Proteomic Profiling of Bacterial Extracellular Vesicles for Exploring Ovarian Cancer Biomarkers.

Eri Asano‐inami : conceptualization, methodology, investigation, funding acquisition, data curation, writing ‐ original draft. Akira Yokoi : conceptualization, methodology, writing ‐ review and editing, supervision, funding acquisition. Kosuke Yoshida : data curation, methodology, formal analysis. Kentaro Taki : data curation, formal analysis. Masami Kitagawa : investigation, methodology. Kazuhiro Suzuki : formal analysis. Ryosuke Uekusa : resources. Yukari Nagao : resources. Nobuhisa Yoshikawa : resources. Kaoru Niimi : resources. Yusuke Yamamoto : formal analysis. Hiroaki Kajiyama : supervision, resources.

The study was approved by the Institutional Ethics Board of Nagoya University (approval number: 2017‐0497). Pre‐existing samples and medical records were used. Thus, we provided disclosure information on the methods of this study and gave the subjects opportunities to reject enroll in this study.

Written informed consent was obtained from all patients, and all participants agreed to publication.

Lactobacillus crispatus  ( L. crispatus , JCM1185),  Lactobacillus gasseri  ( L. gasseri , JCM1131),  Lactobacillus jensenii  ( L. jensenii , JCM15953), Lismosilactobacillus vaginalis ( L. vaginalis , JMC9505) and  Lactobacillus iners  ( L. iners , JCM12513) were obtained from RIKEN BRC and  Fuso nucleatum  (F. nucleatum JNBP_02614) and  Escherichia coli  K‐12 ( E. coli , GTC_2003) were obtained from Gifu University Center for the Conservation of Microbial Genetic Resource, Organization for Research and Community Development, Japan https://pathogenic‐bacteria.nbrp.jp/bacteria/bacteriaAllIt .  L. crispatus , L. gasseri ,  L. jensenii and L. vaginalis were first cultured under the aerobic conditions at 37°C in de Man, Rogosa, and Sharpe (MRS) agar (BD Difco, France). Then, a liquid MRS medium was used for mass culture. L. iners was first cultured under anaerobic conditions at 37°C in BL agar (Nissui Pharmaceutical Co, Tokyo, Japan), and a liquid bifidobacterium medium (referring to DSMZ) was then used for mass culture. F. nucleatum was first cultured under anaerobic conditions at 37°C in Brucella agar (Kyokutoseiyaku, Tokyo, Japan). We used liquid cultures based on brain heart infusion (Nissui Pharmaceutical Co.) broth medium supplemented with hemin (10 µg/mL, Thermo Fisher Scientific, MA, USA), menadione (5 µg/mL, Nacalai Tesque, Japan), L‐cysteine (1 µg/mL, Sigma‐Aldrich, MO, USA) and resazurin (2 µg/mL, FUJIFILM Wako Pure Chemical, Japan) as an anaerobic indicator. E. coli K‐12 was cultured under aerobic conditions at 37°C in L‐Broth agar and liquid medium (MP Biomedicals, CA, USA). Bacterial cell concentrations were measured by assessing the optical density at 660 nm (Miniphoto518R, TAITEC), and colony formation units were counted at each OD measurement. The procedure of separating BEVs used in this study conformed to the standard method of the International Society of Extracellular Vesicles (MISEV2023) (Welsh et al. 2024 ). To obtain an exosome‐free medium, the medium for each cell was ultracentrifuged at 32,000 rpm at 4°C overnight. Approximately 400 mL of the medium was cultured with each bacterium for 16–28 h, centrifuged at 2000 × g for 30 min at 4°C to pellet the bacteria, and then centrifuged at 10,000 × g 4°C for 40 min (KUBOTA Co., Tokyo, Japan). The pellets were washed with Dulbecco's phosphate‐buffered saline (PBS) and centrifuged again at 10,000 × g for 40 min at 4°C to obtain BEVs. The supernatant was passed through a 0.22‐µm filter (Millex‐GV 33 mm, Millipore) and ultracentrifuged at 32,000 rpm for 2 h at 4°C. The protein concentrations of EVs and cell lysates were quantified using a Qubit protein assay kit (Thermo Fisher Scientific) with a Qubit 4.0 Fluorometer (Invitrogen Co., MA, USA), according to the manufacturer's protocol. The process was the same as in dUC up to the point where a 0.22‐µm filter was used. qEV (iZON Science, Oxford, UK) was used to extract EVs using SEC. The samples were concentrated in an ultracentrifuge to 500 µL. Samples were applied to a qEV column, and a total of 1.6 mL of EVs was eluted according to the protocol. Moreover, 1.6 mL of EV was concentrated to 50 µL using Amicon Ultra 3 K (Merck KGaA, Darmstadt, Germany). The size distribution and particle concentration in EV preparations were analysed using a NanoSight NS300 nanoparticle tracking analyser (Malvern Panalytical Ltd., UK). The samples were diluted in PBS and injected at a speed of 100 a. u. into the measuring chamber. The EV flow was recorded in triplicate (30 s each) at room temperature. The equipment settings for data acquisition were kept constant between measurements, with the camera level set at 12. Each bacterial culture was fixed with 2% glutaraldehyde dissolved in PBS at 4°C. The samples were postfixed in 2% osmium tetroxide for 1 h at room temperature and dehydrated using increasing ethanol concentrations (50%–100%). The samples were critical‐point dried, stuck on carbon stubs, and coated with osmium (NL‐OPC80NS, Japan Laser Corporation, Tokyo, Japan). The samples were imaged under a field‐emission electron microscope (JSM‐7610F; JEOL Ltd., Tokyo, Japan). For TEM, BEVs were prepared by negative staining, EV pellets were resuspended in PBS, and 5 µL of EV samples were loaded on a grid with a carbon support film (Nisshin EM, Tokyo, Japan) and blocked with 1% bovine serum albumin for 1 h at room temperature. After washing with PBS, the EV samples were fixed with 1% glutaraldehyde for 10 min. After washing with distilled water, the EVs were stained with 1% uranyl acetate, and the excess liquid was blotted with filter paper and dried at room temperature. Samples were then examined under a JEM‐1400PLUS transmission electron microscope (JEOL Ltd., Tokyo, Japan) Patient ascites samples were obtained from female patients admitted at Nagoya University Hospital, Japan. Informed consent was obtained from each patient before surgery. This study was approved by the Ethics Committee of Nagoya University School of Medicine (Approval number: 2017‐0053). Patient details are shown in Table  1 . Patient characteristics. For mass analysis, EV proteins were treated according to Easy Pep (Thermo Fisher Scientific) protocols and digested with trypsin for 2 h. The peptides were analysed by LC‐MS using a nanoelectrospray ion source with an Orbitrap Fusion mass spectrometer (Thermo Fisher Scientific) coupled to an UltiMate3000 RSLCnano LC system (Dionex Co., Netherlands) with a nano high‐performance LC capillary column (150 mm × 75 µm inside diameter) (Nikkyo Technos Co., Japan). Reversed‐phase chromatography was performed with a linear gradient (5%–40% B for 100 min), solvent A (2% acetonitrile and 0.1% formic acid) and solvent B (95% acetonitrile and 0.1% formic acid) at an estimated flow rate of 300 nL/min. Before the MS/MS analysis, a precursor ion scan was performed using a mass‐to‐charge ratio ( m / z ) of 400–1600. MS/MS was performed with an isolation width of 1.6 m / z with quadrupole, higher‐energy collisional dissociation fragmentation with a normalised collision energy of 35%, and rapid scan MS analysis in the ion trap. Only the precursors with charge states 2–6 were sampled for MS2. The dynamic exclusion duration was set to 15 s with 10 parts per million (ppm) tolerance. The instrument was operated in the top‐speed mode with 3 s cycles. Raw data were processed for protein identification using Proteome Discoverer (version 1.4, Thermo Fisher Scientific) alone or in conjunction with the MASCOT search engine (version 2.7.0, Matrix Science Inc., Boston, MA, USA). Peptides and proteins were identified with reference to the human protein database or bacterial protein database in UniProt (2021_02), with a precursor mass tolerance of 10 ppm and a fragment ion mass tolerance of 0.8 Da. The fixed modification was set to the carbamidomethylation of cysteine, and the variable modification was set to the oxidation of methionine. Up to two missed tryptic cleavages were permitted Homology analysis was performed using the UniProt basic local alignment search tool (BLAST, https://www.uniprot.org/blast ), and UniProtKB Swiss‐Prot was selected as the target database to enter the accession number. Homo sapiens and BLAST searches were performed. A sequence with >50% identity was considered homologous. The score is an indicator of sequence similarity that does not depend on the size of the base emitted. E ‐values < E ‐4 were considered homologous. The peptide sequences were aligned with the amino acid sequences for comparison with multiple sequence alignment using CLUSTALW ( https://www.genome.jp/tools‐bin/clustalw ), and the homology was then analysed using SnapGene ( www.snapgene.com ). Samples (1 µg) lysed with sEV derived from cancer and non‐cancer patients’ ascites were loaded onto polyacrylamide gels and transferred to membranes. After blocking with Blocking One (Nacalai tesque) for 1 h at room temperature, the membranes were incubated overnight at 4°C with the following primary antibodies: anti‐CD9 Antibody, clone MM2/57 CBL162 (Merck, Darmstadt, Germany), CD81 Antibody (B‐11) sc‐166029 (Santa Cruz Biotechnology, Dallas, TX, USA) and GRP94 Antibody (H‐10) sc‐393402 (Santa Cruz Biotechnology), which were diluted 1:100 in 10 % Blocking One/Tris‐buffered saline with 0.1% Tween 20 (TBST). The following day, the membranes were washed three times for 5 min in TBST, then incubated for 4 h at room temperature with the following secondary antibodies: anti‐Mouse IgG, HRP‐Linked Whole Ab Sheep NA931 (Cytiva, Tokyo, Japan) were diluted 1:2000 and used for CD9, CD81 and GRP94. Anti‐Rabbit IgG. The membranes were imaged using ImageQuant LAS 4010 software (GE Healthcare, Atlanta, GA, USA). Analyses were conducted using R version 4.0.3 (R Foundation for Statistical Computing,  http://R‐project.org ). The heatmap.2 function of the gplots package (ver. 3.1.0) or Partek Genomics Suite version 7.0 (Partek Inc., MO, USA) was used for the heatmap and hierarchical clustering analysis. To visualise the volcano plots, log2‐fold change and adjusted  p values for each gene were calculated using the Limma (ver. 1.3.1) and visualised using the EnhancedVolcano (ver1.14.0). The rgl (ver1.3.1) was used for 3D‐PCA generation. Triangular correlation plots and receiver operating characteristic (ROC) curves were generated using the corrplot package version 0.92 and pROC package version 1.18.0, respectively. The best combination model for detecting cancer was developed by logistic LASSO regression analysis using the compute.es package version 0.2‐5, glmnet package version 4.1‐2, hash package version 2.2.6.1, MASS package version 7.3‐54 and mutoss package version 0.1‐12. The optimal cutoff values for each candidate parameter were set based on the maximum point of the summed sensitivity and specificity (Youden index). The significance for all analyses was defined as  p  < 0.05.

We hypothesised that EVs in ascites from patients with ovarian cancer might contain specific bacterial proteins. EVs were separated from the 15 ascites samples using a qEV column (Figure  1A ). NTAs were performed to confirm EV quality. EV particles of approximately 100 nm were observed in both non‐cancer and cancer small‐EVs (sEVs = 30–150 nm in size) (Figure  1B ). The presence of sEV was confirmed by TEM, and vesicles approximately 100 nm in size were observed (Figure  1C ). The expression of EV‐specific proteins was analyzed by immunoblotting. The expression of the EV markers CD81 and CD9 was confirmed, and GRP94, which is not an EV marker, was not observed (Figure  1D ). Characterisation and mass spectrometry analysis of patient ascites EV. (A) Schematics of EV collection from ascites samples. (B) EVs were recovered from the ascites samples of 10 patients with ovarian cancer and five patients without cancer by size exclusion (SEC). Size distribution obtained by NTAs for isolated EVs derived from representative samples of ascites samples. (C) Transmission electron microscopy analysis of isolated cancer and non‐cancer ascites EVs. Scale bar = 100 nm. (D) Immunoblot analyses for CD9, CD81 and GRP94 of EV samples of cancer and non‐cancer ascites samples. Uncropped Western blotting data are shown in Figure S1A . (E) Schematics of mass spectrometry analysis of patient ascites EV with human reference. (F) MS data with human reference was obtained for cancer and non‐cancer ascites EV. According to protein content–based EV characterisation from MISEV2023, EV‐associated proteins were categorised into 1a to 5b. Each protein is identified by its gene symbol. The data were converted to a log10 scale. (G) The heat map shows the protein expression obtained from the same MS data as in Figure  1E (total = 2284 proteins). The data were converted to log2 fold change (log2FC). The details of the protein names are given in the Supplementary Information. (H) PCA of the MS data obtained in Figure  1E . (I) Volcano plot analysis between Cancer and non‐cancer samples of MS data obtained in Figure  1E . The p values for each protein were calculated using the Wald test in Limma. Moreover, five non‐cancer ascites sEV and 15 cancer ascites sEV were isolated using a qEV column, and MS analysis was performed. The results were analysed and identified with reference to the human protein database (Figure  1E ). Initially, the presence of EV‐associated proteins was examined in categories 1–5 of MISEV2023 as a reference. The International Society for Extracellular vesicles highlighted these categories as indicative of the quality of purified EVs, with the following constituents: transmembrane or glycosylphosphatidylinositol‐anchored proteins (category 1), cytosolic proteins (category 2), major constituents of non‐EV structures (category 3), larger EV‐associated proteins (category 4) and secreted or luminal proteins (category 5). Category 1 proteins were found, and no obvious difference was found between cancer‐ and non‐cancer‐derived EVs. In other words, sEVs were isolated from samples of patients with cancer to obtain their proteomic profiles (Figure  1F ). A heat map drawing of the results of the MS analysis of the human protein reference showed differences between the cancer and non‐cancer samples (Figure  1G ). The principal component analysis (PCA) also showed differences in profiling between the cancer and non‐cancer samples (Figure  1H ). Volcano analysis then revealed that in total, 311 of the 2284 proteins were variable in non‐cancer and cancer, 283 of them were highly expressed in cancer, and 28 proteins were lowly expressed in cancer ( p < 0.05) (Figure  1I ). Moreover, we attempted to recover BEVs from cultured vaginal bacteria and identify the protein. Bacteria that were both endogenous and potential pathogens present in the vagina were selected, referring to the article by Chen et al. ( 2017 ), D'Alessandro et al. ( 2021 ), Percival‐Smith et al. ( 1983 ), Ling et al. ( 2010 ), Agarwal et al. ( 2020 ), and Muraoka et al. ( 2023 ). The optimal culture medium for each bacterium was prepared and cultured. Each bacterium was captured by SEM (Figure  2A ). The incubation time to recover EVs from these seven bacteria was then determined. OD values were measured during incubation, and the maximum OD was defined as the point at which the growth of the bacteria reached a plateau. All BEVs were collected approximately 4 h after the bacteria reached their maximum OD. BEVs were collected from seven different bacteria, and the exact BEV recovery was confirmed by TEM and NTA. EVs measuring approximately 100 nm were identified from all bacteria (Figure  2B,C ). Characterisation and mass spectrometry (MS) analysis of the BEVs. (A) Scanning electron microscope images of selected vaginal bacteria. Scale bar = 1 µm (B) Transmission electron microscope images of EVs isolated by size exclusion (SEC) from each bacterium. Scale bar = 100 nm (C) Size distribution obtained by NTAs for isolated EVs derived from representative bacterial samples. (D) Schematics of the mass spectrometry (MS) analysis of BEVs. (E) Circular graph showing the percentage of protein localisation obtained by MS of bacterial EVs obtained by dUC and SEC. (F) Heat map and PCA of the expression of MS data obtained from respective samples (cell body = 7659 proteins, BEV_dUC = 4100 proteins and BEV_SEC = 2112 proteins). The data were converted to log2FC. The details of the protein names are given in the Supplementary file. BEVs were then recovered from each of the seven bacteria by SEC and dUC, respectively, and analysed by MS (Figure  2D ). The proteins identified in each BEV_dUC and BEV_SEC were then classified based on their localisation. BEV_dUC identified approximately 40% of the total 4100 proteins, followed by unknown and cytoplasmic proteins (39%) and plasma membrane proteins (16%) (Figure  2E ). Conversely, of the 2112 proteins identified by BEV_SEC, approximately 40% were unknown proteins, 31% were cytoplasmic proteins, and 22% were plasma membrane proteins (Figure  2E ). As shown in Figure  2F , the protein expression of each bacterium, BEV_dUC, BEV_SEC or the bacterium itself, was analysed by MS. The heat map and PCA showed that E. coli profiled completely differently in the bacterium; however, some similarities were present in the BEVs, with completely different results for the bacterium itself and EV. L. crispatus and L. jensenii showed similar profiles for both the bacteria and EVs, whereas the SEC_EVs L. vaginalis showed different profiles from the others and changes in the EV recovery methods. We attempted to detect EV‐derived bacterial proteins that were variably expressed in patients with ovarian cancer. Bacterial proteins were identified from patient‐derived EVs by proteomics of EVs recovered from ascites samples of patients with and without cancer (Figure  1E ) to a bacterial reference (Figure  3A ). As shown in Figure  3B , 212 proteins were successfully detected, and the cancer and non‐cancer cell profiles were different in heat map analysis. Cancer and non‐cancer profiles were also different in the PCA (Figure  3C ). Volcano analysis identified 25 proteins that were significantly expressed in cancers and 10 proteins that were significantly expressed in non‐cancers ( p < 0.05) (Figure  3D ). Analysis of bacterial‐derived proteins in human ascites EVs. (A) Schematics of the mass spectrometry (MS) analysis of patient ascites EV with bacteria reference. (B) The heat map shows the protein expression obtained from the same MS data as in Figure  3A (total of 35 proteins). The data were converted to log2FC. The details of the protein names are given in the Supplementary Information. (C) PCA of the MS data obtained in Figure  3A . (D) Volcano plot analysis of MS data was obtained in Figure  3A . The p values for each protein were calculated using the Wald test in Limma. (E) Venn diagram analysis of 25 bacterial proteins highly expressed in cancer obtained from human sample EVs and MS analysis of EVs obtained by dUC from the supernatants of cultured bacteria. (F) The same graph as in Figure  3E for BEVs obtained by SEC (G) ROC‐AUC analysis for P15636_Protease1. Values are AUC means (95 % CI). (H) Schematics of the homology analysis of P15636_Protease1. Results of homology analysis of P15636_Protease1. (I) P15636_Protease1 peptide sequences 1–8 were aligned with the amino acid sequence of Q5SSG8_Mucin‐21, with matches indicated in red and mismatches in blue. The amino acid number of Q5SSG8 is shown. Moreover, the Venn diagram was used to search for 25 EV‐derived bacterial proteins highly expressed in patients with ovarian cancer and detected in cultured vaginal bacteria‐derived EVs. As shown in Figure  3E , 4100 proteins identified in cultured vaginal bacteria‐derived bacterial EVs recovered from dUC shared three proteins in the Venn diagram with 25 bacterial proteins from EVs of patients with ovarian cancer. Proteins identified in cultured vaginal bacteria‐derived bacterial EVs recovered by dUC and 25 EV bacterial proteins from patients with ovarian cancer (Figure  3E ). Venn diagram analysis of the 2112 proteins obtained by BEV_SEC (Figure  2E ) revealed two common proteins (Figure  3F ). In addition, one protein (Accession no. P15636_Protease1) was shared by both dUC and SEC (Figure  3E,F ). These proteins were selected as candidate ovarian cancer‐associated BEV markers. To evaluate whether P15363_Protease1 is an effective therapeutic target for patients with ovarian cancer, the ROC area under the curve (AUC) analysis was performed. The resulting AUC (95% CI) of P15363_Protease1 was 0.88 (95 % CI: 0.64–1.00), indicating that it is a candidate therapeutic target (Figure  3G ). In addition, to P15636_Protease1 (Figure  3G ), the B8IHL3_chaperone protein DnaK, A4JPD1_60 kDa chaperonin 2, and A6SY03_ Triosephosphate isomerase (B8IHL3_AUC = 0.96 (95 % CI: 0.87–1.00); A4JPD1_AUC = 0.98 (95 % CI: 0.92–1.00); A6SY03_AUC = 0.84, (95 % CI: 0.61–1.00) were analysed (Figure S1C ). LASSO regression analysis revealed improved diagnostic utility (AUC, 95% CI, 0.88–0.96) Model = (−3.118e‐11)* P15636 + (5.623e‐09)* B8IHL3 + 1.285 (Figure S1D ). In summary, our identified BEVs, which are common in vivo and in vitro, are valid therapeutic targets, and the combination of the identified proteins increases their reliability. To confirm that P15636_Protase1, which was identified as a potential therapeutic target for ovarian cancer, is a protein of bacterial origin, a homology analysis was performed using the identified peptide sequence. First, UniProt BLAST was run on human CD81 identified from the MS results of human ascites EV, and as expected, human CD81 was 100% homologous (Figure S2A ). The four identified CD81 peptides were analysed using UniProt BLAST and identified as 100% identical to the human protein sequence (Figure S2B ). UniProt BLAST analysis of each of the identified CD81 peptides showed 100% homology (Figure S2B,C ). Homology analysis was performed in the same manner for the bacterial proteins. As shown in Figure  3H , the E value of P15636_Protease1 was analysed using UniProt BLAST ( Homo sapiens ) to search for homologous human proteins, and those with E values less than 1E‐4 were considered homologous and listed. All human proteins were <50% homologous, suggesting that P15636_Protease1 was not a human‐derived protein but a bacterial protein. Then, the homology of the P15636_Protease1 peptide sequence identified by MS to the Mucin‐21 sequence listed at the top of the homology analysis (Figure  3H ) was examined. Peptides identified in P15636_Protease1 that had the same sequence (Figure S2D ) were excluded. As shown in Figure S2D , the peptide homology was <50%, indicating that the peptide unit was not derived from a human protein but from a bacterial protein. In addition to. P15636_Protease1, homology analysis of the three EV‐derived bacterial proteins identified in Figure  3E,F showed that some proteins had homology to human proteins, but not the same sequence (= 100% homology), indicating that these proteins are also of bacterial origin. We examined whether the B8IHL3_chaperone protein DnaK, A4JPD1_60 kDa chaperonin 2 and A6SY03_ Triosephosphate isomerase were derived from bacterial proteins in the same way as in Figure  3H . As shown in Figure S2E–I , some peptides are highly homologous to humans; however, no identical peptide sequences were identified. These proteins were of bacterial origin.

BEVs have been the focus of rapidly expanding attention in recent years, and expectations for new therapeutic markers and other applications are growing (Gurunathan and Kim 2023 ). BEVs secreted by pathogenic bacteria act as an infectious system that delivers toxins to host cells. Although the role and pathogenesis of gut microbiota‐derived BEVs have been studied (Peregrino et al. 2024 ; Liang et al. 2024 ), BEVs released by the vaginal microbiota are less known. Research on BEV is expanding, and databases such as EVpedia ( http://evpedia.info ) (Kim et al. 2013 ) compile the proteomics of BEVs. Lipopolysaccharide in Gram‐negative bacteria and lipoteichoic acid in Gram‐positive bacteria are BEV markers (Tulkens et al. 2020 ; Champagne‐Jorgensen et al. 2021 ). However, common BEV markers were fewer than in mammalian cells, and BEV research is currently limited due to the incomplete functional analysis of BEVs. In the present study, BEVs were recovered, and proteomics was performed from seven bacteria by dUC and SEC. A total of 4100 proteins were identified by dUC and 2112 by SEC. Figure S1B shows that dUC and SEC contained 1491 proteins. This result indicates that dUC may be more efficient in EV recovery or the EVs obtained by dUC may contain many proteins not derived from EVs. Even with EV recovery from mammalian cells, which has been extensively studied than from bacteria, which method is better, such as dUC or SEC, is still being debated (Brennan et al. 2020 ; Takov et al. 2019 ) (Gámez‐Valero et al. 2016 ). As shown in Figure  2F , many bacterial proteins have unknown localisation; thus, further analysis is needed to confirm which recovery method is better. Ovarian cancer is a gynaecological cancer with a high fatality rate when metastasis is detected (Coleman et al. 2013 ; Torre et al. 2015 ; Webb and Jordan 2017 ; Goff et al. 2000 ). Ovarian cancer has various subtypes, most notably HGSC, which accounts for approximately 75% of cases and has a lethality rate of nearly 90% (Cancer Genome Atlas Research Network 2011 ). However, no specific and sensitive biomarker has been established for HGSC, and CA125, the most commonly used biomarker in clinical diagnosis, has proven ineffective in the early detection of ovarian cancer (Menon et al. 2021 ). Therefore, ovarian cancer requires a new marker for its early detection and diagnosis. The vaginal microbiota is commonly associated with alterations in immune and metabolic signalling and the aetiology of many gynaecologic events such as vaginitis, pelvic inflammatory disease, endometritis and gynaecologic cancers (Łaniewski et al. 2020 ). Several studies have reported alterations in the vaginal microbiota of patients with ovarian cancer (Nené et al. 2019 ; Jacobson et al. 2021 ). The ovaries are located in the abdominal cavity, and ascites reflect the pathogenesis of ovarian cancer (Ford et al. 2020 ). Despite reports of bacterial changes in ascites fluid 16S‐seq from patients with ovarian cancer and without cancer (Yu et al. 2024 ), no studies have reported on bacteria‐derived proteins in EVs. In this study, a bacterial protein, P15636_Protease1, differentially expressed in patients with and without cancer was identified as a potential candidate for therapeutic prediction (Figure  3G ). Homology analysis was performed by BLAST to confirm the bacterial proteins. No homologous human proteins were found, indicating that the peptides identified were of bacterial origin (Figure  3I ). Homology analysis using BLAST (Altschul et al. 1997 ; Zaru and Orchard 2023 ) is generally popular; however, various other homology analysis tools, such as FASTA (Altschul et al. 1990 ), are available, and which method is better is still being debated (Webber and Barton 2003 ). Therefore, future studies are needed to examine whether proteins such as P15636_Protease1 are expressed in EVs derived from the vaginal microbiota, and more comprehensive patient samples are necessary to generate specific antibodies and further investigate the expression of P15636_Protease1. There have been no cases in which bacterial proteins were used as markers or predictors of treatment in ovarian cancer; thus, functional analysis of this protein may lead to new approaches to ovarian cancer diagnosis. Recently, proteins in BEV have been reported to be promising diagnostic markers for cancer. Su et al. suggested that BEV has potential applications in cancer diagnosis because of its presence in biological fluids and disease‐specific protein profile (Su et al. 2022 ). On the other hand, Zheng et al. found that BEV from F. nucleatum is increased in patients with colorectal cancer and contains proteins that promote tumour attachment, suggesting a link to diagnostic markers and disease progression (Zheng et al. 2024 ). These studies support that the BEV protein found in this study is a potential ovarian cancer biomarker. The bacterial‐derived proteins identified in ascites EVs suggest the presence of BEVs in the peritoneal cavity. Although the precise origin remains unclear, possible sources include ascending migration from the vaginal microbiota due to disrupted local barriers, or translocation from the gut via systemic circulation and leaky gut permeability (Zhang et al. 2024 ). Experimental validation of these hypotheses, such as comparative microbiome analyses or animal models, was not performed in this study and remains a limitation. Future work should focus on elucidating the source of these BEVs by analysing matched patient samples and investigating their effects on the tumour microenvironment. In summary, this study identified a bacterial protein that is highly expressed in the ascites EVs of patients with ovarian cancer and showed that the protein is a candidate marker for ovarian cancer treatment. Currently, no bacterial protein markers for ovarian cancer have been established. This study provides a new potential marker and therapeutic target for ovarian cancer treatment, which may lead to the early detection of ovarian cancer and improvement of prognosis.

Extracellular vesicles (EVs) are enclosed by lipid bilayers, measure 50–100 nm in size, and are released from all cells. EVs carry various molecules, such as proteins, RNAs and DNAs, as cargo and can transport them between cells (Yokoi and Ochiya 2021 ). EVs are abundant in body fluids, reflect pathological conditions, and have become a focus of attention as cancer‐specific diagnostic markers (Xu et al. 2018 ). In recent years, the association with bacteria has become important in tumour biology. Various tissues and organs possess microbiomes with unique characteristics that contribute to population dynamics and the diversity of microbial species and subspecies (Hanahan 2022 ; Nejman et al. 2020 ). This growing recognition of the importance of the microbiome in health and disease has led to recognition of its influence in the acquisition of cancer malignancy. Pathologists have detected bacteria within solid tumours using advanced profiling techniques. The detection of bacteria within solid tumours has long been recognised by pathologists, and these findings are now being demonstrated using advanced profiling techniques (Nejman et al. 2020 ). Bacteria also release EVs, which were discovered more than 60 years ago in Gram‐negative bacteria (Chatterjee and Das 1967 ; Devoe and Gilchrist 1973 ). Since their discovery in Gram‐negative bacteria with outer membranes, they have been recognised as outer membrane vesicles (OMVs) (Jan 2017 ). However, Gram‐positive bacteria also release these vesicles (Toyofuku et al. 2019 ), which have been called by various names (Devoe and Gilchrist 1973 ; Jan 2017 ). EVs released by bacteria are referred to as bacterial EVs (BEVs) (Tulkens et al. 2020 ). Although pathogens release BEV to deliver toxic determinants to local and distal host cells and shape the immune response (Ellis and Kuehn 2010 ; Kaparakis‐Liaskos and Ferrero 2015 ), their roles are still vague. Ovarian cancer is the second most common gynaecological malignancy in developed countries and has a poor prognosis and a high mortality rate (Torre et al. 2015 ). Ovarian cancer has several subtypes, with high‐grade serous carcinoma (HGSC) as the most common (Kossaï et al. 2018 ; Shih et al. 2021 ). However, many uncertainties remain regarding their development, and new molecular targets are required. Vaginal microbiota exists within the vagina (Moreno et al. 2016 ), and the balance in vaginal microbiota can lead to various pathological conditions, such as bacterial vaginosis (Onderdonk et al. 2016 ), preterm birth (Peelen et al. 2019 ), endometriosis (Muraoka et al. 2023 ) and ovarian cancer (Zhao et al. 2023 ). A 16 s analysis in patients with ovarian cancer showed different ovarian cancer and non‐cancer groups have different microbiomes as reflected in ascites (Yu et al. 2024 ). However, the mechanisms underlying the emergence of these conditions remain unknown, and the platform for clinical applications remains unclear. In this study, BEVs in patients’ body fluids were analysed, and they were also recovered from seven strains of bacteria present in the vaginal microbiota. Global proteomic analysis was successfully analysed. To identify BEV‐derived bacterial proteins that are highly expressed in ovarian cancer, ascites samples from 10 patients with ovarian cancer and five without cancer were analysed. Our findings showed that BEV released by vaginal bacteria can be a potential clinical biomarker for ovarian cancer.

The authors declare no conflicts of interest.

Supplementary Materials : jex270073‐sup‐0001‐figuresS1‐S2.docx Supplementary Materials : jex270073‐sup‐0001‐figuresS1‐S2.pdf