Genome-based analyses from four clinically-isolated strains refined the taxonomy of Proteus genomosp. 6 and revealed their underestimated role in gastrointestinal diseases

Four patients were diagnosed with acute appendicitis with peritonitis or abscess, and were accompanied by microbial infection based on clinical blood test diagnosis, which revealed elevated levels of C-reactive protein, increased white blood cell counts, and a heightened neutrophil percentage. During the appendectomy or ultrasound-guided drainage, appendiceal pus was collected from the patients for the isolation of the causative microorganisms. Briefly, appendiceal pus samples were collected and immediately transferred to the department of clinical laboratory for microbial culture. Subsequently, 100 µL aliquots of the pus from each sample were evenly spread onto blood agar plates or Luria–Bertani Agar (1.5%) (LBA) plates and incubated separately under anaerobic and aerobic conditions at 37 °C for 48 h. Colonies that appeared on the plates were subsequently re-streaked to ensure the acquisition of pure cultures.

The obtained pure cultures were subjected to taxonomic identification using an automated bacterial identification instrument (VITEK^®-32; bioMérieux, Marcy-l’Étoile, France) as per the manufacturer’s instructions. One single Proteus strain was exclusively isolated from each sample, with no other aerobes or strict anaerobes recovered (see Table S1 for the detailed identification outcomes). Antimicrobial susceptibility testing was conducted by employing the VITEK^®-2 v9.02.3 automated system with the AST-GN13 card and AST-N33 card. The minimum inhibitory concentration (MIC) and antimicrobial susceptibility of the isolates on the tested antimicrobials were determined as per the Clinical and Laboratory Standards Institute guidelines (CLSI 2023).

Bacterial cells were harvested from LBA plates by gentle washing with 1 mL of sterile phosphate-buffered saline (PBS). Bacteria cells were then pelleted by centrifugation at 10 000 × g for 5 min for DNA extraction. Genomic DNA was promptly extracted using the TaKaRa MiniBEST Bacteria Genomic DNA Extraction Kit (Takara Bio Inc., Shiga, Japan), strictly adhering to the manufacturer’s protocol. The quality of the extracted DNA was assessed employing the Synergy HTX Multi-Mode Reader (BioTek, USA). Whole-genome sequencing was conducted on the Illumina NovaSeq™ (Illumina, Inc., CA, USA) platform. The sequencing library was prepared with a 350 bp insert size and sequenced using the pair-end strategy (PE150). Quality control of the raw sequencing reads was meticulously performed using fastp v0.20.1 [13], which involved adapter trimming and the removal of low-quality reads, ensuring that over 95% of the reads had a quality score higher than Q20 (Phred-score). Genome assembly was performed using Unicycler v0.4.9b 14, and contig sequences shorter than 200 bp (required by the NCBI genome database in the submission process) were removed from the assembled genomes.

The GTDB-Tk software, specifically employing the gtdbtk_wf workflow, was utilized to conduct a rigorous validation of the taxonomic assignments for the isolated strains 15. To identify genomes with high sequence similarity to those of the four sequenced strains, the Similar Genome Finder tool integrated within the PATRIC v3.6.10 web-server was employed. The 50 top similar genome sequences, along with their associated metadata, were retrieved from the NCBI genome database. Subsequently, the whole-genome average nucleotide identity (ANI) between the assembled genomes and these similar genomes was calculated using fastANI 16. Furthermore, the core genome sequences of these genomes were extracted and aligned using the Roary (v3.13.0) pan genome pipeline, configured with the parameters “-e -v -n”, which invokes MAFFT for multiFASTA alignment (a 95% minimum sequence identity were used to cluster orthologs) 17. A maximum-likelihood core-genomic phylogenetic tree was then constructed, employing the generalized time-reverse model with FastTree (v2.1.11) utilizing the parameter “-gtr” 18. The virulence coding gene and antimicrobial resistance gene carried by these genomes were predicted using the ABRicate v1.00 (https://github.com/tseemann/abricate), by searching against the pre-calculated NCBI antimicrobial resistance (AMR) database (AMRFinderPlus database) and the “vfdb” virulence database 19.

Single colonies of the four strains were subjected to three successive re-streaking on LB agar plates t to ensure the isolation of pure colonies. For shotgun whole-genome sequencing, low-quality reads were removed as part of quality control. Moreover, the completeness and contamination of the assembled genome sequences were meticulously evaluated using CheckM v1.0.12 [20]. CheckM v1.0.12 analysis revealed that all assembled genomes had 100% completeness and 0% contamination.

Of the four patients diagnosed with acute appendicitis, three underwent appendectomy (laparoscopic appendectomy) and one underwent ultrasound-guided drainage (Table 1). Bacterial cultures were recovered from each of the postoperative specimens and were identified as Proteus spp. (Table S1). Notably, three patients (Case 1, 3, 4) exhibited severe appendicitis with complications such as peritonitis, abscess formation, or chronic inflammation. One (Case 2) presented with delayed treatment due to a psychiatric comorbidity (schizophrenia), leading to advanced complications (perforation, abscess, and intestinal obstruction) (Table 1). These findings underscore the relevance of Proteus spp. in complicated appendicitis and suggest its potential role in secondary infections. Further genomic and phenotypic analyses of the isolates would be conducted in the future to elucidate their pathogenicity and resistance mechanisms.

NGS of the four strains yielded 8,869,464 to 9,831,412 raw reads. After quality control and adaptor removal, 8,835,548 to 9,774,520 clean reads were obtained (Table S2). Subsequently, draft genomes were generated for the four strains through genomic assembly. The number of contigs in the draft genomes was as follows: DFP240708, 59 contigs; TSJ240517, 42 contigs; LHD240705, 63 contigs; and WDL240414, 56 contigs. The genome sizes ranged from 3.85 to 4.11 Mbp. The GC contents were 38.00%, 38.14%, 38.02%, and 38.20%, respectively. The number of coding DNA sequences (CDSs) encoded were 3,482, 3,656, 3,536, and 3,783, respectively. Notably, strain WDL240414 had the largest genome and the highest number of CDSs (Table S2). GTDBtk analysis revealed that all four strains belongto Proteus genomosp. 6 (ANI cutoff used for species delineation is 95%)rather than Proteus vulgaris or Proteus hauseri, as identified by the VITEK^®-32 system (Table S1). The misidentification by the VITEK^®-32 system is primarily due to the high phenotypic similarity among certain Proteus species, such as nearly indistinguishable biochemical profiles (e.g., urease activity, carbohydrate fermentation patterns, and swarming behavior) between phylogenetically-close Proteus spp. 3. These overlapping traits challenge phenotype-based identification systems that rely on conventional biochemical assays, emphasizing the necessity of genomic methods to resolve such taxonomic ambiguities within the Proteus genus.

Using the Similar Genome Finder tool integrated in the PATRIC web-server (https://www.bv-brc.org/app/GenomeDistance), we identified the 50 closest genomes in public databases to the four strains isolated in our research. After removing duplicates, a total of 52 strains were included for comparative genomic analysis with our strains, with the majority isolated from China (22 strains) and the United States (26 strains) (Table S3). ANI comparisons grouped these genomes into two major clades. Clade I predominantly consisted of P. terrae, including its two subspecies (i.e., P. terrae sub. terrae and P. terrae sub. cibarius). Clade II could be further divided into two sub-clades, namely Clade II-a and Clade II-b (Fig. 1a). Clade II-a was mainly constituted by Proteus genomosp. 6, while clade II-b encompassed P. columbae, P. appendicitidis, P. hauseri, and P. alimentorum (Fig. 1a). The four strains isolated in our study clustered within clade II-a, belonging to Proteus genomosp. 6. ANI values indicated that their closest species were P. columbae and P. appendicitidis. Notably, based on the ANI threshold for species delineation, a high rate of misidentification among P. columbae isolates was revealed. Of the 10 P. columbae strains analyzed (including eight clinically-related isolates), eight were reassigned to Proteus genomosp. 6, forming a distinct clade (Clade II-a; Fig. 1a). Critically, seven out of the eight clinically-related P. columbae strains (87.5%) were reclassified as Proteus genomosp. 6. This finding underscores that the prevalence of Proteus genomosp. 6 in clinical settings may have been significantly underestimated, highlighting the necessity to use genomic-based taxonomic approaches in clinical microbiology. This finding was further supported by the core-genomic phylogenetic tree (Figure S1). As a genomospecies rather than a formally named species, its phenotypic characteristics are typically excluded from comparative analyses during the nomenclature and classification of novel Proteus species, which has resulted in incomplete resolution of phenotypic distinctions between Proteus genomosp. 6 and recently described species such as P. columbae (first identified in 2018) [1]. In this study, only the type strain 08MAS2615^T was confirmed as the authentic P. columbae (Fig. 1a). By performing a literature review [1, 2], we found that the phenotypic profiles of P. columbae 08MAS2615^T and Proteus genomosp. 6 (11 strains) share numerous metabolic traits. For instance, both P. columbae 08MAS2615^T and all Proteus genomosp. 6 strains exhibited positive reactions for indole production, sucrose utilization, and maltose utilization, while being negative for L-rhamnose utilization. However, P. columbae 08MAS2615^T was negative for salicin and aesculin ferric citrate utilization, whereas 9% of Proteus genomosp. 6 strains exhibited positive reactions for these substrates. Additionally, P. columbae 08MAS2615^T was positive for methyl α-D-glucopyranoside and trehalose utilization, whereas only 10% and 20% of Proteus genomosp. 6 strains exhibited positive reactions for these substrates, respectively. Furthermore, the four Proteus genomosp. 6 strains isolated in this study were positive for sucrose and maltose utilization but negative for trehalose utilization (see Table S4). These findings underscore the phenotypic similarity between P. columbae and Proteus genomosp. 6, as well as the variability in substrate utilization among different Proteus genomosp. 6 strains, thereby complicating their differentiation based solely on phenotypic assays.

Using the genome of strain WDL240414 as a reference, a circle genome map comparison with the genomes of the four strains isolated in our study, as well as Proteus genomosp. 6 ATCC51471 and P. appendicitidis HZ0627, showed extremely high genomic sequence similarity. These strains encode multiple drug resistance genes, including arnT (peptide antibiotic resistance), vanG (glycopeptide resistance), qacG (disinfectant resistance), and kpnE (multidrug resistance) (Fig. 1b). Additionally, the genome of strain WDL240414 also possessed some unique gene sequences, such as dfrA1, SAT-2, umuD, and endA, encoding diaminopyrimidine antibiotic resistance, nucleoside antibiotic resistance, translation error-prone DNA polymerase V and transcriptional activator, respectively. These unique sequences are likely associated with genomic mobile elements (e.g., integration, excision, and transfer), suggesting that they may have been acquired through horizontal gene transfer events (Fig. 1b).

Among the 56 analyzed genomes, 20 classes of antibiotic resistance genes (annotated via the AMRFinderPlus database) were predicted, conferring resistance to 12 antibiotics (Table S5). Of these, 12 classes were identified in Proteus genomosp. 6 genomes, encoding resistance to quinolones, beta-lactams, tetracyclines, streptomycin, streptothricin, chloramphenicol, trimethoprim, kanamycin, sulfonamides, florfenicol, gentamicin, and hygromycin. Conversely, eight classes were absent, including those encoding resistance to rifamycins, aminoglycosides (e.g., amikacin/kanamycin/ tobramycin), bleomycin, carbapenems, cephalosporins, and macrolides (Table S5). Except Proteus alimentorum 08MAS0041, all Proteus strains analyzed in this study carried beta-lactam antibiotic resistance genes (Fig. 2). Clinical antibiotic resistance testing of the four strains in this study revealed variable resistance profiles to beta-lactam, cephalosporin, quinolone, and sulfonamide antibiotics, while uniform sensitivity to aminoglycoside antibiotics was observed (Table S5). For instance, all four strains exhibited resistance to cefuroxime but were sensitive to ceftazidime, cefepime, gentamicin, tobramycin and amikacin. Strain LHD240705 showed resistance to cefotaxime, whereas the other three strains were sensitive to cefotaxime (Table S5). All four isolated strains exhibited resistance to at least one beta-lactam or cephalosporin, likely due to the presence of the class A beta-lactamase gene hugA. (Table S5, Fig. 2). The HugA is known to hydrolyze cefuroxime, leading to resistance against this antibiotic 21. This has been observed in clinical isolates, such as Proteus penneri, where the hugA gene and its regulator gene hugR are present, and regulatory changes may lead to increased production of the enzyme, confering resistance to cefuroxime but not ceftazidime, cefepime and imipenem 21. Morever, their susceptibility to the tested aminoglycosides aligns with their antibiotic resistance gene profiles on aminoglycosides (Table S5, Fig. 2). A 2019 systematic review supports aminoglycosides as a clinically effective first-line treatment for Proteus infections. For instance, among 33 Proteus spp. isolates from urinary tract infections (UTIs) analyzed in that study, only two exhibited resistance to aminoglycosides 22. As such, in the clinical management of infections caused by these bacteria, the use of beta-lactam and cephalosporin antibiotics should be avoided when possible, while aminoglycoside antibiotics, such as amikacin, may serve as the preferred therapeutic agents. Moreover, the genome of this strain was found to carry the highest number of resistance genes among the Proteus genomosp. 6 strains, including genes conferring resistance to beta-lactams, quinolones, trimethoprim, streptomycin, streptothricin, tetracyclines, sulfonamides, and chloramphenicol/florfenicol (Fig. 2). Despite the absence of known carbapenem resistance genes, strain WDL240414 exhibited resistance to imipenem and multidrug resistance, including cephalosporins (cefuroxime), quinolones (levofloxacin, ciprofloxacin), sulfonamides (trimethoprim), beta-lactams (ampicillin, cefazolin), and nitrofurans (nitrofurantoin). The imipenem resistance of strain WDL240414 may indicate the presence of porin mutations or efflux-mediated resistance, requiring further investigation. Its resistance to levofloxacin, ciprofloxacin, trimethoprim, and sulfamethoxazole aligns with their antibiotic resistance gene profiles (Table S5). A Kruskal-Wallis non-parametric test was conducted to assess differences in AMR gene profiles across six Proteus species (p = 5.68 × 10^− 5), revealing statistically significant differences among them. Pairwise comparisons revealed that no significant differences between P. terrae and Proteus genomosp. 6, wheras Proteus genomosp. 6 harbored 38.85%, 47.51%, 73.75%, and 73.75% more AMR genes than P. hauseri, P. columbae, P. appendicitidis and P. alimentorum, respectively (p < 0.05). Notably, Proteus genomosp. 6 exhibited a significantly higher AMR gene burden compared to its phylogenetically close Proteus species. This genomic divergence in AMR coding capacity may reflect adaptive evolution under selective pressures (e.g., antibiotic exposure), highlighting the need for functional studies to validate genotype-phenotype linkages and assess clinical implications of these resistance determinants, highlighting the imperative for accurate taxa identification and monitoring of its antibiotic resistance in clinical settings.

Among the 56 genomes, only four strains carried virulence genes (hlyA, hlyB, hlyD) encoding hemolysin (Table S6), a toxin known to damage host cell membranes. Hemolysin has been identified in various pathogenic Gram-negative bacteria, including Proteus spp., Morganella spp., and Escherichia coli23, 24. Two of these four strains, namely TSJ240517 and WDL240414, were isolated and identified in this study, carrying the complete hlyABD gene cluster (Table S6), hemolysin may thus represent one of their virulence factors. Hemolysins (e.g., HlyA, HlyB, HlyD) in Proteus spp. are pore-forming toxins that lyse erythrocytes and other host cells, facilitating iron acquisition and tissue invasion 10. In Proteus genomosp. 6, hemolysin production may enhance pathogenicity by damaging epithelial barriers, triggering inflammation, and promoting bacterial dissemination. While direct evidence linking hemolysins to appendicitis specifically is limited, studies in related P. mirabilis suggest hemolysins contribute to virulence in urinary tract and wound infections 10. The absence of these genes in some strains could reflect niche-specific adaptation or attenuated virulence. In appendicitis, hemolysin-positive strains might exacerbate mucosal damage or secondary infections, though further mechanistic studies are needed to confirm their role.