A 23-year-old male patient with a history of Tourette syndrome and a documented penicillin allergy presented to the emergency department after sustaining a traumatic fall from a second-story balcony. Upon admission, the patient reported severe pain in the cervical, thoracic, and lumbar regions, with additional discomfort in the thorax and both ankles. The initial examination revealed an unstable compression fracture of the fourth lumbar vertebra without evidence of an open wound or external communication. Imaging confirmed the lumbar fracture without additional injuries to the head, thorax, or abdomen. The patient was admitted for surgical treatment of the spinal fracture and underwent transpedicular dorsal stabilization of L2 to L5 at the end of September 2024. The postoperative course was initially uneventful, and the patient demonstrated adequate recovery. However, despite a sudden deterioration in the patient’s infection markers, he was discharged against medical advice 11 days after admission, without undergoing any antibiotic treatment. Three days after being discharged, the patient presented at a nearby hospital with worsening general condition and progressive pain. As a result, the patient was readmitted to our clinic for further treatment. Upon readmission, the patient exhibited signs of infection, including fever and elevated inflammatory markers, with a C-reactive protein (CRP) level of 282 mg/L and leukocytosis with a white blood cell count 11.9 × 10⁹/L. A series of blood cultures were collected for microbiological examination. Further, a computed tomography (CT) scan was performed, revealing signs of spondylodiscitis at the L4 vertebral level (Figure 1). Consequently, the patient underwent revision surgery, which revealed pus in the wound extending to the osteosynthesis. Following meticulous surgical debridement and lavage, five tissue samples were obtained from different areas. The transpedicular osteosynthesis was retained as the fracture had not yet consolidated.

Bacterial growth was observed after overnight cultivation of the tissue samples, with colonies appearing white and haemolytic on Columbia agar plates with 5% sheep blood (BD Diagnostics, Heidelberg, Germany). All tested colonies were positive in the catalase test, while the clumping factor test using the Pastorex Staph-Plus (Bio-Rad, Marnes-la-Coquette, France) was negative. Identification of the bacterial colonies from these samples was performed using MALDI-TOF MS utilizing the MALDI Biotyper® sirius system (Bruker Daltonics, Bremen, Germany) with MBT Biotargets 96 (Bruker Daltonics). Identification was initially performed using two protocols, direct transfer (DT) and extended direct transfer (eDT), following the manufacturer’s instructions and a previous study (Letunic and Bork, 2024). Some spots showed no peaks or identification scores below 2.00 and therefor the median identification score was 1.70 (range = 1.60–2.12). To improve sensitivity and accuracy, a protein extraction (PE) procedure was applied before repeating MALDI-TOF MS measurements (Idelevich et al., 2023). This resulted in a marked increase in the identification scores, with a median value of 2.22 (range = 2.08–2.37), and all isolates were unequivocally identified as S. pseudintermedius. For quality control and standardization, a 1 μL aliquot of IVD Bacterial Test Standard (IVD BTS, Bruker) was included on the MBT Biotarget during all three procedures, per manufacturer guidelines.

Of the six-vial blood culture set initially collected, only one aerobic vial (BACTEC™ Plus; BD Diagnostics) tested positive. Gram staining showed gram-positive cocci consistent with staphylococci. This vial was analyzed using the loop-mediated isothermal amplification-based eazyplex® MRSA kit (Amplex Diagnostics), which detects S. aureus complex species (S. aureus, S. argenteus, S. schweitzeri, S. roterodami), S. epidermidis, and mecA/mecC resistance genes. No target genes were detected in this case. The positive blood culture vial was subcultured on Columbia blood agar (BD Diagnostics, Heidelberg, Germany), yielding white, hemolytic colonies similar to those from tissue samples, both negative for clumping factor (Pastorex Staph-Plus, Bio-Rad, Marnes-la-Coquette, France). After protein extraction procedure (PE), MALDI-TOF MS identified all isolates from tissue and blood cultures as S. pseudintermedius with median score of 2.22 (range = 2.11–2.28). To exclude contamination, two additional blood culture sets were collected earlier; one aerobic vial was positive, and its colonies were identified as S. pseudintermedius. Except for these two positive aerobic vials, all other blood cultures remained negative after 6 days of incubation. No other bacteria or fungi were detected in clinical or screening samples throughout treatment.

The antimicrobial susceptibility testing (AST) of the isolates was performed using the VITEK 2 system (bioMérieux, Marcy l’Etoile, France) and VITEK software (Version 9.04) following the manufacturer’s instructions with the AST-P-654 test cards. Phenotypic methicillin resistance was further assessed by disk diffusion testing using oxacillin (1 μg), as recommended by the European Committee on Antimicrobial Susceptibility Testing (EUCAST) for screening in S. pseudintermedius, S. intermedius, S. schleiferi, and S. coagulans. AST results were interpreted according to the EUCAST Breakpoint Tables for MICs and zone diameters, version 14.0. All isolates were found to be susceptible to all tested antibiotics, except for erythromycin and clindamycin (Table 1).

Bacteria were cultured on blood agar plates (Becton, Dickinson and Company [BD], USA) at 37 °C for 16–24 h. Cells were harvested using a 1 μL inoculation loop, and genomic DNA was extracted with the NucleoSpin® Microbial DNA Kit (Macherey-Nagel, Germany) according to the manufacturer’s instructions. DNA purity and quality were assessed using the NanoDrop™ One/OneC Spectrophotometer (Thermo Fisher Scientific, USA), while DNA concentrations were determined using the Qubit™ 1X dsDNA Assay Kit and Qubit™ 4 Fluorometer (Thermo Fisher Scientific, USA). DNA was either used immediately for library preparation or stored at −80 °C until further processing.

Due to the use of two different sequencing platforms, separate library preparation protocols were followed. For Illumina short-read sequencing, libraries were prepared using the Illumina DNA Prep Kit following the manufacturer’s instructions, and sequencing was performed on a MiSeqDx system (Illumina, USA) using a 2 × 300 bp paired-end configuration.

For Oxford Nanopore (ONT) long-read sequencing, libraries were prepared using the Rapid Barcoding Kit SQK-RBK114.24 (Oxford Nanopore Technologies, UK) as per the manufacturer’s protocol. Sequencing was conducted using a MinION device equipped with an R10.4.1 flow cell (FLO-MIN114).

Raw Illumina reads were processed using the platform’s default basecalling settings, and the resulting FASTQ files were subsequently used for hybrid assemblies. For Oxford Nanopore Technologies (ONT) sequencing, raw signal data (POD5 files) were basecalled using the super-accurate (SUP) model dna_r10.4.1_e8.2_400bps_sup@v5.0.0 with the Dorado v0.9.1 basecaller¹. The resulting ONT FASTQ reads were first used for genome size estimation with Long Read-based Genome size Estimation (LRGE) (Hall et al., 2024), followed by downsampling to 100 × coverage using Rasusa (Randomly Subsample Sequencing Reads or Alignments) (Hall, 2022). Downsampled FASTQ files were then assembled using the Autocycler pipeline v0.2.1 in fully automated mode, utilizing the following long-read assemblers: Flye, Miniasm, NextDenovo, and Raven. Resulting consensus sequences were polished with Medaka v2.0.1, applying the bacterial methylation model r1041_e82_400bps_bacterial_methylation². Polished assemblies were reoriented using Dnaapler (Bouras et al., 2024), aligning sequences to start at the dnaN gene for chromosomes or the rep gene for plasmids, thereby standardizing orientation across all sequenced isolates. Final assemblies were hybrid-polished using high-quality Illumina short reads. These reads were quality-trimmed with Trimmomatic v0.39, and final polishing was performed using Polypolish v0.6.0 (Bouras et al., 2024; Wick and Holt, 2022; Bolger et al., 2014). The resulting high-quality consensus sequences were used in all downstream analyses.

Initially, the consensus sequences were analyzed using the TORMES pipeline v1.3.0³ (Quijada et al., 2019) for genome annotation via Prokka (Seemann, 2014) and identification of antimicrobial resistance genes using the CARD, ResFinder, and ARG-ANNOT databases (McArthur et al., 2013; Zankari et al., 2012; Gupta et al., 2014) (Supplementary Table S1). In addition, the sequences of S. pseudintermedius HGW2412 and pHGW2412 were annotated using the Bakta pipeline⁴ (Schwengers et al., 2021) (Supplementary Table S1).

Subsequently, Prokka-annotated sequences were employed for virulence gene identification. To this end, the complete protein dataset of the Virulence Factor Database (VFDB⁵) was downloaded, and Staphylococcus-specific virulence factors were extracted. These were used in a BLAST+ (v2.16.0+) search⁶ against the HGW2412 proteome. Hits were filtered using the following thresholds (Supplementary Table S1): alignment coverage (= align_l/s_end) ≥ 50%, sequence identity ≥ 30%, and bit score ≥ 100.

To assess the phylogenetic context of HGW2412, 5,500 S. pseudintermedius genomes were retrieved from the NCBI Reference Sequence Database (RefSeq)⁷. Pairwise genomic similarity was computed using the Fast Average Nucleotide Identity algorithm (FastANI)⁸ (Jain et al., 2018), and the 100 most closely related genomes were selected for downstream analysis.

These 100 genomes, along with HGW2412, were first subjected to MLST profiling in SeqSphere+ v10.5.0 using the Solyman et al. (2013) scheme, and a minimum spanning tree was constructed.

In parallel, core SNP-based phylogenetic analysis was conducted. An optimal k-mer size of 17 was determined using Kchooser4, and SNP calling was performed with kSNP4.1 (Hall and Nisbet, 2023). A maximum likelihood (ML) phylogenetic tree was then inferred from the core SNP matrix using RAxML v8.2.12 (Stamatakis, 2014), applying the GTRGAMMA substitution model and 1,000 bootstrap replicates. The annotated ML tree, including information on isolation source, country, and bootstrap values, was visualized using Interactive Tree of Life (iTOL v7.1.1) (Letunic and Bork, 2024).

To investigate AMR determinants, a total of 5.500 S. pseudintermedius genomes and their associated metadata were downloaded from the RefSeq database (NCBI⁹) as described above. AMR gene detection was performed using ABRicate¹⁰ with four reference databases: NCBI AMRFinderPlus, CARD, ResFinder, and ARG-ANNOT (Zankari et al., 2012; Gupta et al., 2014; Feldgarden et al., 2019; Jia et al., 2017). Because the different databases can report partially distinct AMR genes, multiple hits corresponding to the same gene within a genome were collapsed into a single entry, ensuring that each AMR gene was counted only once per isolate. The results from all databases were then merged into a unified presence/absence matrix reflecting the similarity of each detected gene to its respective reference sequence (Supplementary Table S2). A minimum sequence identity threshold of 80% was applied, and gene names were standardized by replacing outdated nomenclature with the currently accepted designations. To calculate AMR gene prevalence across the isolates, all hits were binarized (presence = 1 or 100%). This matrix was subsequently used to identify AMR gene clusters employing the Self-Organizing Tree Algorithm (SOTA) implemented in MeV v4.9.0 (TM4 Software Suite, USA) using default settings exceptional the number of cycles which was adjusted to in total 10 cycles. Fewer cycles resulted in overly coarse clustering with high within-cluster variability, whereas more cycles produced overly fine clusters, some of which contained only very few isolates. The SOTA dendrogram depicts the hierarchical relationships among clusters; however, the accompanying heatmap represents relative internal node values rather than true gene prevalence. Therefore, AMR gene prevalence was recalculated for each cluster and visualized as a heatmap adjacent to the dendrogram in Figure 2. The whole ABRicate results table and all matrices (complete heatmap table of 5,501 sequences and individual SOTA clusters 1–10) are provided in Supplementary Table S2.

A 23-year-old man presented with a closed, unstable compression fracture of the fourth lumbar vertebra following a fall from a second-story balcony. He initially underwent transpedicular stabilization from L2 to L5, and his early postoperative recovery was uneventful. Despite elevated infection markers, he left the hospital against medical advice without receiving antibiotics. Three days later, he returned with worsening back pain, fever, and markedly increased inflammatory markers; imaging revealed spondylodiscitis at L4. During revision surgery, purulent material was observed around the osteosynthesis, and five tissue samples were collected for microbiological analysis. Culture of these samples produced white, hemolytic colonies that were catalase-positive but negative for clumping factor. After protein extraction, MALDI-TOF MS confirmed all isolates as S. pseudintermedius. Blood cultures obtained during readmission also grew S. pseudintermedius in two aerobic vials, while other cultures remained negative. Notably, the patient had no contact with pets either prior to the fall or after discharge, a detail specifically inquired about by the treating physicians given that S. pseudintermedius is typically associated with animals. No mecA or mecC resistance genes were detected, and antimicrobial susceptibility testing indicated that the isolates were susceptible to all tested agents except erythromycin and clindamycin. Collectively, these findings identify S. pseudintermedius as the pathogen responsible for the patient’s postoperative spondylodiscitis. Detailed patient history and microbiological analyses information’s are provided in the Materials and Methods section.

The S. pseudintermedius infection posed a therapeutic challenge due to the patient’s documented penicillin allergy. Although the strain was methicillin-susceptible, beta-lactam antibiotics were contraindicated, necessitating alternative treatment. Empirical therapy with intravenous clindamycin (600 mg TID) was started for 3 days. Following species identification and antimicrobial susceptibility testing, therapy was changed to intravenous daptomycin (8 mg/kg body weight, corresponding to 1,250 mg once daily) for 3 weeks, in accordance with the AST results and the patient’s clinical condition. Five days after the first revision surgery, a second extensive debridement and lavage were performed. The patient responded well, with significant decreases in CRP and leukocyte counts within 2 weeks, no new neurological deficits, and stable follow-up imaging showing no progression of spondylodiscitis. Upon completion of intravenous therapy, the patient was discharged on oral levofloxacin and rifampicin to ensure eradication of residual bacteria, considering their pharmacokinetics and broad-spectrum activity.

At six-week follow-up, the patient was symptom-free with no signs of infection. The transpedicular dorsal spondylodesis was safely removed 6 months later after CT scans confirmed fracture healing.

To complement the microbiological diagnostics, comprehensive WGS was performed on three S. pseudintermedius isolates of the patient. Two isolates were obtained from infected tissue and blood during the initial surgery, and a third isolate was collected 5 days later during wound revision. Hybrid assemblies were generated using both short-read (Illumina) and long-read (Oxford Nanopore Technologies) platforms. All three isolates contained an identical chromosome of 2,626,791 bp and a 3,100 bp plasmid (pHGW2412) (Figure 3). No SNPs or indels were detected between the isolates (data not shown), indicating infection by a single clonal strain. Therefore, all subsequent analyses were conducted using the genome of the initial isolate recovered from tissue.

Initial sequence analysis was performed using the TORMES pipeline (Quijada et al., 2019), including gene annotation with Prokka (Supplementary Table S1 – annotations tabs) and resistance gene screening via ABRicate (Zankari et al., 2012; Gupta et al., 2014; McArthur and Tsang, 2017) (Supplementary Table S1 – antibiotic resistance). The isolate harbored chromosomal genes encoding proteins responsible for resistance to various antibiotics, including β-lactams (blaZ), chloramphenicol (cat), macrolides (ermB), and aminoglycosides [aadE = ant(6)], aph-Stph, aph(3″)-III, and sat4A. Virulence factor screening with the Virulence Factor Database VFDB, see footnote⁵ revealed high-identity protein homologs of S. aureus toxins including beta-hemolysin (Hlb, 74.09%), staphylococcal enterotoxin C (Sec, 57.58%), superantigen-like protein SSL11 (Set26, 44.68%), and leukocidins LukS and LukF from S. intermedius (98.71 and 99.39% respectively) (Supplementary Table S1). Additional gene homologs were detected for iron acquisition systems (sfaABCD, sbnABCDEFGHI, and sirABC), capsule biosynthesis (cap8OP), adhesion (icaABC, vWb, and fnbB), tissue invasion (lip, geh, nuc, and aur), and immune evasion (sbi) (Supplementary Table S1). Furthermore, an autolysin gene homolog (aae) has been detected. Besides other functions, autolysins play a pivotal role in adherence and biofilm formation and have been described for several species such as S. aureus and various coagulase-negative staphylococci (Hussain et al., 2015; Heilmann et al., 1997; Biswas et al., 2006). The mere presence of homologs to these virulence factors suggests a considerable virulence potential of this isolate, which was supported by the course and severity of the infection.

The plasmid pHGW2412 encodes six distinct proteins, none of which resemble known antibiotic resistance determinants. Two proteins were associated with plasmid replication (Rep_1 and a mobilization protein) (Supplementary Table S1 – annotation pHGW2412). BLAST analyses revealed that pHGW2412-like sequences occur in other S. pseudintermedius genomes, often fragmented and unannotated, likely due to older, less accurate assembly methods. However, the exact biological function of the plasmid remains unknown.

For phylogenetic classification, our isolate was compared with 5,500 S. pseudintermedius genomes from the NCBI RefSeq database, identifying GCA_905169555 as the closest related genome using FastANI (Supplementary Table S1) (Jain et al., 2018). The 100 most similar genomes and our isolate were assigned to Multilocus Sequence Typing (MLST) (Solyman et al., 2013), which were visualized in a minimum spanning tree (Figure 4). Notably, none of the 100 closest genomes shared the sequence type (ST2051) of HGW2412, indicating high diversity. The most similar isolate GCA_905169555 differed by two housekeeping alleles, placing it in close proximity to our isolate (Figure 4). However, only one other ST2051 isolate was identified in the PubMLST database, derived from a cat in Wrocław (Poland) in 2017, but unfortunately, no associated whole-genome sequencing data are available.

As no core-genome MLST scheme is available for S. pseudintermedius, phylogeny was reconstructed using a core-SNP approach, and the resulting tree was annotated with host and geographic metadata (Figure 5). Results revealed extensive genomic diversity and global distribution, hindering precise phylogenetic placement. HGW2412 clustered with isolates from North America, the Netherlands, Canada, Slovenia, and Kenya; German isolates appeared more distantly related. These findings were consistent with the MLST-based Minimum Spanning Tree (MST) (Figure 4), highlighting the substantial genetic heterogeneity even among the 100 most similar genomes.

To contextualize the AMR gene repertoire of our isolate HGW2412 within the genetically diverse species S. pseudintermedius, we compared it with all RefSeq genomes available in June 2025 (n = 5,500). In total, 34 AMR genes were identified (Figure 2A and Supplementary Table S2 – results ABRicates/heatmap_5,501 sequences) and assigned to functional gene families. Thirteen genes were associated with aminoglycoside resistance; seven, including regulatory genes, with β-lactam inactivation; four with macrolide resistance; three with tetracycline resistance; two with phenicol resistance; and one gene each with resistance to fosfomycin, fusidic acid, trimethoprim, mupirocin, and oxazolidinones. Aminoglycoside resistance was dominated by aph-Stph (99.7%), while ant6, aph(3″)-III, and sat were detected in approximately 40% of genomes, and aac6-aph2 and aph2 in about 30%. Among β-lactam resistance genes, blaZ (78.4%), mecA (39.5%), and blaTEM-116 (0.14%, n = 8) were identified. MLS resistance genes included erm (42.3%) and mefA (27.0%), whereas lnu and lsaE were rare (2.4 and 0.8%). Tetracycline resistance was mainly associated with tetM or tetO (45.5%), while tetK or tetL occurred in 6.8%. Phenicol resistance genes included cat (14.4%) and fexA (0.4%). Dihydrofolate reductase genes (dfr) (trimethoprim resistance) were present in 37% of genomes. All remaining AMR genes occurred in fewer than 1% of isolates (Figure 2A and Supplementary Table S2 – heatmap_5,501 sequences).

Cluster (cl) analysis of AMR gene prevalence revealed three major groups (Figure 2B and Supplementary Table S2 – SOTA_cl:1 – SOTA _cl:10). Group 1 (clusters 1–2) exhibited a multidrug-resistant gene profile, with six aminoglycoside resistance genes present in >98% of genomes, high frequencies of blaZ (>98%) and mecA (100% in cluster 1, 83.2% in cluster 2), and erm genes in 98.4%. dfr genes were also common (>97% in cluster 1; 82.3% in cluster 2). Tetracycline resistance was dominated by tetM/tetO in cluster 2 (>91%), whereas tetK/tetL were more frequent in cluster 1 (33.8%). The mecA regulators mecI and mecR were detected exclusively in cluster 1. Group 2 (clusters 3–7) was characterized by the generally low prevalence of aac6-aph2 and the mecA regulatory genes mecI and mecR. Clusters 3–5 also showed low levels of blaI/blaR, and blaZ was absent in clusters 4 and 5. Our isolate HGW2412 belonged to cluster 7, which displayed high prevalences of ant6, aph(3″)-III, aph-Stph, and sat (99.7, 99.0, 100.0, and 99.0%), as well as blaI, blaR, blaZ, and erm (96.2, 85.3, 94.9, and 91.1%, respectively). Approximately 30% of cluster 7 genomes carried tetM/tetO, and nearly 60% carried the cat gene, which was also found in HGW2412. Group 3 (clusters 8–10) contained isolates with few AMR genes. Cluster 8, the largest cluster (n = 2,059; 37.4%), harbored aph-Stph (99.7%), the blaIRZ locus, and ~30% mefA, tetM, or tetO. Cluster 9 exhibited the lowest AMR gene content, containing only aph-Stph (100%) despite representing 13.3% of all genomes (n = 732). Cluster 10 resembled cluster 9 but showed higher prevalences of blaI (26.8%), blaZ (25.7%), and mefA (100%). In sum, a total of 42.5% (n = 2,337) of isolates (clusters 1–7) had gene combinations that predicted resistance to aminoglycosides, β-lactams, macrolides, sulfonamides/trimethoprim, and/or tetracyclines. By contrast, most genomes (57.5%, n = 3,164) harbored only a limited number of AMR genes, primarily predicting resistance to aminoglycosides, β-lactams, and/or macrolides, and were represented largely by clusters 8–10. Notably, the mecA gene was rarely detected in clusters 8–10. Finally, the most frequent AMR gene combinations were analyzed and computed for each cluster and are summarized in Table 2 and Supplementary Table S2. As expected, these patterns mirrored the overall AMR gene distribution; however, they also revealed substantial heterogeneity and a broad range of possible AMR gene constellations, particularly within clusters 1–7.