Profiling of Extracellular Small RNAs Highlights a Strong Bias towards Non-Vesicular Secretion

The extracellular environment consists of a plethora of molecules, including extracellular miRNA that can be secreted in association with extracellular vesicles (EVs) or soluble protein complexes (non-EVs). Yet, interest in therapeutic short RNA carriers lies mainly in EVs, the vehicles conveying the great majority of the biological activity. Here, by overexpressing miRNA and shRNA sequences in parent cells and using size exclusion liquid chromatography (SEC) to separate the secretome into EV and non-EV fractions, we saw that >98% of overexpressed miRNA was secreted within the non-EV fraction. Furthermore, small RNA sequencing studies of native miRNA transcripts revealed that although the abundance of miRNAs in EVs, non-EVs and parent cells correlated well (R2 = 0.69–0.87), quantitatively an outstanding 96.2–99.9% of total miRNA was secreted in the non-EV fraction. Nevertheless, though EVs contained only a fraction of secreted miRNAs, these molecules were stable at 37 °C in a serum-containing environment, indicating that if sufficient miRNA loading is achieved, EVs can remain delivery-competent for a prolonged period of time. This study suggests that the passive endogenous EV loading strategy might be a relatively wasteful way of loading miRNA to EVs, and active miRNA loading approaches are needed for developing advanced EV miRNA therapies in the future.


Introduction
The constant drive to find novel carriers of RNA therapeutics has witnessed a shift from artificial delivery systems to nature-inspired biomolecule carriers, such as extracellular vesicles (EVs). Owing to their pivotal role in cell-to-cell communication as well as the inherent ability to deliver bioactive lipids, proteins and nucleic acid, EVs hold a spotlight position in the research of advanced therapeutic carriers.
EVs contain a wide range of nucleic acid molecules, and though stretches of up to several kilobases in size have been detected [1], the majority of the encapsulated material does not exceed 200 nucleotides [2,3]. Therefore, most studies focus on loading EVs with therapeutic miRNAs and other short miRNA-like RNAs, such as siRNAs and shRNAs [4][5][6][7][8].
In order to load EVs with RNA, exogenous loading as well as endogenous overexpression of therapeutic genes has been utilized, the latter often being preferred [9][10][11]. Nevertheless, despite significant effort, the enrichment of EVs with the nucleic acid of interest still remains challenging.
It is evident from previous studies that most of the extracellular nucleic acids are not associated with vesicular structures, being either bound to protein complexes or lipoprotein particles [12][13][14][15][16]. Yet, as biological effects of EV-depleted specimens are minimal [17][18][19], it is generally accepted that practically all the biological activity is rather conveyed by EVs than other circulating nucleic acid carriers. Early studies on nucleic acid transfer via EVs were published more than a decade ago [20][21][22]. These have now been complemented with numerous reports describing the transfer of EV-RNA, which, for example, modulates metabolic- [23,24] and immune responses [25], combats cancer [4,6,7,26,27], ameliorates liver diseases [28,29] and even acts across the blood-brain barrier [5,8].
The presence of a distinct miRNA per EV has been estimated to be approximately 1 in 100, or even 1 in 10,000 [30][31][32], being largely insufficient to ensure solid therapeutic delivery [33]. This underscores the urgency to understand miRNA (and more broadly short RNA) accumulation in EVs and non-EVs and could provide important background information on which strategies to explore for enriching the desired short RNAs to EVs. Different loading strategies have already been employed [10], but frequently without considering the context of global miRNA secretion.
Here, we analyze the relative abundance of a selection of miRNA molecules in the EV and non-EV secretory fractions at their basal level as well as after endogenous overexpression. These results are viewed in the global context of basal miRNA secretion to provide extended knowledge on extracellular sorting of short RNA sequences.

EV Isolation for Overexpression and Validation
For miRNA overexpression studies, after 48 h of media conditioning, the EVs were isolated by using the Amicon Ultra-15 100 kDa molecular weight cut-off spin-filters (Millipore, Darmstadt, Germany), discriminating the retentate and flow-through/filtrate. Otherwise, the retentate was subjected to ultrafiltration SEC fractionation using the S-400 as described in [36] (fractionation of overexpressed miRNA secretome) or fractionated by Tricorn 10/300 columns (Sigma-Aldrich, Saint-Louis, MO, USA) packed in-house with Sepharose Fast Flow 4 resin (GE Healthcare, Pittsburgh, PA, USA) (miRNA validation). In the case of miRNA overexpression, the A280 and RNA quantification measurements (Quant-iT Ribo-GreenRNA Assay Kit; Thermo Fisher Scientific, Waltham, MA, USA) across the elution profile were used to set the location of fractions 1 to 5 (each spanning 10 mL) in a way that the cross-talk between the set ranges would be minimal, yet would cover the major changes in the protein and RNA profiles. For miRNA validation, the 'non-EV' fraction contained the non-vesicular fractions from the SEC as well as the flow-through from the 100 kDa ultrafiltration step.

miRNA Quantification by qRT-PCR
For miRNA quantification, an appropriate amount of Trizol LS (Thermo Fisher Scientific, Waltham, MA, USA) was used to prepare the samples for subsequent RNA extraction according to the manufacturer's protocol. For miRNA validations, to account for the variation in RNA extraction efficiencies, 3 µL of a 5 µM synthetic miRNA oligonucleotide, cel-miR-39 (5 -UCACCGGGUGUAAAUCAGCUUG) (Integrated DNA Technologies, Leuven, Belgium), was added to each sample at the phenol extraction stage. In all cases, cDNA synthesis was performed by using the TaqMan MicroRNA Reverse Transcription Kit and the respective hairpin primers from TaqMan MicroRNA Assays (let-7a-5p assay ID:000377 and let-7b-5p assay ID:000378, miR-186-5p assay ID:002285; miR-210-3p assay ID: 000512; miR-16-5p assay ID:000391; miR-21-5p assay ID:000397 and miR-100-5p assay ID:000437; all from Thermo Fisher Scientific, Waltham, MA, USA). The quantification was performed by using the TaqMan Gene Expression Master Mix together with the respective TaqMan MicroRNA Assays on a Step-One Real-Time PCR instrument (Thermo Fisher Scientific, Waltham, MA, USA). shRNA detection was performed by synthesizing cDNA as above, but using custom-designed hairpin RT primers and amplified using KiCqStart SYBR Green qPCR Ready Mix (Sigma Aldrich, Saint-Louis, MO, USA) on a Step-One Real-Time PCR instrument, as above. The PCR efficiency of each reaction was calculated with the LinRegPCR program (developed by Dr. Jan Ruijter, Heart Failure Research Center, The Netherlands), while the Ct values were obtained from StepOne Software (Applied Biosystems, Waltham, MA, USA). In the case of overexpression studies, an equal amount of input RNA was used for reverse transcription. For miRNA validations, the total miRNA quantity was normalized to cel-miR-39 levels and further corrected for the volume of the respective fractions (Sepharose Fast Flow 4 fractionation volumes of 3-3.5 mL for EVs and 131-133 mL for non-EV (non-EV fraction + flow-through of 100 kDa ultrafiltration units)). For EV miRNA stability studies, EVs were isolated and purified as above, mixed with FBS at a final serum concentration of 10% and incubated at 37 • C for indicated times. At the end of each time point, Trizol LS was added to samples, and the RNA was extracted and analyzed by qPCR, as above.

EV Isolation for Small RNA Sequencing
Neuro2a, hTERT-MSCs and HEK293T cell-lines were maintained in high-glucose (4.5 g/L) DMEM GlutaMax (Thermo Fisher Scientific, Waltham, MA, USA; HEK293T) or RPMI medium 1640 (Thermo Fisher Scientific, Waltham, MA, USA; hTERT-MSCs) supplemented with 10% fetal bovine serum (FBS, Thermo Fisher Scientific, Waltham, MA, USA). All cells were cultured in a humidified atmosphere at 37 • C and 5% CO 2 . A total of 6 × 10 6 cells were first seeded onto 15 cm cell culture dishes. After 24 h, cells were washed with 0.01 M phosphate-buffered saline (PBS) to ensure removal of any residual FBS before a media change to OptiMEM (Thermo Fisher Scientific, Waltham, MA, USA) supplemented with 1% Antibiotic Antimycotic Solution (Sigma-Aldrich, Saint-Louis, MO, USA). The supernatant was collected after 48 h, and EVs were harvested by ultrafiltration SEC fractionation as described in [36]. To obtain the non-vesicular pool, flow-through from the 100 kDa ultrafiltration step was combined with the non-vesicular fractions from the SEC fractionation and thereafter concentrated on the 10 kDa molecular weight cut-off (MWCO) Amicon spin filters (Merck Millipore, Burlington, MA, USA), the latter of which was also used to concentrate the vesicular fractions. In some control experiments, when specifically indicated, 100 kDa MWCO spin filters were used instead. Immediately after concentrating, all samples were mixed with Trizol LS (Thermo Fisher Scientific, Waltham, MA, USA) and stored at −80 • C until RNA extraction for small RNA sequencing analysis. The abovementioned workflow was used to generate three biological replicates from each cell line.

EV Characterization
Nanoparticle tracking analysis (NTA) was performed by using the NanoSight NS500 nanoparticle analyzer (Malvern Instruments, Malvern, UK) to measure the size distribution and concentration of the purified EVs. The EV samples were diluted in PBS, and the movement of the particles was recorded within five 30 s videos that were subsequently analyzed using the NTA Software 2.3 (Malvern Instruments, Malvern, UK) at detection threshold 5, camera gain 350 and shutter setting 800. Transmission Electron Microscopy (TEM) was performed as described in [37].

Small RNA Sequencing and Data Analysis
Library preparation (from 60ng ± 10% of total RNA),small RNA sequencing and data analysis regarding mapping and annotation were performed as outlined in [38]. For data visualization (principal component analysis, correlation scatterplots, heatmaps), raw read counts were transformed in R-studio software (R version 3.4.2) [39] by using the approach of variance stabilizing transformation (vsd; blinded = F). Differential expression analyses of miRNAs were performed by using the R package DESeq2 [40], including mature miRNAs Cells 2021, 10, 1543 5 of 17 exceeding a raw read count cut-off across the studied samples (RowMeans > 1 within the studied cell line). Heatmaps of differentially expressed miRNAs (vsd counts, local zero-centered normalization) were generated by using the Multiple Experiment Viewer program (Version 4.9.0) [41]. The miRNAs were hierarchically clustered by using the average linkage clustering method, Pearson distance metric and optimized gene leaf order.

Statistical Analysis
The statistics for RNA stability experiments were performed using simple linear regression analysis using GraphPad Prism Version 6 (GraphPad Software, Inc., La Jolla, CA, USA). The same software (Version 7) was used for statistical fold enrichment analyses with two-way ANOVA, Tukey post-hoc correction. The statistics from the differential expression analysis of miRNAs were generated by the R software package DESeq2 [40].

Results
It is known that cells constantly release miRNA into the extracellular space. A fraction of secreted miRNA is encapsulated in EVs, yet according to previous reports [12,13], a large amount can be detected outside vesicular structures. To be able to separate the secretome into vesicular (EVs) and non-vesicular fractions (non-EVs), this study has employed sizeexclusion chromatography (SEC) (Figure 1a). We have previously validated SEC workflow for its suitability to separate EVs from the non-vesicular secreted material and found that the total yield of the entire EV isolation workflow was high (>70%) [36]. Furthermore, the isolated EVs had high purity as EV protein markers were detected only in the SEC eluate, corresponding to the void volume peak (i.e., the EV peak) and not in fractions that eluted later (i.e., the non-EV peak) [36].
In the present study, SEC-purified EVs were characterized by Western blot, Nanoparticle Tracking Analysis (NTA) and transmission electron microscopy (TEM) (Supplementary Material Figure S1). Further analysis of eluted SEC fractions confirmed that high molecular weight particles, as detected by NTA, eluted all in the void volume peak (Supplementary Material Figure S2A,B). Also, as expected, most of the protein and RNA were found to elute in the non-EV area of the chromatogram, with the latter being susceptible to RNAseA treatment (Supplementary Material Figure S2C,D). Additional control experiments with 100 kDa MWCO ultrafiltration spin filters showed that all detectable particles remained in the retentate while most of the miRNA was found in the filtrate (Supplementary Material Figure S2E,F), corroborating with the RNA elution profile of the SEC chromatogram. Due to the widely recognized concerns on bovine serum-derived miRNA contamination (even when EV-depleted serum is used) [42,43], the conditioned medium was prepared using serum-free Opti-MEM, being suitable for EV production as shown previously [44,45].

Pri-miRNA and U6-shRNA Overexpression Leads to RNA Secretion in the Non-EV Fraction
Many studies focusing on developing EVs as therapeutic RNA carriers aim to enrich miRNA in EVs by overexpressing the respective transcripts in EV-producing cells. Yet, little is known about miRNA secretion patterns when using such a strategy. To investigate to what extent miRNAs of interest get released to the extracellular space, we transiently overexpressed pri-let-7a and pri-let-7b in HEK cells (yielding a 7-10× and 5-7× increase in expression level, respectively (data not shown)) and quantified the respective mature miRNA levels in the EV and non-EV portion of the secretome. This analysis revealed that the majority of overexpressed let-7a and let-7b mature transcripts were secreted in the non-EV fraction, whereas the EV fraction contained <2% of the respective miRNAs ( Figure 1b). This suggests that while miRNA overexpression in source cells does lead to its increased level in EVs, the majority of extracellular transcripts are predominantly found in the non-EV fraction. To clarify, throughout the study, the non-EV secretory fraction was obtained by combining the non-EV fraction from SEC with the 100 kDa ultrafiltration flow-through, allowing us to look at the global non-EV material.
SEC workflow for its suitability to separate EVs from the non-vesicular secreted material and found that the total yield of the entire EV isolation workflow was high (>70%) [36]. Furthermore, the isolated EVs had high purity as EV protein markers were detected only in the SEC eluate, corresponding to the void volume peak (i.e., the EV peak) and not in fractions that eluted later (i.e., the non-EV peak) [36].  For CM fractionation, it was first cleared from cell debris and large particles using two centrifugation steps (5 min at 300× g and 10 min at 1200× g) followed by 0.22 µm sterile filtration. The pre-cleared CM was concentrated by ultrafiltration on 100 kDa MWCO filters, followed by retentate fractionation into EV and Non-EV fractions using SEC with Sephacryl S-400 resin. Finally, to obtain the global non-EV pool used in the study, the Non-EV pool from SEC was combined with the flow-through from the ultrafiltration step. (b,c) miRNA sequences were transiently overexpressed in cells using pri-miRNA or U6-shRNA expression plasmids, followed by miRNA quantification using RT-qPCR. Strikingly >98% of overexpressed miRNAs were found to be secreted to the non-EV fraction (relative RNA abundance in % over the total is shown). Nevertheless, despite that only a minor fraction of miRNA was found in EVs, these miRNAs showed high serum stability for up to 24 h (p = 0.66, simple linear regression analysis by GraphPad Prism Version 6) (d). N = 3 for all experiments.
The pri-miRNA expression vectors used above were driven by the CMV promoter, which is a Pol II promoter and requires full processing of the overexpressed pri-miRNA transcripts by respective miRNA processing machinery (DROSHA, DICER1, etc.). However, in many cases, when aiming to load EVs with small RNA, the transcription is initiated using the U6 (Pol III) promoter. This promoter facilitates highly efficient RNA transcription with well-defined transcription start and termination sites, being extensively used for expressing short hairpin RNA (shRNA) for siRNA-like gene silencing purposes. Having this in mind, we also overexpressed a shRNA transcript (shRNA targeted against Firefly luciferase, shLuc) in HEK cells to measure its effect on shRNA secretion in EVs. Interestingly, even though the U6-promoter-driven shRNA expression level was significantly higher than the miRNA expression level using the CMV-driven pri-miRNA expression vectors (data not shown), we observed that the shRNA secretion profile remained unchanged. Namely, <2% of all secreted shLuc was found in the EVs, where similarly to naturally secreted miR-21 transcripts, it remained stable in a serum-containing environment for 24 h (Figure 1c,d). This indicates that even though only a minority of miRNA is secreted with EVs, it remains Cells 2021, 10, 1543 7 of 17 stable over a long period of time and may thus be biologically active when delivered to recipient cells.

Small RNA Secretome Is Predominantly Composed of miRNA and piRNA-Like Sequences
After observing that the overexpressed miRNAs are predominantly secreted via a non-vesicular route, we were curious to investigate whether the same trend applies to miRNAs at their basal expression level. Hence, we sequenced RNA from cells, SEC-derived EV and non-EV fractions from HEK and MSC cell lines and focused on the analysis of reads between 17 and 35 nucleotides in length. This restriction was applied in order to not only cover full-length miRNAs but also to include other small RNAs (sRNA) and check for the presence of larger RNA fragments. The latter have been found in EVs by the authors and others even when focusing on short reads (Figure 2a) [38,46,47].
unchanged. Namely, <2% of all secreted shLuc was found in the EVs, where similarly to naturally secreted miR-21 transcripts, it remained stable in a serum-containing environment for 24 h (Figure 1c,d). This indicates that even though only a minority of miRNA is secreted with EVs, it remains stable over a long period of time and may thus be biologically active when delivered to recipient cells.

Small RNA Secretome Is Predominantly Composed of miRNA and piRNA-Like Sequences
After observing that the overexpressed miRNAs are predominantly secreted via a non-vesicular route, we were curious to investigate whether the same trend applies to miRNAs at their basal expression level. Hence, we sequenced RNA from cells, SEC-derived EV and non-EV fractions from HEK and MSC cell lines and focused on the analysis of reads between 17 and 35 nucleotides in length. This restriction was applied in order to not only cover full-length miRNAs but also to include other small RNAs (sRNA) and check for the presence of larger RNA fragments. The latter have been found in EVs by the authors and others even when focusing on short reads (Figure 2a) [38,46,47]. Principal component analysis (PCA) showed a clear cluster separation between cell, EV and non-EV samples, indicating significant differences in their RNA composition (Figure 2b). Upon detailed analysis of the sRNA secretome, we found that miRNA and Principal component analysis (PCA) showed a clear cluster separation between cell, EV and non-EV samples, indicating significant differences in their RNA composition (Figure 2b). Upon detailed analysis of the sRNA secretome, we found that miRNA and piRNA-like sequences together constituted the majority of sRNA biotype reads (Figure 3a,b) and, depending on the secretory fraction and cell type, covered between 88-98% of sRNA sequences. When comparing with piRNA and miRNA levels in parent cells, piRNA reads were enriched in both EV and non-EV samples, whereas miRNAs were not (Supplementary Material Figure S3A,B). Due to sequence similarities and conflicting data in piRNA databases, the sequences which aligned to piRNA loci were also checked for the presence of piRNA features (such as read length distribution and the presence of 5 uracil) and potential overlap with tRNA annotations (Supplementary Material Figure S3C-H). The highest level of tRNA-overlapping sequences was found in non-EV samples, yet even when the potentially tRNA-derived sequences were removed from the analysis, the trend of EVs being enriched in piRNA sequences (compared to their parent cells) remained largely unchanged (Supplementary Material Figure S3G,H). and potential overlap with tRNA annotations (Supplementary Material Figure S3C-H). The highest level of tRNA-overlapping sequences was found in non-EV samples, yet even when the potentially tRNA-derived sequences were removed from the analysis, the trend of EVs being enriched in piRNA sequences (compared to their parent cells) remained largely unchanged (Supplementary Material Figure S3G,H).
Overall, it must be noted that the proportion of sRNA biotype sequences in the RNA secretome was relatively low (<20% of total reads within the 17-35 nt read length range). The majority of the annotated sequences were 'rRNA' and 'otherRNA' fragments (Supplementary Material Figure S4), the latter mainly covering fragments of tRNA and protein-coding sequences (Supplementary Material Figure S5). Secreted miscellaneous RNA in both cell types consisted primarily of Y RNA sequences. Yet, small cytoplasmic RNAs and vault RNAs, which were prominently observed in EVs, remained practically undetected in the non-EV secretome (Supplementary Material Figure S5).  Overall, it must be noted that the proportion of sRNA biotype sequences in the RNA secretome was relatively low (<20% of total reads within the 17-35 nt read length range). The majority of the annotated sequences were 'rRNA' and 'otherRNA' fragments (Supplementary Material Figure S4), the latter mainly covering fragments of tRNA and protein-coding sequences (Supplementary Material Figure S5). Secreted miscellaneous RNA in both cell types consisted primarily of Y RNA sequences. Yet, small cytoplasmic RNAs and vault RNAs, which were prominently observed in EVs, remained practically undetected in the non-EV secretome (Supplementary Material Figure S5).

The Bulk of Secretory miRNAs Reflect the Content of the Source Cells
After observing that miRNA and piRNA were the most abundant sRNA species both in EVs and non-EVs, we next decided to analyze whether there are any differences in miRNA secretion between the different fractions. We observed that nearly all miRNA reads fell in the range of 20-24 nucleotides in length and represented mature miRNA sequences (Figure 3c). Interestingly, the number of distinct miRNAs detected in the HEK secretome was lower than in HEK cells, whereas in MSC samples, the miRNA numbers were relatively similar in cells, EVs as well as non-EVs (Figure 3d).
The analysis of expression levels of individual miRNAs revealed a good correlation between the EV and non-EV samples (Pearson R 2 = 0.78 and R 2 = 0.84 for HEK and MSC samples, respectively; Figure 3e,f). When comparing each of the secretory fractions to cells, miRNA expression in both EVs and non-EV fractions correlated highly to parent MSCs (R 2 = 0.86 and R 2 = 0.87, respectively). The same was seen for the HEK non-EV sample (R 2 = 0.78), yet miRNAs in HEK EVs showed a somewhat lower correspondence to their source cells (R 2 = 0.69). Nevertheless, these generally high correlation coefficients indicate that, for most miRNAs, a significant contributor to their secretory level is their expression in the cells, suggesting an average stochastic secretion mechanism with evidence on selective sorting in the case of some miRNAs.

Enriched Secretory miRNAs Have Low Expression Levels in Cells
Even though miRNA expression was highly correlated between cells, EVs and non-EVs, there were several individual miRNA sequences that seemed to be enriched in different fractions of the secretome. To confirm that, we performed differential expression (DE) analysis of mature miRNA levels in both cell types by comparing miRNA expression between EVs and non-EVs as well as against their background level in their source cells. As a general observation, miRNAs that were enriched in EVs and in the non-EV fraction displayed low cellular expression levels (Figure 4a,b). In HEK-derived samples, a total of 460 distinct miRNAs were detected, out of which, as compared to their source cells, 38 and 31 miRNAs were differentially expressed in EV and non-EV samples, respectively (Figure 4a; Supplementary Material Table S1). Out of 516 mature miRNAs that were included in the analysis for MSC-derived samples, we detected 78 and 83 DE miRNAs in EV and non-EV samples, respectively, as compared to source cells (p adj < 0.05, log2 fold change ≥ 1.5; Figure 4b; Supplementary Material Table S2). As exemplified by the heatmaps and hierarchical clustering analysis, miRNAs with decreased secretion levels clustered together for both EV versus cells and non-EV versus cells comparisons (Figure 4c-f, bright blue branches). This suggests that certain high-expressing miRNAs are likely to be retained in cells via an unknown mechanism.
proportion of total secreted miRNAs was found to be associated with EVs. More specifically, only 0.1-0.4% and 0.9-3.8% of miRNA sequences were detected in EVs of HEK-and MSC-derived conditioned medium, respectively (Figure 5a,b). Thus, a striking 96.2-99.9% of extracellular miRNAs were secreted in association with non-EV carriers, underpinning the need for active strategies to enforce vesicular miRNA loading.   . (a,b) miRNA relative enrichment analysis revealed that those miRNAs that were enriched in EVs and Non-EVs had a low average read count in cells. This is true in both HEK as well as MSC samples. (c-f) Heat maps of hierarchical clustering analysis further suggested that certain miRNAs are restricted from secretion, as miRNAs enriched in cells were reduced in both the EV and Non-EV fractions. Also, the enrichment of miRNAs in EVs was generally not mirrored in the Non-EV fraction. The connecting lines in c-f denote pairwise sample comparisons.

The Bulk of Extracellular miRNAs Are Secreted via Non-EV Route
From the sequencing results, we could conclude that the enrichment of short RNA sequences at their basal level into EVs can mainly be seen for lowly expressed sequences. To validate these findings with an alternative methodology, we focused on miRNAs with a wide basal expression range (and hence potentially different secretion enrichment characteristics). We selected five miRNAs from our data set (miR-100, miR-210, miR-16, miR-21 and miR-186) and quantified their secretion level by RT-qPCR. As compared to source cells, the level of miR-100 and miR-21 was higher in the extracellular space for HEK cells and lower for MSCs (Supplementary Material Tables S1 and S2), whereas no significant changes were detected in secreted miR-210, miR-16 and miR-186 (Supplementary Material Tables S1  and S2). Nevertheless, regardless of the basal expression and secretory enrichment level of tested miRNAs, similarly to overexpression experiments, only a very small proportion of total secreted miRNAs was found to be associated with EVs. More specifically, only 0.1-0.4% and 0.9-3.8% of miRNA sequences were detected in EVs of HEK-and MSCderived conditioned medium, respectively (Figure 5a,b). Thus, a striking 96.2-99.9% of extracellular miRNAs were secreted in association with non-EV carriers, underpinning the need for active strategies to enforce vesicular miRNA loading.

Discussion
The extracellular space consists of multiple types of nucleic acid carriers, in EVs, which are one of the most actively studied fractions of the secretome. EVs many different types of RNA and, as widely demonstrated, transfer it in a bio active form [48][49][50]. Owing to the RNA transfer capacity, EVs are actively deve carriers of biotherapeutic RNA molecules, such as miRNAs and shRNAs. Since th abundance of specific miRNA molecules in EVs is low and only a small fraction o contain a given miRNA sequence [30][31][32], a lot of effort is directed towards enrich with therapeutic miRNA and shRNA sequences, often by overexpressing them in ducer cells. Owing to studies often using ultracentrifugation (UC) for EV isolat known that certain RNA stretches are rather secreted independently of EVs [14,15 must be noted that incomplete EV pelleting in centrifugation-based approaches p the interpretation of those results, indicating the need to employ more refined m ogies for dissecting the secretome [51,52].
The study at hand employed a well-established secretome fractionation SEC, in order to explore the relative distribution of short RNA stretches in EVs a EV secretory fractions. Our preliminary observations in the RNA overexpression showed that <2% of analyzed miRNAs were secreted via EVs, while the great ma overexpressed miRNA ended up in the non-EV fraction, corroborating with a rece showing similar findings from virus-infected cells [53]. Moreover, the low EV amount was observed irrespective of the overexpression type, using either CM promoter and pri-miRNAs/shRNAs for loading the EVs, as often used for the RNA loading into EVs [35,[54][55][56][57]. This indicates that the passive overexpression

Discussion
The extracellular space consists of multiple types of nucleic acid carriers, including EVs, which are one of the most actively studied fractions of the secretome. EVs contain many different types of RNA and, as widely demonstrated, transfer it in a biologically active form [48][49][50]. Owing to the RNA transfer capacity, EVs are actively developed as carriers of biotherapeutic RNA molecules, such as miRNAs and shRNAs. Since the overall abundance of specific miRNA molecules in EVs is low and only a small fraction of all EVs contain a given miRNA sequence [30][31][32], a lot of effort is directed towards enriching EVs with therapeutic miRNA and shRNA sequences, often by overexpressing them in EV producer cells. Owing to studies often using ultracentrifugation (UC) for EV isolation, it is known that certain RNA stretches are rather secreted independently of EVs [14,15]. Yet, it must be noted that incomplete EV pelleting in centrifugation-based approaches perplexes the interpretation of those results, indicating the need to employ more refined methodologies for dissecting the secretome [51,52].
The study at hand employed a well-established secretome fractionation strategy, SEC, in order to explore the relative distribution of short RNA stretches in EVs and non-EV secretory fractions. Our preliminary observations in the RNA overexpression context showed that <2% of analyzed miRNAs were secreted via EVs, while the great majority of overexpressed miRNA ended up in the non-EV fraction, corroborating with a recent study showing similar findings from virus-infected cells [53]. Moreover, the low EV-miRNA amount was observed irrespective of the overexpression type, using either CMV or U6 promoter and pri-miRNAs/shRNAs for loading the EVs, as often used for therapeutic RNA loading into EVs [35,[54][55][56][57]. This indicates that the passive overexpression strategy could be a rather wasteful way for loading EVs with RNA cargo molecules. Also, similarly to our previous findings [16], the overexpressed EV miRNA was fully processed and stable in a serum-containing environment for up to 24 h. Yet, despite this greatly exceeding the half-life of EVs in vivo [58][59][60][61], the low miRNA level is still unlikely to have satisfactory therapeutic potential.
Owing to the insufficient systematic knowledge on how short RNA sequences are secreted in the global context on the basal level, it is difficult to design robust universal strategies for miRNA enrichment in EVs. Though a number of small RNA sequencing studies have been performed to define the exact nucleic acid composition of EVs, the majority thus far exploit UC-based EV enrichment protocols [32,[62][63][64]. Further complexity in studying EV-RNA arises from the low overall abundance of specific miRNA molecules in EVs [30][31][32]53] and the knowledge that sRNAs cover only the minority of reads, even when concentrating on studying short sequences [38,46]. Another aspect in such global analysis is the presence of bovine serum-derived miRNAs, which can contaminate both EV and non-EV samples when EV-depleted serum is used [42,43,65]. Though this can successfully be overcome by using Opti-MEM-based serum-free media, it must be considered that altering the culturing conditions might change the transcriptome of cells and therefore also affect the contents of EVs, as seen previously [44,66]. To minimize the potential risks of changes in EV content and contaminating RNA originating from EV-depleted serum [42,43], the study at hand conducted parallel comparisons (i.e., the cell, EV and non-EV samples were collected from within the same experiment) and used Opti-MEM-based serum-free media for EV production, respectively.
To follow up on the aforementioned overexpression work, we continued to use SEC, separated the secretome of HEK cells and MSCs into the EV and non-EV fractions and aimed to characterize the global sRNA sorting at the basal expression level. The preliminary clustering analysis showed a clear separation between cellular, EV and non-EV samples. While this suggested global differences, the content of EVs and non-EVs appeared to be more similar to each other than to their source cells. A closer look at different short RNA biotypes revealed that the similarity might be attributed to piRNA sequences, of which the cells were relatively depleted, as well as to miRNAs, being highly prevalent in cells and less abundant in the secretory fractions. Nevertheless, the miRNA expression both in EVs and non-EVs correlated highly to that of their parent cells, suggesting that in general, the most important factor determining the level of secreted miRNA is its expression level in the source cells. This does not, however mean that miRNA sequences cannot be enriched in EVs, as certain proteins mediating miRNA sorting into EVs, such as hnRNP-Q, hnRNPA2B1 and major vault protein, have been suggested [67][68][69]. Yet, according to a recent report, "classical" EVs are devoid of these RNA-binding proteins [70], with parallel studies suggesting Lupus La protein involvement in miRNA sorting and the simultaneous presence of selective as well as non-selective miRNA sorting mechanisms, possibly depending on the origin of EVs [71]. Similarly, to those observations, compared to source cells, we also noticed a relative enrichment of certain miRNA sequences in EVs as well as in the non-EV fraction. Interestingly, these sequences were expressed in HEK cells and MSCs at a low level, whereas the secretion of highly-expressed miRNAs was generally not enriched. Hence, in accordance with recent reports, the data at hand suggests an overall average stochastic secretion mechanism with evidence on selective sorting in the case of some miRNAs.
Importantly, we must underline that even if a specific sequence was enriched in EVs versus non-EVs in relative terms, the total abundance of miRNAs in the non-EV fraction, in fact, greatly exceeded that of EVs. Though similar evidence has been reported previously by others, often by using UC for EV isolation [14,15], we were now able to confirm these observations on SEC-purified secretome fractions by using RT-qPCR. In the cell culture supernatants of HEK cells, we found less than 2% of miRNAs in EVs, whereas over 98% of miRNA was found in the non-EV fraction. MSC-derived samples contained slightly more miRNAs in EVs; however, over 96% of miRNAs were still secreted in the non-EVs fraction. This could, on the one hand, reflect the inherent differences in cell line-dependent packaging mechanisms, or else hint to mechanisms modulating the preservation of basal therapeutic potential of MSC EV miRNAs [72,73]. Moreover, we saw that the basal expression and secretory enrichment level of miRNAs are irrelevant to their overall accumulation in EVs, as the secretion is nonetheless heavily inclined towards the non-EV pool. Considering our aforementioned observation that even miRNA/shRNA overexpression is unable to reroute the miRNAs to the EVs, transient overexpression strategy seems to be uneconomic for loading EVs with RNA cargo. Though sequences with low basal expression levels ought to give better results in terms of expression level increase, and even if significant enrichment is achieved, the bulk material will still not end up in EVs. Overall this pinpoints that without active EV-targeted loading strategies, the level of a given miRNA in the vesicles will most probably not reach a therapeutically relevant scale.
Supplementary Materials: The following are available online at https://www.mdpi.com/article/ 10.3390/cells10061543/s1, Figure S1: EV characterization. Figure S2: Further validation of the SEC workflow for separating EV and non-EV secreted material using Neuro2a cells after overexpression of let-7a miRNA. Figure S3: sRNA composition, characteristics and enrichment in secretory fractions. Figure S4: Read alignment, annotation and length statistics across the studies samples. Figure S5: The prevalence of 'other RNA' biotypes. Figure S6: Length distribution histograms for reads aligning to miRNAs. Figure S7: Correlation plots of all mature miRNAs. Figure S8: Heatmaps of differentially expressed miRNAs detected between the EV and Non-EV fraction of the MSC (A) and HEK (B) secretome. Table S1: Differentially expressed miRNAs in EVs vs. Cells. Table S2: Differentially expressed miRNAs in Non-EVs vs. Cells. Table S3: Differentially expressed miRNAs in EV vs. Non-EV samples.