Cell lines PC3MLuc2a obtained from Xenogen (Perkin Elmer, UK) and hOB cells - hfOb1.19 obtained from ECACC (Culture Collections, UK) were cultured in Dulbecco’s modified medium (#D5796, Sigma, United Kingdom) supplemented with 10% fetal bovine serum (#B9433, Sigma, United Kingdom), 50 U mL−1 penicillin and 50 µg mL−1 streptomycin (#15140122, Gibco, UK). C42-4b (provided in collaboration with Dr Penny Ottewell, University of Sheffield) and PNT1A (were provided in collaboration with Professor Nigel Mongan, University of Nottingham) were cultured as described (Kashyap et al., 2013).

Cells were seeding 1 × 10⁶ cells per 10 cm² dish and incubated overnight. Following the overnight incubation, cells were washed in phosphate-buffered saline (PBS) and fresh media containing 10% exosome-free fetal bovine serum (#A2720801, Gibco, United Kingdom) applied. Cells were incubated for 48 h under standard conditions. After 48 h, the media was collected and subjected to centrifugation at 300 × g for 10 min, 2000× g for 10 min, 100,00× g for 30 min at room temperature. The resulting supernatant was concentrated to 0.5 mL using a Vivaspin‐20 (100 MWCO) device (Sigma Aldrich, United Kingdom). The resulting concentrated media was treated with Vibrant Dil Cell Labelling Solution (#V22885, Thermo-Fisher, UK) at 1µL/100 µL (v/v) and incubated at 37°C for 15 min. The labelled supernatant was then separated into fractions by gravity flow chromatography (qEV1-70, Izon Science, United Kingdom) following the manufacturers’ protocol.

EVs were characterized by electrophoresis and Brownian motion analysis using laser scattering microscopy (ZetaView, ParticleMetrix, Germany) and visualized by electron microscopy. Briefly, EV samples were fixed with 2% formaldehyde and 2 μL applied to glow discharged amorphous carbon transmission electron microscopy grids (200 mesh; Agar Scientific) and then placed onto a piece of parafilm in a humidity chamber for 30 min. The grids were then washed twice with 2 μL of ddH2O and then blotted dried. For negative staining, the TEM grids were placed onto a 10 μL droplet of 2% filtered uranyl acetate for 1-min, excess solution was blotted dry, and the grid allowed to air dry for 5 min. Transmission electron micrographs (Technai T12; FEI) were then taken of the grids containing the EV fractions at 4.8kx, 9.9kx, and 20.5kx magnifications (Supplementary Figure S1).

EVs were further characterized for EV cargo proteins. Total protein was determined using the Qubit Protein Assay Kit (ThermoFisher, United Kingdom) and the Qubit 3.0 instrument (ThermoFisher, United Kingdom) prior to Western blot. 20 μg of protein per EV sample and 10 μg per cell lysate were separated by SDS-PAGE on 10% polyacrylamide gels (BioRad, United Kingdom). Proteins were transferred to nitrocellulose membrane using the BioRad Transblot Turbo System. Membranes were blocked for 1 h in 5% (w/v) bovine serum albumin (BSA) in TBST. Primary antibodies were incubated for 18 h at 4°C at a 1:1,000 dilution in 3% (w/v) BSA in TBST as follows: CD9 (D801A), Alix (3A49), and GM130 (D6B1) (Cell Signaling, United States). Membranes were washed three times in TBST for 10 min, followed by a 60-min incubation with either HRP-conjugated anti-rabbit or anti-mouse secondary antibody (Cell Signaling, United States) at room temperature. Membranes were washed three times in TBST for 10 min, followed by incubation with Clarity ECL (BioRad, United Kingdom) and visualization using the BioRad GelDoc Go Imaging System (Supplementary Figure S1).

The in vivo studies were carried out as described below. Briefly, CD1 NuNu male adult mice were injected via the tail vein with Dil labelled EVs in100 µL of PBS (equivalent to 60 µg total protein, median EV concentration of 2 × 10⁷), according to the schedule described below. Animals and organs were imaged at the appropriate time points using the IVIS Spectrum (Perkin Elmer, United Kingdom). Organs were snap frozen and stored at −80°C prior to downstream processing.

In vivo biodistribution experiments were performed to assess distribution and organ accumulation of the EVs. The experiments were conducted under the UK Home Office License number PPL P435A9CF8 with approval from the University of Nottingham AWERB. LASA good practice guidelines, FELASA working group on pain and distress guidelines and ARRIVE reporting guidelines were also followed.

10–12-week-old male CD-1 NuNu mice were purchased from Charles River UK. It was appropriate to only use males to model patient relevance. Mice were maintained in Individually Ventilated Cages (Tecniplast, United Kingdom) within a barriered unit, illuminated by fluorescent lights set to give a 12-h light-dark cycle (on 07.00, off 19.00), as recommended in the guidelines to the Home Office Animals (Scientific Procedures) Act 1986 (United Kingdom). The room was air-conditioned by a system designed to maintain an air temperature range of 21ºC ± 2°C and a humidity of 55% + 10%. Mice were housed in social groups, 4 per cage, during the study, with irradiated bedding and autoclaved nesting materials and environmental enrichment (Dates and, United Kingdom). Sterile irradiated 5V5R rodent diet (IPS Ltd., United Kingdom) and irradiated water (Baxter, United Kingdom) was offered ad libitum. The condition of the animals was monitored throughout the study by an experienced animal technician. After a week of acclimatization, the mice were randomly allocated by weight to the study groups of 2 mice per injection schedule. As this was a biodistribution study, no power calculation was required, minimum numbers used.

Prior to EV injection, the mice were optically imaged in the IVIS^® Spectrum imaging system, PerkinElmer (MA, United States), to obtain background control images. Prior to imaging, the mice were anaesthetized with an injectable anesthetic combination (Anaestemine [ketamine]/Sedastart [medetomadine], Animalcare Ltd. United Kingdom) before being placed in the imaging system. The mice were then injected with the EVs according to the schedule described below. After warming the mice in a thermostatically controlled heating box (Date sand United Kingdom), they were injected via the tail vein with Dil labelled EVs in100 µL of PBS (equivalent to 60 µg total protein). We selected the dosage based on previous optimization studies, attempting to tie together physiological relevance within the technical limitations of detection in our assays. In our hands, 60 µg of protein equated to a median EV count of 2 × 10⁷ per injection per mouse. Based on studies of plasma from human patients, circulating EVs have been reported as 2.92 × 10¹⁰ EV/mL plasma secreted from a variety of cellular sources (Johnsen et al., 2019). To date, the contribution of cancer EVs to the circulating population has not been defined due to variability in individuals and disease stage. However, our injection of EVs is within the physiological range for total amounts of circulating EVs and their accumulation remained detectable by IVIS.

Mice were imaged as above immediately post injection and then again prior to termination. No adverse effects were observed following the injections or for the duration of the study. At the appropriate study time point, the mice were terminally anesthetized, and blood was collected by intracardiac sampling and allowed to clot for serum collection. The mice were then killed by cervical dislocation and the organs harvested and imaged ex vivo. The organs excised and imaged are detailed in Table 1. Following imaging, organs and serum were snap frozen in liquid nitrogen. All imaging data was analyzed using the Living Image^® software (PerkinElmer, United Kingdom).

Recipient cells (hOBs) were seeded into six-well plates at 2 × 10⁵ cells per well. Following an overnight incubation, the media was replaced with media containing 10% exosome-free fetal bovine serum (#A2720801, Gibco, United Kingdom) and cells were grown for 24 h prior to the application of isolated EVs.

Femur, tibia, humerus, radius and ulnar limb bones were used after removing patella, fibula, and entirety of the feet. The bones were cleaned by cutting away all soft tissues, blood clots and blood vessels and other visibly contaminating material followed by aspiration with PBS. Visual observation was used to determine complete removal of any contaminating non-bone material. The tissue was cut into approximately 100 mg pieces and placed in the pre-packed FastPrep-24 2 mL microcentrifuge tubes (MP Biomedicals, United Kingdom). 0.6 mL of RNA protection reagent was added to each tube and the material homogenized at a speed of 4 m/s for 20 s in a Fast Prep Classic Homogenizer (MP Biomedicals, United Kingdom). If tissue was not completely homogenized, a subsequent 20 s pulse was performed. Homogenized samples were centrifuged at 12,000 × g (15 min, 4°C) to settle the protein precipitates and the supernatant was collected for RNA extraction. RNA was extracted using the Qiagen miRNAeasy kit according to the manufacturer’s instructions (Qiagen, United Kingdom).

For RNAseq, EV-treated mouse bone samples (bones from one fore and one hind leg per mouse) were sequenced by Novogene Co., LTD. (Cambridge, United Kingdom) using the Illumina NovaSeq platform to generate 150bp paired end reads. To determine if the potential molecular modulations of bone were detectable within the circulation, RNAseq data from circulating EVs isolated from whole blood from castration resistant PCa patients was analyzed [published (Huang et al., 2015b) accessed at GSE58410].

Data was analyzed using the Galaxy bioinformatics platform (Galaxy, 2022). RNAseq data of the EV-treated mouse bones, the sequence reads were trimmed with the TrimGalore wrapper for CutAdapt and mapped against the mm39 full genome and human hg19 using HISAT2 (Kim et al., 2015). 1.8%–2.29% of reads map to human. Cutoffs of <15% mapped reads to indicate sample contamination and <5% expected mapping when aligning to any similar species were used to concluded that human RNA sequences (contributed by EVs) fell within background levels and were without substantial contribution. Analysis was continued to determine differential expression based on mapping to the mm39 genome. Low-read filtering was applied to mapped data and differential expression determined using EdgeR on samples representing control and day 27 animals (Robinson et al., 2010). Due to n = 2 group numbers, stringent cutoffs for changes in genes were employed as a log2 fold change of ± 5 and FDR <0.1. Pathway enrichment analysis was performed using of QIAGEN IPA (QIAGEN Inc., https://digitalinsights.qiagen.com/IPA) and FunRich (Kramer et al., 2014; Pathan et al., 2015).

RNAseq data from circulating EVs isolated from whole blood from twenty-four castration resistant PCa patients was downloaded from GSE58410 (Huang et al., 2015b). The raw data was reanalyzed using the Galaxy bioinformatics platform. Sequence reads were trimmed with the TrimGalore wrapper for CutAdapt and mapped against the human hg19 using RNA STAR and FPKM used to calculate raw read count, FPM (fragments per million), and FPKM (fragments per million mapped reads per kilobase exon) for each gene (Wang et al., 2012; Dobin et al., 2013; Liao et al., 2014).

Additional statistical analyses were performed using GraphPad Prism version 10.0 for Windows (GraphPad Software, United States).

Upper and lower limb tissue was fixed in 4% paraformaldehyde for 2 h, dehydrated through an ethanol series, embedded into paraffin blocks, and sectioned at 7 μm. Immunohistochemistry was carried out on serial sections using a Leica Novolink Polymer Detection Kit (Leica, Germany) according to manufacturer’s protocols with primary antibodies diluted in 1:100 in fetal calf serum with an overnight incubation at 37°C. Rabbit monoclonals ab218237 [EPR21138] osteopontin (secreted phosphoprotein 1, SPP1, in humans), ab283654 [EPR23917-164] CD68, and ab93876 osteocalcin, rabbit monoclonals ab81289 [EP373Y] CD34 and ab290636 [EPR25122-122] SPARC, and mouse monoclonal ab88147 [3G3] Collagen I - BSA and azide free (Abcam, United Kingdom) were used to stain proteins of interest. In addition to immunohistochemistry, hematoxylin and eosin staining was conducted (2.5 min in hematoxylin, 15 s in 1% acetic industrial methylated spirits, 15 s in ammoniated water and 4 min in eosin). Picrosirius red staining was also performed to visualize collagen types I and III and undifferentiated collagens under polarized light (Polysciences, Inc., United States).

Microscopy was carried out to confirm positive staining (Leica CTR500 microscope, Leica Microsystems, Germany). Negative controls received no primary antibody and were incubated in fetal calf serum only, whilst skin, adipocytes, and muscle tissues were used as positive controls.

To determine the biodistribution of EVs and the organs they accumulate in overtime, Dil labelled EVs secreted from the PCa cell line PC3Mluc2a were injected via the tail vein of adult male CD1 NuNu mice. EVs were injected twice per week for up to 4-week and the accumulation of Dil signal in different organs was determined by IVIS imaging (Figures 1A–K). We identified a significant accumulation of signal in the liver and variable accumulation in the spleen from 24 h (Figures 1D, E). We also saw accumulations of signal in the lung after 2 weeks of EV exposure and in the lymph nodes and limb bones after three and 4 weeks of EV exposure. The later accumulation of signal in the lungs, lymph nodes and bone may represent a slower accumulation of the Dil signal in these organs compared to the liver and spleen or that the mechanism of accumulation in the non-elimination organs requires a specific homing process that required a longer period of EV exposure to manifest. We did not directly compare IV injection of EVs from a non-cancer cell line, as existing data from previous studies confirms non-cancer EVs do not commonly accumulate in bone. Grange et al., performed comparable studies using male CD1 nude mice injected with Dil labelled EVs isolated from non-cancer mesenchymal stem cells, no biodistribution to bone was detected using EVs in this model, whereas accumulation in the elimination organs was comparable to our data set (Grange et al., 2014). Furthermore, a meta-analysis of 38 EV biodistribution studies across other mouse models also confirms non-cancer EVs do not commonly accumulate in bone (Kang et al., 2021).

The limb bones from animals treated with EVs for 4 weeks were analyzed by RNAseq. Over 97% of reads mapped to the mouse transcriptome indicating a minimum contribution from potential human PCa EV transcripts. The data was further filtered to remove low reads, using only genes exhibiting a log2 fold change of ± 5 and FDR <0.1 for the final differential expression analysis (Figures 2A, B). The top fifty differentially expressed genes are displayed in Figure 2C and showed predominantly increased expression. The cellular localization of dysregulated genes was predicted to be across all cellular compartments, with many proteins not confined to a single cellular space (Figure 2D). Of genes showing increased expression in bone following PCa EV exposure, 4.8% were linked with surface receptors and 2% with downstream kinase signaling, and 22% with enzymatic, transporter and gene regulatory activity (Figure 2E).

Pathway analysis identified gene enrichment that aligned with 17 pathways, of which the direction of change in gene expression was attributed a significant positive Z score for activation in four pathways (Paxillin signaling p = 0.0163, Estrogen Receptor signaling p = 0.0271, RHOA signaling p = 0.0287 and Ribonucleotide reductase signaling p = 0.0307) and a negative Z score for inhibitory activity in one pathway (ERK/MAPK signaling p = 0.0299) (Figure 2F). Taken together these predicted changes in pathway activation and inhibition suggested an increase in cell organization, cell-contact and dendritic protrusion formation (a factor that may be pertinent to osteocyte activity), together with changes in cell signaling networks for which phosphoinositide-3-kinase regulatory subunit alpha (PIK3R1) was identified as a central node (Figure 2G). Whilst not unique to bone, these pathways have been linked with bone homeostasis/remodeling, but these findings require further validation.

Based on the pathways determined, selected genes and regulators within these pathways were chosen to enable validation of the RNAseq data in a human in vitro model system. Expression of a subset of the genes identified (Table 2) were validated using human osteoblasts (hOB) grown in vitro. hOB cells were chosen as we had previously demonstrated PC3 EVs change the phenotype of hOB cells indicating they are a target recipient cell (Liu et al., 2016). hOB cells were exposed to EVs isolated from PC3MLuc2a and C42-4b prostate cancer cells and compared to treatment with EVs from PNT1A healthy prostate epithelial cells and EVs isolated from hOB cells.

PC3MLuc2a and C42-4b were chosen as both cell lines have bone metastatic capacity in vivo, although this propensity for bone differs as do the effects on bone due to the original derivation of the C42 line from an LNCaP (lymph node met cell line) subcutaneous xenograft. However, we were keen to see if these early EV-mediated changes were recapitulated by an alternative advanced PCa cell line. Despite published data indicating non-cancer EVs are unlikely to home to bone, we also used EVs from the non-cancer prostate epithelial cell line PNT1A. The use of the in vitro system enabled the forced interaction between PNT1A EVs and bone cells, that otherwise may not occur in vivo, determining if the EV-mediated changes were prostate cancer specific. Finally, EVs derived from hOB cells were used as a control for the addition of EV material, as this is the addition of the cells own EVs at the same concentration as the other EV types, the use of this control informed us if the changes observed were simply the result of the addition of a higher concentration of EVs rather than being specific to the origin of the EVs.

The expression of SOS, ESR1, SP7, CSNK1A1, and DKK1 were confirmed by qPCR. For SOS, we confirmed a significant increase in expression following hOB exposure to PC3MLuc2a EVs (p = 0.01) (Figure 3A). ESR1 and SP7 were significantly downregulated following hOB exposure to PC3MLuc2a EVs (p = 0.04 and p = 0.0019 respectively) (Figures 3B, C), and whilst not reaching significance CSNK1A1 and DKK1 were also downregulated (Figures 3D, E). Exposure of hOBs to C42-4b EVs showed similar changes, except for SOS which showed no change in expression (Figures 3A–E). PNT1A EVs did not cause an increase in SOS expression or a significant reduction in expression of ESR1, SP7, CSNK1A1 or DKK1. However, we did observe that PNT1A EVs had the opposite effect for DKK1, causing a significant increase in DKK1 expression (p = 0.048) (Figure 3E).

In addition to changes in mRNA expression, two microRNAs were also identified as showing significantly increased expression in PCa EV treated bone. miR3965 and miR3470a, which have sequence similarity to human miR21-5p and miR1285-5p, respectively. The downstream targets of both miRNAs were identified in the bone RNAseq dataset, demonstrating the activity of these enhanced miRNAs to modulate key regulators of bone remodeling (Figures 4A, B). To validate these findings, we identified a selection of genes which are targeted by these miRNAs and their human counterparts in both mice and humans (STAG2, SLC35E3, NDNF, ZCCHC4, METTL21A, ZFX, ZBTB25, COX7B, and CLIP3) and determined if the in vitro treatment of human osteoblasts (hOBs) with EVs isolated from PC3MLuc2a cells compared to EVs from control PNT1A cells resulted in the same repression of these targets. Except for METTL21A and CLIP3, all targets showed significant repression (p < 0.0001) (Figure 4C).

Given the genes with altered expression in bone following PCa EV exposure were enriched in pathways linked with cell-cell contact and protrusion formation, we wanted to confirm if there was evidence to support osteocytes being a contributor to the molecular changes being induced by the PCa EVs. Osteocyte processes contact walls of canaliculi at discrete points within the mineralized matrix and relay cell signaling in response to bone stress, making them central in remodeling processes (McNamara et al., 2009). Youlten et al. (2021) characterized the gene expression of mouse osteocytes and defined a unique gene signature using RNAseq data (Youlten et al., 2021). We determined whether genes known to be expressed in enriched mouse osteocyte populations featured within the aberrant gene expression profile identified in our PCa EV treated mouse bones. 39.6% of the dysregulated genes in the PCa EV treated bones aligned with the transcriptome of mouse osteocytes (Figure 5Ai–ii). When comparing with the unique gene signature identified by Youltan et al., 52.2% of the genes overlap with our dataset from the PCa EV treated bone (Figure 5Aiii). These findings suggested osteocytes may be a major contributor to the molecular changes mediated by PCa EVs.

To identify what contributions may be made by this subset of genes which potentially originate in the osteocyte population, we performed pathway enrichment analysis on this subset of genes only. We identified enrichment of genes in thirteen pathways for which the associated Z-score suggests activation or inhibition (Figure 5B). Positive Z-scores were attributed to Myelination signaling (p = 0.0487), ribonucleotide reductase signaling (p = 0.0292), Paxillin signaling (p = 0.00462), Integrin signaling (p = 0.0209), Estrogen receptor signaling (p = 0.000972), EIF2 signaling (p = 0.028), D-myo-inositol-phosphate metabolism (p = 0.048) and 3-phosphoinositide degradation (p = 0.0446). Negative Z-scores for ERK/MAPK signaling (p = 0.00632), Osteoblasts in rheumatoid arthritis signaling (p = 0.0378), Osteoarthritis pathways (p = 0.0329), Sirtuin signaling (p = 0.01) and mitochondrial dysfunction (p = 0.0243).

Given the evidence of molecular changes, we sought to ascertain if these changes had yet translated into histological alterations within the bone. Qualitative analysis by H&E and with markers osteopontin, CD68, osteocalcin, CD34, collagen I and picrosirius red did not identify any changes in the bone architecture at the histological level (Figure 6; Supplementary Figure S2). Bone, bone marrow, and hypertrophic chondrocyte immunohistochemistry staining for osteopontin (secreted phosphoprotein 1, SPP1, in humans), osteocalcin and SPARC (osteonectin), no staining was observed in the non-hypertrophic chondrocytes (Figure 6; Supplementary Figure S2). CD34 and CD68 were present in bone and cartilage tissue, whereas collagen I was not expressed in the cartilage but was expressed in bone tissues, as confirmed by IHC and picrosirius red staining, within the EV treated mice (Figure 6; Supplementary Figure S2). These results indicated normal histological development of both hypertrophic and non-hypertrophic chondrocytes, and bone tissues, within the EV treated mice. We did not observe any gross changes in osteocyte peri-lacunocanalicular remodeling. However, due to limited material, specific studies to explore this further were not conducted and we could not conclusively rule out early changes in the lacunocanalicular network. These findings suggested the molecular changes identified are early changes occurring prior to any substantial bone remodeling activity. Therefore, if these molecular changes are creating a pre-metastatic niche, screening for these early changes in patients may provide a window of opportunity to block the progression of the metastatic cascade.

To determine the potential for using the molecular changes identified as early markers to identify patients at risk of bone metastasis, we sought to determine whether any of these changes were detectable within the circulation of patients with castration resistant PCa. RNAseq of circulating EVs isolated from twenty-four patients with castration resistant PCa, expressed 55 RNAs and one miRNA previously found to be increased in mouse bone exposed to PCa EVs for 27 days (Figure 7). Whilst these molecular changes are not unique to the creation of a bone-metastatic niche, the existence within the circulation warrants further investigation as to their utility as early indicators of the development of a pre-metastatic niche, either via sampling blood or bone directly.

This study has used an immunodeficient mouse model to enable the use of human PCa cell lines. A major limitation within the field is generating reliable bone metastases in vivo, which only occur with the use of selected human prostate cancer cell lines. However, our data has uncovered the alteration of pathways that are involved not only in bone-remodeling, but also potentially in the immune response if it were present. Whilst the use of immunocompromised mice does not negate the findings of the study, an awareness that a major element of the metastatic process is absent within these animals needs to be kept in mind. Future studies need to account for how the immune system may be altered by PCa EVs, to ensure key signals have not been missed and to understand the growing role of immune-based therapy in PCa. Further studies are also needed to evidence the mechanistic insights uncovered by the study, particularly validating the role of osteocytes and contributions of other bone cells in response to PCa EVs.