Circulating cell-free micro-RNAs (miRNAs) have emerged as promising biomarkers for the development of blood-based assays for early detection, prognosis and monitoring of cancer. In 2008, Mitchell et al. demonstrated for the first time that miRNAs are released from prostate cancer (PC) cells into the bloodstream, where they exist in a remarkably stable form [1]. miRNAs were shown to remain stable after incubation of plasma or serum at room temperature for up to 24 h and to resist RNase A digestion, HCl and NaOH treatment or multiple freeze-thaw cycles [1, 2]. Subsequently, the levels of circulating miRNAs have been studied in patients with various cancers, including PC, resulting in the discovery of individual miRNAs or miRNA signatures with diagnostic and/or prognostic value [3].

PC is the most frequently diagnosed cancer in males in Europe and the United States [4, 5]. Currently, the serum PSA test is the most commonly used tool for organised screening programs, opportunistic screening and monitoring of PC [6]. However, PSA is not cancer specific and the high false-positive rate and low specificity leads to large numbers of unnecessary prostate biopsies and emotional morbidity [7]. Furthermore, PC is characterised by a highly heterogeneous course - one part of the patients develops a high-grade disease with extracapsular spread and distant metastases requiring aggressive treatment, while others have a relatively indolent, slowly progressing disease that could have been managed by active surveillance [8]. The current standard of care analyses, however, do not predict whether a histologically proven tumour will give rise to a clinically significant disease, leading to overtreatment of indolent PC. Hence, the greatest unmet clinical needs in the management of PC are sensitive and reliable non-invasive tools for differentiating between PC and benign prostatic diseases, and between potentially fast progressing PC requiring aggressive treatment and a relatively indolent disease that can be managed by active surveillance.

More than 20 studies investigating levels of cell-free miRNAs in plasma or serum of PC patients have been published up to date [9, 10]. The majority of these studies were focused on the identification of circulating miRNAs that differentiate between patients with PC and benign prostatic hyperplasia (BPH) or healthy controls. Some of these studies have shown remarkably high diagnostic value. For example, Chen et al. identified a 5 miRNA panel that could differentiate PC from BPH with an AUC of 0.924 and PC from healthy controls with an AUC of 0.860 [11]. Some other studies have reported cell-free miRNAs that differentiate between localised and metastatic castration resistant prostate cancer (mCRPC) or between low-grade and high-grade PC. For example, Mihelich et al. developed a “miRScore” that based on the serum levels of 14 miRNAs could predict absence of high-grade PC among men with PC and BPH with a negative predictive value of 0.939 [12]. However, relatively few miRNA biomarkers have been validated by several independent studies, while many other miRNAs either have been reported in a single study or show conflicting results [3, 10]. Therefore, the analysis of cell-free miRNAs is regarded as a poorly reproducible technique [3, 13, 14].

Cell-free miRNAs circulating in the bloodstream have been found to be enclosed into extracellular vesicles (EVs) [15, 16], or to exist in a vesicle-free form associated with high-density lipoproteins [17], Ago2 protein [18, 19] or other RNA binding proteins [20]. The majority of the studies has used whole plasma or serum as a source of cell-free miRNAs. However, it has recently been hypothesised that cancer-derived EVs may be enriched with miRNA signatures reminiscent of their cell of origin, contain rare yet highly specific RNA biomarkers and protect their RNA cargo from degradation in the bloodstream and, therefore, the analysis of EV-enclosed miRNAs may be superior to whole plasma/serum analysis [10, 21, 22]. Nevertheless, to the best of our knowledge, a direct comparison of miRNA detection assays in whole plasma and plasma EVs has not been reported so far.

In this study, we evaluated the performance of 9 miRNA biomarkers previously reported to have a diagnostic or prognostic significance in PC by quantifying them in the whole plasma and plasma EVs in a cohort of 50 PC and 22 BPH patients.

Cells can release miRNAs to the extracellular space either incorporated into EVs [31, 32] or in a vesicle-free form bound to various protein and lipoprotein complexes [17–20]. Quantification of these miRNAs in blood from cancer patients may offer new opportunities for diagnosis, prognosis, monitoring of treatment response and early detection of recurrence in a minimally invasive way. However, human blood contains a complex mixture of miRNAs derived from various cell types and, therefore, robust quantification of cancer-derived cell-free miRNAs has turned out to be a challenging task [14]. Currently, it is still debated if the EV-based miRNA detection assays are superior to the whole plasma-based assays. miRNA profiles in cancer-derived EVs have been found to be reminiscent of their cell-of-origin [31, 33], though due to selective RNA sorting mechanisms they may be enriched or depleted of some specific miRNAs [34]. The EV membrane protects the RNA cargo from degradation in the bloodstream and the intraluminal RNA content is thought to be relatively stable, therefore EVs may provide a more consistent source of miRNA biomarkers than whole plasma [15, 30]. On the other hand, it has been calculated that there is far less than one molecule of a given miRNA per EV [35], which raises the question of whether all EVs contain miRNAs and if the amounts are biologically meaningful. Moreover, it can be argued that the EV isolation step may introduce a higher variation and result in a low RNA yield that in turn would lead to lower sensitivity, higher standard deviations and poor reproducibility of the EV-based miRNA assays as compared to whole plasma assays.

Here, we have performed a systematic comparison of miRNA levels in whole plasma and EVs isolated from the same plasma samples in a well-characterised cohort of PC and BPH patients. Our results show that EV-incorporated miRNA constitutes only a minor fraction of whole plasma miRNA. This is in line with a study by Chevillet et al. showing that exosome fractions contained a small minority of the miRNA content of plasma [35]. Nevertheless, the miRNA levels in EVs and whole plasma were poorly correlated, and the EV-incorporated and whole plasma miRNA profile was clearly different. This finding is consistent with a NGS-based study by Cheng et al. that compared small RNA profiles in EVs, plasma and serum of 3 healthy individuals and showed that the miRNA levels differ remarkably between plasma and serum EVs and between EVs and cell-free plasma and serum [30].

Three out of 9 miRNAs analysed could differentiate between PC and BPH patients in our cohort. MiR-375 showed a better diagnostic performance when tested in whole plasma as compared to EVs. MiR-375 is an oncogenic miRNA that is overexpressed in tumours with high Gleason score and more advanced pathological stage [36]. Increased plasma or serum levels of miR-375 in patients with PC vs BPH or metastatic CRPC vs localised PC have been reported before in several studies (Table 2), and the AUC obtained in our study was similar to that reported before [37]. MiR-375 had one of the lowest EV to whole plasma ratios among the miRNAs analysed in this study and it was undetectable in a significant portion of EV samples. It still may have diagnostic properties in cases where it is detectable, though proving its diagnostic value would require a larger cohort of samples. Two studies have reported the presence of miR-375 in blood EVs from PC patients. Bryant et al. showed that its level is increased in serum EVs from patients with metastatic PC as compared to non-recurring PC [38], and Huang et al. reported that high EV-miR-375 level is associated with a poor prognosis in CRPC [39]. Hence, increased levels of EV-incorporated miR-375 appear to be associated with metastatic disease. As only 3 of the patients in our cohort had a metastatic disease at the time of the blood draw, we reasonably detected it only in a minority of PC patients in our cohort. Moreover, as these studies did not describe treatment of EVs with proteinase K, it is possible that the EV preparations also contained protein-bound miRNAs co-isolated with EVs.

On the contrary, EV-incorporated miR-200c-3p and miR-21-5p showed better diagnostic performance than in whole plasma. Increased plasma or serum levels of miR-200c-3p have been found before in patients with metastatic CRPC as compared to localised PC or healthy controls [40, 41]. Similarly, miR-21-5p has been reported to be increased in plasma or serum of patients with PC as compared to healthy controls and patients with CRPC as compared to localised PC [40, 42, 43]. However, to the best of our knowledge, an association of EV-incorporated miR-200c-3p and miR-21-5p with PC has not been reported before. Hence, our study shows for the first time that EVs provide a better source for testing these miRNAs as PC biomarkers than whole plasma.

The only miRNA biomarker that could differentiate between PC patients with high vs low Gleason score was EV-incorporated Let-7a-5p, whose level was decreased in patients with Gleason score ≥ 8. This is in line with a study by Mihelich et al. showing that serum levels of Let-7a were decreased in PC patients with Gleason 4 + 5 grade tumours as compared with Gleason grade 3 [12]. Our study, though, shows that the whole plasma and EV levels of Let-7a-5p are poorly correlated and that EV-incorporated Let7a-5p level is more informative than Let7a-5p in whole plasma.

The cellular origin of circulating miRNAs is unclear. Although it seems likely that oncogenic miRNAs such as miR-375, miR-200c-3p and miR-21-5p that are overexpressed in PC tissues are released in the bloodstream from the tumour tissues, direct evidence for this is still lacking. On the contrary, Let-7a-5p is a tumour suppressive miRNA that is downregulated in PC tissues as compared to normal or BPH tissues [44, 45]. Hence, the decrease in Let-7a-5p plasma level in patients with aggressive PC is unlikely to be due to the release from cancer tissue. More plausibly, lower expression level or reduced release of Let-7a-5p is genetically associated with PC. Alternatively, it could be possible that signalling molecules produced by cancer cells actively downregulate the expression or release of this miRNA from normal tissues. In fact, a recent study by Chen et al. has demonstrated that breast cancer cells can downregulate the expression and release of miR-486 from cardiac and skeletal muscle in a TNFα-dependent manner [46], thus providing evidence that miRNAs released from non-tumour cells can have a diagnostic significance.

We did not observe diagnostic properties for circulating miR-30c-5p, miR-106a-5p, miR-141-3p, miR-223-3p and miR-210-3p in our patient cohort. The main reasons for this could be a relatively low sample size, the usage of different RNA isolation methods and different sample storage and processing conditions that may affect miRNA abundance and stability, and different normalisation methods for RT-qPCR results. In most of these studies, the results were normalised to the RNA input. Here, we normalised the RT-qPCR data against plasma volume and spike-ins that allow controlling for experimental variation. We reasoned that the quantity of EV-RNA in our samples is by far too small to be reliably measured by the currently available RNA quantification methods (e.g. Nanodrop, Qubit or Agilent Bioanalyzer), therefore the normalisation against RNA input may lead to biased results. Moreover, as EV levels have been found to be increased in cancer patients as compared to healthy controls [47], it seems likely that the levels of EV-enclosed RNA may also be increased, hence normalisation against the RNA input may result in the loss of diagnostically relevant information. Alternative approaches for RT-qPCR data normalisation include normalisation to an individual endogenous reference gene or the geometric mean of a set of normalisers. While the selection of housekeeping genes is relatively straight-forward for miRNA expression analysis in cells or tissues, the most commonly used internal control genes have turned out to be highly variable in biofluids [48–50], therefore, there is currently no consensus on appropriate normalisers in biofluids. Possibly, the most reliable normalisation strategy is a global geometric averaging of multiple genes, however this is applicable only when large panels of miRNAs are analysed. It should also be considered that EV-miRNA and cell-free miRNAs may require different normalization genes.

Furthermore, miR-141-3p and miR-106a-5p had discordant results across publications. Increased miR-106a serum levels were shown to correlate with increased CAPRA scores in one study [51], while another study showed that it is decreased in sera from PC patients with Gleason grades 4 + 5 as compared to grade 3 and BPH [12]. Increased serum or plasma levels of miR-141 have been found in PC patients as compared to healthy controls or BPH [1, 52], and in patients with metastatic CRPC as compared to localised PC [40, 53, 54]. At the same time, other studies reported that miR-141 was detectable in less than 50% of patients or had similar levels in PC patients and healthy controls [43, 54].

To the best of our knowledge, this is the first study providing a head-to-head comparison of diagnostically relevant miRNA detection assays in whole plasma and plasma EVs from cancer patients. We show that only a minor fraction of the total cell-free miRNA could be recovered from the plasma EVs, however the EV-incorporated and whole plasma cell-free miRNA profiles were clearly different. Whole plasma MiR-375 could differentiate between PC and BPH, while miR-200c-3p and miR-21-5p performed better when analysed in EVs. EV-incorporated but not whole plasma Let-7a-5p level could distinguish patients with aggressive and indolent PC. This shows that EVs provide a more consistent source of RNA than whole plasma for the analysis of some miRNA biomarkers, while, possibly due to specific sorting mechanisms, the abundance of other miRNAs in EVs is very low and they show better diagnostic performance in whole plasma.