Cellular Disposal of miR23b by RAB27-Dependent Exosome Release Is Linked to Acquisition of Metastatic Properties

Bladder cancer is the fifth most common malignancy among men. Approximately 75% of patients with bladder cancer have frequently recurring non–muscle-invasive tumors (Ta, T1, or CIS). The remaining 25% of patients are diagnosed with muscle-invasive cancer (T2–T4) with eventual metastatic spread to lymph nodes, bones, liver, and lungs. The mechanism of metastasis is largely unknown and patients are, therefore, treated with nontargeted combination chemotherapy. However, only half of the patients respond, leading to poor patient survival (1). There is, therefore, an urgent need for improved understanding of the mechanisms leading to metastatic bladder cancer.

Carcinoma cells acquire distinct properties during metastatic colonization. These include migration and invasion in the primary tumor setting, intravasation and extravasation of the bloodstream, resistance to anoikis when in circulation, and finally outgrowth from micrometastases in distant organs. To address molecular events of metastatic bladder cancer, isogenic human bladder carcinoma cell lines with different metastatic capacity have previously been established through iterative injections of parental cell lines into mice (2, 3). A metastasis expression signature was deduced and specific genes were identified that suppress (RhoGDI2/ARHGDIB) or stimulate (LAMC2 and sialoglycoprotein CD24) metastatic progression (2–4). Tumor exosomes have been implicated in promoting metastatic spread of breast cancer and melanoma in mouse models (5–7). We, therefore, speculated if the release of exosomes and their specific content contribute to the metastatic behavior of the isogenic bladder cancer cells.

Exosomes are small vesicles (30–100 nm) that are secreted by various cells. They originate from inward budding into multivesicular bodies (MVB; ref. 8). Upon MVB fusion with the plasma membrane, the intraluminal vesicles are released into the extracellular space as exosomes, a process regulated, for e.g., by small GTPases RAB27A/B, as well as SMPD3/NSMASE2 (9–11). The vesicles were first discovered in differentiation of reticulocytes, during which obsolete transferrin receptors (TFR) were removed by exosome release (12). Exosome pathways may also facilitate the exocytosis of key molecules in cancer. Exosomal tetraspanins CD82 and CD9 restrain intracellular WNT signaling by targeting β-catenin/CTNNB1 for exosome release (13); viral LMP1 (encoded by Epstein–Barr virus) is secreted in CD63-exosomes as an antagonizing mechanism for LMP-1–mediated constitutive NF-κB activation (14), and active tumor-suppressor PTEN is exocytosed and transferred to other cells via exosomes (15). These studies highlight a possible cellular disposal role by exosomes. On the other hand, exosomes can transfer functional miRNAs to recipient cells (16, 17) and tumor exosomes may promote oncogenic signaling in recipient cells upon delivery of cargo (7, 18).

miRNAs are approximately 21 nt regulatory molecules that repress mRNA target translation. They are transcribed as long primary transcripts, processed into hairpin precursors by DROSHA, cleaved by DICER into mature miRNAs that then may bind argonaute (AGO) proteins for loading into the miRNA-induced silencing complex (miRISC). Exosomes contain precursor- and mature miRNAs at altered relative abundance compared with their cellular expression (17, 19), and modulate gene expression in recipient cells (10, 17). The vesiculation of miRNAs may depend on sequence motifs and protein sumoylation (20, 21). miRNAs with tumor-suppressor function are often lost in cancer (22). In this study, we analyzed whether the release of exosomes and their specific miRNA cargo were coupled to a metastatic behavior of bladder carcinoma cells and if miRNA exocytosis could be of intrinsic advantage to cancer cells themselves. Our results demonstrate that exocytosed miRNAs are decoupled from mRNA target and miRISC association. Moreover, miRNA secretion can be inhibited by RAB27A and -B knockdown that consequently lead to the intracellular upregulation of functionally active miR23b. This miRNA suppressed various steps of the metastatic process (invasion, angiogenesis, anoikis resistance, and lung colonization). Data from clinical samples linked RAB27B with poor prognosis of bladder cancer patients and loss of miR23b during lymph node colonization of tumor cells. In conclusion, our data suggest that exosome release may be a route to cellular disposal of tumor-suppressor miRNAs with intrinsic advantages to the parental cell.

All patients gave written consent, and the study was approved by the Central Denmark Region Committees on Biomedical Research Ethics (1994/2920 and 20040110). Primary tumors and patient-matched lymph node metastases were laser microdissected (>90% carcinoma cells) using the PALM laser microbeam system. Human primary urothelial cells (obtained from Prof. J. Catto, Department of Oncology, University of Sheffield, United Kingdom) and transitional carcinoma cells T24 (purchased from ATCC in 2009), SLT4, FL3, UMUC3, and LUL2 (obtained from Prof. Dan Theodorescu, University of Colorado Cancer Center, Aurora, CO, in 2010) were used in the study. Cell line authentification was conducted by Promega Cell-ID kit (IdentiCell, MOMA; Supplementary Fig. S1C). Last testing was conducted in January 2014. Metastatic SLT4 and FL3 cell lines are isogenic derivatives of the poorly metastatic T24T and its nonmetastatic parental cell line T24. SLT4 metastasizes primarily to the liver (2). FL3 and LUL2 have been generated from tail vein injections using T24T and UMUC3 cells, respectively, with resultant lung metastases (2, 3).

Neovascularization was examined by CAM assay. NCr nude mice (Taconic) were tail-vein injected with 1 × 10⁶ cells. At 4/6 weeks, mice were killed (4 at each time point), lungs were harvested, and processed for immunohistochemical staining or ddPCR. Three human-specific DNA sequences at chromosome 3, 7, and 12 were quantified by ddPCR using QX100 Droplet Digital system (Bio-Rad Laboratories Inc.).

Cells were propagated in 3 × 150 cm² flasks or CELLine Adhere1000 bioreactors (Integra Biosciences) for exosome isolation. Cells were propagated in DMEM supplemented with 10% FCS until 90% confluency, washed three times in PBS and replenished with advanced DMEM (Invitrogen) devoid of FCS or in DMEM + 10% exosome-depleted FCS (by 100,000 × g overnight centrifugation). The media were harvested and samples were processed when viability exceeded 95% (Trypan blue exclusion, Cedex XS; Roche Innovatis AG). Bioreactor cultivation was conducted as previously described (23).

Culture media were subjected to differential centrifugation of 3,200 × g (10 minutes) to remove residual cells and cellular debris, 16,000 × g (30 minutes) to remove apoptotic bodies and microparticles, and 100,000 × g (2 hours) to pellet exosomes (rotors: Type 70 Ti and Sorvall surespin 630 used in ultracentrifuges Beckman coulter and Sorvall Discovery 100SE by Hitachi, respectively). Exosome pellets were washed in PBS and centrifuged at 100,000 × g. Pelleted exosomes were added to 1,100 μL Qiazol for RNA extraction, PBS for nanoparticle tracking analysis (NTA), and SDS sample buffer for immunoblotting.

Vesicle size/concentration was analyzed by NanoSight LM10 system and NTA software v2.3 (NanoSight Ltd), calibrated at 183 nm/pixel using 100-nm calibration beads. Video recordings of 60 to 90 seconds and approximately 500 to 1,500 tracks were analyzed per sample.

Quantification of secreted protein was performed using the MicroBCA assay (Thermo Fischer). Immunoblotting of cellular and exosomal protein content was conducted using antibodies against RAB27A (Abcam; ab55667), RAB27B (Immuno-Biological Laboratories; 18973), Alix (Cell Signaling Technology; clone 3A9), CD63 (Pharmingen BD), Hsp90, CD81, and Tsg101 (Santa Cruz Biotechnology), and VDAC and β-actin (Abcam).

Trypan blue exclusion followed by automatic cell counting and MTT (Sigma-Aldrich) reduction assay was used to measure the cell viability, essentially as described (24). Cellular morphology and plasma membrane integrity was monitored by phase-contrast microscopy using a Zeiss Axiovert 40 CFL microscope.

Real-time invasion assays were conducted on CIM-16 plates using xCELLigence System (Roche). Invasion was quantified by calculation of the slope increment of a period of time (T0–T9 hours or T0–T20 hours) using RTCA software. Resistance to anoikis was quantified by plating 1.6 × 10⁵ cells on Corning costar ultra-low attachment 24-well plates (Corning Life Sciences).

RNA was extracted using the miRNeasy Mini Kit (Qiagen Sciences) by inclusion of MS2 carrier RNA (Roche) and cel-miR39 spike-in (DNA Technology A/S) for technical validation. RNA integrity was assessed using a 2100 Bioanalyzer (Agilent Technologies). miRNA expression profiling of exosomes and parental cells was performed using the stem loop RT-qPCR based TaqMan Low Density Array Human MicroRNA A+B cards (Applied Biosystems) v2.0 for exosomes/cultivated cells and v3.0 for primary tumors versus metastases, essentially as described (25). mRNA profiling was conducted using GeneChip Human Gene 1.0ST, 2.0ST, and U133+ Arrays (Affymetrix).

Scr- and RAB27B siRNA-transfected cells were scraped off culture flasks in gentle lysis buffer, processed, and incubated with AGO2 antibody (11A9; Sigma-Aldrich)-bound Protein G–coupled Magnetic Dynabeads (Life technologies), or anti-FLAG as a negative control (F1804; Sigma-Aldrich). Total RNA from input, AGO2-IP, and FLAG-IP were purified and miR23b and selected mRNAs were subsequently quantified by RT-qPCR TaqMan assays.

Statistical testing was conducted by linear regression, parametric (t test), or nonparametric (Wilcoxon–Mann–Whitney, Kolmogorov–Smirnov) testing when appropriate. P < 0.05 was considered significant. miRNA target site analysis was conducted using TargetScanS, Gene Set Enrichment Analysis (GSEA), and UCSC genome browser.

(Additional details are provided in Supplementary Information).

To investigate possible differences in exocytosis of miRNAs related to metastatic propensity, exosomes were isolated from isogenic bladder carcinoma cells with different metastatic capacity in mice by differential centrifugation. Electron microscopy of T24 and FL3-secreted vesicles revealed round-shaped double-lipid membrane vesicles upon epon-embedment, whereas cup-shaped upon negative stain (Fig. 1A). Immunoblotting demonstrated vesicular enrichment of tetraspanins CD63 and CD81, and presence of HSP90, ALIX, TSG101, and β-actin (Fig. 1B and Supplementary Fig. S1A). Minimal presence of contaminating mitochondrial VDAC was observed.

The size and concentration of vesicles was examined by NTA. The profiles were more homogenous from 16,000 × g supernatants than from 100,000 × g pellets, likely due to vesicle aggregates upon 100,000 × g sedimentation (Fig. 1C and Supplementary Fig. S1B). Subsequent quantitation was, therefore, performed on 16,000 × g supernatants. Increased number of secreted vesicles was observed for the metastatic cell line SLT4 (P < 0.01), but not significantly for FL3 (P = 0.15) in comparison with their nonmetastatic isogenic parental cell line T24 (Fig. 1D, left). The mean vesicle size (∼120 nm) was equivalent (Fig. 1D, right). MVBs are sites of exosome biogenesis and we, therefore, analyzed for possible differences in MVB volume percentage. Small vesicles included in MVBs were observed and a slight increase in MVB volume density in FL3 and SLT4 cells, although not significant (Fig. 1E), was also observed. Overall, the results indicated elevated, but not profound, exosome release from metastatic versus nonmetastatic cells.

We next profiled the expression of 667 unique human miRNAs in cells and their exosomes released over 24 hours by RT-qPCR. In addition to T24, SLT4, and FL3, isogenic bladder carcinoma UMUC3 (low-metastatic) and LUL2 (high-metastatic) cells, and primary human urothelial cells were included. The miRNA expression was significantly altered in carcinoma versus primary cells and exosomes, and correlated more between FL3 and SLT4 exosomes (from metastatic cells) than between T24 (nonmetastatic) and FL3/SLT4 exosomes (Fig. 2A; Supplementary Fig. S1D; Supplementary Table S1).

Next, cell-to-vesicle miRNA abundances were compared by nonparametric ranking analysis. A delta-rank score (vesicles versus cells) was calculated to identify relative highly exocytosed miRNAs (Fig. 2B and C, Supplementary Fig. S1E). Interestingly, some miRNAs were exclusively observed in vesicles and not in the parental cells (e.g., miR921), and some miRNAs were exclusively retained in cells (e.g., miR330-5p; Supplementary Table S2).

We analyzed whether the export of miRNAs was coupled to binding of miRISC complex proteins and putative target transcripts. First, from our recent quantitative proteomic study on exosomes (23), we observed no spectral counts for miRNA processing proteins (DICER and DROSHA) or miRISC-associated proteins (TNRC6A/GW182 and TARBP2P/TRBP; Supplementary Table S3A). AGO1 and AGO2 were detected in one of three replicates. However, immunoblotting revealed no detectable AGO2 (Fig. 2D). Second, mRNA profiling was conducted on cells and exosomes. The vesicles were deprived in transcripts containing target sites for the top 30 highly exported miRNAs (Fig. 2E). Instead, a significant higher number of target sites was detected in the cellular-enriched pool of transcripts (P < 2E10⁻¹⁶ for all six cell types; Fig. 2E and Supplementary Fig. S1F). Together, the results suggest that exocytosed miRNAs are not engaged in miRISC activity, and that exocytosis of specific miRNAs is coupled to high intracellular levels of their putative target transcripts.

We noticed that miR143 and -145 were relatively highly secreted from T24, FL3, and SLT4 cells (Fig. 2C). We previously identified these as tumor suppressors in bladder cancer (26, 27). Exosome-mediated transfer of miR143 inhibits proliferation of recipient cells (28) and we speculated if exosomal pathways could be a means of miRNA tumor-suppressor elimination by the producing cell. Two groups of miRNAs were selected for further analysis (Supplementary Table S3B) based on (i) their exclusive exocytosis from carcinoma versus normal cells and/or (ii) their high relative secretion from metastatic versus nonmetastatic cells.

Ectopic expression of miR145, miR223, miR885-5p (group 1) and miR23b, miR921, miR224 (group 2) was conducted by miRNA mimic transfection in T24 and FL3 cells (these cell lines were chosen as two extremes with respect to metastatic capacity). As expected, miR145 markedly reduced cell viability, whereas the remaining miRNAs did not (Fig. 3A and Supplementary Fig. S2A). We then focused on relative highly secreted miRNAs from metastatic cells (group 2). Tissue dissemination is a crucial initial step during the metastatic process. We, therefore, examined the capacity of cells to invade Matrigel upon miRNA overexpression by xCELLigence. Cellular invasion was inhibited by miR23b, miR224, and miR921 in FL3 cells (Fig. 3B, Supplementary Fig. S2B–S2D), the latter of which was diminished by LNA-miR921 knockdown (Supplementary Fig. S2C). Similar effects were observed in (low)-metastatic UMUC3 cells (Supplementary Fig. S2F). On the contrary, cellular invasion was stimulated by miR23b and inhibited by LNA knockdown (Fig. 3B, Supplementary Fig. S2E) in nonmetastatic T24 cells.

To further investigate this, we cloned miR23b and miR921 into pEGFP-C1 vectors. Stable cell lines expressing miR921 could unfortunately not be expanded, but T24 and FL3 cells with 10 to 20-fold elevated miR23b expression were obtained (Supplementary Fig. S2G). mRNA profiling revealed enrichment of miR23b seeds and/or predicted targets among downregulated transcripts (Fig. 3C). The overall transcriptional changes in T24 and FL3 cells were, however, predominantly opposite with respect to up- and downregulation (Supplementary Fig. S2H). Consistent with transient expression, cellular invasion was reduced for FL3 cells and increased for T24 cells upon stable eGFP-miR23b expression (Fig. 3D and Supplementary Fig. S2I). During the metastatic cascade, cancer cells avoid anoikis when arrested in circulation. In line with this, metastatic FL3 cells were more resistant to anoikis than nonmetastatic T24 cells (Fig. 3E). The anoikis resistance of FL3 cells was attenuated by stable expression of miR23b as previously documented for this miRNA (29).

We next addressed if miR23b acted as a tumor suppressor in vivo. miR23b significantly reduced angiogenesis in a CAM assay (Fig. 3F and G). When injected into the tail vein of mice, FL3 cells form lung metastases, whereas T24 cells do not (30, 31). Early lung colonization can be detected by a molecular PCR assay after 5 weeks, which correlates with subsequent macroscopic lung metastatic disease and Kaplan–Meier survival curves of mice (30). We examined early lung colonization in NCr nude mice upon tail vein injection of FL3 eGFP-ctrl and eGFP-miR23b cells. After 4 weeks, ddPCR of three human-specific DNA regions detected little or no human DNA (from FL3 cells) in mouse lungs. At 6 weeks, lung colonization was evident and a reduced tumor cell burden was observed upon injection of FL3 eGFP-miR23b versus eGFP-ctrl cells (Fig. 3H). Together, the results indicate that miR23b acts as a tumor suppressor at different steps of the metastatic process [invasion (in FL3), anoikis, angiogenesis, and lung colonization].

We wished to investigate intrinsic functional consequences of blocking miRNA tumor-suppressor exocytosis. For this, we aimed to establish a model system of diminished exosome release. This has previously been accomplished by knockdown of RAB27A and -B GTPases, or by pharmacologic inhibition (GW4869) or knockdown of NSMASE2/SMPD3 (9, 10, 32).

Exome sequencing identified none or only one missense mutation in RAB27A and RAB27B, respectively (Supplementary Fig. S3A; ref. 33). Immunohistochemical staining of tumor cores revealed RAB27A staining in less than 1 of 40 biopsies. For RAB27B, a diffuse cytoplasmic staining of varying degree was observed in the carcinoma cells (Supplementary Fig. S4B). High RAB27B expression correlated to progression from non–muscle-invasive to muscle-invasive disease (Fig. 4A; P = 0.011, log-rank test), and with reduced cancer-specific survival among patients with advanced disease (Fig. 4B; P = 0.046, Supplementary Fig. S3B).

All the examined cell lines showed detectable RAB27A, RAB27B, and SMPD3 (Fig. 4C, Supplementary Fig. S3C). The steady-state level of RAB27A or -B did not correlate directly with the metastatic behavior of the cells. Previously, a 20-fold reduction in RAB27B expression upon cultivation of urothelial cells has been reported (34). Thus, direct comparisons of expression levels in clinical samples and cell lines may be difficult. Specific knockdown of RAB27A and -B was obtained (Fig. 4D and Supplementary Fig. S3D), whereas knockdown of SMPD3 was inconsistent and excluded from further analysis. RAB27B knockdown was accompanied by several transcriptional changes (Supplementary Table S3C; Fig. 4E).

The phenotypic consequence of RAB27A or RAB27B knockdown was subsequently examined. Cell morphology and viability was unaffected 72 hours after transfection (with exception of one RAB27A siRNA in FL3 cells; Fig. 4F; Supplementary Fig. S3E) and at later time points (Supplementary Fig. S3F). RAB27A or -B knockdown did, however, attenuate cellular invasion (Fig. 4G and Supplementary Figs. S3G and S4A). Combined knockdown showed no additive effects (Supplementary Fig. S4B). Overexpression of RAB27B increased migration and invasion of UMUC3 cells (Supplementary Fig. S4C). Finally, GW4869 pretreatment also inhibited cellular invasion (Supplementary Fig. S4D–S4F). The data indicate that RAB27A, RAB27B, and NSMASE2 are involved in regulation of cellular invasion, and that RAB27B is associated with poor prognosis of bladder cancer.

We next established a model system to attenuate exosomal miRNA secretion. RAB27A and RAB27B knockdown reduced the number of vesicles present in 16,000 × g cell culture supernatant as detected by NTA (Fig. 5A), whereas GW4869 failed to inhibit secretion from FL3 cells (Fig. 5B). Knockdown also reduced the 100,000 × g pelleted vesicle fraction as measured by protein amount (Fig. 5C).

We then quantified the levels of exocytosed miRNAs. Knockdown of RAB27A and RAB27B decreased the exocytosis of miR23b and miR921. The effect of attenuation was similar and not additive upon combined knockdown (Fig. 5D). Treatment with GW4869 did not attenuate the release in accordance with no inhibitory effects on vesicle secretion (Fig. 5B and D). Exosomes are mainly pelleted at 100,000 × g. Accordingly, approximately 70% and 95% of the miR23b and miR921 signals were cleared upon 100,000 × g ultracentrifugation (Fig. 5E, Scr).

We next investigated intrinsic effects of attenuated miRNA exocytosis. Target transcript cleavage in miRISC relies on miRNA and target site abundances, sequence accessibility, and negative- and positive regulator expression (35, 36). We found that cellular transcripts most differentially regulated between FL3 and T24 also contained the largest number of (conserved) miRNA target sites (Supplementary Fig. S4G; P < 0.01). RAB27A and RAB27B knockdown was accompanied by an intracellular rise of (mature) miR23b, but not of pri-miR23b, in FL3 cells (Fig. 6A). We also analyzed T24 cells that secrete miR23b RAB27A/B-dependently but at a reduced relative rate (Figs. 2C and 5D). An intracellular rise in exact miR23b copies was observed upon RAB27B knockdown, which was not the case for cellular-retained miR330-5p (Fig. 6B).

Intracellular accumulation of miRNAs destined for exocytosis can either be functionally inactive due to encapsulation in intraluminal vesicles of MVBs or may accumulate in the cytosol to actively bind AGO2/miRISC proteins and direct target transcript degradation. To address this, we performed AGO2 immunoprecipitation experiments (Fig. 6C). RAB27B knockdown resulted in elevated miR23b association with AGO2 (with varying degree in three experiments; Fig. 6D). The increase was accompanied by general elevated levels of four putative target transcripts (Fig. 6E), whereas ubiquitin was unchanged (Supplementary Fig. S4H).

On the basis of the observations that (i) RAB27B knockdown and miR23b overexpression inhibited cellular invasion of FL3 cells (Fig. 3B and D and 4H) and (ii) RAB27B knockdown increased intracellular functional miR23b, we finally tested if miR23b knockdown could rescue the diminished invasion observed by RAB27B knockdown. An LNA molecule against miR23b had no effect in itself (Supplementary Fig. S2E), however, reverted the attenuated invasion induced upon RAB27B knockdown (Fig. 6F). Together, the results indicated that attenuated exocytosis of miR23b was associated with intracellular accumulation and increased functional activity.

The expression of the group 1 and 2 exocytosis miRNAs was finally examined in patient-matched laser-dissected primary tumors and lymph node metastases. Unsupervised clustering analysis largely separated primary tumors from lymph node metastases (Fig. 7A). A general reduced expression, including miR23b expression, was observed in lymph node metastases compared with primary tumors (Fig. 7A and B, box plots of selected miRNAs; of note, miR921 was not included in the assay platform version used here). Next, we performed GSEA of transcriptome profiles from 10 primary tumors and 12 lymph node metastases. A significant enrichment (FDR < 0.25 as defined by GSEA) of transcripts with target sites for exocytosis miR23b, miR224, miR144*, and miR143 was observed in the metastases. The data indicate a selective pressure for reduced expression of these tumor-suppressor miRNAs accompanied by increased expression of their putative targets during metastatic colonization.

Three major findings are presented in this study. First, the intracellular level and activity of miR23b with tumor-suppressor function was influenced by exocytosis. Second, RAB27A and RAB27B were identified as regulators of exosomal miRNA secretion, of which high RAB27B expression correlated to poor prognosis of bladder cancer. Third, metastatic colonization in vivo was accompanied by loss of miRNAs, which were also identified as highly exocytosed from cultivated cells with a metastatic capacity. Our study highlights a possible intrinsic advantage for cancer cells to exit superfluous tumor-suppressor miRNAs via exosomes, a process that may support the metastatic cascade (modeled in Fig. 7C).

RAB27A and/or -B have been found previously to regulate secretion of secretory granules, exosomes, and soluble proteins (9, 32, 37). Functional redundancy has been reported in some model systems (38), whereas nonredundancy in others (9, 39). We observed comparable (and nonadditive) inhibition of exosome secretion, miR23b/miR921 exocytosis, and cellular invasion upon RAB27A and -B knockdown. This may suggest functions along the same pathway in our model system. Urothelial cells predominantly express RAB27B over RAB27A (32, 34). In accordance, we found little expression of RAB27A in bladder cancer tumor cores. Previously, high RAB27B expression has been associated with presence of lymph node metastases among ER-positive breast cancer patients (40). In analogy, high RAB27B expression correlated with poor prognosis of bladder cancer. Whether this is mainly ascribed to increased RAB27B-dependent vesiculation or other identified RAB27B-mediated transcriptomic changes cannot be explicitly ascertained.

We identified a novel pathway for RAB27A/B-dependent cellular export of miRNAs. The majority of miR23b and miR921 was associated with 100,000 ×g pelleted material, suggesting predominant exosome association rather than a soluble released form. We cannot entirely exclude that a fraction of 100,000 ×g-associated miR23b may be soluble and potentially contribute to the route of cellular disposal. miR23b was still secreted via the 100,000 × g fraction after RAB27A knockdown, which was not the case upon RAB27B knockdown (after which the remaining miR23b was present in the soluble fraction). This may suggest that subpopulations of miRNA-containing exosomes exist, and/or that inhibition of one exit route may redirect miRNAs destined for exocytosis to other secretory routes.

What could be the mechanism by which tumor-suppressor miRNAs are sorted into exosomes? Our data indicate that exocytosed miRNAs, in general, are decoupled from their repressive activity based on absence of miRISC proteins and depletion of their target transcripts in exosomes. It does not appear to be a passive consequence of increased MVB formation, as the MVB volume was only marginally increased in the metastatic cells. The cargo enclosure may be specified by tethering proteins that recruit excessive miRNAs. One study has identified miRNA sequence motifs and sumyolation status of hnRNPA2B1 to direct the sorting (20). We have identified several RNA-binding proteins of elevated abundance in exosomes from metastatic FL3 and SLT4 cells versus nonmetastatic T24 cells (e.g., hnRNPA3, -U, M, C1/C2; ref. 23), which could sequester specific miRNAs in metastatic cells. miRNAs may also be sorted into exosomes following 3′tail modifications (M. Pegtel and D. Koppers-Lalic, personal communication), of which some have been implicated in miRNA turnover (41). miRNAs are generally long-lived (average ∼5 days, 10× fold that of mRNAs; ref. 42). It is, therefore, plausible that cellular export serves an auxiliary route to efficient elimination of undesired miRNAs. In line with this, selective exosomal disposal of miR150 was recently proposed as a rapid means to regulate gene expression during lymphocyte activation (43). Thus, a balance of “intracellular active” versus “exocytosis inactive” miRNAs is likely influencing the transcriptome as a whole.

The main focus of this study was the exocytic trafficking and function of miR23b. It reduced invasion (in FL3), resistance to anoikis, and in vivo angiogenesis and lung colonization in line with previous reports of miR23b being a pleiotropic metastasis suppressor (29). miR921 also showed remarkable metastatic-cell–specific and RAB27A/B-dependent exocytosis, but stable miR921 cell lines could unfortunately not be obtained. Overexpression of miR23b surprisingly inhibited invasion of FL3 cells and stimulated invasion of T24 cells, although target transcripts were downregulated in both cell lines. Variable secondary miRNA effects due to cellular context have previously been described (36), which could explain the overall opposite transcriptional changes and phenotype observed. miR23b displayed several metastasis-suppressive effects most prominent in the metastatic cell line (FL3), from which it was more highly exported.

Previous studies have documented reduced expression of miR23b, miR224, and miR921 in bladder- and prostate cancer (44–46), and reduced miR143/145 upon bladder cancer lymph node metastasis (47). We confirmed this latter finding and observed reduced expression of other identified exocytosis miRNAs accompanied by significant upregulation of their putative target transcripts in the metastases. The mechanism behind reduced expression is difficult to prove in vivo. Tumor-suppressor miRNAs are commonly lost due to epigenetic silencing, acquired mutations, or alterations in effector/repressor expression balance. We identified no or little evidence of miRNA gene mutations (Supplementary Fig. S4I; ref. 33) or epigenetic silencing of exocytosis miRNAs miR144, miR451, miR143, and miR145 in the TCGA bladder cancer dataset (Supplementary Fig. S4J). For miR23b, a modest correlation was observed, indicating epigenetic silencing as a likely contributing factor, but not in all cases. We did not attempt to mechanistically address miRNA vesiculation in vivo, which may be a limitation to the study. However such analysis is not trivial as circulating miRNAs originate from a range of cell types (and possibly from both primary tumor and metastases) and under influence of normal physiologic processes.

Our work may suggest a novel role for exosome release as a route to cellular disposal of miRNAs with the consequence to dampen intracellular effects on target transcripts. In this case, an intrinsic selective pressure for disposal of superfluous miR23b, which in turn may contribute to the transcriptomic changes associated with a cellular metastatic state. The model does not argue against exosomes as conveyors of cell-to-cell communication, but rather suggests the existence of a duality. On one side, exosome release may limit abundance of specific intracellular content and on the other side may affect other cells when the vesicular content is transferred.

No potential conflicts of interest were disclosed.

Conception and design: M.S. Ostenfeld, M.H. Rasmussen, T.F. Ørntoft

Development of methodology: M.S. Ostenfeld, D.K. Jeppesen, A.T. Boysen, B. Primdal-Bengtson, D. Theodorescu, K.A. Howard, L. Dyrskjøt

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.S. Ostenfeld, A.T. Boysen, B. Primdal-Bengtson, A. Hendrix, F. Dagnaes-Hansen, M.H. Rasmussen, K.H. Bui, N. Fristrup, E.I. Christensen, I. Nordentoft, J.P. Morth, J.B. Jensen, M. Borre, L. Dyrskjøt, T.F. Ørntoft

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M.S. Ostenfeld, J.R. Laurberg, A.T. Boysen, A. Hendrix, P. Lamy, N. Fristrup, E.I. Christensen, I. Nordentoft, J.S. Pedersen, L. Dyrskjøt

Writing, review, and/or revision of the manuscript: M.S. Ostenfeld, B. Primdal-Bengtson, A. Hendrix, P. Lamy, M.H. Rasmussen, E.I. Christensen, I. Nordentoft, J.B. Jensen, M. Beck, M. Borre, K.A. Howard, L. Dyrskjøt, T.F. Ørntoft

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M.S. Ostenfeld, A.T. Boysen, B. Primdal-Bengtson, M. Borre, L. Dyrskjøt

Study supervision: M.S. Ostenfeld, M. Beck

Other (experimental work): J.B. Bramsen

The authors thank Kasper Thorsen and Thomas Reinert for fruitful technical discussions and Susanne Bruun, Margaret Gellett, and Martin Rasmussen (Department of Clinical Pathology, Odense University Hospital) for technical assistance.

This work was supported by the Novo Nordisk Foundation, Toyota Foundation, Carlsberg Foundation, John and Birthe Meyer Foundation, Danish Research Council for Natural Sciences, Lundbeck Foundation, The Danish Cancer Biobank, Danish Cancer Society, and the European Community's FP7/2007-2011 (grant no. 201663).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.