Extracellular vesicles in nipple discharge for breast cancer screening Authors: Yuri Ozaki, Kosuke Yoshida, Takahiro Ichikawa, Masami Kitagawa, Masamichi Kato, Ravi Velaga, Madoka Iwase, Yuko Takano, Dai Takeuchi, Toyone Kikumori, Kazuyoshi Murata, Takao Yasui, Masayuki Nagahashi, Tomoki Ebata, Hiroaki Kajiyama, Norikazu Masuda, Akira Yokoi Published: 2026 ABSTRACT AbstractPurposeNipple discharge (ND) is one of its major symptoms of breast cancer. However, the low volume of NDs has limited their diagnostic capability. This study aimed to identify extracellular vesicles (EVs) miRNAs in ND using cellulose nanofiber (CNF) sheets to develop clinical biomarkers.MethodsPatients with ND from two independent institutions between 2023 and 2024 were included. ND-EVs were isolated from 23 samples [normal, n = 8; intraductal papilloma, n = 1; ductal carcinoma in situ (DCIS), n = 8; invasive ductal carcinoma, n = 6] using CNF sheets. Additionally, miRNAs were also analyzed from intracystic fluid, tumor tissue, and tissue surface EVs from the same patients, and miRNA sequencing was performed.ResultsmiRNA sequencing for ND-EVs revealed distinct clustering between cancer and normal groups, identifying miR-342-3p, miR-20b-5p, and miR-550a-5p as candidate biomarkers. ND-EV miRNA profiles reflected tumor tissue characteristics, especially in DCIS. Four additional miRNAs (miR-92b-3p, miR-375-3p, miR-182-5p, miR-96-5p) were commonly upregulated in DCIS tissue, tissue surface, and ND-EVs. These miRNAs were validated by qRT-PCR. ROC curve analyses demonstrated that combining the five miRNAs yielded high diagnostic performance, with AUCs of 0.902 for breast cancer and 0.938 for DCIS. A subset of three miRNAs also showed robust discrimination ability.ConclusionND-EV miRNAs collected via CNF sheets show great potential as noninvasive biomarkers for early breast cancer detection.Supplementary InformationThe online version contains supplementary material available at 10.1007/s10549-026-07975-9.Keywords: Extracellular vesicles, Exosomes, Nipple discharge, Breast cancer, microRNA Background Breast cancer is the most common cancer in women, and approximately 2.3 million are diagnosed annually worldwide, resulting in 666,000 deaths in 2022 [1]. Breast cancer mortality from 1975 to 2019 decreased by 58% in the United States. Although mammography-based breast cancer screening is recommended and 25% of the decrease is attributable to improved screening [2, 3], earlier detection is desirable because of the consistently high breast cancer death rate. Nipple discharge (ND) is the third most common breast complaint after pain and masses [4, 5]. Although ND is secreted physiologically, some causes are pathological. Physiological ND includes normal milk production, prolactin-producing tumors, hypothyroidism, and side effects of medications [6–8]. Pathologic ND is frequently caused by a benign lesion, such as papilloma, ductal hyperplasia or ductal ectasia; however, an underlying malignancy can be present in 5%–33% of the cases [6–9]. Clinical investigations of the cause of these secretions are crucial in breast cancer detection. However, some patients with ND do not show any apparent abnormalities on mammography or ultrasonography and may further require magnetic resonance imaging or ductography [10, 11]. Even with these imaging modalities, a definitive diagnosis may still be difficult to obtain, and the National Comprehensive Cancer Network guidelines still recommend surgical duct excision [12]. To detect breast cancer noninvasively from ND, the cytology of ND has been used for over 70 years. However, its diagnostic accuracy has been reported to be limited [7, 13, 14]. Although other biomarkers such as tumor markers [15], genetic alterations [16, 17], miRNA [18–21] or proteins [22] have been investigated, their diagnostic abilities are also inadequate. Therefore, new diagnostic biomarkers are still needed. This study focused on extracellular vesicles (EVs), including exosomes. EVs are lipid bilayer membrane-delimited, nano- to micro-sized particles which appear to be released by all cell types. They contain proteins, lipids, and nucleic acids inside a lipid bilayer membrane. They are present in all biofluids, such as blood, breast milk, urine, ascites, and saliva, and mediate communication between cells, making them a potential noninvasive biomarker [23–25]. The function of EVs in cancer initiation, progression, and metastasis, as well as in response to therapy, has been widely reported in recent years [23, 24]. Breast cancer studies have reported an association of EVs isolated from blood with cancer incidence [26, 27]. Although analysis of EVs in NDs can be an ideal biomarker, the small sample volume, sometimes only a single droplet, is one of the reasons that complicates ND biomarker analysis [13]. Thus far, only two studies have isolated EVs from NDs [28, 29], and its contents or function is unclear. To overcome this problem, EV analysis was performed using cellulose nanofiber (CNF) sheets, which can capture EVs within porous nanopaper and require only 10 µL of liquid [30]. In the present study, we show that CNF sheet can capture EVs from NDs and that the miRNAs in these EVs reflect the miRNAs possessed by breast cancer. Furthermore, new breast cancer biomarkers were investigated by examining the differences in profiles between cancer-related and non-cancer-related NDs. Methods Patient selection Patients with ND who presented at Nagoya University Hospital and KATO Surgery-Gynecology-Obstetrics-Breast Clinic between October 2023 and October 2024 were included. The fluid produced in the ducts may be expelled from the teats as ND, or it may be encapsulated in the ducts [31]. Patients with intracystic fluid (ICF), the fluid inside the intracystic tumor, that corresponded to the latter category were also included. All breast cancers and intraductal papillomas were resected and pathologically diagnosed by expert pathologists. Normal cases included those without findings suggestive of malignancy on imaging or cytology and those with porous, non-bloody discharge in women who had breastfed within the past year. The patient’s characteristics are shown in Table 1. Table 1Patient characteristicsPatient No.AgeDiagnosisStageER statusHER2 statusSample type177IDC1+0.0ND267IDC1+2+ (FISH no amplification)ND351IDC1+2+ (FISH no amplification)ND451DCIS0+2+ (FISH not performed)ND538DCIS0+2+ (FISH not performed)ND668DCIS0+2+ (FISH not performed)ND769IDC1−0.0ND845IDC1+2+ (FISH no amplification)ND978IDC1−3+ICF1048IDP−ND1147DCIS0+2+ (FISH not performed)ND1268DCIS0+0.0ICF1359DCIS0+2+ (FISH not performed)ND, T-Ti, T-Sur, N-Ti, N-Sur1469DCIS0+1+ND1543No malignancy−ND1643No malignancy−ND1754No malignancy−ND1846No malignancy−ND1940No malignancy−ND2051No malignancy−ND2144No malignancy−ND2246IDP−ICF, T-Ti, T-Sur, N-Ti, N-Sur2348No malignancy−ND2443IDC1+1+ND2549DCIS0−3+ND, T-Ti, T-Sur, N-Ti, N-Sur2675DCIS0+1+ND2729DCIS0་1+ICFDCIS, ductal carcinoma in situ; ICF, intracystic fluid; IDC, invasive ductal carcinoma; IDP, intraductal papilloma; ND, nipple discharge; N-Ti, normal tissue; N-Sur, normal tissue surface; T-Ti, tumor tissue; T-Sur, tumor tissue surface Patient characteristics DCIS, ductal carcinoma in situ; ICF, intracystic fluid; IDC, invasive ductal carcinoma; IDP, intraductal papilloma; ND, nipple discharge; N-Ti, normal tissue; N-Sur, normal tissue surface; T-Ti, tumor tissue; T-Sur, tumor tissue surface This study was approved by the Ethics Committee of Nagoya University School of Medicine (Approval no. 2023-0066). Informed consent was obtained for each participant. Consent to publish clinical information potentially identifying individuals (e.g., age, gender, histological subtype, etc.) was also obtained. Sample collection Before ND collection, the nipple was first cleaned with alcohol pad. Once ND appeared from the nipple after manual pressure on the breast, a CNF sheet was used, and the fluid was spread evenly on the sheet for absorption (Fig. 1A and B). To obtain the EV profile of the tissue surface, the tumor and normal mammary gland areas of the excised specimen were devided and a CNF sheet was applied to collect the wetted portion of the surface. Tumor and normal mammary tissues were also collected and immediately stored at − 80℃. The CNF sheet was stored and air-dried at room temperature for 7–11 days. ICFs were collected by echo-guided puncture drainage following surgical excision and applied to CNF sheets in the same manner as NDs. Sample collection and characteristics of extracellular vesicles (EVs) in nipple discharge (ND). (A) Schematic representation of the sample collection. ND from breast cancer is adsorbed onto the EV sheet to collect ND-derived EVs (ND-EVs). (B) Representative images of sample collection. The nipple secretion fluid is adsorbed onto the EV sheet. (C) Size distribution and representative images of the ND-EVs. The size distribution was analyzed by a nanoparticle tracking analysis. Video data were collected three times, each with a duration of 10 s. (D) Representative images of extracellular vesicles (EVs) obtained by cryo-transmission electron microscopy. The scale bars show 100 nm. (E) Protein expression of EV markers in ND-EVs. DCIS ND-EV was positive for CD9 and normal ND-EV was positive for CD9 and CD81 Isolation of EVs from the CNF sheet and RNA extraction To isolate EVs from CNF sheets, the sheets were washed with 0.22 μm-filtered PBS in a Peri dish for 10 s and placed in 1mL of 0.22 μm-filtered PBS. The solution was vortexed for 10 s and left to stand at room temperature for 5 min. The total RNA was extracted using the Urine microRNA Purification Kit (Norgen Biotek, Ontrio, Canada) following the manufacturer’s instructions. The CNF sheet was transferred to a mixture of 1.5 mL of lysis buffer A (Norgen Biotek) and 15 µL of β-mercaptoethanol. The mixture was vortexed for 15 s and left to stand at room temperature for 5 min. After 5 min, the CNF sheets was removed. Next, 1.5 mL of 96%–100% ethanol was added, and the solution was vortexed for 10 s. An aliquot of 650 µL of the solution was added to the Norgen Kit’s column and centrifuged at 8,000 rpm for 1 min to discard the waste solution. This step was repeated until all samples were processed. The column was washed with 400 µL of the wash solution prepared by mixing 18 mL of Wash Solution A (Norgen Biotek) and 42 mL of 96%–100% ethanol. The column was centrifuged at 14,000 rpm for 1 min, and the waste solution was discarded. This washing step was repeated three times, followed by dry spinning at 14,000 rpm for 2 min. The column was placed on a new tube, and 50 µL of Elution Solution A (Norgen Biotek) was added. The column was centrifuged at 2,000 rpm for 2 min and 14,000 rpm for 2 min to elute the RNA. The RNA concentration was measured using a Qubit 2.0 Fluorometer (Thermo Fisher Scientific, Waltham, MA). Nanoparticle tracking analysis The size distribution and particle concentration in the isolated EVs were analyzed using a NanoSight NS300 (Malvern Panalytical, Malvern, UK). The cancer sample was not diluted. The non-cancer sample was diluted 1/10 before measurement. Video data were collected three times, each with 10-s duration. Cryotransmission electron microscopy The CNF sheets were washed by 0.22 μm-filtered PBS in a Peri dish for 10 s and placed in 1mL of 0.22 μm-filtered PBS. The solution was vortexed for 10 s and left to stand at room temperature for 5 min. The solution was filtered using a 0.22-µm filter and ultracentrifuged at 55,000 rpm for 2 h at 4℃ using an TLA-55-19U rotor (Beckman Coulter, Brea, CA). The isolated EVs were visualized under a cryotransmission electron microscope (Terabase, Okazaki, Japan). A 3-µl aliquot of the sample was applied onto a glow-discharged Quantifoil Mo grid R2/1 (Quantifoil Micro Tools, Grosslöbichau, Germany) and plunged-frozen in liquid ethane using a Vitrobot Mark IV (Thermo Fisher Scientific) at 4 °C and 95% humidity. The frozen grid, mounted on a 626 side-entry cryo-specimen holder (Gatan, Pleasanton, CA) was loaded into a 200 kV electron microscope equipped with an Omega-type energy filter, JEM-2200FS (JEOL, Tokyo, Japan), and imaged at a suitable magnification on a direct electron detector, DE-20 (Direct Electron LP, San Diego, CA) using a low dose system. Western blotting analysis The EV samples were loaded onto polyacrylamide gels for the electrophoretic separation of proteins at 30 mA. After blocking with Blocking One (Nacalai Tesque, Kyoto, Japan) for 1 h at room temperature, the membranes were incubated overnight at 4℃ with the following primary antibodies: mouse monoclonal anti-CD9 antibody (dilution 1:100, CBL162, Merck, NJ), rabbit monoclonal anti-CD63 antibody (dilution 1:1,000, EXOAB-CD63A-1, System Biosciences, LLC, CA) and mouse monoclonal anti-CD81 antibody (dilution 1:1,000, SHI-EXO-M03, Cosmo Bio, Tokyo, Japan). The membranes were washed three times for 5 min each using Tris-buffered saline with 0.1% Tween® 20 and incubated for 2–3 h at room temperature with secondary HRP-conjugated anti-mouse IgG (dilution 1:2,000, NA931-1ML, Cytiva, Marlborough, MA) or anti-rabbit IgG (dilution 1:5,000, NA934-1ML, Cytiva) antibodies. The protein ladder was MagicMarkTM XP Western Protein Standard (Thermo Fisher Scientific). The membranes were imaged using ImageQuant LAS 4010 (Cytiva). RNA extraction from tissues To obtain tissue RNA, tissues were shredded and crushed with QIAzol (Qiagen, Hilden, Germany) in a gentleMACS M tube (Miltenyi Biotec, Bergisch Gladbach, Germany). The total RNA was extracted using the miRNeasy® Mini Kit (Qiagen) following the manufacturer’s instructions. The RNA concentration was measured by NanoDrop (Thermo Fisher Scientific). miRNA sequencing The RNA extracted from CNF sheets and freshly frozen tissues were prepared using the NEBNext Multiplex Small RNA Library Prep Set for Illumina (New England Biolabs, Ipswich, MA). Polymerase chain reaction (PCR) products were purified using the QIAquick PCR Purification Kit (Qiagen). Electrophoresis was performed using 6% TBE gel at 120 V for 60 min. DNA fragments corresponding to 140–60 base pairs were trimmed. The DNA concentration was measured using a Qubit dsDNA HS Assay Kit and a Qubit 2.0 Fluorometer (Thermo Fisher Scientific). Single-end reads were performed on the Illumina MiSeq (Illumina, San Diego, CA). Quantitative polymerase chain reaction CDNAs were synthesized using the TaqMan Advanced miRNA cDNA Synthesis Kit (Thermo Fisher Scientific) according to the manufacturer’s instructions. TaqMan Fast Advanced Master Mix and TaqMan Advanced miRNA Assay (Assay ID_478043_mir for hsa-miR-342-3p, Assay ID_478074_mir for hsa-miR-375-3p, Assay ID_ 477823_mir for hsa-miR-92b-3p, Assay ID_477935_mir for hsa-miR-182-5p and Assay ID_478215_mir for hsa-miR-96-5p) were used. Then, qPCR was performed using QuantStudio 3 (Thermo Fisher Scientific). The PCR condition involved enzyme activation at 95℃ for 20 s followed by 40 amplification cycles of 95℃ for 1 s and 60℃ for 20 s. Each experiment was performed in triplicate. Data was normalized based on RNA concentration, which was measured using a bioanalyzer (Agilent Technologies, Santa Clara, CA). The relative expression was calculated by comparing it with the mean of normal ND-EVs. Statistical analysis The data analysis of miRNA sequencing (miRNA-seq) was performed using the CLC Genomics Workbench version 9.5.3 (Qiagen). First, adaptor trimming and mapping to the miRbase 21 database were performed, allowing two base mismatches. After normalization using reads per million mapped reads, miRNAs with low expression (< 10 reads in all samples) were excluded from further analyses. Subsequently, R software (ver. 4.2.2) and RStudio (Posit, Boston, MA) were used. For the heatmap analysis, data were converted to base 10 logarithms and z‑scores, and the heatmap.2 function of the gplots package (ver. 3.1.3.1) was used. To calculate and visualize principal component analysis (PCA), the prcomp and the plot3d functions of the rgl package (ver. 1.3.1) were used. For volcano plots, the p-values and log2 fold change for each miRNA were calculated using the Wald test in DESeq2 (ver. 1.36.0). The receiver operating characteristic (ROC) curve analysis was performed using the pROC package (ver. 1.18.5). Welch’s t-test was used to assess significant differences between the means of two groups. For comparisons involving more than two groups, Dunnett’s test was used with the normal group as the control. Moreover, Pearson’s correlation coefficient was used to calculate the correlation. A p-value < 0.05 was considered significant. Results Characterization of ND-EVs ND-EVs were collected using an EV sheet (Fig. 1A and B). EVs extracted from the EV sheet were identified using nanoparticle tracking analysis, and ND-EVs mainly consisted of small EVs (sEVs) because the particle size of was < 200 nm in diameter (Fig. 1C). Cryotransmission electron microscopy confirmed the presence of sEVs (Fig. 1D). Moreover, Western blot analyses showed that DCIS ND-sEV was positive for CD9, and normal ND-sEV was positive for CD9 and CD81, which were major EV markers (Fig. 1E). Identification of predictive biomarkers by miRNA-seq for ND-sEVs Twenty-three ND samples were included in the analysis: normal (n = 8), intraductal papilloma (IDP, n = 1), ductal carcinoma in situ (DCIS, n = 8), and invasive ductal carcinoma (IDC, n = 6; Fig. 2A). None of the breast cancer cases had lymph node or distant metastasis. Of the 17 breast cancer cases, 14 were positive and 3 were negative for estrogen receptor negative (Table 1). Initially, a comprehensive miRNA-seq was performed, and the miRNA profiles were analyzed (Supplementary Table S1). The heatmap and PCA showed that the miRNA profiles of the normal and cancer groups were placed into different cluster (Fig. 2B and C). The volcano plot showed that 15 miRNAs were significantly upregulated in IDC-derived sEVs compared with normal-derived sEVs (Fig. 2D and Supplementary Table S2). Similarly, 62 miRNAs were significantly upregulated in DCIS-derived sEVs compared with normal-derived sEVs (Fig. 2D and Supplementary Table S3). The Venn diagram showed that miR-20b-5p, miR-342-3p, and miR-550a-5p were commonly upregulated in IDC and DCIS (Fig. 2E). ROC curve analyses were performed to stratify patients with breast cancer and healthy women for the three candidate miRNAs, and the areas under the curve (AUCs) for miR-20b-5p, miR-342-3p, and miR-550a-5p were 0.794, 0.831, and 0.835, respectively (Fig. 2F). Moreover, the expression of the miRNAs in each case were visualized, showing that the baseline expression of miR-342-3p was higher than miR-20b-5p and miR-550a-5p (Fig. 2G). Therefore, we selected miR-342-3p as a candidate to stratify patients with breast cancer and healthy women with higher AUC and baseline expression. miRNA profiles of the extracellular vesicles (EVs) in nipple discharge (ND). (A) Patient characteristics. Twenty-three ND samples were included in the analysis: normal (n = 8), intraductal papilloma (IDP, n = 1), ductal carcinoma in situ (DCIS, n = 8), and invasive ductal carcinoma (IDC, n = 6). (B) Heatmap analysis for ND-EVs. After excluding miRNAs with low expression (< 10 reads in all samples), 809 miRNAs were used for the analysis. The data were converted to base 10 logarithms and z‑scores. (C) Principal component analysis for ND-EVs. After excluding miRNAs with low expression (< 100 reads in all samples), 344 miRNAs were used for the analysis. (D) Volcano plot showing differentially expressed miRNAs between cancer-derived and normal mammary-derived ND-EVs. The p-values and log2 fold change for each miRNA were calculated using the Wald test in DESeq2 (ver. 1.36.0). An absolute Log2 fold change (Log2FC) of > 1 and a p-value of < 0.05 were used as cutoff values. (E) Venn diagram showing significantly upregulated miRNAs in both IDC- and DCIS-derived ND-EVs. (F) Receiver operating characteristic curves distinguishing breast cancer and normal mammary for miR-20b-5p, miR-342-3p, and miR-550a-5p in ND-EVs. (G) Dot plots showing normalized read counts of miR-20b-5p, miR-342-3p and miR-550a-5p in ND-EVs. Error bars represent standard errors of the mean miRNA profiles of various samples Subsequently, the miRNA profiles of various samples were investigated. In addition to 23 ND samples, tissue miRNAs from three tumor samples (DCIS, n = 2; IDP, n = 1) and EV miRNAs on the respective tissue surface were analyzed. Moreover, tissue miRNAs and EV miRNAs from normal mammary samples from corresponding patients were also collected. Furthermore, four ICFs (DCIS, n = 2; IDC, n = 1; IDP, n = 1) were also analyzed. Consequently, miRNA profiles of a total of 39 samples were obtained by miRNA-seq (Fig. 3A). The PCA showed that the sample type was a critical factor, and miRNA profiles of the ICF samples were similar to those of the ND samples (Fig. 3B). However, the differences due to the disease were not apparent. Comparison of miRNA profiles in different samples from patients with breast cancer. (A) Schematic representation of sample collection. In addition to the 23 ND samples, tissue miRNAs from three tumor samples (DCIS, n = 2; IDP, n = 1) and EV-miRNAs on the respective tissue surface were analyzed. Moreover, tissue miRNAs and EV-miRNAs from normal mammary samples from corresponding patients were also collected. Furthermore, four intracystic fluids (ICFs) were also analyzed. In total, 39 samples from 27 patients were used for analysis. (B) Principal component analysis (PCA) of the 39 samples. The left panel shows the PCA for the sample types, whereas the right panel shows the PCA for IDC, DCIS, IDP, or normal mammary. (C) Representative images for collecting tissue-surface EVs and corresponding histology. Scale bars showed 100 μm. (D) Heatmap analysis of the miRNA profiles of the tissue and tissue-surface EVs. After excluding miRNAs with low expression (< 10 reads in all samples), 564 miRNAs were used for the analysis. The data were converted to base 10 logarithms and z‑scores. (E) Heatmap analysis of ND or ICF, tissue, and tissue surface for each case. Comparing cancer tissue and normal tissue in each patient, miRNAs with log2 fold change > 1 and mean read count > 100 were extracted and used for the analysis. The data were converted to base 10 logarithms and z‑scores Then, the miRNA profiles of the tissue-surface were analyzed, and the representative images of the sample collection are shown in Fig. 3C. The heatmap analysis showed that the sample type (tissue or tissue surface) was a critical factor and some miRNAs were highly expressed in the tissue surface of DCIS and IDP compared with those in the tissue surface of normal mammary tissue (Fig. 3D). This finding may indicate the abundance of tumor tissue-derived EV miRNAs on the tissue surface. Therefore, the miRNA profiles of various samples from the same patients were investigated. Tumor tissue-abundant miRNAs were extracted (log2FC > 1 and mean read count > 100). The heatmap analysis showed that tumor tissue-abundant miRNAs were also highly expressed in DCIS-surface EVs and ND-EVs, suggesting that ND-EVs might reflect the presence of DCIS (Fig. 4C). On the other hand, the miRNA profiles of ICF did not reflect those of the tissue miRNAs in the patient with IDP (Fig. 3E). Investigation of DCIS-specific nipple discharge extracellular vesicles (ND-EVs) miRNAs. (A) Venn diagram showing the commonalities of upregulated miRNAs across DCIS tissues and DCIS tissue surface EVs. A Log2 fold change > 0.7 was used as the cutoff, identifying 25 miRNAs. (B) Heatmap analysis for the 25 miRNAs in ND-EVs. The data were converted to base 10 logarithms and z‑scores. (C) Principal component analysis of the 25 miRNAs in ND-EVs. (D) Dot plots showing the performance of the 25 miRNAs. The X-axis represents the log10 base mean in DCIS-derived ND-EVs, whereas the Y-axis represents the AUC discriminating DCIS patients and normal women. (E) Receiver operating characteristic curves for the four selected miRNA (miR-92b-3p, miR-375-3p, miR-182-5p, and miR-96-5p). (F) Dot plots showing normalized read count of the four miRNAs in DCIS and normal. Error bars represent standard errors of the mean Identification of ND biomarkers for DCIS Considering that the NDs of DCIS may reflect the characteristics of DCIS tissue, analyses were performed to identify ND biomarkers of DCIS. Thus, miRNAs that are commonly upregulated in tissues and on tissue surfaces in two patients with DCIS were investigated, and 25 miRNAs were identified (Fig. 4A). The heatmap and PCA revealed that the expression of the 25 selected miRNAs in the ND tended to stratify DCIS and normal mammary tissue (Fig. 4B and C). To narrow down the miRNA candidates, the AUC of each miRNA was calculated. The baseline expression levels of each miRNA were also considered. As a result, four upregulated miRNAs (miR-92b-3p, miR-375-3p, miR-182-5p, and miR-96-5p) in the DCIS were selected (Fig. 4D and Supplementary Table S4). The AUCs for miR-92b-3p, miR-375-3p, miR-182-5p, and miR-96-5p stratified by DCIS and normal mammary tissues were 0.875, 0.875, 0.859, and 0.797, respectively (Fig. 4E). Moreover, the normalized reads of miR-92b-3p, miR-182-5p, and miR-96b-5p were significantly upregulated in NDs from DCIS compared with those from normal mammary tissue (p = 0.007, 0.020, and 0.007, respectively; Supplementary Fig. 4F). The expression of miR-375-3p was also upregulated in NDs from DCIS compared to those from normal mammary tissue; however, no significant difference was observed (p = 0.084). Therefore, these four miRNAs, reflecting the information from DCIS tissue and tissue surface, were considered excellent ND biomarker candidates. Validation of the diagnostic performance of miRNA candidates Finally, the diagnostic performance of five miRNA candidates (miR-342-3p, miR-92b-3p, miR-375-3p, miR-182-5p, and miR-96-5p) was assessed using qRT-PCR (Supplementary Table S5). In ND, qRT-PCR validated that the expression of the five miRNAs tended to be upregulated in IDC and DCIS compared to that in normal mammary tissue (Fig. 5A). Moreover, the normalized reads from miRNA-seq and the relative expression by qRT-PCR of these miRNAs showed a tendency for positive correlation, and miR-342-3p showed a significantly positive correlation (R = 0.887 and p < 0.001; Supplementary Fig. S1). Then, the diagnostic performance of these miRNAs was evaluated. The AUC for cancer versus normal ND for miR-342-3p was 0.643 (Fig. 5B). In addition, the AUCs for DCIS versus normal ND for miR-92b-3p, miR-375-3p, miR-182-5p, and miR-96-5p were 0.531, 0.672, 0.563, and 0.766, respectively (Fig. 5C). The diagnostic performance of miR-342-3p for distinguishing DCIS from normal ND was also assessed, but the AUC was modestly reduced to 0.578 (Supplementary Fig. S2A). In contrast, the AUCs for discriminating cancer from normal ND using miR-92b-3p, miR-375-3p, miR-182-5p, and miR-96-5p were 0.536, 0.563, 0.603, and 0.750, respectively (Supplementary Fig. S2B), indicating that these miRNAs may detect not only DCIS but also IDC. Diagnostic performance of target miRNAs. (A) Relative expression levels of miR-342-3p, miR-92b-3p, miR-375-3p, miR-182-5p and miR-96-5p in nipple discharge extracellular vesicles (ND-EVs). The expression of each miRNA was evaluated using qRT-PCR. The data was normalized based on RNA concentration, and the relative expression was calculated by comparing it with the mean of normal ND-EVs. Error bars represent standard errors of the mean. (B) Receiver operating characteristic (ROC) curve for miR-342-3p discriminating breast cancer (IDC and DCIS) from normal mammary–derived ND-EVs. (C) ROC curves for miR-92b-3p, miR-375-3p, miR-182-5p, and miR-96-5p discriminating DCIS from normal mammary–derived ND-EVs. (D) ROC curves for the combination of all five miRNAs discriminating breast cancer (IDC and DCIS) from normal mammary–derived ND-EVs. (B) ROC curves for the combination of miR-342-3p, miR-92b-3p, and miR-96-5p discriminating breast cancer (IDC and DCIS) from normal mammary–derived ND-EVs Furthermore, the diagnostic ability of the combination of the five miRNAs was investigated. The AUCs for all five miRNA combinations were 0.902 for breast cancer and 0.938 for DCIS (Fig. 5D). In addition, the AUCs for the combination of miR-342-3p, miR-92b-3p, and 96-5p were 0.804 for breast cancer and 0.813 for DCIS (Fig. 5E). Discussion In this study, EVs were successfully captured from NDs using CNF sheet, and their internal miRNAs were analyzed. In addition, five miRNAs were identified as potential biomarkers of early-stage breast cancer. A diagnostic platform that can noninvasively isolate EVs using NDs is reasonable for breast cancer screening and is expected to have high applicability. By analyzing EVs using CNF sheet, even a small sample can be evaluated, and the sheet can be stored at room temperature after collection, making this technique easy to apply clinically [30]. In breast cancer, ND may be more suitable for biomarker searches than other liquid samples such as blood; ND emerges from the ducts and may directly reflect the ductal epithelium and the surrounding microenvironment. EVs, which contain miRNAs and other contents, are released from cells and mediate cell-to-cell communication in the local and distant microenvironment [23, 24, 26, 27]. To our knowledge, very few studies of EVs from ND have been conducted [28, 29], and no previous reports have examined EV contents. Thus, the analysis of EVs from NDs using our method is expected to deepen our understanding of the breast cancer microenvironment. In IDC, a coexisting DCIS component lines the ductal lumen; therefore, EVs derived from in situ lesions can be directly released into the ductal lumen and reach the ND. Furthermore, given that EVs are extremely small particles (less than 200 nm) that can pass through the stroma and ductal epithelium, EVs derived from invasive cancer cells may also reach the ND. Therefore, ND-EVs may serve as biomarker in breast cancer. miRNAs are small, noncoding RNA crucial in posttranscriptional gene regulation by binding to complementary sequences in target mRNAs, which results in mRNA degradation or translation repression [32]. The miRNAs contained in EVs are covered by the lipid bilayer, so they are stable when released from the cell into the circulating body fluids [33]. Despite studies on miRNAs [18–21], none have been applied to clinical practice. In the present study, five EV-encapsulated miRNAs that were upregulated in breast cancer were identified. Previous studies have reported the association of these miRNAs with breast cancer, and miR-182-5p is involved in carcinogenesis, proliferation, metastasis, and sensitivity to tamoxifen [34–37]. miR-96-5p is also involved in proliferation and migration [38–40], and is useful for breast cancer detection using plasma [41]. miR-92b-3p was involved in proliferation in cell line study [42], and was reported as a potential biomarker for the diagnosis and prognosis when using the serum of patients with breast cancer [43]. miR-375-3p is associated with responsiveness to neoadjuvant therapy [44] and advanced stage [45]. Regarding miR-342-3p, as in our study, its levels in serum are higher in patients with breast cancer compared to those in healthy controls [19]. Moreover, studies have reported its suppressive effect on tumors [46–51], and three of these studies reported on miR-342-3p encapsulated in EVs [46, 47, 50]. Previous studies have suggested that mesenchymal stem cell-derived exosome miR-342-3p inhibits metastasis and chemotherapy resistance in metastatic breast cancer [46], EVs extracted from bone marrow mesenchymal stem cells carrying miR-342-3p suppress the growth and metastasis of breast cancer [47], serum miR-342-3p level correlates inversely with staging in patients with brain metastases [50], and miR-342-3p upregulation is a good prognostic marker in patients with metastatic disease [51]. Our study differs from these previous studies in that it only targeted early breast cancer without lymph node or distant metastasis. Since it has been reported that miR-342-3p inhibits epithelial-to-mesenchymal transition by interacting with inhibitor of differentiation 4 (ID4) [46], we may be observing a state in which metastasis is suppressed by the expression of miR-342-3p in this study. This study has several limitations. First, the sample size was small, which resulted in insufficient investigation into the biology of breast cancer such as the presence or absence of estrogen receptor expression. Furthermore, the small number of IDP cases was insufficient to investigate the differentiation between IDP and breast cancer and determine the need for surgical intervention. On the contrary, overdiagnosis and overtreatment of DCIS have become issues in recent years [52, 53]. Therefore, biomarkers not only for early diagnosis but also for risk assessment are desirable. Because ND highly reflects the state of the ductal epithelium and its microenvironment, our method may be applied in a large number of samples to stratify the risk of early breast cancer. In conclusion, five novel biomarkers were identified that could enable the early diagnosis of breast cancer through noninvasive testing which analyze miRNAs in ND-EVs using CNF sheets. The findings are promising for allowing early diagnosis and risk stratification, with potential application in future clinical trials. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 2 Supplementary Material 3 Supplementary Material 4 Supplementary Material 5 Supplementary Material 6 Supplementary Material 7 Acknowledgements We are grateful to the Division for Medical Research Engineering, Nagoya University Graduate School of Medicine, for equipment usage and technical support. We thank Mr. Mitsuru Ikeda for technical assistance with cryo-EM. We also deeply appreciate staff of the department of breast and endocrine surgery, pathologists (especially Dr. Satoko Shimada), and all patients who agreed to participate in the study. Abbreviations Cellulose nanofiber Extracellular vesicles Intracystic fluid Nipple discharge Principal component analysis Polymerase chain reaction Receiver operating characteristic Author contributions Y.O., T.I. and A.Y. conceived the project, designed the experiments, and interpreted the results. Each author contributed to this work as follows: Y.O. and M.K.: Isolation of EVs, RNA extraction and library preparation for sequencing. T.Y.: CNF sheet creation and optimization. K.M: Cryo-electron microscopy. K.Y.: Bioinformatics analysis. Y.O., T.I., M.K., M.I., Y.T., D.T., T.K., and N.M: Patient sample collection. R.V., M.N., H.K. and N.M.: supervision. Y.O., Y.I., K.Y. and A.Y. wrote the manuscript, and all authors critically reviewed the paper. Funding Open Access funding provided by Nagoya University. This work was financially supported by the following grants: a Project for Cancer Research and Therapeutic Evolution (P-PROMOTE) grant number: 25ama221433h0002 (to A.Y.) and Program for Accelerating Medical Research grant number: JP256f0137010 (to A.Y.) from the Japan Agency for Medical Research and Development (AMED); the Fusion Oriented Research for disruptive Science and Technology (FOREST; JPMJFR204J to A.Y.) and CREST (JPMJCR2576 to A.Y. and Y.Y.) from Japan Science and Technology Agency (JST); JSPS KAKENHI grant numbers 24K02586 (to A.Y.); BINDS from AMED under grant number JP24ama121005 (to K.M.). Moreover, the Princess Takamatsu Cancer Research Fund, the Uehara Memorial Foundation, Tokai Pathways to Global Excellence (T-GEx) as part of MEXT Strategic Professional Development Program for Young Researchers, the Aichi Cancer Research Foundation and the Nagoya University Research Fund. Data availability All small RNA sequencing data generated in this study have been deposited in the Gene Expression Omnibus database under accession codes “GSE293809”. All other data supporting the findings of this study are available within the article and its supplementary files. Declarations Ethical approval This study was performed in line with the principles of the Declaration of Helsinki and approved by the Ethics Committee of Nagoya University School of Medicine (Approval no. 2023-0066). Consent to participate Written informed consent was obtained from all participants prior to enrollment. Consent to publish Written informed consent for publication was obtained from all participants. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Yuri Ozaki and Kosuke Yoshida contributed equally to this work. References Associated Data Supplementary Materials Supplementary Material 1 Supplementary Material 2 Supplementary Material 3 Supplementary Material 4 Supplementary Material 5 Supplementary Material 6 Supplementary Material 7 Data Availability Statement All small RNA sequencing data generated in this study have been deposited in the Gene Expression Omnibus database under accession codes “GSE293809”. All other data supporting the findings of this study are available within the article and its supplementary files.