Microbiota in Chronic Suppurative Otitis media: association with Postoperative Tympanic membrane outcomes

Abstract

Abstract

Chronic suppurative otitis media (CSOM) is a prevalent condition with global health implications due to its impact on hearing and quality of life. Conventional treatments often fail because of bacterial biofilms and antimicrobial resistance. Effective treatment of CSOM depends on the precise determination of the middle ear microbiota; however, current microbial detection methods do not meet this need. Postoperative reperforation may compromise surgical outcomes. If the risk of perforation can be predicted immediately after surgery, sensitive antibiotics could be administered proactively for early intervention to optimize treatment efficacy. This study introduces 2b-RAD sequencing for the microbiome (2b-RAD-M), a novel technology designed to provide a comprehensive profile of the CSOM microbiota and identify diagnostic biomarkers that predict postoperative outcomes. We analyzed ear swabs from patients with postoperative perforation (PO), nonperforation (NPO), and otosclerosis (CON) using microbial diversity, relative abundance, and composition analyses. Bacillus_A_bombysepticus and Pseudomonas aeruginosa were identified as potential biomarkers, with Bacillus_A_bombysepticus demonstrating superior diagnostic accuracy (area under curve (AUC) = 0.92) compared to P. aeruginosa (AUC = 0.25). Functional predictions revealed that biological activities related to gene regulation, substance metabolism, and DNA repair were more prominent in the PO group. This study offers new insights into CSOM pathogenesis and progression, proposing Bacillus_A_bombysepticus as a novel biomarker for predicting postoperative outcomes that can indicate an increased risk of tympanic membrane reperforation for the first time.

Key points

• 2b-RAD-M technology enables comprehensive CSOM microbiota profiling and biomarker identification.

• Bacillus_A_bombysepticus (AUC = 0.92) outperforms Pseudomonas aeruginosa in diagnostic accuracy.

• Bacillus_A_bombysepticus predicts postoperative tympanic membrane reperforation via functional activity analysis.

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at 10.1007/s00253-026-13770-9.

Introduction

Chronic suppurative otitis media (CSOM) is one of the most common otolaryngological diseases, characterized by persistent ear discharge through a perforated tympanic membrane for over 6 weeks. It is a global health issue, with more than 30 million new cases annually and an estimated global prevalence that exceeds 200 million cases (Chadha et al. 2021). The hearing impairment and tinnitus associated with CSOM have long-term negative effects on patients’ quality of life and mental health, creating a substantial financial burden (Sidam et al. 2024). Although surgery and antibiotics are effective for CSOM treatment, the formation of bacterial biofilms and development of antimicrobial resistance can lead to treatment failure, which ultimately results in the recurrence of tympanic membrane perforation (Nguyen et al. 2012; Santa Maria et al. 2021). Furthermore, an inadequate understanding of the pathogenesis has hindered the development of new therapies (Bhutta et al. 2017, 2024; Dhingra et al. 2023). Thus, CSOM management remains a challenge for otologists.

The pathogenesis of CSOM is believed to be intricately linked to changes in the middle ear microenvironment (Frank et al. 2022; Neeff et al. 2016). Due to the limitations of early detection technologies, the middle ear was considered a sterile space (Hall-Stoodley et al. 2006; Westerberg et al. 2009). Consequently, CSOM was thought to be caused by pathogens that entered the middle ear from either the nasopharynx or the external auditory canal (Schilder et al. 2016). With the development of sequencing technology, however, it has been confirmed that the middle ear is not a sterile environment, suggesting that the occurrence and development of CSOM are related to the disruption of the microbial balance (Neeff et al. 2016). Although numerous bacteria have been confirmed to play significant roles in the pathology of CSOM, the involvement of fungi and other pathogens remains uncertain (Daniel 2012; Shangali et al. 2023). The correlation between microbiome research findings and the treatment outcomes of CSOM patients also requires exploration (Coleman et al. 2018; Fujikawa et al. 2022). Therefore, a deeper understanding of the CSOM microbiota is needed for clinicians to develop optimal treatment strategies and prevent potential complications.

Currently, hospitals routinely use culture techniques to identify the microbiota of CSOM. However, these methods are limited to the detection of specific microorganisms, overlooking other potentially significant pathogens. Moreover, many microorganisms are extremely difficult to culture under standard conditions, leading to possible omission of these crucial pathogens (Chen et al. 2022; Suttle et al. 2024). The collection of ear secretion samples also poses a risk of contamination by microorganisms from the external auditory canal, compromising the accuracy of detection results. To overcome these limitations, gene sequencing technology has been introduced to characterize the microbiota community composition in the middle ear. Although 16S rRNA sequencing can provide more comprehensive information about the middle ear microbiota, it generally offers only genus-level classification and cannot effectively identify other microorganisms, such as fungi (Frank et al. 2022; Neeff et al. 2016). Moreover, 16S rRNA sequencing cannot directly provide information about microbial metabolic functions. Although whole metagenome shotgun sequencing can provide precise, reliable microbial information at the species level, this technique is time-consuming and costly. Additionally, ear swab samples often fail to meet the quality and quantity requirements necessary for whole metagenome shotgun sequencing (Chen et al. 2023; Ordinola-Zapata et al. 2024; Schlegel et al. 2023). As a result, there is an urgent need for alternative techniques in CSOM microbiota research to overcome these challenges.

Here, we introduce 2b-RAD sequencing for the microbiome (2b-RAD-M), a novel sequencing method, to the CSOM microbial community study. This method uses type IIB restriction enzymes to digest the genomic DNA of the samples, producing DNA fragments of uniform length. These fragments are then amplified for sequencing and mapped to species-specific 2b-RAD markers for microbial characterization and quantification (Huang et al. 2024; Lam et al. 2022; Xiao et al. 2024). By overcoming the limitations of conventional sequencing methods and microbial culture, 2b-RAD-M sequencing provides a comprehensive, species-resolved microbial profile of the CSOM microenvironment and enables prediction of therapeutic outcomes, allowing for prophylactic treatment in patients at risk of tympanic membrane perforation. We believe this innovative technological approach holds the potential to prognosticate surgical outcomes in CSOM and provide novel insights into therapeutic strategies.

Methods

Ethical considerations

The Ethics Committee of the Eye and ENT Hospital, Fudan University, Shanghai, China, approved the study protocol (approval number: 20200423) on April 23, 2020. The study was registered with the Chinese Clinical Trial Register (ChiCTR: 2000038981).

Patient selection and diagnostic criteria

Patients with Chronic Suppurative Otitis Media (CSOM) were enrolled and subsequently categorized into either the perforation (PO) or non-perforation (NPO) group, based on the occurrence of postoperative perforation recurrence. The definition of CSOM used for this study is persistent inflammation of the middle ear with associated otorrhea through a tympanic membrane perforation, lasting beyond 6 weeks despite medical treatment (Levi and O'Reilly 2023). High-resolution computed tomography (HRCT) of the temporal bone was routinely performed to assess the extent of middle ear pathology.

For the control (CON) group, patients diagnosed with otosclerosis were recruited. To ensure these subjects served as strictly noninflammatory controls, candidates with a history of otitis media or ear trauma were excluded. Diagnostic criteria for otosclerosis included progressive conductive hearing loss, an intact tympanic membrane, and a type A or As tympanogram with absent stapedial reflexes, alongside the presence of a Carhart’s notch at 2000 Hz in bone conduction thresholds. All otosclerosis diagnoses were definitively confirmed intraoperatively by visualizing stapes fixation during stapedectomy. Furthermore, the absence of active inflammation was verified through negative intraoperative bacterial and fungal cultures, combined with the lack of visual indicators of mucosal inflammation, such as hyperemia or edema.

Outcome measures and follow-up

The primary outcome of this study was the recurrence of tympanic membrane perforation. This was defined as the clinical observation of an unrepaired perforation with associated ear discharge during the early postoperative period. To ensure objective and consistent assessment, all patients were evaluated by a senior otolaryngologist using endoscopic otoscopy. A standardized follow-up protocol was implemented, with the primary assessment of perforation status performed for every patient at 3 weeks postoperatively. This uniform timing across both the PO and NPO groups ensured that the assessment of predictive markers was not confounded by variations in the evaluation period.

Sample collection

Surgery was performed under general anesthesia and sterile conditions. Swabs for microbial collection (Copan Italia, Brescia, Italy) were rubbed against the middle ear mucosa during the operation. Effort was made to minimize contact with external skin to avoid contamination with skin flora. The swab stems were cut with sterile scissors; the collection tube was closed, sealed, and labeled. All samples were stored at −40 °C until shipment to OE BioTech, Qingdao, China, on dry ice for 2b-RAD-M.

Genomic DNA extraction, library preparation, and metagenomic sequencing

Genomic DNA was extracted using the TIANamp Micro DNA Kit (Tiangen, Beijing, China) with the addition of carrier RNA to enhance yield. The extracted DNA was eluted in 20 μL of RNase-free water. Four units of BcgI restriction enzyme (NEB, Ipswich, MA, USA) were used to digest the genomic DNA at 37 °C for 3 h. Subsequently, ligation with 0.2 μM of library-specific adaptors (Ada1 and Ada2) was performed at 4 °C for 16 h, followed by heat inactivation at 65 °C for 20 min. The ligation products were amplified by polymerase chain reaction (PCR) using 7 μL of ligated DNA, 0.1 μM of primers, 0.3 mM of dNTPs, 1× Phusion HF buffer, and 0.4 U of Phusion High Fidelity DNA Polymerase (NEB, Ipswich, MA, United States). The PCR protocol consisted of 16–28 cycles of 98 °C for 5 s, 60 °C for 20 s, and 72 °C for 10 s, with a final 10-min extension at 72 °C. The library products were purified using the QIAquick PCR Purification Kit (QIAGEN, Hilden, Germany) and sequenced on an Illumina HiSeq X™ Ten platform (Illumina, San Diego, CA, USA).

Sequence processing and analysis

Reads were scanned for the type IIB restriction enzyme recognition site, and sequences containing enzyme fragments were extracted. Clean reads were generated by the removal of reads that either contained more than 8% ambiguous bases or were classified as low-quality (more than 20% of bases with a quality value below Q30). Taxonomic classification was performed using a custom 2b-RAD-tag database, which included species from 173,165 microbial genomes (bacteria, bacteriophages, and archaea). Read coverage was determined for each identified genome to estimate the relative abundance of each taxon. Taxa were included in the downstream analysis if they had at least five taxon-specific 2b-RAD-tags and at least 15 sequenced reads. To assess potential cross-contamination during preparation, the ATCC® MSA-1002™ mock microbial community, consisting of 20 bacterial species with a genomic DNA abundance of ~ 5 ng, was processed along with the experimental samples. To control for false-positive species identification, a G score was calculated for each species as follows: G score species i = √(Si × ti), where Si is the number of reads assigned to all 2b-RAD-tags belonging to species i in the sample and ti is the number of all possible 2b-RAD-tags of species i sequenced in the sample. The G score, which represents the harmonic mean of the read coverage of the 2b-RAD-tags of a species and the number of all possible 2b-RAD-tags for that species, was used to identify false positives with a threshold of 10. Finally, a secondary, sample-specific 2b-Tag-DB was constructed using only the identified candidate taxa; each taxon contained more specific 2b-RAD-tags compared with the default database. All reads were then re-mapped against this refined database to provide a more accurate assessment of the relative abundance of the candidate taxa. Sequences for these specimens are available in the NCBI Sequence Read Archive (SRA) under BioProject accession PRJNA1295442.

Microbial diversity analysis

Based on the classification abundance profile, the Chao1, Shannon, and Simpson alpha diversity indices were calculated using the “vegan” package and visualized as box plots. Beta diversity was estimated using Bray–Curtis, Binary Jaccard, and Euclidean distance metrics calculated by the “vegan” package, and visualized using Principal Coordinate Analysis (PCoA). Venn diagrams were used to visualize the unique and shared species between the PO and NPO groups. Linear discriminant analysis effect size (LEfSe) was used to identify differentially abundant taxa between the PO and NPO groups, with a log linear discriminant analysis (LDA) score threshold of 2.0. Functions were predicted using Clusters of Orthologous Groups (COG) (https://www.ncbi.nlm.nih.gov/research/cog) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses (https://www.genome.jp/kegg/).

Statistical analysis

Statistical analyses were performed using IBM SPSS Statistics (ver. 26, IBM Corp., Armonk, NY, United States) and R (ver. 4.1.1, University of Auckland, Auckland, New Zealand). The Wilcoxon test was used for pairwise comparisons of alpha diversity between groups, whereas permutational multivariate analysis of variance was used to compare beta diversity. The Kruskal–Wallis test was used to compare microbial communities among the PO, CON, and NPO groups. For functional predictions (COG and KEGG), the Wilcoxon test was used to analyze differences between groups. The threshold for statistical significance was set at p < 0.05.

Results

Patient characteristics and study design

Middle ear swabs were obtained during surgery from 20 patients who underwent tympanoplasty between May 2020 and November 2021 at the Eye and ENT Hospital of Fudan University, Shanghai, China. The same treatment team evaluated the preoperative imaging and performed the surgical resection in all patients. Patients with CSOM were divided into perforation (PO) and nonperforation (NPO) groups based on whether perforation recurred postoperatively. Patients with otosclerosis and no inflammation were used as controls (CON). Patient demographic information is presented in Table 1. The baseline clinical characteristics, including age and preoperative ear discharge status, were compared between the PO (n = 4) and NPO (n = 13) groups. In the NPO group, 3 out of 13 patients (23.1%) had dry ears (absence of preoperative otorrhea), while all 4 patients in the PO group presented with wet ears. Fisher’s exact test showed no statistically significant difference in the proportion of dry ears between the two groups (p = 0.541). Furthermore, a Mann–Whitney U test was conducted to compare age as a continuous variable due to the small sample size; no significant difference was observed between groups (p = 0.6943). These findings indicate that the groups were comparable regarding these known clinical risk factors.

Table 1.

Demographic and clinical characteristics of enrolled patients

Group	Patient	Sex	Age	Side	Pre-op otorrhea (Yes/No)	Pre-op EAC status	Intra-op middle ear status
NPO	NPO-1	F	21	Left	No	Dry	Dry, normal mucosa
NPO	NPO-2	M	63	Right	Yes	Moist, purulent discharge	Moist, scant viscous discharge
NPO	NPO-3	M	66	Right	No	Dry	Moist, viscous discharge
NPO	NPO-4	F	34	Left	Yes	Moist, purulent discharge	Moist, viscous discharge
NPO	NPO-5	F	45	Left	No	Dry	Dry
NPO	NPO-6	F	54	Left	Yes	Moist, purulent discharge	Moist, viscous discharge
NPO	NPO-7	M	38	Right	Yes	Moist, purulent discharge	Moist, viscous discharge
NPO	NPO-8	M	71	Left	Yes	Moist, purulent discharge	Moist, viscous discharge
NPO	NPO-9	F	70	Left	Yes	Moist, purulent discharge	Moist, purulent discharge
NPO	NPO-10	F	66	Left	Yes	Moist, purulent discharge	Moist, purulent discharge
NPO	NPO-11	F	62	Right	Yes	Moist, purulent discharge	Moist, purulent discharge
NPO	NPO-12	M	53	Right	Yes	Moist, purulent discharge	Moist, purulent discharge
NPO	NPO-13	M	30	Left	Yes	Moist, purulent discharge	Moist, purulent discharge
PO	PO-1	F	44	Right	Yes	Moist, purulent discharge	Moist, viscous discharge
PO	PO-2	F	70	Left	Yes	Moist, purulent discharge	Moist, viscous discharge
PO	PO-3	F	56	Right	Yes	Moist, purulent discharge	Moist, viscous discharge
PO	PO-4	F	19	Right	Yes	Moist, purulent discharge	Moist, viscous discharge
CON	CON-1	M	38	Left	No	Dry	Dry, normal mucosa
CON	CON-2	F	36	Left	No	Dry	Dry, normal mucosa
CON	CON-3	F	43	Right	No	Dry	Dry, normal mucosa

Microbial diversity of the middle ear specimens

The obtained reads were aligned using 2b-Tag-DB and subsequently classified into 875 distinct microbial categories at the species level. Venn diagram analysis (Fig. 1A) revealed 54 microbial species common to the three groups, whereas 320, 215, and 99 species were exclusive to the NPO, PO, and CON groups, respectively. For alpha diversity, the Shannon and Simpson indexes were calculated to evaluate community diversity; Chao1 was calculated to evaluate the community richness of the middle ear microbiome. These three indexes showed no significant difference among the NPO, PO, and CON groups. Both the PO and NPO groups with middle ear infection differed from the CON group without inflammation, but these differences were nonsignificant. The control group had greater richness and diversity. Although minimal differences were observed between the PO and NPO groups, the NPO group had greater richness and diversity, approaching that of the control group without inflammation. For beta diversity, PCoA revealed no differences in bacterial composition among the three groups based on Bray–Curtis, binary Jaccard, and Euclidean distances.

Fig. 1.

Differences in microbial diversity and composition within the middle ear among the CON, NPO, and PO groups. A The Venn diagram shows the species distribution among the CON, NPO, and PO groups, as well as interrelationships among these groups. B Comparison of alpha diversity (Chao1, Simpson, and Shannon) of the three groups. C The three groups were compared using the Bray–Curtis, binary Jaccard, and Euclidean distance beta diversity metrics. The results of this analysis were visualized using 3D-PCoA plots, where each sample is represented by a point, and samples from the same group are shown in the same color. These plots provide insights into differences among the three groups based on the three distance matrices

Microbial community composition in the three groups

Fungi (e.g., Agaricomycetes, Sordariomycetes, and Blastocladiomycetes), Archaea (e.g., Halobacteria), and Eubacteria (e.g., Chlamydiia and Bacteroidia) were detected at the class level. Nineteen phyla were identified in the three groups, and Proteobacteria was the dominant phylum in all three groups (Fig. 2A). At the class level, Gammaproteobacteria were the most prevalent in all three groups (Fig. 2B). Moraxella was the dominant genus in the control group, whereas Pseudomonas was most prevalent in both the PO and NPO groups, comprising 74.7% of the perforated group (Fig. 2C). In the three groups, 875 different species were detected. Pseudomonas aeruginosa, Sphingomonas sp. 000797515, Porphyromonas uenonis, and Fenollaria massiliensis were the most common species in the PO group. Staphylococcus aureus, P. aeruginosa, and Achromobacter xylosoxidans were the most common species in the NPO group. Finally, Moraxella catarrhalis, Cutibacterium acnes, and Ralstonia sp. 000620465 were the most common species in the control group. M. catarrhalis was present in 26.9% of the normal group but absent from both the NPO and PO groups (Fig. 2D).

Fig. 2.

Characterization of microbial composition and relative abundance in the three groups. A Phylum- and B class-level relative abundances in the three groups. Bar charts show the top 30 most abundant C genera and D species in the three groups

Differential abundances of microbial taxa

To characterize the differential representation of microorganisms (individual-level relative abundance data of microbial taxa is shown in Supplemental Table S1), we performed Kruskal–Wallis analysis of taxonomic differences. This analysis revealed significant increases in the abundances of Acinetobacter johnsonii, Escherichia flexneri, Ralstonia pickettii, and other species at the species level in the CON group relative to both the PO and NPO groups (Supplemental Fig. S1). For the NPO and PO groups, at the phylum level, the abundances of Actinobacteriota, Ascomycota, and Basidiomycota significantly increased in the NPO group, whereas the NPO group had a significantly lower abundance of Chlamydiota compared with the PO group. Regarding genera, Acinetobacter, Bacillus, Corynebacterium, Cutibacterium, and Malassezia were significantly more abundant in the NPO group; Chlamydophila was more abundant in the PO group (Supplemental Fig. S2). At the species level, Bacillus bombysepticus, Burkholderia oklahomensis, Corynebacterium kefirresidentii, and C. acnes were more enriched in the NPO group, whereas P. aeruginosa, Pseudomonas alcaligenes, Pseudomonas protegens, Pseudomonas qingdaonensis, Pseudomonas sp. 003696305, Pseudomonas thermotolerans, Pseudomonas nitroreducens, and Pseudomonas panipatensis were more enriched in the PO group (Fig. 3A).

Fig. 3.

Comparison of differences in the abundance of microbial taxa between the NPO and PO groups. A Top 10 species with significant differences. B Differential species annotated branching plots represent the taxonomic hierarchy of the discriminatory biomarkers identified with LEfSe. C Differential species score plots present the species with relatively high abundances in each of the two groups

We also used LEfSe to identify differentially abundant taxa in the three groups. Nodes in each layer represent phylum/class/order/family/genus from inside out, respectively; the annotations of species markers in each layer represent phylum/class/order/family/genus (Fig. 3B). LEfSe identified 15 discriminative features (LDA score ≥ 2.0) with significantly different relative abundances between the PO and NPO groups (Fig. 3C). At the genus level, the microbiome of the NPO group was enriched with Bacillus, Corynebacterium, Cutibacterium, and Acinetobacter, whereas that of the PO group was enriched with Chlamydophila. At the species level, taxa that differentiated the two groups were Bacillus_A_bombysepticus, B. oklahomensis, C. kefirresidentii, and C. acnes in the NPO group and P. aeruginosa, P. thermotolerans, P. nitroreducens, and P. panipatensis in the PO group. These significantly different bacterial taxa can be used as indicators to distinguish the two groups (Fig. 3B, C).

Model prediction analysis in the three groups

The indicator values of the species in each group were calculated, and statistical analysis was performed on the indicator values between the groups (p < 0.05); this approach revealed the index species of each group, which were species with high group specificity. In the NPO group, the indicators were Bacillus_A_bombysepticus, Ralstonia sp. 000620465, C. acnes, and Sphingomonas sp. 000797515; conversely, P. protegens, P. alcaligenes, P. qingdaonensis, Pseudomonas sp. 003696305, P. aeruginosa, P. panipatensis, P. thermotolerans, and P. nitroreducens were indicators in the PO group (Fig. 4A). Several species were correlated between the NPO and PO groups, indicating close associations among them, including A. xylosoxidans, Porphyromonas sp. 900548415, F. massiliensis, Metamycoplasma salivarium, Porphyromonas somerae, Prevotella timonensis, Campylobacter rectus, and P. uenonis. Notably, Porphyromonas sp. 900548415 had the strongest association with the other species (Fig. 4B).

Fig. 4.

A Indicator species diagram of the PO and NPO groups. B Species correlation network diagram depicting species correlations between the PO and NPO groups. The figure shows species with a default threshold of |SpearmanCoef|> 0.8 and p < 0.01. C Variable importance plot from the Random Forest model ranking the top 30 bacterial taxa. The x-axis represents the mean decrease in accuracy, which measures the contribution of each taxon to the model’s overall predictive accuracy. The color gradient indicates the mean decrease in Gini index, defined as the total reduction in node impurity (heterogeneity) resulting from splits on a specific variable. The size of each point reflects the mean relative abundance of the corresponding taxon across all samples. D ROC curves evaluating the diagnostic efficacy of Bacillus_A_bombysepticus (AUC = 0.92) and Pseudomonas aeruginosa (AUC = 0.25) in discriminating between the postoperative perforation (PO) and non-postoperative perforation (NPO) groups. The dashed line represents the reference for a random classifier (AUC = 0.50)

To identify the most discriminative microbial features associated with postoperative perforations (PO), we utilized a Random Forest model and subsequently validated the diagnostic efficacy of key taxa using Receiver Operating Characteristic (ROC) curves (Fig. 4C, D). The Random Forest analysis prioritized features based on their Mean Decrease Accuracy (MDA) and the Mean Decrease Gini index, the latter of which is defined as the total reduction in node impurity (heterogeneity) resulting from splits on a specific variable. Among the top-ranked taxa, P. aeruginosa and Bacillus_A_bombysepticus emerged as the most significant predictors, though they exhibited distinct importance profiles. Specifically, Bacillus_A_bombysepticus demonstrated a substantial Gini index, suggesting a high capacity for partitioning data into homogeneous groups. This was confirmed by its superior diagnostic performance in the ROC analysis, where it achieved an area under the curve (AUC) of 0.92, indicating its robustness as a primary biomarker for PO. In contrast, while P. aeruginosa yielded the highest MDA (approximately 3.5), its individual AUC was 0.25. In the context of binary classification, an AUC significantly below the 0.5 random-chance threshold indicates a strong inverse correlation with the target condition. This suggests that while P. aeruginosa is a critical feature for maintaining the overall accuracy of the multi-variable model, it likely serves as a characteristic signature for the NPO group rather than a positive indicator for PO. Consequently, our findings indicate that while P. aeruginosa possesses the greatest global influence on model stability, Bacillus_A_bombysepticus provides the most precise and direct individual diagnostic power for identifying postoperative perforations.

The control group had numerous indicator species that distinguished it from the NPO and PO groups, including Methylobacterium fujisawaense, Ralstonia sp. 001078575, Brevundimonas aurantiaca, Malassezia globosa, A. johnsonii, Corynebacterium resistens, Nocardioides sp. 009699265, C. acnes, Blastomonas ursincola, Afipia broomeae, Tardiphaga sp. 002256345, and Xanthomonas campestris (Supplemental Fig. S3A). These indicator species likely play important roles in the maintenance of a normal microbial microenvironment in the middle ear. Moreover, positive correlations were observed in the correlation network diagram (Supplemental Fig. S3B). Pseudomonas, Aspergillus, and Achromobacter were identified as the most important genera in the NPO, PO, and CON groups, respectively, consistent with the three most important species in the NPO and PO groups. This finding indicates close relationships of these three species with postoperative outcomes and the presence of middle ear infections (Supplemental Fig. S3C).

Differences in predictive functions among the three groups

For the identified microbiota, we predicted their functions and used the Wilcoxon test to compare the NPO, PO, and CON groups. We identified COGs that differed among the three groups; the top five COGs were COG2207 (AraC-type DNA-binding domain and AraC-containing proteins), COG1269 (A/V-type ATP synthase), COG4974 (site-specific recombinase XerD), COG1289 (uncharacterized membrane protein YccC), and COG0564 (pseudouridine synthase RluA, 23S rRNA- or tRNA-specific) (Fig. 5A). Analyses were performed based on the KEGG database of gene protein sequences (KEGG Genes), chemicals with endogenous and exogenous properties (KEGG Ligand), molecular interactions and metabolic pathways (KEGG Pathway), and hierarchical relationships among various organisms (KEGG Brite). The top three pathways with significant differences included putative ABC transport system permease protein (K02004), TatD DNase family protein (K03424), and nuclear receptors (K03310) (Fig. 5B).

Fig. 5.

Differences in microbial function among groups according to functional annotation analysis. Results of the A COG and B KEGG function prediction for the top 10 most significant differences

Discussion

Early research suggested that a healthy middle ear is typically sterile, highlighting the crucial role of pathogens from adjacent organs, such as the external auditory canal or eustachian tube, in the occurrence and development of CSOM. Given that these studies relied primarily on culture techniques and early molecular identification techniques and that most microorganisms in nature are difficult or impossible to culture, hospital laboratory diagnoses may overlook unidentified bacteria, viruses, and fungi. Additionally, the limited quantity of DNA from ear secretions can also lead to false-negative PCR test results. The emerging DNA sequencing technology has unique advantages for the characterization of microbial communities in the middle ear. 16S rRNA sequencing-based research revealed that the healthy middle ear is not sterile but has a complex microbial ecology consisting of various microorganisms, which suggests a different pathogenesis theory for CSOM. Nevertheless, these studies may not accurately reflect the true middle ear microbial community, thereby affecting the accuracy and reliability of clinical diagnoses. Furthermore, the limited resolution of 16S rRNA sequencing, which identifies microorganisms only at the genus level, may have difficulty distinguishing closely related species, which is crucial for the development of effective treatment strategies. Here, 2b-RAD-M was used to provide a more detailed and comprehensive depiction of the middle ear microbiome in patients with otosclerosis (as an analogy to a healthy middle ear environment) and in CSOM patients with or without postoperative tympanic membrane perforation.

Preoperative otorrhea is widely recognized as a potential risk factor for the recurrence of tympanic membrane perforation, as active inflammation may impair graft vascularization and healing. In our study, we analyzed this potential confounder and found no statistically significant difference in the distribution of dry versus wet ears between the PO and NPO groups (p = 0.541). While the small sample size may limit the power to detect subtle differences, our preliminary analysis suggests that preoperative discharge status did not disproportionately influence the outcomes of our predictive models. Future studies with larger cohorts are warranted to further clarify the synergistic effects of these clinical factors alongside molecular biomarkers.

In this study, the detection of numerous microorganisms in healthy middle ears supported the evidence that the healthy middle ear is not sterile. Additionally, no significant differences in the alpha or beta diversity of microorganisms were observed among the CON, PO, and NPO groups, indicating that the middle ear microbiota of CSOM patients is similar to that of healthy individuals. The conventional view is that the middle ear should normally be free of fungi, and the presence of a fungal infection in CSOM significantly increases the difficulty of treatment. Intriguingly, fungi were detected in the CON group in this study, suggesting their involvement in the maintenance of microbial balance in the healthy middle ear (Chen et al. 2022). We also observed that Aspergillus, one of the most common pathogens in otomycosis, was significantly more abundant in the NPO group than in the PO group, suggesting that it can serve as an indicator to distinguish between these two groups (Bojanović et al. 2022, 2023). One possible explanation for this is that tympanic membrane perforation may lead to better ventilation and drainage, thereby inhibiting fungal growth. Another explanation is that patients with recurrent tympanic membrane perforations tend to use antibiotics more frequently, which suppresses the growth of fungi. Archaea, a unique class of single-celled microorganisms, also were detected in the middle ear of the CON group. To our knowledge, however, no studies have examined the presence and role of Archaea in the middle ear microenvironment. Whether Archaea play a role in the maintenance of microbial balance and immune regulation in the middle ear, similar to their role in the gut, merits further investigation (Shi and Mu 2017; Wei et al. 2024).

At the genus level, Pseudomonas was the dominant genus in both CSOM groups; its proportion was higher in PO (74.7%). Notably, other than P. aeruginosa, which is considered the most common pathogen associated with CSOM, the abundances of Pseudomonas species such as P. alcaligenes, P. protegens, P. qingdaonensis, Pseudomonas sp. 003696305, P. thermotolerans, P. nitroreducens, and P. panipatensis were significantly higher in the PO group than in the NPO group, suggesting that these species play important roles in the recurrence of tympanic membrane perforation. Because these Pseudomonas species are primarily found in the environment rather than in the human body, their presence in the middle ear suggests compromised immune function in CSOM patients with tympanic membrane perforation (Garrido-Sanz et al. 2023; Lin et al. 2023; Miloloža et al. 2022). Our findings show that 2b-RAD-M can provide species-level microbial information, offering valuable guidance for the treatment of CSOM because these Pseudomonas species may have different sensitivities to antibiotics.

Based on Kruskal–Wallis and LEfSe analyses, Acinetobacter guillouiae, Bacillus_A_bombysepticus, B. oklahomensis, C. kefirresidentii, and C. acnes were significantly more abundant in the NPO group than in the PO group. However, the involvement of these bacteria in CSOM remains unclear because few relevant reports have been published. We note that Bacillus_A_bombysepticus and C. kefirresidentii were not detected in either the CON or PO groups. We hypothesize that the relatively closed stable environment provided by the repaired tympanic membrane may allow these exogenous bacterial species to colonize and proliferate, ultimately contributing to the reconstruction of middle ear microbial homeostasis in CSOM patients after surgical intervention and systemic antibiotic therapy. The other three bacteria, A. guillouiae, B. oklahomensis, and C. acnes, were significantly more abundant in the NPO group than in the PO group; they were less abundant in the CON group. This finding suggests that major CSOM pathogens, such as P. aeruginosa, disrupt the microbial balance in the middle ear during an infection and significantly inhibit the growth of these three bacteria. During recovery, these bacteria may aid in the restoration of microbial homeostasis as resident flora, potentially preventing the recurrence of tympanic membrane perforation. M. catarrhalis, C. acnes, and Ralstonia sp. 000620465 were the dominant bacteria in the CON group. Of note, M. catarrhalis had a prevalence of 26.9% in the CON group, whereas it was entirely absent from both the NPO and PO groups. Although reports have suggested that M. catarrhalis, as a resident bacterium in the middle ear, constitutes a potential endogenous source of infection in CSOM, our results indicate that CSOM-induced dysbiosis of the middle ear microbiota may inhibit the survival of M. catarrhalis (Mittal et al. 2014). Furthermore, M. catarrhalis may have a protective role in maintaining a healthy middle ear environment.

Previous studies often have neglected to connect middle ear microbiota data with the clinical outcomes of CSOM. We performed random forest analysis on the 2b-RAD-M microbial data and found that Bacillus_A_bombysepticus had a high NPO indicator value, whereas P. aeruginosa had a high PO indicator value. ROC curve analysis was used to evaluate the reliabilities of these bacteria in distinguishing between NPO and PO. Intriguingly, although P. aeruginosa was significantly more abundant in the PO group, its ability to distinguish between the two groups was poor (AUC = 0.25). In contrast, Bacillus_A_bombysepticus may be a potential diagnostic biomarker for preventing tympanic membrane perforation after surgery for CSOM, considering its impressive AUC of 0.92.

Functional annotation analyses were performed to compare differences in microbial function among the three groups. The COG and KEGG function predictions indicated enhancement of specific microbial functions in the PO group relative to the NPO and CON groups. Specifically, the increased abundances of COG4771, COG1269, COG0534, K02004, K03310, K03561, and K03327 suggest enhanced gene regulation and expression. The elevated levels of COG2207, COG4974, COG0564, COG1476, and COG0492 indicate improved efficiency in substance transport and metabolism. Additionally, the higher abundances of K03324 and K03630 imply enhanced DNA repair capabilities. The synergistic actions of these pathways ensure that microorganisms can efficiently perform metabolic activities, safeguard the stability of genetic material, maintain normal physiological activities and homeostasis, and ultimately better adapt and survive in harsh environments.

Several limitations of this study should be acknowledged. First, the sample size was relatively small and characterized by a class imbalance (13 vs. 4). Although we employed Leave-One-Out Cross-Validation (LOOCV) to mitigate the risk of overfitting and to maximize the utility of the available data, the limited number of cases may still constrain the generalizability of the Random Forest model and the statistical precision of the ROC analysis. Consequently, our findings should be interpreted as exploratory and hypothesis-generating. Future multicenter studies with larger, more balanced cohorts are required to confirm these preliminary predictive markers and refine the model’s diagnostic accuracy.

Conclusion

This is the first study to use 2b-RAD-M to analyze the middle ear microbiota in patients with CSOM and otosclerosis. The results provide more comprehensive data concerning the middle ear microbiota, suggesting potential mechanisms by which various microorganisms contribute to the pathogenesis and progression of CSOM. Importantly, we identify Bacillus_A_bombysepticus as a potential diagnostic biomarker that can indicate an increased risk of tympanic membrane reperforation. We believe that our study offers a new target for predicting postoperative outcomes in CSOM patients.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

XF and YC have contributed equally to this work and share first authorship. Conceptualization, BC, JX and DR; Data curation, XF and YC; Formal analysis, XF and YW; Funding acquisition, DR; Investigation, JL, HS and MC; Methodology, XF and YW; Project administration, JX and DR; Resources, DR; Supervision, ZW, GL and DR; Visualization, DR; Writing – original draft, XF and YC; Writing – review & editing, JX, GL and DR.

Funding

This study is supported by the National Natural Science Foundation of China (grants 82571313/H1304 and 82271166/H1304) to DR.

Data Availability

The datasets used and analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval

The Ethics Committee of the Eye and ENT Hospital, Fudan University, approved the study protocol (approval number: 20200423) on April 23, 2020. The study was registered with the Chinese Clinical Trial Register (ChiCTR: 2000038981).

Consent to participate

Written informed consent has been obtained from the patients to publish this paper.

Competing interests

The authors declare no conflict of interest.