Lipidomic profiling of extracellular vesicles derived from prostate and prostate cancer cell lines

Background Extracellular vesicles (EVs) are produced and secreted from most cells of the body and can be recovered in biological fluids. Although there has been extensive characterisation of the protein and nucleic acid component of EVs, their lipidome has received little attention and may represent a unique and untapped source of biomarkers for prostate cancer diagnosis and prognosis. Methods EVs were isolated from non-tumourigenic (RWPE1), tumourigenic (NB26) and metastatic (PC-3) prostate cell lines. Lipids were extracted and subsequently used for targeted lipidomics analysis for the quantitation of molecular lipid species. Results A total of 187 molecular lipid species were quantitatively identified in EV samples showing differential abundance between RWPE1, NB26 and PC-3 EV samples. Fatty acids, glycerolipids and prenol lipids were more highly abundant in EVs from non-tumourigenic cells, whereas sterol lipids, sphingolipids and glycerophospholipids were more highly abundant in EVs from tumourigenic or metastatic cells. Conclusions This study identified differences in the molecular lipid species of prostate cell-derived EVs, increasing our understanding of the changes that occur to the EV lipidome during prostate cancer progression. These differences highlight the importance of characterising the EV lipidome, which may lead to improved diagnostic and prognostic biomarkers for prostate cancer. Electronic supplementary material The online version of this article (10.1186/s12944-018-0854-x) contains supplementary material, which is available to authorized users.


Background
Cells release vesicles into the extracellular environment that include exosomes, microvesicles and apoptotic bodies, collectively termed extracellular vesicles (EVs). Exosomes (30-150 nm diameter) are formed via the endosomal pathway and released by fusion of a multivesicular body with the plasma membrane, whereas microvesicles (100-1000 nm) and apoptotic bodies (> 1000 nm) bud directly from the plasma membrane. EVs transport cargos of mRNAs, miRNAs, proteins, lipids and metabolites to target cells throughout the body, protecting them from degradation in the extracellular environment. Thus, EVs are involved in cell-cell communication, allowing for the activation of signalling pathways, and the transfer of cargo to affect cellular function [1]. Through this mechanism, EVs have been shown both in vivo and in vitro to promote tumour growth and metastasis in many cancers, including pancreatic [2,3] and prostate cancers [4,5]. However, EV-mediated communication is an important component of normal physiological function, with roles in priming the immune system and modulation of stem cell plasticity [6].
Due to the overlapping biophysical properties of the various EVs, analysis of a homogeneous population is difficult without the use of multiple isolation and purification steps, including differential ultracentrifugation, which are time consuming and are often still unable to yield a truly homogeneous population. Despite this limitation, EVs have received quite extensive characterisation of both their protein and nucleic acid (mRNA and miRNA) content. Indeed, there are now many proteins that show an enrichment in EV populations, including tetraspanins (CD9, CD63 and CD81), integrins, Tsg101 and Alix [7], with these proteins regularly used as markers of EV populations. However, one class of biomolecules, lipids, have been largely overlooked in studies characterising the composition of EVs.
Lipids have been implicated in multiple aspects of EV biogenesis and function. Due to the presence of lipid raft-associated proteins, including flotillin-1, in EVs, it is thought that lipid rafts may be influencing selective protein sorting into EVs [8]. Lipid rafts are areas of the plasma membrane that are rich in cholesterol and glycosphingolipids, acting as platforms for lipid raft-associated protein signalling. Further, the tetraspanins CD9 and CD81 are present in most EV preparations and both have known associations with cholesterol [9], suggesting that their specific incorporation and enrichment in EVs may be due to their associations with lipid rafts. Cholesterols, sphingomyelins and phosphatidylserine are the major components of lipid rafts, with all three lipids showing increased abundance in EVs when compared to their secreting cells [10][11][12].
Due to the unique architecture of the prostate and renal system, there is an increased interest in the use of urinary EVs for biomarker discovery in prostate cancer [13][14][15]. Prostatic secretions, which contain EVs, make their way into the urine and can be collected non-invasively and in large quantities compared to blood, making this technique ideal for assessment of prostate-specific vesicles and the identification of biomarkers for the diagnosis and prognosis of localised cancer [16,17]. In addition to the potential of utilising EVs for diagnosis, their use as targeted therapies, and for the monitoring of pathological conditions, including cancer, has captured the attention of many researchers [18][19][20]. Clinical trials have already been performed, assessing the safety and efficacy of using exosomes as a therapeutic agent [21], however, there are many unanswered questions that remain pertaining to the complete biological composition of EVs. There is a lack in the understanding of the lipid composition of EVs and whether the lipid composition of EVs becomes altered during different pathological conditions. Further, it is not known how these alterations may influence, or be related to, the cargo recruitment and biological function of EVs. Herein, we have used a simple ultrafiltration approach for EV isolation, and targeted lipidomics to characterise the relative abundance of lipid species in three prostate cell line-derived EV populations. We have identified differences in the abundance of lipid species between non-tumourigenic, tumourigenic, and metastatic prostate cell line-derived EVs, highlighting the potential importance of the EV lipidome to cancer progression and diagnosis.

EV isolation and nanoparticle tracking analysis
Cells were grown to approximately 50% confluency in T175 flasks before media was aspirated and cells washed twice in sterile phosphate buffered saline (PBS; Thermo Fisher). KSFM, without growth supplements, was added to RWPE1 and NB26 cells. Serum-free Ham's F-12 K was added to PC-3 cells. Cells were incubated in a humidified incubator at 37°C with 5% CO 2 for 48 h before media was collected for EV isolation.
EVs were collected using a modified ultrafiltration protocol, as previously described [22]. Briefly, 150 mL of media per collection was centrifuged for 20 min, 4°C, 2000 × g to pellet cells and cellular debris. Clarified media was then sequentially passed through a 0.22 μm polyethersulfone (PES) syringe filter (Merck Millipore, Bayswater, VIC, Australia) and a 0.1 μm PES vacuum filter unit (Nalgene, Thermo Fisher). The 0.1 μm filtered media was then concentrated with an Amicon 100,000 MWCO Ultra-15 centrifugal filter unit (Merck Millipore) at 4°C, 4000 × g. Flowthrough was discarded and centrifugation was repeated until the entire sample had been processed. Retentates were washed twice with 0.1 μm filtered PBS and collected into Protein LoBind tubes (Eppendorf, North Ryde, NSW, Australia) and stored at − 80°C until use.
EV size and concentration was determined as previously described [22] using Nanoparticle Tracking Analysis (NTA) with a NanoSight NS300 and NTA v3.1 software (Malvern, ATA Scientific, Taren Point, NSW, Australia). Briefly, samples were diluted in 0.1 μm-filtered PBS and illuminated using a 405 nm (violet) laser. A scientific CMOS camera recorded 3 × 60s videos for each sample. NTA software was used to analyse the Brownian motion of particles in the captured videos and to generate analysis reports.

Lipid extraction
Six biological replicates of EVs from each cell line were collected and batch processed to minimise variation in lipid extraction between samples. Prior to lipid extraction, a total of 1 × 10 10 EVs per sample were pelleted at 18,000 rpm, 3 h, 4°C using a Hettich 1195-A fixed angle rotor in a Hettich Micro 220R refrigerated centrifuge. The Centrifugation Parameters Calculator, developed by the Laboratory of Molecular Human Genetics, Research Institute of Physical-Chemical Medicine, Moscow, Russia (http://vesicles.niifhm.ru [23]) was used to calculate the pelleting time, using a vesicle density of 1.08 g/ cm 3 , and a complete sedimentation "cut-off" size of 50 nm diameter.
Lipids were extracted from EVs using a modified Folch extraction protocol [24]. Briefly, EV pellets were resuspended in an ice cold chloroform:methanol mix (2:1) with 10 mg/L of internal standards PC(19:0/19:0) and PG(17:0/17:0) (Avanti Polar Lipids, Alabama, USA). Samples were vortexed and then mixed at 950 rpm, 20 min, 22°C with a Thermomixer C (Eppendorf ). Samples were centrifuged at 15,000 rpm for 10 min at room temperature and the supernatant collected into a LoBind tube. Samples were completely dried using a centrifugal evaporator and stored at − 80°C.

MS analyses
Extracted lipids were processed and detected by Metabolomics Australia (Bio21 Institute, Melbourne, VIC, Australia) as previously described [25]. Liquid chromatography electrospray ionisation-tandem mass spectrometry (LC ESI-MS/MS) was performed using an Agilent 1290 LC system and an Agilent 6490 triple quadrupole mass spectrometer (Agilent Technologies, Mulgrave, VIC, Australia). Equal volumes of each EV lipid sample were combined to create a pooled quality control (QC) sample. An injection volume of 5 μL was used for EV lipids, with QC samples run after every five EV samples. LC separation was performed using ZORBAX eclipse plus C18 column (2.1 × 100 mm, 1.8 μm) (Agilent) maintained at 60°C. Mobile phase solutions were water, acetonitrile and isopropanol (5:3:2 v/v for A and 1:9:90 v/v for B), both with 10 mM ammonium formate. Columns were run using a gradient of A and B solutions over 18 min at a flow rate of 0.4 mL/min. Metabolites were detected in positive ionisation mode using multiple reactions monitoring.

Data processing
The MS data were processed with Agilent Mass Hunter software. The resultant raw data, indicative of relative abundance (AUC) was further processed using Microsoft Excel for Mac v16.12 and filtered to remove metabolites with a CV > 20%. Missing values were imputed, and the raw data Log 2 transformed and normalised to the QC samples using MetaboAnalyst 4.0 [26], a web-based tool for the comprehensive analysis of metabolomics data. MetaboAnalyst was further used for principal component analysis (PCA) and the generation of heatmaps. Pie charts were created using the base functions in RStudio (RStudio, MA, USA), and the Metabolomics R package used to create boxplots of metabolites.
Lipids are named according to the LIPID MAPS classification system 'Headgroup(sn1/sn2)' , where sn refers to the radyl side-chains [27,28]. The side-chain structures are denoted as carbon chain length:number of double bonds and are provided for each chain where they could be determined, or as a total number of all carbons and double bonds where individual chains could not be determined.

Statistics
To determine differences between individual lipid species between EV samples, the Log 2 transformed data were analysed using an unpaired t-test, correcting for comparisons using the Benjamini and Hochberg false discovery rate with a q value of 5% (analyses available as Additional file 1). Data were considered statistically significant when q ≤ 0.05 with p values reported. These statistical analyses were performed using GraphPad Prism v7.0d (GraphPad Software, CA, USA).

Results
EVs isolated from non-tumourigenic (RWPE1), tumourigenic (NB26) and metastatic (PC-3) prostate cell lines had mean and mode sizes in the range of typically defined exosomes, as assessed by NTA (Fig. 1a). RWPE1 EVs had the lowest overall concentration compared to NB26 and PC-3 EVs, with NB26 having the highest overall concentration, in accordance with previous reports [29,30]. All EVs were positive for the common EV markers CD29/ITGB1, Alix, CD63 and CD9 by western blot (Fig. 1b), however PC-3 EVs had minimal CD9 expression in comparison to RWPE1 and NB26 EVs, which is expected as CD9 expression is lower in PC-3 cells compared to RWPE1 [31].

Characterisation of EV lipids
Targeted lipidomics was performed on EVs, using six replicate collections each, allowing for the quantitation of 187 lipid species using a log-transformed, QC normalised relative abundance that included lipids from the fatty acid, sterol lipid, sphingolipid, glycerolipid and glycerophospholipid lipid categories. Principal component analysis was performed on the dataset showing clear separation between the cell lines and clustering of the samples within the cell lines (see Additional file 2). The relative composition of EVs in the identified lipid classes differed slightly between RWPE1, NB26 and PC-3 cell lines (Fig. 1d-f; Table 1), where a decrease in abundance of glycerolipids, and an increase in abundance of sphingolipids and glycerophospholipids was observed in NB26 and PC-3 EVs compared to RWPE1. Further, MetaboAnalyst was used to generate a heatmap, using the Euclidean distance measure and Ward clustering, displaying the top 50 features as determined by t-test, highlighting the enrichment and clustering of various lipid species in EV groups (Fig. 1c).
There were two acylcarnitine species identified, acylcarnitines C17:0 and C18:1. Acylcarnitine C17:0 showed a − 1.73 FC decrease in NB26 EVs compared to RWPE1 EVs and a further − 1.86 FC decrease in PC-3 EVs compared to NB26 EVs (see Additional file 1). For acylcarnitine C18:1, there was a − 1.66 FC decrease in NB26 EVs compared to RWPE1 EVs and a further − 1.95 FC decrease in PC-3 EVs compared to NB26 EVs (see   showing the relative abundance of each lipid species in EVs. c TG species with a C15 or C17 acyl chain length at the sn1 position showed similar or increased abundance in PC-3 EVs compared to NB26 EVs. d TG species with a C14, C16 or C18 acyl chain length at the sn1 position had decreased abundance in NB26 and PC-3 compared to RWPE1 EVs   (Fig. 4e).

Discussion
Although there is increasing interest in the use of EVs for both diagnostic and therapeutic potential in a variety of pathologies, the understanding of the complete  [32,33], as well as the accumulation of triacylglycerols as energy reservoirs [34]. As EVs are known to transfer lipids and lipid-related proteins to affect target cell function [35][36][37], we employed a targeted lipidomics approach on our prostate cell-derived EVs, allowing for the specific identification and quantitation of lipid species. We detected relatively high numbers of DG, TG and CE species. It has been previously reported that the detection of large numbers of these species in EVs is due to the presence of contaminating lipoproteins and/or lipid droplets in EV isolates [12]. We utilised ultrafiltration over the traditional ultracentrifugation method for the isolation of EVs, as it has higher yields of vesicles reported [38]. Further, ultrafiltration is better suited for translation into a clinical setting due to the relative speed of the protocol, the larger volumes of sample able to be processed and the fact that no specialist equipment is required [39]. Ultrafiltration is more likely to co-isolate other particles, resulting in a heterogeneous mixture, and as such the DG, TG and CE species detected here may have arisen from lipid droplets. However, far from being considered a confounding factor, the detection of these species may have clinical relevance in terms of diagnosis of disease.
The accumulation of CEs in prostate cancer is associated with the progression and metastasis of the tumour, with this CE accumulation being identified in lipid droplets within prostate cancer cells [40], suggesting that the co-isolation of lipid droplets with EVs may be beneficial for prostate cancer diagnosis. Further, it has recently been shown that CEs, specifically CE (18:1) (cholesteryl oleate), can discriminate between prostate cancer, non-tumour, and benign prostatic hyperplasia [40,41]. Although we have shown that CEs were higher in abundance in metastatic PC-3 EVs compared to the normal RWPE1 EVs we saw a decrease in CE abundance in NB26 EVs compared to RWPE1 EVs, however, this may be a result of the lineage of the NB26 cell line. The NB26 cell line is a derivative of the RWPE1 cell line, formed by exposure of RWPE1 cells to N-methyl-N-nitrosourea and two generations of injection and growth in nude mice [42]. Due to this lineage, and although the NB26 cell line has a tumourigenic and invasive phenotype, it may retain many similarities to the RWPE1 cell line. In addition to the potential diagnostic value of CE detection in EV isolates, it has also been demonstrated that the metastatic prostate cancer cell line DU145 accumulates unsaturated TG species when undergoing the epithelial to mesenchymal transition [34]. Therefore, the detection of these lipid species, whether they be from EVs, lipoproteins or lipid droplets, may provide important information relating to the progression and prognosis of prostate cancer.
An increase in ether-linked lipids has been shown to cause higher numbers of EVs (exosomes) to be secreted from cells. PC-3 cells treated with an ether lipid precursor, hexadecylglycerol, resulted in increased levels of ether-linked lipids in both cells and their secreted exosomes, and alterations to the protein composition of the exosomes [43]. Here, we observed a large fold change increase in the abundance of ether-linked PC species in PC-3 EVs compared to RWPE1 and NB26 EVs, with NB26 and PC-3 also having the highest concentration of EVs isolated. Increased PC abundance has been identified in breast cancer, with higher levels identified in cancerous tissue compared to healthy tissue [44]. PC species have also been associated with poor prognosis, and inclusion of a PC species in a biomarker panel consisting of the lipids Cer (d18:1/24:1), SM (d18:2/16:0) and PC (16:0/16:0) had prognostic significance and this lipid signature could identify patients with poor prognosis [45]. In accordance with this three-lipid signature, the present study identified Cer (d18:1/24:1) in all samples, with lower abundance in RWPE1 (non-tumourigenic) and higher abundance in NB26 and PC-3 (tumourigenic, metastatic) EVs, suggesting a role for this ceramide in the progression of prostate cancer. As we were unable to determine the individual acyl chains of SM and PC species, it is unknown whether SM (d18:2/ 16:0) or PC (16:0/16:0) were detected in our EV samples.
A large proportion of the sphingolipids identified in this study belonged to the Cer and SM classes. Sphingolipids, including ceramides have been implicated in the production and release of EVs, however there is conflicting evidence regarding the role of ceramide in this process. Using the oligodendroglial precursor cell line, Oli-neu, treated with three unrelated sphingomyelinase inhibitors to inhibit the formation of ceramide from SM, it was found that there was a reduction in the number of exosomes released [46]. Conversely, in PC-3 cells treated with a glucosylceramide synthase inhibitor, there was no change in the number of exosomes released [47]. Additionally, acylcarnitines are utilised by cancer cells for energy production where they are shuttled from the cell membrane to the mitochondria to undergo fatty acid oxidation for downstream use in the TCA cycle. Although there were only two acylcarnitine species detected in this study, we saw that there was a significant stepwise decrease in abundance in EVs from non-tumourigenic, tumourigenic and metastatic prostate cell lines. In a prospective study of prostate cancer, an association between plasma acylcarnitine concentration and cancer progression was seen [48]. Acylcarnitine concentration was shown to decrease in patients with high grade prostate cancer and increase in patients with more advanced prostate cancer or who had succumbed to their disease. Although the associations were statistically significant using conventional statistics of linearity, when correcting for multiple comparisons using the false discovery rate, the associations were no longer significant. Similar findings have been seen in hepatocellular carcinoma where there was an increase in the accumulation of acylcarnitines with long FA chains (>C14) during disease progression [49]. As we saw a decrease in the concentration of long chain acylcarnitines in our EVs, this may suggest that there is a higher cellular utilisation of these metabolites that restrict their abundance in EVs.
Together, and as identified by heatmap analysis, the differences in abundance of lipid species identified between EVs from non-tumourigenic, tumourigenic and metastatic prostate cells highlights that these differences may be a source of biomarkers for prostate cancer diagnosis and prognosis. Further, the known functional roles of particular lipid species in cancer cells suggests that, in addition to providing encapsulation for EV cargo, the vesicle membrane composition along with the EV cargo may possibly have important roles in driving cancer progression.

Conclusions
In summary, we have identified differences in the abundance of the molecular lipid species of EVs from non-tumourigenic, tumourigenic and metastatic prostate cells in vitro. These differences highlight the potential for the use of EV-derived lipids for diagnostic and prognostic evaluation of prostate cancer. This work increases our understanding of the biological makeup of EVs and may pave the way for the development of a lipid signature biomarker for different disease states. By understanding the molecular lipid composition of EVs, we may better understand the roles of the EV lipidome in cell signalling events.

Additional files
Additional file 1: BH Corrected Multiple Comparisons of EV Lipid Species. Lipid species data were analysed using an unpaired t-test, correcting for multiple comparisons using the Benjamini and Hochberg FDR with a q value of 5%. The workbook contains 3 sheets, one for each comparison: RWPE1 v NB26, RWPE1 v PC-3 and NB26 v PC