Revisiting extracellular RNA release, processing, and function

Abstract

It is assumed that RNAs enriched in extracellular samples were selected for release by their parental cells. However, recent descriptions of extracellular RNA (exRNA) biogenesis and their differential stabilities question this assumption, as they could produce identical outcomes. Here, we share our opinion about the importance of considering both selective and nonselective mechanisms for RNA release into the extracellular environment. In doing so, we provide new perspectives on RNA-mediated intercellular communication, including an analogy to communication through social media. We also argue that technical limitations have restricted the study of some of the most abundant exRNAs, both inside and outside extracellular vesicles (EVs). These RNAs may be better positioned to induce a response in recipient cells compared with low abundance microRNAs.

Extracellular RNA dynamics

It has been known for decades that RNA exists outside of cells. However, functionality of extracellular RNAs (exRNAs; see Glossary) was initially elusive until the functional transfer of messenger RNAs (mRNAs) and microRNAs between mammalian cells was reported [1-4]. Additionally, exRNAs are a promising source of disease biomarkers in minimally invasive liquid biopsies [5].

exRNA is thought to exist inside membranous nanoparticles known as extracellular vesicles (EVs), which can be derived from the cell plasma membrane (i.e., ectosomes, microvesicles) or from endosomal compartments (i.e., exosomes) [6]. To date, most studies have focused on the “beginning” and the “end” of the exRNA transfer process. The “beginning” of the process is the packaging and release of exRNA in cell-derived particles. The “end” of said process is uptake into cells, which is a presumed prerequisite for several mechanisms of action of exRNA. However, recent observations on exRNA processing [7,8] prompt us to ask two questions that have not been widely addressed: i) apart from selective sorting into EVs, what other mechanisms are responsible for the release of RNAs into the extracellular space, and, ii) what is happening after release and before uptake to drive exRNA dynamics?

Herein, we argue that the mechanisms of RNA release are intertwined and probably more heterogeneous than previously thought, and that post-release processing contributes importantly to abundance of exRNAs. One-size-fits-all models do not seem to be entirely consistent with experimental evidence. While packaging of specific RNAs into EVs has been widely studied, abundant intracellular RNAs might also be released following concentration gradients or in a passive, vesicle-independent manner due to cell death. Extracellular processing is also a common and diverse set of phenomena, with different exRNAs protected from degradation to vastly different degrees. Consequently, an observed extracellular enrichment of a given RNA does not necessarily mean that selective release has occurred, nor does the lack of selective packaging preclude function.

Sorting RNAs to the extracellular space: how?

To understand the mechanisms responsible for their release into the extracellular space, RNAs can be labeled and tracked by fluorescence microscopy or similar techniques. However, mechanisms identified by cell biology-based approximations are not necessarily representative of the bulk of RNAs present in the extracellular milieu. A widely used population-level approach is to sequence RNAs inside and outside cells and perform differential expression analysis. Following this approach, several groups have reported that extracellular microRNA profiles are rather distinct from their expression in parental cells, suggesting that cells actively select specific microRNAs for export [9-11]. Further, while performing motif-enrichment analysis on microRNAs enriched in or depleted from EVs, several short motifs have been identified that are more prominently present in one environment compared with the other [12]. Motif enrichment strongly suggests that RNA-binding proteins (RBPs) can recognize specific microRNAs and facilitate their release. Indeed, a decade of research has identified several RBPs as potential mediators of selective microRNA export [13,14].

The selective export hypothesis has been influential, especially in the literature of extracellular microRNAs. However, evidence to the contrary also exists. For instance, two independent studies have found that intracellular and extracellular microRNA levels are strongly correlated. Both studies were performed with human MCF-7 cells grown in serum-free medium [15,16]. Remarkably, a later analysis of microRNAs present in EV-depleted fetal bovine serum (FBS) suggested that many of the previously identified EV-enriched microRNAs were probably derived from the FBS in which cells had been cultured [17,18]. The high degree of sequence conservation of microRNAs makes it difficult, if not impossible, to discriminate between FBS-derived microRNAs and their human counterparts [19]. FBS is not the only possible source of contaminating exogenous microRNAs. Recombinant protein additives used in serum-free medium formulations can also include microRNAs, such as miR-122-5p and miR-451a [20]. As a result, if differential abundance analysis of intracellular and extracellular samples is done without considering this bias, it will inevitably and erroneously identify contaminating RNAs as being selectively released from cells (Figure 1).

Figure 1: The impact of contamination on extracellular RNA enrichment analysis.

A) A widely used approach to infer RNA release mechanisms is to sequence the RNA content of cells and extracellular fractions, including extracellular vesicles (EVs) and nonvesicular RNAs or ribonucleoproteins (RNPs). Intracellular RNAs are shown in green, EV-RNAs in blue, and nonvesicular RNAs in black. Contaminating RNAs derived from reagents, labware, or environmental contamination are shown in red. B) Ligation-dependent small RNA sequencing (miRNA-seq) is applied to the hypothetical samples in A and results are exemplified with ten reads. One sequence (represented as a stem-loop structure) is strongly enriched in the two extracellular fractions (center and right panels). However, its enrichment in EVs is artefactual due to contamination (red) and inefficient separation of EVs and RNPs.

Different mechanisms are consistent with extracellular enrichment

Contamination can therefore inflate the list of extracellularly enriched RNAs. Surely, there are other RNAs that will still show genuine extracellular enrichment after accounting for contamination and removing problematic sequences. Can we directly assume that these remaining RNAs constitute cases of selective release?

RBP-mediated RNA selection and encapsulation inside EVs is one possibility, but not the only one. We believe that even in a case where the silencing of an RBP is shown to affect the levels of specific exRNAs, there are alternative explanations. The question is whether RBPs directly transport the RNA to the sites of EV biogenesis or indirectly change the subcellular localization of the RNA and affect its local concentration in neighboring sites (Figure 2A, B). For instance, several of the RBPs that have been suggested to mediate the sorting of specific RNAs into EVs [14,21,22], are also known to regulate RNA subcellular localization [23,24]. Because these RBPs impact multiple aspects of RNA biology, their silencing could thus affect RNA sorting into EVs in an indirect manner.

Figure 2:

Different mechanisms can explain the enrichment of specific RNAs in extracellular samples. (A) Apart from contamination or other technical issues (see Figure 1), the extracellular enrichment of an RNA does not necessarily imply that an RNA binding protein (RBP, red) selects this sequence for release (selective sorting). (B) RNAs could also be released at a rate that is dependent on local concentration (nonselective sorting) and still show extracellular enrichment when compared with cells. This is because intracellular RNA levels are usually estimated from cell lysates. However, cells can be strongly polarized and compartmentalized. An RNA might seem to be enriched in extracellular samples compared with cells because of technical limitations on measuring relative RNA abundance in the vicinity of multivesicular bodies or the plasma membrane. (C) Third, extracellularly enriched RNA fragments might have been generated from precursors in the extracellular space and not directly released as fragments from cells. The source of these nonvesicular RNAs is still unclear. Possible mechanisms include fusion of autophagic vesicles with the plasma membrane, cytoplasmic extrusion, or passive release from damaged or dead cells. For simplicity, figures depict only vesicles derived from budding of the plasma membrane and not endosome-derived exosomes. Images were created with BioRender.com.

There is a third possibility that we feel should be considered before assuming selective RNA release based on extracellular enrichment analysis: post-release processing coupled to differential extracellular stability [7,8] (Figure 2C). However, this is mostly relevant when studying extracellular nonvesicular RNAs and will be discussed later.

Nonselective sorting does not preclude functionality

Having proposed several mechanisms for extracellular RNA enrichment as alternatives to cell-based specific sorting, we now also like to divorce the concept of selective RNA sorting from that of RNA-mediated intercellular communication. To be sure, it seems to be intuitive that evolutionary pressure could support the release of a specific RNA, perhaps to decrease its intracellular levels in the parent cell or to achieve a desired change (beneficial to the organism) in a recipient cell after local or long-range communication. Examples include T cell activation [25] and parasite infection [26], respectively. Evolution could also ensure mass action and stability-based mechanisms. Once the RNA is found in the extracellular space and available for interaction with or uptake by a target cell, the specific mechanism explaining its presence may be of limited importance.

In sum: the mechanisms that cells use to release RNAs into the extracellular space do not necessarily determine exRNA functionality (Box 1). Both selective and nonselectively sorted RNAs could be functionally relevant, but the latter category probably contains those RNAs that are highly expressed both in cells and in EVs [27]. There is also a tendency to assume that only RNAs showing a high fold change in expression between EVs and their parental cells are promising mediators of intercellular communication. Yet, a highly abundant cellular RNA could even be depleted from EVs relative to the parent cell and still constitute one of the most abundant EV RNAs with the potential to induce a potent effect in recipient cells. For example, unbiased analysis by gel electrophoresis and Northern blot [7] or refined sequencing techniques [22] clearly showed that full-length transfer RNAs (tRNAs) are among the most abundant constituents of EVs, far above the levels of tRNA-derived fragments, microRNAs, or other small RNAs. Thus, for every microRNA that a cell imports via EVs, this same cell is probably exposed to many copies of extracellular tRNA molecules. Yet, research on this topic has been scarce compared with studies of relatively low-abundance microRNAs.

Box 1: nonselective RNA release and the interferon analogy.

MicroRNAs are minor constituents of extracellular vesicles (EVs) [48,49], although this affirmation should be checked on a case-by-case basis given the involvment of oncogenes in the regulation of AGO2/microRNA sorting into EVs [50]. By contrast, the number of transfer RNA (tRNA) halves per EV is much higher [15], albeit lower than the levels of full-length tRNAs [7,22]. Additionally, tRNA halves are known to be upregulated in response to stress [51] and they can interact with translation initiation factors [52] and induce translational arrest [53]. We have recently shown that overexpression of tRNA halves by transfection of synthetic RNAs is followed by a proportional increase in their levels inside EVs and in receptor cells exposed to those EVs [27]. Thus, we have hypothesized that tRNA halves encapsulated inside EVs could play a role as signaling molecules capable of amplifying a stress response program beyond the limits of a single cell [54]. This is analogous to the role of interferons in the context of viral infections, as has been previously suggested for Angiogenin, the enzyme responsible for stress-induced tRNA cleavage in mammals [51]. If cells tend to release a statistical sample of their transcriptome in a dynamic and continuous fashion, this could constitute a form of intercellular communication that enables cells to sense gene expression changes in their neighbors (see Outstanding Questions).

Beyond stress, others have asked how transcriptomic changes induced in immune cells upon activation are reflected in the RNA content of their EVs [25,55], finding many RNAs that did not follow a reflective trend between cells and EVs [55]. Thus, there is not really a dichotomy between a strictly selective and a strictly nonselective model of RNA sorting into EVs; both mechanisms seem to coexist [14,26] and both mechanisms can be involved in intercellular communication.

The extracellular space as a biogenetic compartment

There have been reports suggesting RNA processing inside EVs [28,29]. However, EVs seem to confer a privileged microenvironment where exRNAs are protected from the action of extracellular ribonucleases (RNases). In fact, RNase protection assays are usually the methods of choice to demonstrate that exRNAs are present inside and not outside EVs [6,30].

But not all exRNAs are associated with EVs and these nonvesicular exRNAs are highly prone to the action of RNases (Box 2). Indeed, most microRNAs in cell culture media or circulating in human blood plasma are present in the form of ribonucleoprotein (RNP) complexes with proteins such as Argonaute 2 (AGO2) [16,31]. Thus, nonvesicular RNA constitutes a significant fraction, if not the majority, of the extracellular RNAome (exRNAome) [15]. It is thus nothing short of astonishing that nonvesicular exRNAs have attracted so little attention compared with their EV counterparts. However, two recent reports have focused on these nonvesicular RNAs [7,8] and offer innovative perspectives and new tools to understand exRNA metabolism.

Box 2: Survivorship bias defines nonvesicular exRNA profiles.

Survivorship bias is the logic error of assuming that the composition of a system can be inferred from the observation of that system in the present, ignoring the effects of a selection process in the past. It is often explained by the work of the statistician Abraham Wald, who suggested adding extra armor to the areas of US bombers that showed the least damage when the planes returned to the base after a mission. This could sound counterintuitive, but the areas where holes were never observed in returning bombers represented the areas that, if hit, produced critical damage and plane lost.

Our initial characterization of nonvesicular extracellular RNAs (exRNAs) showed most sequencing reads corresponding to glycine and glutamic acid 5’ transfer RNA (tRNA) halves [15]. This was also observed by later studies performed in cell culture [56] and in different biofluids [57]. Based on these observations, one could be tempted to speculate about selective release of nonvesicular tRNA halves into the extracellular space. However, we found that these specific tRNA halves have a self-complementary sequence and can form RNA dimers that are highly stable against single-stranded RNases [32]. Oligomerization capacity seems to be a characteristic of 5’ tRNA halves, as those derived from alanine and cysteine can form highly stable tetramers stabilized by G-quadruplex structures [58].

By recognizing that the most abundant nonvesicular RNAs are also highly stable, we realized that the composition of the extracellular RNAome (exRNAome) could be affected by survivorship bias, especially in the nonvesicular fraction. What was lost along the way? To solve this question, we developed a method called RI-SEC-seq, consisting of separating RNPs by size in extracellular samples treated or not with RNase inhibitors (RI) [7]. Surprisingly, RI-treated samples were highly heterogeneous and contained all major RNA families and RNPs found inside cells, including ribosomes and full-length tRNAs. By modulating extracellular RNase activity, we observed cleavage of these non-coding RNAs (ncRNAs), and the generation of different types of fragments. However, only those fragments with the capacity to form self-protecting dimers accumulated in samples with high RNase activity such as those containing serum. In addition to RNA oligomerization, protein binding is probably also a driver of survivorship bias (Figure 2C). AGO2/microRNA complexes are very stable and they have been observed outside extracellular vesicles (EVs) [16, 39], even in human plasma [31]. It remains to be elucidated whether they are predominantly present in free form or whether they are part of larger aggregates that have been called exomeres [59, 60].

By inhibiting extracellular RNases [7] or mutating the RNase 1 gene by CRISPR/Cas9 [8], these studies have shown that extracellular, nonvesicular samples contain ribosomes [7], full-length tRNAs, and YRNAs [7,8], suggesting that the action of extracellular RNases is responsible for their fragmentation into shorter RNAs. Some of these short non-coding RNA (ncRNA) fragments accumulate as a consequence of their high extracellular stability, and define the nonvesicular RNAome under standard conditions [7]. At least for some of these fragments, their differential extracellular stabilities seem to be a consequence of their capacity to form dimers that are resistant to single-stranded RNases [27,32]. Taken together, these studies provide strong evidence for the extracellular biogenesis and differential stability of nonvesicular ncRNA fragments. These factors can contribute to faulty conclusions of selective RNA sorting, especially in the nonvesicular fraction (Figure 2C).

The dark matter of the extracellular RNAome

Not only must we update our notions of selective packaging as the only explanation for exRNA abundance, but we must also recognize that technical limitations mean that RNA sequencing data do not always reflect the true composition of extracellular samples. Because of known biases of sequencing methods [33], we are essentially fishing in a lake and estimating its biodiversity without realizing that the output is selected by the bait. There is a whole “dark matter” of RNA molecules that are not studied because they are not amenable to the most popular sequencing library preparation methods. When technical variations are introduced (e.g., demethylation, end-healing, avoidance of ligation steps, thermostable reverse transcriptases), the composition of extracellular samples changes as well [8,22,34-36]. Normalization is also not trivial when working with exRNAs, where all the RNAs in the sample are affected by the variable under study (i.e., release into the extracellular space). Thus, a combination of state-of-the-art sequencing approximations and orthogonal “old-school” methods, such as Northern blot and RNA immunoprecipitation, are needed to get a reliable estimate of the RNA and RNP composition of extracellular samples. An additional layer of “dark matter” exists because of the differential stabilities of exRNAs. Indeed, we have shown the presence of full ribosomes in nonvesicular fractions, and these are invisible unless RNase inhibitors are added to the media [7]. Extracellular ribosomes are probably not the only elephants left in the (extracellular) room.

Nonvesicular exRNAs: a message from the dead?

By inhibiting extracellular RNA degradation, we have shown that nonvesicular exRNA profiles closely mirror those of intracellular lysates [7]. Although different reported mechanisms could explain this pattern [37-39], it is reasonable to assume that cell death is also a contributor to the exRNAome.

We argue that the main reason why cell death has not been seriously considered as a relevant source of exRNAs is because highly abundant species such as full-length ribosomal RNAs (rRNAs) and tRNAs have not been widely detected in extracellular preparations. However, it is now evident that ribosomes and tRNAs are initially present in the extracellular milieu and extracellular RNases are responsible for their rapid clearance, destroying the initially positive correlation between intracellular and (nonvesicular) exRNA profiles [7,8].

Interestingly, the presence of extracellular ribosomes has also been suggested in biofluids, such as blood serum [35], although it has still not been directly demonstrated. We think it is possible that the contribution of cell death to exRNA profiles is not simply an artifact of cell culture but rather a relevant process occurring in vivo, especially in the hypoxic inner mass of solid tumors or in cells exposed to lytic viruses. If this were the case, it would make sense for immune cells to perform exRNA surveillance. Pattern recognition receptors are important constituents of the cell’s antiviral machinery but are also thought to play a role in the recognition of self-RNAs that could act as damage-associated molecular patterns (DAMPs) [40]. In support of this hypothesis, when we added chromatographic fractions containing extracellular ribosomes to dendritic cells, we obtained preliminary data suggesting that exRNAs could play a role in sterile inflammation [7].

Beyond their potential use as biomarkers, what are the biological functions of nonvesicular exRNAs? One possibility is that RNA release serves to maintain steady state levels of RNAs under homeostatic conditions (in a balance between synthesis, degradation, excretion, and reuptake). Additionally, rapid clearance of specific RNAs could serve as a switch in a cellular gene expression program upon stimuli [25]. RNA degradation in the extracellular space could avoid uncontrolled RNase activation in the cytoplasm, which can be cytotoxic [41] and fuel cellular energetics by the uptake of degradation products, as shown for proteins [42]. Additionally, nonvesicular exRNAs could play a role in intercellular communication. In this case, extracellular RNases could act by erasing certain signals and generating others. RNase 1, the enzyme responsible for the conversion of extracellular tRNAs into extracellular tRNA halves [8] is actively released by endothelial cells into the bloodstream [43]. Could it play a role as a shaper of the exRNAome, preventing acute inflammation induced by exRNAs acting as DAMPs? What about other biofluids where RNase activities are expected to be lower than in serum and plasma? Note that mutations in RNase A family members such as Angiogenin (RNase 5) are causally linked to neuroinflammatory diseases [44]. EVs are thought to mediate intercellular RNA transfer, but some specific tRNA halves are also incorporated spontaneously by cells when present in the extracellular milieu [45]. What about extracellular ribosomes and mRNAs [7]? Could they also have a noncanonical message to convey? Could this message be a last communication from the dead?

Concluding Remarks

Our main motivation for this work is to draw attention to nonvesicular exRNAs, which constitute a significant fraction of the exRNAome despite receiving far less attention than EV-associated RNAs (see Outstanding Questions). At the moment, we can only speculate about the possible roles of nonvesicular exRNAs in intercellular communication, but preliminary results suggest that extracellular ribosomes are probably not silent from an immunological perspective [7]. The problem with studying nonvesicular exRNAs is that they are subjected to strong survivorship bias (Box 2) and their study thus requires the development of specific analytical techniques that account for this effect.

OUTSTANDING QUESTIONS BOX.

1. What is the relative contribution of selective and nonselective (i.e., concentration-driven) RNA release into different types of extracellular vesicles (EVs)?

2. In the cases were silencing of an RNA-binding protein (RBP) affects the extracellular abundance of specific RNAs, is this effect direct or indirect?

3. Are subtle changes in the intracellular transcriptome (e.g., during cell activation or the stress response) reflected in extracellular RNA (exRNA) carriers, such as EVs? Are other cells able to sense these changes? Are RNAs capable of communicating cell metabolic status to other cells in real time?

4. Is cell death the main source of nonvesicular extracellular transfer RNAs (tRNAs) and ribosomes? If not, what is the mechanism for their release?

5. What is the physiological relevance of extracellular nonvesicular YRNAs, tRNAs, and ribosomes? In case they can exert function on recipient cells, is that function mediated by full-length RNAs or their extracellular fragmentation products?

6. Apart from encapsulation in EVs or lipoproteins, why are certain non-coding RNA (ncRNA) fragments highly stable in extracellular samples? What is the relative contribution of post-transcriptional modifications, protein binding, and intermolecular RNA interactions? Could these stable nonvesicular exRNAs be used as disease biomarkers?

It is still not clear whether cell death or unknown active release mechanisms explain the majority of these nonvesicular exRNAs, but, although more study is needed to address this question, RNA release mechanisms and exRNA functionality are not conceptually linked. An RNA could play an important role in intercellular communication regardless of its selective or nonselective release from healthy cells or even passive release from damaged or dying cells. Indeed, it is becoming evident that the autophagy machinery, functionally linked to degradation and recycling, is deeply involved in EV biogenesis and in nucleic acid release into the extracellular space [39,46,47]. Are exRNAs cellular junk or one of the ways in which cells communicate? These possibilities are not mutually exclusive, according to the logics of recycling. Cells can convey a message even if they do not have the “intention” to do so. As an analogy on the societal level, modern humans, as consumers of social media, are continuously communicating inner thoughts and feelings to companies specialized in big data analysis, even though these pieces of information are released with a different purpose entirely. When we think about RNA-mediated intercellular communication, we tend to think about the transfer of a message from one cell to another that has been prepared by selective packaging and transport. Perhaps we should think, instead, in terms of “social media-based communication”: cells, with intention or not, are releasing a fraction of their dynamic transcriptome into a space where even trash can be valuable information.

HIGHLIGHTS.

• A significant proportion of extracellular RNAs (exRNAs) are not associated with extracellular vesicles (EVs)

• Extracellular small RNAs can be generated in the extracellular space from longer precursors like transfer RNAs (tRNAs) and ribosomes

• Extracellular RNA biogenesis coupled to differential stability destroys positive correlations between intracellular and extracellular RNA profiles

• Contamination with cell culture additives is also a source of RNAs that appear to be enriched in extracellular samples

• Abundant RNA species, such as full-length tRNAs, are also abundant inside EVs

GLOSSARY

Argonaute 2 (AGO2)

one of the effector proteins that directly associates with microRNAs and is capable of mediating degradation or the translational silencing of specific messenger RNAs (mRNAs)

Autophagy

a complex molecular process used by cells to recycle organelles, consisting in the formation of a lipid membrane surrounding the organelle and directing its delivery into the lysosome for degradation

Damage-associated molecular patterns (DAMPs)

molecules from the host, usually hidden from the immune system in viable and healthy cells, capable of signaling via pattern recognition receptors when they are released into the extracellular space

Exomeres

a type of extracellular particle recently described, smaller in size than the so-called exosomes or small extracellular vesicles; their exact composition and function is still under investigation

Extracellular RNAs (exRNAs)

RNA molecules detected in extracellular samples, including cell-conditioned media and biofluids (blood plasma or serum, urine, saliva, etc.)

Interferons

group of proteins with intercellular signaling potential released by cells into the extracellular space upon viral infection or other activating stimuli

microRNAs

a group of small RNAs of about 22 nucleotides which regulate gene expression by sequence-dependent recognition of mRNAs

Northern blot

a technique used to detect RNAs of specific sequence and length by performing electrophoresis, transfer of size-separated RNAs to a membrane, and hybridization with a labelled probe

Pattern recognition receptors

innate immunity receptors, found in different types of cells, specialized in the recognition of the molecular hallmarks of invading pathogens (e.g., lipopolysaccharides) and the triggering of a cellular response; they are also involved in the recognition of DAMPs

Ribonucleases (RNases)

enzymes specialized in RNA degradation

Ribonucleoprotein (RNP)

a complex between a RNA-binding protein (RBP) and its cognate RNA

RNAome

a way of referring to the universe of RNA species that are present in a particular simple

tRNA-derived fragments

a class of small regulatory RNAs generated by endonucleolytic cleavage of mature tRNAs, usually in response to stress

YRNAs

a class of noncoding RNAs, ranging 83 – 112 nucleotides, with still unclear molecular functions but linked to DNA replication and formation of RNPs; they tend to be frequently detected in exRNA studies