Small Extracellular Vesicle Enrichment of a Retrotransposon-Derived Double-Stranded RNA: A Means to Avoid Autoinflammation?

Small extracellular vesicles (SEVs) such as exosomes are released by multiple cell types. Originally believed to be a mechanism for selectively removing unwanted cellular components, SEVs have received increased attention in recent years for their ability to mediate intercellular communication. Apart from proteins and lipids, SEVs contain RNAs, but how RNAs are selectively loaded into SEVs remains poorly understood. To address this question, we profiled SEV RNAs from mouse dendritic cells using RNA-Seq and identified a long noncoding RNA of retroviral origin, VL30, which is highly enriched (>200-fold) in SEVs compared to parental cells. Bioinformatic analysis revealed that exosome-enriched isoforms of VL30 RNA contain a repetitive 26-nucleotide motif. This repeated motif is itself efficiently incorporated into SEVs, suggesting the likelihood that it directly promotes SEV loading. RNA folding analyses indicate that the motif is likely to form a long double-stranded RNA hairpin and, consistent with this, its overexpression was associated with induction of a potent type I interferon response. Taken together, we propose that preferential loading into SEVs of the VL30 RNA containing this immunostimulatory motif enables cells to remove a potentially toxic RNA and avoid autoinflammation. In this way, the original notion of SEVs as a cellular garbage bin should not be entirely discounted.


Introduction
Small extracellular vesicles (SEVs) such as exosomes are extracellular, membranebound vesicles that originate from the multivesicular bodies of the cell's endocytic compartment. Originally considered a mechanism by which cells excrete unwanted materials [1], SEVs have since been recognized for their ability to mediate intercellular communication and influence the fate of recipient cells via a selective cargo of proteins, lipids and RNAs. Consistent with this, SEVs have received substantial attention in recent years as a novel means of delivering therapeutic payloads into the body for the treatment of various diseases [2,3].
Using SEVs to deliver therapeutic RNAs is seen as a particularly promising strategy [4,5]. However, it remains unclear how best to load SEVs with particular RNAs of Biomedicines 2021, 9, 1136 2 of 14 interest. In this regard, knowing the mechanisms by which endogenous RNAs are naturally shuttled into SEVs would be beneficial, but to date few studies have examined this topic. Koppers-Lalic et al. previously reported that miRNAs are not randomly incorporated into exosomes but that the addition of bases at the 3 end of a miRNA influences its sorting into exosomes [6]. Specifically, they showed that miRNAs rich in uridines (Us) at the 3 end were enriched in exosomes compared to cells, and conversely that miRNAs rich in adenines (As) at the 3 end are more likely to be sorted into cells. Villaroya-Beltri et al. on the other hand reported that certain motifs govern the sorting of miRNAs into exosomes [7]. They found that the protein hnRNPA2B1 binds specific sequences in miRNAs and facilitates their loading into exosomes [8], while other RNA-binding proteins have also been implicated in the sorting of miRNAs into SEVs (reviewed in [9]). A small number of studies have examined longer RNAs. One of these claimed that the presence of a "molecular zipcode" in the 3' UTR of mRNAs serves as a "docking site" for miR-1289 and results in an enrichment of the mRNAs in the multivesicular body [10]. However, this enrichment was only modest (two-fold) and the authors did not examine whether this zipcode sequence led to enrichment within exosomes themselves. Another study identified three motifs bioinformatically (ACCAGCCU, CAGUGAGC and UAAUCCCA) that are enriched in exosomal mRNAs and long noncoding RNAs (lncRNA) [11]. These motifs exhibit double-stranded stem-loop structures and were subsequently reported to be recognised by the exosomal proteins YB-1 and NSUN2 [12].
To better understand RNA loading into SEVs, we assessed the mRNA and long noncoding RNA content of dendritic cell (DC) SEVs, which are already being explored therapeutically in clinical trials [13]. Using next generation sequencing, we found that one of the most highly enriched SEV transcripts was VL30, an endogenous RNA derived from a retroviral element. The abundance of VL30 within SEVs was several hundred times higher than in the parental DCs, a result that was subsequently confirmed in multiple other cell types. Bioinformatic analysis revealed that SEV-enriched isoforms of VL30 RNA contain a repetitive motif whose secondary structure is strongly predicted to form an extended double stranded RNA (dsRNA) hairpin. Expression of this motif alone resulted in its being preferentially loaded into SEVs and induced a strong type I interferon response and cell death. Taken together, we speculate that the preferential SEV loading of VL30 RNAs containing this dsRNA motif is a means by which cells remove a potentially toxic retroviral RNA and avoid damage.

Mice and Ethics
Mice were bred and maintained in the animal facilities at the Walter and Eliza Hall Institute of Medical Research (WEHI) according to national and institutional guidelines for animal care. All experimental procedures were approved by the WEHI Animal Ethics Committee (project number 2014.019, 5 September 2014).

The VL30 lncRNA Is Enriched in DC SEVs
To first isolate SEVs, we performed ultrafiltration and differential ultracentrifugation of culture supernatant from mouse BMDCs, which are a rich source of SEVs [24]. The resultant extracellular vesicles appeared round and membrane-bound via electron microscopy ( Figure 1a) and their size ranged from 50-200 nanometers (mean:~120 nm) (Figure 1b), in keeping with the expected appearance and size of SEVs [25]. Western blotting was also consistent with these vesicles being SEVs, as indicated by the presence of classical SEV markers such as transferrin receptor, flotillin-1 and TSG101 and the absence of markers for the Golgi (GM130), endoplasmic reticulum (Calnexin) and mitochondria (VDAC-1/Bcl-2) (Figure 1c). showing the average log expression of each gene (horizontal axis) plotted against the log-fold change in gene expression between cells and SEVs (vertical axis), based on RNA sequencing of DCs and their SEVs (n = 6 each). Each dot in the graph represents one RNA. Transcripts that were differentially expressed between cells and SEVs were identified (adjusted p-value < 0.05; see Methods) and are shown either in magenta (upregulated in cells) or blue (upregulated in SEVs). The RNA that showed the greatest enrichment in SEVs (~200-fold) was VL30 (big blue dot).
To identify RNAs enriched in SEVs, we next isolated total RNA from these SEVs and their parental DCs, prepared cDNA libraries, and performed RNA-Seq. A total of 7968 genes were identified as differentially expressed when comparing the cell and SEV transcriptomes ( Figure 1d). Of the 3496 genes upregulated in SEVs, >90% showed only a mild increase in relative abundance (log2 fold change <2). Even among the top 20 RNAs that were most enriched in SEVs, there was one RNA known as VL30 (gene symbol:  To identify RNAs enriched in SEVs, we next isolated total RNA from these SEVs and their parental DCs, prepared cDNA libraries, and performed RNA-Seq. A total of 7968 genes were identified as differentially expressed when comparing the cell and SEV transcriptomes (Figure 1d). Of the 3496 genes upregulated in SEVs, >90% showed only a mild increase in relative abundance (log 2 fold change <2). Even among the top 20 RNAs that were most enriched in SEVs, there was one RNA known as VL30 (gene symbol: A130040M12Rik) that stood out based not only on its strong enrichment (~200-fold) but also its overall abundance (Figure 1d, Supplementary Table S1).

The VL30 lncRNA Is Enriched in SEVs from Multiple Cell Types
VL30 is a long noncoding RNA that is derived from a mouse-specific endogenous retrovirus and functions as a transcriptional regulator in steroidogenesis and oncogenesis [26,27]. To explore whether VL30 is also enriched in SEVs from other cell types, we next cultured a variety of different mouse cell lines, including immortalized DCs (DC2.4), T cells (EL4), B cells (WEHI-231) and fibroblasts (NIH/3T3), and isolated SEVs from each of these lines. VL30 abundance in SEVs was then compared to that of each parental cell line by qRT-PCR, using β-actin for normalization purposes because of its high abundance and consistent average expression in both cells (~8400 CPM) and SEVs (~6300 CPM) as observed in our original RNA-Seq data. Consistent with our data from primary DCs, VL30 was significantly enriched in SEVs from each of the tested cell lines, with an abundance in SEVs ranging from 500-to 300,000-fold higher than the parental cells ( Figure 2). A130040M12Rik) that stood out based not only on its strong enrichment (~200-fold) but also its overall abundance (Figure 1d, Supplementary Table S1).

The VL30 lncRNA Is Enriched in SEVs from Multiple Cell Types
VL30 is a long noncoding RNA that is derived from a mouse-specific endogenous retrovirus and functions as a transcriptional regulator in steroidogenesis and oncogenesis [26,27]. To explore whether VL30 is also enriched in SEVs from other cell types, we next cultured a variety of different mouse cell lines, including immortalized DCs (DC2.4), T cells (EL4), B cells (WEHI-231) and fibroblasts (NIH/3T3), and isolated SEVs from each of these lines. VL30 abundance in SEVs was then compared to that of each parental cell line by qRT-PCR, using β-actin for normalization purposes because of its high abundance and consistent average expression in both cells (~8400 CPM) and SEVs (~6300 CPM) as observed in our original RNA-Seq data. Consistent with our data from primary DCs, VL30 was significantly enriched in SEVs from each of the tested cell lines, with an abundance in SEVs ranging from 500-to 300,000-fold higher than the parental cells ( Figure 2).

VL30 RNA Isoforms Enriched in SEVs Contain a Repeated Sequence Motif
Consistent with its retrotransposon origin, the VL30 gene has multiple copies (>400) throughout the mouse genome [28]. Over time, these sequences have diverged considerably. To understand the possible sequence requirements for the VL30 RNA to be efficiently incorporated into SEVs, multiple VL30 isoforms from across the genome were

VL30 RNA Isoforms Enriched in SEVs Contain a Repeated Sequence Motif
Consistent with its retrotransposon origin, the VL30 gene has multiple copies (>400) throughout the mouse genome [28]. Over time, these sequences have diverged considerably. To understand the possible sequence requirements for the VL30 RNA to be efficiently incorporated into SEVs, multiple VL30 isoforms from across the genome were evaluated based on the number of counts present in the SEV and cell-based libraries within our original RNA-Seq data. Specifically, the ten most SEV-enriched VL30 isoforms were selected and their sequences assessed using the Multiple Expectation maximizations for Motif Elicitation (MEME) tool, which enables motif discovery among related sequences. This revealed a 26 nucleotide motif (Figure 3a) that exists in tandem repeats within SEV-enriched VL30 isoforms (Figure 3b) but is absent from the ten VL30 isoforms that showed the least SEV enrichment.
Biomedicines 2021, 9, x FOR PEER REVIEW 8 of 14 evaluated based on the number of counts present in the SEV and cell-based libraries within our original RNA-Seq data. Specifically, the ten most SEV-enriched VL30 isoforms were selected and their sequences assessed using the Multiple Expectation maximizations for Motif Elicitation (MEME) tool, which enables motif discovery among related sequences. This revealed a 26 nucleotide motif (Figure 3a) that exists in tandem repeats within SEV-enriched VL30 isoforms (Figure 3b) but is absent from the ten VL30 isoforms that showed the least SEV enrichment.

A VL30 Sequence Containing the Repeated Motif Alone Is Efficiently Incorporated into SEVs
To test whether this repetitive motif was important for packaging into SEVs, a fulllength cDNA clone of VL30 (C730003K16) containing nine tandem copies of the motif was obtained, and a truncated, "motif-only" construct representing ∼20% of the full-length sequence and containing the repetitive motif alone was generated (Figure 4a). The full length and motif-only constructs were then separately cloned into a doxycycline-inducible lentiviral vector and transduced into the SH-SY5Y human neuroblastoma cells, which are a rich source of SEVs and lack VL30 expression.  Figure 3. SEV-enriched VL30 isoforms contains a tandemly-repeated motif. (a) Using MEME, a 26 nucleotide motif was identified within the ten VL30 isoforms that showed the greatest SEV enrichment. (b) The location of this motif (red box) is shown within each of the ten VL30 isoforms that showed the greatest SEV enrichment. The chromosomal location of the 5'-end for each isoform is shown (mm10).

A VL30 Sequence Containing the Repeated Motif Alone Is Efficiently Incorporated into SEVs
To test whether this repetitive motif was important for packaging into SEVs, a fulllength cDNA clone of VL30 (C730003K16) containing nine tandem copies of the motif was obtained, and a truncated, "motif-only" construct representing~20% of the full-length sequence and containing the repetitive motif alone was generated (Figure 4a). The full length and motif-only constructs were then separately cloned into a doxycycline-inducible lentiviral vector and transduced into the SH-SY5Y human neuroblastoma cells, which are a rich source of SEVs and lack VL30 expression.
To first test whether expression of the full length construct was associated with enrichment of VL30 in SH-SY5Y SEVs, SEVs from SH-SY5Y cells were isolated and the relative abundance of VL30 RNA in cells and SEVs was compared by qRT-PCR (Figure 4b).
In the absence of doxycycline, VL30 was readily detected within cells (presumably due to 'leakiness' of the doxycycline-inducible promoter as commonly occurs) and enriched >5000-fold in SEVs. A similar SEV enrichment (~20,000 fold) was observed following doxycycline treatment, which as expected increased overall VL30 levels within the cells themselves. Together, these results suggested that the full length VL30 construct could be successfully overexpressed and efficiently packaged into SEVs. To first test whether expression of the full length construct was associated with enrichment of VL30 in SH-SY5Y SEVs, SEVs from SH-SY5Y cells were isolated and the relative abundance of VL30 RNA in cells and SEVs was compared by qRT-PCR (Figure 4b).
In the absence of doxycycline, VL30 was readily detected within cells (presumably due to 'leakiness' of the doxycycline-inducible promoter as commonly occurs) and enriched >5000-fold in SEVs. A similar SEV enrichment (~20,000 fold) was observed following doxycycline treatment, which as expected increased overall VL30 levels within the cells themselves. Together, these results suggested that the full length VL30 construct could be successfully overexpressed and efficiently packaged into SEVs.
We next turned our attention to the motif-only construct. Here, we observed that, similar to the full length construct, VL30 containing the repeated motif alone was detected in cells even in the absence of doxycycline and highly enriched in SEVs (~600-fold) ( Figure  4c), consistent with the tandem repeat of the motif being itself sufficient to promote SEVs loading. However, when doxycycline was added to induce overexpression of the motifonly construct, widespread cell death was unexpectedly observed, which was not the case for the full length construct (Supplementary Figure S1).
To investigate this further, we examined the likely secondary structure of our motifonly VL30 RNA using Mfold, and found that this RNA is strongly predicted to form a long dsRNA hairpin (Figure 5a). Given that dsRNA is a potent pathogen associated We next turned our attention to the motif-only construct. Here, we observed that, similar to the full length construct, VL30 containing the repeated motif alone was detected in cells even in the absence of doxycycline and highly enriched in SEVs (~600-fold) (Figure 4c), consistent with the tandem repeat of the motif being itself sufficient to promote SEVs loading. However, when doxycycline was added to induce overexpression of the motifonly construct, widespread cell death was unexpectedly observed, which was not the case for the full length construct (Supplementary Figure S1).
To investigate this further, we examined the likely secondary structure of our motifonly VL30 RNA using Mfold, and found that this RNA is strongly predicted to form a long dsRNA hairpin (Figure 5a). Given that dsRNA is a potent pathogen associated molecular pattern that induces a type I IFN response and cell death, we performed qRT-PCR for IFN-β as well as several common interferon-stimulated genes (ISGs), including IFIT1, IRF7, MDA5 and RIG-I 48 h after doxycycline treatment. While we could not reliably detect any IFN-β, which is often produced at very low levels and is notoriously difficult to detect, each of the ISGs showed strong up-regulation upon doxycycline induction of the motif-only VL30 RNA but not the full-length construct (Figure 5b). molecular pattern that induces a type I IFN response and cell death, we performed qRT-PCR for IFN-β as well as several common interferon-stimulated genes (ISGs), including IFIT1, IRF7, MDA5 and RIG-I 48 h after doxycycline treatment. While we could not reliably detect any IFN-β, which is often produced at very low levels and is notoriously difficult to detect, each of the ISGs showed strong up-regulation upon doxycycline induction of the motif-only VL30 RNA but not the full-length construct (Figure 5b).

Discussion
Since the identification of RNA in SEVs [29] and the discovery that SEVs can facilitate the transfer of RNAs into recipient cells [30,31], there has been growing interest in the use of SEVs as a 'natural delivery system' for therapeutic RNAs. To date, however, there have been very few studies examining the loading requirements for RNAs to be selectively packaged into SEVs. The original motivation for this study was therefore to better understand why certain RNAs are loaded into SEVs. Our subsequent identification of VL30 as an RNA that is highly enriched in SEVs (up to several thousand-fold) provided an opportunity to examine the features of this RNA that promote SEV loading. In this regard, we observed that highly enriched isoforms of VL30 contained multiple copies of a 26 nucleotide motif, and we then demonstrated that a small fragment of the VL30 RNA containing tandem repeats of this motif was sufficient for strong SEV enrichment.
In theory, our identification of an RNA sequence that promotes SEV loading could assist efforts to selective package therapeutic RNAs into SEVs. However, overexpression of the repeated motif led to induction of a strong type I IFN response, consistent with its dsRNA structure. If the motif were added to the sequence of a therapeutic RNA as a means of promoting SEV loading, it would therefore be important to avoid this innate immune response; otherwise the health of the parental cells and their SEV production would be compromised, as we observed. In this regard, it is interesting to note that overexpression of the full length VL30 sequence-despite it containing the same repeated motif in its entirety-did not induce a type I IFN response. Why this should be the case is unclear, but one possibility is that the tertiary folding of the full length VL30 RNA either prevents the repeated motif's dsRNA structure from forming in the first place or else hides it internally so as to prevent recognition by innate immune receptors such as MDA5, RIG-I and TLR3 that recognize dsRNA. Whether adding a therapeutic RNA to the repeated motif enables selective loading of the RNA into SEVs and/or helps to similarly avoid a type I IFN response remains to be seen.
Our observation that VL30 RNA was enriched in SEVs from a wide variety of cell types as well as from different species (mouse and human) suggests that the features of VL30 that promote its selective packaging into SEVs utilize a cellular mechanism that is widespread and evolutionarily conserved. What that mechanism might be is something for future study, but we can speculate as to the features of the VL30 RNA that promote its SEV loading. Firstly, by conducting a BLAST search of the 26 nucleotide motif against the NCBI nucleotide database (and excluding inevitable hits to VL30 itself within the mouse genome), we found that the motif matched the viral packaging signal (Psi) that is contained within various retroviral/retrotransposon-based vectors. This signal, originally derived from the VL30 retrotransposon, is believed to efficiently direct the packaging of recombinant RNAs, such as those from the reporter gene lacZ, into virions [32,33]. At first glance, the observation that the same signal is involved in virion and SEV packaging might seem surprising, but it would be entirely consistent with the "Trojan exosome hypothesis" which proposes that retroviruses have come to exploit our bodies' exosome biogenesis pathways for the purposes of producing retroviral particles [34,35]. Secondly, the predicted dsRNA structure of the repeated motif is in keeping with an earlier report from Botagov and colleagues that RNAs enriched in SEVs contain dsRNA hairpin structures [11]. It is also consistent with previous observations that human immunodeficiency virus (HIV) transactivating response (TAR) RNA, which also contains a dsRNA structure [36], is highly enriched in SEVs [37]. Taken together with our own findings, these previous studies therefore suggest the dsRNA might itself be a signal for selective SEV loading.
If dsRNA is indeed a signal for selective SEV loading, a question that arises is what purpose this might serve. During viral infection, the extracellular transfer of viral dsRNA from infected cells has been proposed as a means of activating the innate immune response within noninfected bystander cells, thus augmenting antiviral immunity in the face of the various immunosuppressive mechanisms that viruses employ within infected cells [38]. Consistent with this, SEVs from HIV-infected cells contain TAR RNA that activates TLR-3 and stimulates proinflammatory cytokine production [39]. Similarly, cells infected with hepatitis virus C produce SEVs that transfer viral RNAs to recipient cells and trigger the production of type I IFN [30], although in this case it was unclear whether the immunostimulatory viral RNAs were double-stranded. Nevertheless, the selective loading of dsRNA into SEVs represent a feasible strategy for infected cells to augment antiviral immunity. But what about if there was no viral infection? Cells produce an abundance of endogenous dsRNAs, and there are a variety of mechanisms to ensure that these do not cause unwanted autoinflammatory responses [40]. In this regard, it is tempting to speculate that the selective packaging of dsRNA into SEVs might provide an additional mechanism for cells to remove this material and therefore avoid autoinflammation. Such a role would be in keeping with the abundance of SEVs that are continually excreted in urine and other bodily fluids and would hark back the original notion of SEVs as a cellular waste bin [1].
Our study is not without several important limitations. Firstly, we are aware that it would have been helpful to further define the minimum VL30 sequence that facilitates SEV loading. In this regard, it is notable that previous work with the VL30 Psi sequence in relation to retroviral packaging suggests that a Psi sub-sequence of as little as 61 nucleotides is sufficient for promoting RNA encapsidation (albeit with somewhat reduced efficiency) [32]. It would therefore be interesting to determine if the same minimal sequence can facilitate SEV loading in the future. At the same time, removing the motif sequence from VL30 and testing whether VL30 is still enriched within SEVs would provide further confirmation that the motif is required for SEV loading. Secondly, another limitation of the present study is that we did not test whether our putative VL30 SEV loading sequence was able to facilitate the loading of a reporter RNA. This would have allowed us to provide direct proof-of-concept that therapeutic RNAs can be more efficiently packaged into SEVs through the addition of a VL30 sequence. Finally, looking ahead, it would be important to identify the cellular components that facilitate VL30 RNA loading into SEVs. In this regard, RNA pull-down assays using tagged VL30 RNA to isolate and identify the proteins that interact with VL30 would be informative and ultimately facilitate a proper understanding of why and how VL30 is so efficiently loaded into SEVs.

Conclusions
In this study, we observed that the VL30 RNA is highly enriched in SEVs from multiple cell types, and identified a tandemly-repeated motif that appears to help promote the selective loading of VL30 into SEVs. Further study into whether this repetitive motif can help promote loading of therapeutic RNAs into SEVs is warranted.
Supplementary Materials: The following are available online at https://www.mdpi.com/article/10 .3390/biomedicines9091136/s1, Figure S1: Overexpression of the motif-only construct in SH-SY5Y cells results in widespread cell death, Table S1

Institutional Review Board Statement:
The study was conducted according to the according to relevant national and institutional guidelines for animal care, and all experimental procedures were approved by the relevant animal ethics committee at the WEHI.

Informed Consent Statement: Not applicable.
Data Availability Statement: The data presented in this study are available on request from the corresponding author.