median of two per CC, maximum of five per CC, and range from zero to five per genome within 120 samples that are having more than one NI-Siderophore in their genomes similar to what is observed in the references, indicating a strong siderophore biosynthetic potential in these clonal complexes. This suggests enhanced capabilities for iron acquisition, which may contribute to virulence and adaptability. Additionally, NRPS, and Opine-like-metallophore clusters showed moderate to high presence across several CCs, with a median of one per CC, maximum of three per CC, and range of zero to three per sample within six samples that are having more than one NRPs in their genomes which is more than what was observed in the references, reflecting their widespread distribution and possible roles in SM production. On the other hand, four GCs -lassopepetide, betalactone, NRPS-like, and LAP- were the least abundant GCs. These four lowest GCs were not detected in the reference genomes, suggesting limited involvement in the metabolic repertoire of the studied isolates. Clonal complexes CC1 and CC121 demonstrated greater GC diversity, whereas CC5, CC8, and CC97 showed moderate diversity with specific GC enrichments.

Fig. 2.

Predicted metabolism GCs in the clonal complexes of CA-MRSA isolates. A Specialized metabolism GCs. B Primary metabolism GCs. The heatmaps illustrate the presence and distribution of various GCs across different clonal complexes (CCs) in community-acquired MRSA isolates. The color gradient from light blue (low presence) to black (high presence) represents the GC presence count

Fig. 3.

Metabolism Gene clusters detected per sample in the clonal complexes. A Specialized metabolism GCs. B Primary metabolism GCs. The heatmaps represent the distribution of the detected GCs across different samples. The color gradient from light blue (low presence) to black (high presence) represents the GC presence count

Furthermore, lanthipeptide-class-I, T3PKS_terpene, and cyclic-lactone-autoinducer had much lower abundance of 3%, 2%, and 6% in the 123 isolates compared to 16% in the three reference genomes, respectively. New GCs were also found in the isolates. Terpene, T3PKS, RiPP-like, terpene_ T3PKS, NRPS-like, REE-containing GCs were uniquely detected in the isolate genomes only. Terpene and T3PKS represented about 19% of GCs compared to less than 12% represented by Lanthipeptide-class-I, (T3PKS, terpene), and Cyclic-lactone-autoinducer in the GCs detected in the isolates.

Six GC types were detected in the reference genomes. The most abundant GC was NRPS-independent-siderophore with 25%, while the least abundant was NRPS with 10% in the entirety of the GCs (Fig. 3A). Surprisingly, the four other GC (Opine-like-metallophore, Lanthipeptide-class-I, (T3PKS, terpene), and Cyclic-lactone-autoinducer) were equally distributed with 16% each in the three reference genomes.

As shown in the NRPS-based phylogenetic tree in Figure S7A, almost each of the screened isolates in addition to the references has a single NRPS GC except for isolates (CA_24, CA_44, CA_94, CA_148) that have two NRPS GCs per isolate, and isolate CA_101 that has three NRPS GCs. Yet the isolates were very different from the references and from each other in both the length and sequence of the NRPS GCs. Almost the same was observed in Figure S7B, for the phylogenetic tree based on NI-Siderophore GCs. The three references contain two NI-Siderophore GCs, while the isolates differ in their NI-Siderophore GCs content with several of them having up to four NI-Siderophore GCs per isolate such as CA_68, CA_107, CA_129, and CA_146.

Fumarate to succinate, arginine to Hcarbonate pathways and nitrate reductase energy-capturing GC were the most abundantly detected primary metabolism GCs

Upon analyzing the isolates, a higher diversity of primary metabolism GCs were detected, (Figs. 2B and 3B) [15]. Nineteen different primary metabolism GCs were detected in the isolate genomes. On the other hand, the reference genomes had six primary metabolism GCs. All of them had the same relative abundance of %16—%17 per reference genome. They were Fumarate2succinate, TPP_AA_metabolism, Nitrate_reductase, TPP_fatty_acids, Arginine2_Hcarbonate, and GR_fatty_acids, Pyruvate2acetate-formate, fatty_acids_unassigned.

In the isolates, the primary metabolism GCs Fumarate2succin, Arginine2Hcarbonate, and Nitrate_reductase showed the highest prevalence, with higher prevalence in CC1, CC121, and undetermined (UD). This pattern may indicate an enhanced capacity for alternative energy metabolism, particularly anaerobic respiration and nitrogen metabolism, which may support survival under diverse environmental conditions. Pathways like TPP_FA and TPP_AA_metabolism were consistently present across multiple CCs, highlighting their essential roles in fundamental metabolic processes. In contrast, Succin2prop,Putt2sper and Oxidative_glycerol exhibited limited presence, suggesting specialized functions or restricted distribution among specific lineages.

Interestingly, combinations of metabolic clusters, such as GR_FA, Pyr2acet-formate, FA-unasgd, and Caffeate_respir., frequently co-occurred, implying potential co-regulation or metabolic interdependence. Overall, CC1 and CC121 displayed the highest metabolic diversity, while CC5, CC8, and CC97 exhibited more specialized profiles. These findings highlight the metabolic versatility and potential adaptive advantages of specific MRSA clonal complexes, which may influence their ecological fitness and pathogenicity in community settings.

De novoGC prediction detected mostly unknown and NRP types

NRP, RiPP, Saccharide, and Terpene GCs were identified by a de novo GC prediction tool in the isolates (Fig. 4A), besides identifying two known GCs in the reference genomes; RiPP, and NRP (Fig. 4B). In both the reference and the screened genomes, several GCs were not specifically labeled and were hence labeled Unknown.

Fig. 4.

De novo specialized metabolism GCs detected by GECCO. A GCs detected in the isolate genomes. B GCs detected in the reference genomes

Using BiG-SCAPE to cluster BGCs into gene cluster families, several distance matrixes were generated to show the clustered families on the horizontal axis and the samples in each cluster on the vertical axes (axis and trees) with each BGC per sample represented as a red square (Supplementary Figures S2-S6). The reference genomes had three NRPS families clustered with an average number of one GC per family with a max number of one GC per family. Two RiPPs families were clustered with an average number of two GCs per family with a max number of two GCs per family and one family of MIBiG reference GCs. Moreover, three Terpene families were clustered with an average number of one GC per family and a max number of one GC per family. Additionally, three PKSother families were clustered with an average number of one GC per family and a maximum number of one GC per family. And lastly, another GC family with an average number of three GCs per family and a max number of one GC per family as in of GCF clusters based on absence and presence of genomes and families.

Then, upon running BiG-SCAPE on the CA-MRSA isolate genomes, twelve NRPS families were detected, with an average number of ten GCs per family with a maximum number of 35 GCs per family. Also, nine RiPPs families were found with an average number of nine GCs per family with a maximum number of 24 GCs per family. Ten further Terpene families were detected with an average number of ten GCs per family with a maximum number of 32 GCs per family. Moreover, eighty-one PKS other families were found with an average number of two GCs per family with a maximum number of 18 GCs per family. Finally, eleven Others GC family was found with an average number of eleven GCs per family with a maximum number of 35 GCs per family as in of GCF clusters based on absence and presence of genomes and families. No GCFs were detected from MIBiG reference GCs.

Being among the top abundant GC groups detected in both the isolates and the references, NRPS GCs were selected to display their alignments in both the reference and isolate genomes in Supplementary Figs. 3, S1.

Predicted SM structures

PRISM 4 was run on all of the isolates, including the three genomic reference sequences (Table S4). 580 GC clusters were detected in the comprised isolates which were pertaining to the following SM types: 312 NRPS-independent siderophore synthase clusters, 132 non ribosomal peptide clusters, 71 autoinducing peptide clusters, 29 class I lanthipeptide clusters, 24 Class II/III Confident Bacteriocin clusters, 5 bacterial head-to-tail cyclized peptide clusters, 5 of the unknown thiotemplated cluster type, 1 class II lanthipeptide cluster and 1 polyketide cluster.

Statistical analysis

Several BGC types showed enrichment in CC8 (TPS, odds-ratio = 4, p-value = 0.0208), CC15 (Ripp-Like, odds-ratio = 11.235, p-value = 0.0225), and CC121 (Ripp-Like, odds-ratio = 8.951, p-value = 0.0369), Figure S10 and Table S9. Jaccard co-occurrence showed that NRPS, NI-Siderophore, T3PKS, opine-like Metallophore, and TPS have strong correlations as BGCs and are found together in most of the samples, Figure S11 and Table S10. While it also shows that some specialized GCs are found together as well like Fumarate to succinate, Figure S12 and Table S11. Logistic regression analysis showed non-significant correlation between clonal complexes, and the presence of absence of metallophone or NI-Siderophore even after normalization by genome size Tables S12 and S13, respectively.

Discussion

Comprehensive genomic analysis of bacterial pathogens is essential for computationally predicting the molecular determinants of virulence and for identifying potential novel therapeutic targets to combat antimicrobial resistance (AMR). Among the pathways of interest are BGCs that encode SMs [46]. These metabolites can act as virulence factors, as shown for staphyloxanthin in S. aureus [47], and we have previously reported lineage-specific BGC signatures in Gram-negative pathogens [48]. Here, we extend this framework to a Gram-positive pathogen, namely community-acquired MRSA (CA-MRSA), for strains isolated from Egypt, and demonstrate that both specialized and primary metabolism GCs contribute to a diverse metabolic repertoire. Such studies, when later coupled with experimental validation, pave the way for discovery of novel antimicrobial agents targeting multidrug-resistant (MDR) pathogens. Homology-based detection indicated that non-ribosomal siderophore (NIS) clusters were the most abundant specialized metabolism GCs in both screened isolate and reference genomes (Figs. 2 and 3). This result, consistent across antiSMASH and PRISM, may highlight the central role of iron acquisition in MRSA. NIS clusters encode IucA/IucC-like synthetases that produce high-affinity ferric iron chelators, enabling survival in host environments with restricted iron [49–52]. The consistent presence of aerobactin-, desferrioxamine-, and rhizobactin-like clusters may suggest functional redundancy and highlights siderophores as candidate virulence factors requiring experimental validation. Moreover, clustering of reference genomes -on different branches of the phylogenetic trees- away from clinical isolates based on their NIS and NRPS repertoires (Figure S7) may indicate lineage-specific metabolic variation, a pattern consistent with reports of variability in Staphylococcus pathogenicity islands [53].

Primary metabolism analysis revealed that the majority of the clusters were shared between isolate and reference genomes (Figs. 2B and 3B), with fumarate-to-succinate, arginine deiminase (ADI), and nitrate reductase pathways particularly abundant. The rare detection of hybrid clusters combining glycerolipid synthesis, pyruvate-to-acetate/formate, and fatty acid biosynthesis [54, 55] is notable, -which is in accordance with other previous experimental studies, in which lipid modification and altered fermentation pathways have been linked to biofilm formation, immune evasion and AMR, and is yet to be proven experimentally in the included isolates [56–58]. Similar associations between biofilm traits and population structure have been observed in S. epidermidis [59]. The fumarate-to-succinate cluster represents a key tricarboxylic acid (TCA) cycle reaction catalysed by succinate dehydrogenase [60], while the predicted ADI pathway would potentially support anaerobic survival by generating ATP and bicarbonate [61]. The frequent detection of nitrate reductase clusters [27, 62–65] further underscores the predicted possible adaptation of CA-MRSA to oxygen-limited environments. Thiamine pyrophosphate (TPP)-dependent amino acid and fatty acid modules were also common, reflecting their potential essential role in central metabolism [66–68]. Together, these findings may point to metabolic plasticity that may enhance survival in abscesses, inflamed tissue and biofilm niches.

De novo BGC prediction with GECCO revealed NRPS and RiPP clusters as the most abundant classes (Fig. 4), which is consistent with their roles in producing antibiotics, immunomodulators and other bioactive SMs in previous studies [69, 70]. Occasional detection of saccharide and terpene clusters may further extend the potential metabolic repertoire [48, 71, 72], with possible applications in drug discovery and biotechnology. Importantly, BiG-SCAPE clustering revealed many GCFs lacking MIBiG hits (Figures S2–S6), supporting the presence of potentially novel or under-characterized SM pathways. Similar expansions of metabolic diversity linked to clinical outcome have been reported in S. epidermidis device-related infections [73, 74]. The observed enrichment of NRPS and RiPP families in Egyptian CA-MRSA compared with reference genomes may suggest that local lineages could potentially harbour unique metabolic traits relevant to virulence or community adaptation.

Alignments with Clinker (Fig. 5 and Fig. S1) with the distance matrix plots (Figures S2 – S6) suggest that NRPS clusters are colocalized across all the isolates. On the other hand, clusters like (RiPPs, and Terpenes) differ in their composition across the isolates. In addition, the raincloud plots (Figures S8 – S9), together with Fisher’s exact test (Figure S10), and the logistic regression tables (S12 – S13) suggest that BGCs -subsequent to normalizing for genome size- are not distributed randomly across the different clonal complexes. Specific CCs are enriched in specific BGCs, especially the iron acquisition BGC types; metallophores and NISs.

Fig. 5.

Selected NRPS GC alignment in the isolate and reference genomes

The Jaccard metrices (Tables S10 – S11, Figures S11 – S12) show that these BGCs (NIS, NRPS and T3PKS) tend to be found together (Jaccard ~ 1) in the same genomes. This may be due to either co-selection pressure or functional linkage. This is reinforced by the pairwise Fisher’s test (Table S9) that shows statistical significance of non-random occurrence between specific BGC pairs. Our findings align with prior reports of variable BGC content in bacterial pathogens [25, 29, 75] but extend this work by showing lineage-specific enrichment of siderophores and NRPS clusters in CA-MRSA clonal complexes. The identification of hybrid and rare primary metabolism GCs (e.g., GR-fatty-acid, pyruvate formate lyase modules) may suggest potential links to virulence and AMR, though functional significance requires validation. Persistent S. aureus isolates often display enhanced fitness and immune evasion [76], and traits that may interact with the metabolic adaptations described here. Zoonotic and livestock-associated S. aureus lineages also carry distinctive genomic features [77], underscoring the need to examine how environmental and host contexts shape BGC repertoires. Metabolic flexibility may further support persister and dormant cell survival [78], complementing SM strategies in chronic infection.

For BGC prioritization, experimental validation is needed, however, we recommend focusing on the unique BGCs, especially if they are shared between the CA-MRSA isolates, and possible wider comparison studies with other MRSA strains and also with non-pathogenic bacteria. Thus, concerning clinical and translational implications, focusing on the enriched BGCs, as well as unique BGCs that are only detected in the samples -perhaps also primary metabolism GCs as well- stand to be the most promising BGCs as anti-virulence targets of CA-MRSA isolates.

Comparison of detection tools underscored complementary strengths. AntiSMASH [12] provided detailed classifications of known BGCs, while GECCO [17] identified additional, less characterized clusters with less confident class labels, albeit with more “unknown” labels. In a future needed step, the BGCs that were classified as unknown would need first perhaps functional prediction such as using NCBI BLAST tool or similar function prediction tools. GutSMASH [15] enabled specific mapping of primary metabolism modules, and BiG-SCAPE [27] facilitated GCF-level comparisons. Collectively, these approaches detected 934 clusters across isolates (median 8 per genome), revealing both conserved modules and lineage-specific diversity. Horizontal gene transfer between S. aureus and coagulase-negative Staphylococci [79] may explain the mosaic distribution of BGCs, consistent with broader patterns of mobile element exchange.

Nonetheless, each of the tools used in this research has its own bias and limitations. Despite being the gold standard for the BGC computational detection research field, antiSMASH is highly biased for known classes as a result of being rule-based. This decreases its ability to detect novel BGCs and may lead to false cluster boundaries. Pertaining to the present study, the bias of the BGCs detected by antiSMASH would possibly be that the BGCs with alignment to known and well-characterized BGCs would be more detected and thus more abundant than others, and longer contigs would have more BGCs detected. GECCO has higher probabilities for detection of novel BGCs yet still has bias towards any data similar to its training data. This causes higher rate of false positives than antiSMASH. Predicted bias in GECCO-predicted BGCs would be that more BGCs similar to the data it has been based on would be more detected than others. GutSMASH has a narrow scope that limits its ability to detect novel GCs. Hence, the primary metabolism GCs detected by gutSMASH tool would generate bias towards known GCs in anaerobic gut microbes and other GCs would remain undetected. Clinker has intensive computational needs upon using larger datasets, although false alignments can occur if non-similar proteins were aligned together, and would hence be possible in the case of the included dataset. Finally, BiG-SCAPE can group different clusters into similar families, if the cutoff is set too high. This would possibly result into categorizing clusters with different chemical products into the same gene cluster family, which is also an inherent possible grouping occurrence in our present study.

The use of short-read assemblies is a limitation, as reads can fragment multi-kilobase BGCs and obscure structural predictions, while Nextera XT library preparation may introduce locus-specific bias. Assembly completeness was evaluated using BUSCO against the bacteria_odb10 lineage dataset, which provides a more biologically meaningful measure of gene space integrity than contiguity metrics such as N50 alone [80]. While N50 reflects assembly fragmentation and size distribution, a high BUSCO score can occur independently of a high N50, indicating that an assembly may exhibit gaps in contiguity yet still retain the majority of expected conserved single-copy orthologs; conversely, a high N50 does not guarantee gene completeness [81]. Nonetheless, the bacteria_odb10 marker set (n = 124) represents a narrow and conserved subset of genes, and a high completeness score does not preclude assembly errors, collapsed repeats, or missing genomic regions outside the marker set. Furthermore, BUSCO scores may be influenced by phylogenetic distance between the query organism and the reference lineage, and the tool does not detect misassemblies or chimeric contigs that nonetheless harbor intact marker genes, requiring cautious interpretation of completeness estimates alongside complementary quality metrics. Predictions remain computational; no expression or metabolite or phenotypic data were measured, and functional roles cannot be assumed. Also, no correction for multiple testing was applied, therefore, causation cannot be implied from the reported associations in infection outcomes. In addition, our dataset represents a single geographic region and timeframe, and results may not extrapolate to other CA-MRSA populations. Nonetheless, the consistent detection of NIS, NRPS, and metallophore clusters across diverse clonal complexes suggests these features are robust and biologically relevant. Additional constraints also exist, including assembly fragmentation, as they are currently draft MRSA genomes, and this would possibly affect NRPS modularity. Further phenotypic assays need to be conducted to confirm the functions of the BGCs.

Future studies would optimally combine genomics with transcriptomics, metabolomics, and phenotypic assays to confirm the activity of predicted clusters. Linking BGC content to clinical data such as infection type or treatment outcome would clarify their role in pathogenicity. Expanding analyses to MRSA from other regions and to related pathogens will provide comparative context and may reveal convergent metabolic strategies. Ultimately, targeting BGC-encoded products, particularly those involved in iron acquisition and anaerobic survival, offers a promising anti-virulence strategy. Such approaches could attenuate virulence without imposing strong bactericidal selection, thereby helping to reduce the emergence of further resistance.

Conclusions

In conclusion, Egyptian CA-MRSA isolates harbour a rich and varied repertoire of specialized and primary metabolism BGCs. Prominent siderophore, NRPS, and metallophore clusters, combined with versatile anaerobic metabolic pathways, likely enhance adaptation to host environments and persistence in the community. This work highlights the importance of BGC mapping in understanding MRSA biology and opens new avenues for therapeutic exploration.

Abbreviations

MRSA

Methicillin-resistant Staphylococcus aureus

CA-MRSA

Community-Acquired MRSA

SMs

Specialized Metabolites

GCs

Gene Clusters

NI

NRPS-independent

NRP

Non-Ribosomal Peptide

NIS

Non-ribosomal peptide synthetase-independent siderophore

MSSA

Methicillin-Sensitive S. aureus

SCCmec

Staphylococcal cassette chromosome mec

PBP2a (PBP2′)

Penicillin-Binding Protein

HA-MRSA

Hospital-associated MRSA

PVL

Panton–Valentine leucocidin

MGEs

Mobile Genetic Elements

NRPS

Non-Ribosomal Peptide Synthetase

RiPPs

Ribosomally Synthesized and Post-translationally Modified Peptides

T3PKS

Type III Polyketide Synthase

MDR

Multi-Drug Resistant

CDC

US Centers for Disease Control and Prevention

NCBI

National Center for Biotechnology Information

antiSMASH

Antibiotics and Secondary Metabolite Analysis SHell

gutSMASH

Specialized Metabolic Analysis of Specialized gene clusters from the Human gut microbiome

CCs

Clonal Complexes

PRISM

Prediction of genetically encoded natural product structures based on microbial genomes

GECCO

Biosynthetic Gene Cluster prediction with Conditional Random Fields.

CRFs

Conditional Random Fields

Prokka

Prokaryotic Genome Annotation Tool

ST

Sequence Type

BiG-SCAPE

Biosynthetic Gene Similarity Clustering and Prospecting Engine

GCFs

Gene cluster families

MAFFT

Multiple Alignment program for amino acid or nucleotide sequences.

NI-siderophore

NRPS-independent, IucA/IucC-like siderophores (siderophore prior to 7.0)

lanthipeptide-class-I

Class I lanthipeptides like nisin

UD

Undetermined

NA

Unclassified

MIBiG

Minimum Information about a Biosynthetic Gene cluster

AMR

Antimicrobial Resistance

NIS synthetases

NI-siderophore synthetases

TCA

Tricarboxylic Acid cycle

SDH

Succinate Dehydrogenase

ADI

Arginine deiminase

DNR

Dissimilatory Nitrate Reductase

TPP

Thiamine pyrophosphate

PCP

Peptidyl Carrier Protein

gDNA

Genomic DNA

CDSs

Coding Sequences

LAP

Linear azol(in)e-containing peptides

REE-containing

RRE-element containing cluster

FA

Fatty Acid

ANR

Assimilatory nitrate reductase

PKSs

Polyketide Synthases

TPS

Terpene

cyc-lac-autoinducer

Cyclic-lactone-autoinducer

FA

Fatty Acid

Pyr

Pyruvate

Acet

Acetate

Unasgd

Unassigned

Acet

Acetate

But

Butyrate

Arg

Arginine

Succin

Succinatte

Caf

Caffeate

Nit_Reduct

Nitrate Reductase

Putr2sper

Putrescine2spermidine

Oxid_gly

Oxidative_glycerol

HMMs

Hidden Markov Models

Authors’ contributions

L.Z. & M.E. generated the main ideas and workflow of the study, and supervised the team, and they also revised the manuscript. A. A. generated the figures for the BGCs and wrote the main manuscript draft with N. A. S., S. Salem and S. S. E. Additionally, R. A. E. analyzed the raw data from the WGS. N. A. S. performed data analysis, data visualization & writing. B. P. and S. Sheppard did the sequencing and reviewed the manuscript. All authors reviewed the manuscript.

Funding

This work was partially supported by the Jesor Program of the Academy of Scientific Research and Technology, and by the Science, Technology and Innovation Funding Authority (STDF) through the One Health Research and Innovation Grant (OHRIG), Project ID: 48072.

Data availability

All the data associated with the manuscript are included in the figures and supplementary figures. Additionally, the used codes are stored in a GitHub repository (https://github.com/AhmedAboushanab/Genomic-Insights-into-Specialized-and-Primary-Metabolism-Gene-Clusters). For the assembly sequences, they can be accessed from the following: BioProject PRJNA1135495 (https://www.ncbi.nlm.nih.gov/bioproject/?term=PRJNA1135495).

Declarations

Ethics approval and consent to participate

This retrospective study involved genomic characterization of historical strains collected during the study period, with no collection of patient data. The Ethics approval for this study was waived by the Ethics committee Zewail City IRB in compliance with the Declaration of Helsinki. Written informed consent was obtained from all participants included in the study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.