What goes up must come down: off switches for regulatory RNAs

In this review, McJunkin and Gottesman discuss the mechanisms of RNA degradation and turnover that control the stability and function of small regulatory RNAs (sRNAs in bacteria and miRNAs in eukaryotes). These include RNA-directed mechanisms like RNA decoys and sponges, ribonucleases, and target-mediated miRNA degradation, which together orchestrate post-transcriptional gene expression regulation.

Abstract

Small RNAs base pair with and regulate mRNA translation and stability. For both bacterial small regulatory RNAs and eukaryotic microRNAs, association with partner proteins is critical for the stability and function of the regulatory RNAs. We review the mechanisms for degradation of these RNAs: displacement of the regulatory RNA from its protein partner (in bacteria) or destruction of the protein and its associated microRNAs (in eukaryotes). These mechanisms can allow specific destruction of a regulatory RNA via pairing with a decay trigger RNA or function as global off switches by disrupting the stability or function of the protein partner.

The last 25 years have led to an appreciation of the critical roles of noncoding regulatory RNAs in gene regulation and development in bacteria and eukaryotes. While many of the steps of biosynthesis and presentation of these regulators are very different, microRNAs (miRNAs) in eukaryotes and the most abundant class of small regulatory RNAs (sRNAs) in bacteria such as Escherichia coli carry out parallel roles, regulating the translation and stability of mRNAs by pairing with target mRNAs. Here, we discuss an essential step for any regulator, including these regulatory RNA molecules: How and when does the cell rid itself of them when they are no longer needed? How do studies across kingdoms highlight what we know and do not yet know? This aspect of RNA regulation has become an active area of research in recent years.

RNAs in complex with proteins can be very stable

Free RNAs in the cell are rapidly degraded by endonucleases and exonucleases, with mRNAs having half-lives on the order of minutes in bacteria such as E. coli or on the order of hours in mammalian cells; in both cases, this is significantly more rapid than the doubling time of the cell (Mohanty and Kushner 2016; Herzog et al. 2017; Eisen et al. 2020; Jenniches et al. 2024). However, RNAs in complex with proteins—from the ribosome to the chaperones and protein partners associated with regulatory RNAs discussed here—can be extremely stable. As a result, much of what is understood about turnover reflects what occurs when the RNAs either fail to associate or lose their association with their protein partners.

Given the stability of regulatory RNAs when in complex with their protein partners, much of the analysis here outlining how these regulatory RNAs are degraded focuses on how they are displaced from the proteins, a common feature for both prokaryotes and eukaryotes. Chaperone protein binding appears to be reversible for many prokaryotic sRNAs, disrupted by pairing of the sRNAs with mRNA targets and/or specific “decoy” RNA regulators. For eukaryotic miRNAs to be destabilized, degradation of the protein partners is often a necessary step, with wholesale degradation found in some specific situations, and miRNA-specific decay triggered by interaction with specialized target RNAs found in others.

Bacterial sRNA turnover

Bacteria such as E. coli and Salmonella, which are the focus of discussion here, grow in a variety of environments and have complex mechanisms for rapidly and reversibly adjusting gene expression. sRNAs, usually 50–200 nt in length, can be very rapidly synthesized, do not require translation to be active, and have been found to play significant roles in the adaptation of bacteria to stress. We first describe our current understanding of how these sRNAs work and then discuss how they are destroyed when appropriate.

Bacterial sRNAs: brief outline of synthesis/function

We focus here on the large class of sRNAs that use the Hfq chaperone to promote sRNA–mRNA pairing, which have been well studied in E. coli and Salmonella (for reviews, see Updegrove et al. 2016; Hor et al. 2020). These sRNAs act by pairing with mRNAs, leading to a variety of outcomes, depending on characteristics of the mRNA target. For regulation to occur, the mRNA must also be able to bind to Hfq, and the sRNA pairing region needs to be accessible (Beisel et al. 2012). Negative regulation of mRNA translation results from the sRNA or Hfq blocking ribosome entry and/or recruiting ribonucleases for rapid mRNA degradation. Positive regulation usually reflects sRNA-dependent restructuring of the target mRNA to free ribosome entry sites. The same sRNA can mediate all of these effects, with the differences more in the target mRNA and its pairing region rather than dependent on the sRNA (for example, see Feng et al. 2015).

The sRNAs either are made as freestanding transcripts or are processed from mRNAs, most commonly from the 3′ untranslated regions (UTRs); they generally terminate with a factor-independent terminator, a stem–loop, and a run of U residues at the 3′ end, important for binding Hfq (Morita et al. 2015, 2017). Transcription of the sRNAs, whether from their own promoters or from the mRNAs from which they are processed, is generally tightly regulated, with promoters activated in response to stress or metabolic signals, resulting in expression of the sRNA only when it is needed. Hfq, a homohexamer of the Lsm/Sm family, has three faces capable of binding RNAs, with the proximal face critical for binding the stem–loop and run of Us that serve as a factor-independent transcription terminator (Fig. 1A). Binding stabilizes the sRNAs, as measured by following the fate of the full-length RNA on Northern blots after treatment of the cell with rifampicin to stop new transcription. Although mRNAs in Salmonella or E. coli generally have very short half-lives when measured in this fashion (Jenniches et al. 2024), the majority of studied sRNAs have half-lives of >30 min in most studies (Schu et al. 2015). Hfq is not essential in E. coli, and sRNAs are rapidly degraded in cells deleted for Hfq (Massé et al. 2003) or in cells carrying an Hfq mutant in the proximal binding site that binds the sRNAs (Zhang et al. 2013; Schu et al. 2015). There is also evidence that overexpressing some sRNAs by displacing others from Hfq can lead to degradation of the less tightly bound sRNAs (Papenfort et al. 2009; Moon and Gottesman 2011). Although the RNases responsible for degradation have only been examined in a few cases, the endonuclease RNase E has been shown to play the primary role in decay of many sRNAs in the absence of Hfq (Massé et al. 2003). RNase E cuts single-stranded RNA with a preference for AU sites (Chao et al. 2017) and is found as part of a larger degradosome complex associated with helicases, nucleases, metabolic enzymes, and Hfq (for reviews, see Carpousis 2007; Mackie 2013). However, the ability of RNase E to degrade many sRNAs in cells in which Hfq has been inactivated or deleted (Massé et al. 2003) demonstrates that, at least in these mutants, Hfq is not necessary to deliver the sRNAs for RNase E degradation.

Figure 1.

Function and turnover of bacterial Hfq-binding sRNAs. (A) The Hfq homohexamer chaperones bacterial sRNA function. The structure of Hfq shown is based on data from Sauter et al. (2003). Hfq has three known RNA binding faces with different binding preferences, as labeled here (for review, see Vogel and Luisi 2011). sRNAs bind to the proximal face of the Hfq RNA chaperone, with targets generally binding to the distal face. (B) sRNAs are degraded when displaced from Hfq. Hfq helps to promote base pairing, as well as binding to both the sRNA and mRNA, leading to negative or positive regulation of mRNA translation and stability. sRNA transcription is induced under conditions specific for each sRNA; the sRNA is stabilized by binding to Hfq. After pairing, the mRNA and most sRNAs are displaced from Hfq, and the sRNA becomes subject to RNase E-dependent degradation. If pairing mRNAs are not available (during treatment with a transcription inhibitor such as rifampicin or in cells carrying a mutation in the mRNA binding face of Hfq), sRNAs are stable as long as they can bind to Hfq (not shown). This has led to a model for turnover in which, after pairing with the target, the sRNA is displaced from Hfq, becoming sensitive to degradation by RNase E, as shown here, or by other nucleases (“use and lose” degradation). Decoy RNAs can mimic the effect of a target, binding the sRNA, or, in other cases (not shown), can disrupt sRNA structure to prevent or reduce Hfq binding, both leading to sRNA degradation.

Coupled use and degradation of most bacterial sRNAs

Although we do not have good data on how efficiently Hfq binds sRNAs as they are transcribed, immunoprecipitation with Hfq captures essentially all of the detectable sRNAs. Furthermore, judging by Northern blots done in the presence of the transcription inhibitor rifampicin, these sRNAs are very stable (for example, see Schu et al. 2015). If so, how does the cell restore homeostasis after the inducing stress is no longer present or has been adequately dealt with? Our laboratory first considered this question with respect to the shutoff of the sRNA RyhB. RyhB is induced when iron is limiting and helps the cell deal with iron starvation by shutting down translation of many iron-containing proteins, shifting the cell's metabolism to use the remaining iron for essential processes (Massé and Gottesman 2002). We found, to our surprise, that when the RyhB promoter was specifically shut off without altering global mRNA transcription (by restoring iron or by using an ectopic regulatable promoter), RyhB sRNA was now unstable, as were two other sRNAs tested at that time (Massé et al. 2003). Thus, at least these three sRNAs are all stable when examined under conditions where all cellular transcription is turned off (with the antibiotic rifampicin) but are degraded when only the promoter driving synthesis of the sRNA is turned off. The model arising from this result was that the sRNAs are degraded as used, dependent on pairing with target mRNAs (Fig. 1B). Because the mRNAs have short half-lives, in the presence of rifampicin they are rapidly depleted from the cell, leaving the sRNAs unable to pair and thus stable. RNase E was also needed for the degradation of RyhB in the promoter-specific chase condition (Massé et al. 2003).

One question is whether all of the Hfq-binding sRNAs behave similarly with respect to turnover during use. Studies of sRNA turnover in cells carrying point mutations in the binding faces of Hfq led to assignment of the sRNAs into two classes. The majority, class I, bind to Hfq on both the proximal face and the rim and behave like RyhB (Fig. 1). These sRNAs are stable in the presence of rifampicin and are destabilized by mutations in the proximal face or rim. However, they are unstable under conditions of continued mRNA synthesis—with a half-life of 3–5 min—in bacteria doubling every 20 min. Hfq-chaperoned sRNA regulation requires that the mRNA also be capable of binding the chaperone, and for these class I sRNAs, the mRNA binds to the distal face of Hfq. Mutations in the distal face, blocking mRNA binding, lead to stabilization of the class I sRNAs even when other transcription is active (Schu et al. 2015). These results all support a model in which the class I sRNAs are used stoichiometrically, pairing with a target on Hfq and then being displaced from Hfq after pairing, leading to their destruction. This simple cycle ensures that the sRNAs are available for only a short time after their synthesis ceases, providing a direct mechanism for the return to homeostasis.

The second class of sRNAs, class II, bind to the proximal face and distal face of Hfq, and for these sRNAs, their mRNA targets bind the rim. Many of these class II sRNAs are turned over similarly to the class I RNAs, and for these, mutations in the rim lead to stabilization by blocking target mRNA binding; mutations in the distal face lead to destabilization by impairing the stabilizing interaction of the sRNA with Hfq (Schu et al. 2015).

RNA-directed sRNA-specific degradation in bacteria

For sRNAs that are degraded as they are used, after pairing with target mRNAs (Fig. 1), one can to some extent consider all targets as mediators of turnover. However, clearly some targets will be more abundant, and we do not currently know whether all targets are equally active in pairing, leading to the displacement from Hfq and degradation of the sRNA, or whether the fate of the mRNA matters. For instance, does pairing with both negatively and positively regulated mRNA targets lead to sRNA turnover? The multiple targets for most sRNAs makes sorting this out a challenge.

Interestingly, a subset of class II sRNAs that contain strong distal face binding sequences is not displaced after pairing. These sRNAs are stable even when transcription of mRNA targets is active. The best studied of these, ChiX, has been shown to be used catalytically in vivo, with each ChiX supporting multiple rounds of pairing and target degradation (Overgaard et al. 2009). Furthermore, the chiX promoter, unlike that of most studied sRNAs, is apparently constitutive (Figueroa-Bossi et al. 2009). ChiX abundance is instead regulated by a fragment of RNA, called a decoy RNA, that interacts with ChiX, destabilizing the 3′ stem–loop, presumably stripping ChiX off Hfq and leading to its rapid degradation. ChiX is made at a level that keeps the expression of its target, chiP, at very low levels under normal growth conditions. chiP encodes a porin for import of chitobiose fragments into the bacteria. When chitobiose is present in the environment, it induces operons for genes involved in chitobiose degradation, including the chbBC operon, encoding the decoy in an intergenic region. Thus, for ChiX, the default situation is silencing targets, with relief of that silencing under the conditions where the target mRNA is needed (Plumbridge et al. 2014). Similarly, a portion of the 5′ UTR of the chiP mRNA, the target for ChiX pairing, can also accumulate and, when overproduced, reduce ChiX levels (Adams et al. 2021). In the context of understanding sRNA turnover, ChiX provides a counterexample to the sRNAs turned over with use discussed above—the off switch for ChiX is the inducible “decoy” RNA, promoting ChiX degradation (for review, see Mandin and Gottesman 2009). The identification of a ChiX decoy provides a very specific example and at the same time demonstrates that defining a decoy as distinct from a target may be difficult.

The role of the decoy is particularly important in understanding the turnover of ChiX, because it is otherwise exceptionally stable. However, multiple RNAs capable of pairing with and promoting decay of other sRNAs continue to be found (for review, see Denham 2020). New approaches to identify RNAs interacting with sRNAs and studies to identify inhibitors of sRNA signaling have uncovered new examples of these (Adams et al. 2021; Chen et al. 2021). These are frequently referred to as sponges for the sRNAs. Although expression of the decoys often is associated with lower levels of the sRNA, in some cases they may act more like a true sponge, sequestering the sRNA but not necessarily leading to its destruction (Matera et al. 2022). Thus far, the terms sponge and decoy have been used interchangeably in the literature. We propose that in the future the term decoy should be used when degradation of the sRNA is induced, whereas sponge should be used to refer to cases of sRNA inhibition without decay.

An intriguing example of a decoy that works on an sRNA that is also subject to “use and lose” degradation is derived from the processed product of a tRNA operon (leuZ 3′ ETS). This transcript is abundantly expressed under most growth conditions, and the RNA can pair with two different sRNAs, likely acting to keep the basal level of expression of these sRNAs low (Lalaouna et al. 2015). In some cases, sponges and decoys can be sRNAs that carry out direct regulation of mRNA targets as well as act to sponge or direct the degradation of other sRNAs. For instance, sRNA CyaR negatively regulates translation of the general stress σ factor RpoS by pairing with the 5′ UTR of rpoS; sRNA ArcZ positively regulates the same target. Both sRNAs depend on Hfq for their stability and activity. However, ArcZ acts both by direct pairing to the 5′ UTR of the rpoS mRNA to free the ribosome binding site and thus stimulate RpoS translation and as a decoy by interacting with CyaR, leading to its degradation (Kim and Lee 2020). In Vibrio cholerae, where quorum sensing is dependent on a set of Hfq-dependent sRNAs, a newly identified sponge contributes to the ability of the cell to rapidly transition from one growth condition to another (Huber et al. 2022). Additional sponges or decoys, referred to by the investigators as anti-sRNAs, were found in studies of the 0157 pathogenic EHEC E. coli strain, which differs from the reference E. coli K12 strain by addition of multiple pathogenicity islands and prophages, many of them encoding RNAs (Tree et al. 2014). One of these, like the ChiX decoy, binds to an sRNA, disrupting the terminator and leading to sRNA instability; the other acts more like a pure sponge, binding stably to the sRNA as if it were a target.

The examples found thus far suggest that there is likely a continuum between target mRNAs that drive degradation of sRNAs upon pairing and decoys or sponges, which in the extreme case only drive degradation (or inhibition) without any regulatory consequences for themselves or the transcript from which they are derived.

Ribonucleases and ribonuclease adaptors for targeted bacterial sRNA destruction

In addition to the growing list of base-pairing RNAs that can promote degradation of specific sRNAs, a few cases in which protein mediators lead to sRNA destruction have been identified. In the best-studied case, regulation focuses on the sRNA GlmZ; GlmZ binds Hfq and activates translation of glmS, which encodes a protein that synthesizes glucosamine-6-phosphate (GlcN6P), an essential precursor for cell wall biosynthesis. When GlcN6P is abundant, GlmZ is cleaved to an inactive form, dependent on RapZ, a protein adaptor for RNase E (Islam et al. 2023). When GlcN6P is low (so more GlmS activity is needed), a second sRNA, GlmY, is synthesized at high levels. GlmY is also recognized by RapZ (but not by Hfq), freeing GlmZ to bind Hfq and activate GlmS expression (Göpel et al. 2013). Thus, in this case, adaptor protein RapZ keeps GlmZ low when not needed, and the antiadaptor sRNA GlmY provides the regulatory input to free GlmZ. Intriguingly, recent work demonstrated that RapZ itself senses cytoplasmic GlcN6P levels and feeds that information into the regulatory cascade, changing GlmY synthesis (Khan et al. 2020). This example suggests that other sRNA-specific nucleases or adaptor proteins may exist and will play important regulatory roles. At least one other ribonuclease, YicC, is able to destroy sRNA RyhB, but not a number of other sRNAs, when overproduced (Chen et al. 2021). Although YicC and its homologs can display endonuclease activity in vitro (Ingle et al. 2022), in E. coli, destruction of RyhB required PNPase, a 3′-to-5′ exonuclease, suggesting that YicC may also act as an adaptor (Chen et al. 2021).

General conclusions for bacterial sRNA decay

1. Hfq is protective, and removing the sRNAs from Hfq, by whatever mechanism, leads to their degradation. Since turnover is tied to Hfq occupancy, anything that limits Hfq availability will affect sRNA degradation/half-life.

2. For stress-induced sRNAs, “use and lose” is the most common driver of sRNA degradation (Fig. 1). If the sRNAs rapidly turn over, abundance is directly dependent on continued synthesis, and therefore the sRNA disappears as the inducing condition (frequently associated with stress) ends.

3. For a subset of “silencing” sRNAs, use and decay are not linked; in this case, the default may be that the sRNA is present, with specific mechanisms of decay induced under appropriate conditions to allow targets to be expressed.

4. Both the “use and lose” pathway (pair and decay) and decoy RNA pairing with sRNAs depend on RNA/RNA pairing and thus provide specificity in terms of which sRNA is being degraded. Mechanisms that more generally interfere with binding to Hfq (reduce Hfq or block it in some fashion) or sRNAs that successfully compete for Hfq may exist and would likely be less sRNA-specific, although these types of mechanisms should reinforce a hierarchy of sRNA binding to Hfq, depending on their affinities for the chaperone.

5. Thus far, no evidence of wholesale sRNA degradation or shutdown of Hfq-dependent sRNA regulation has been reported, but given the growing examples of ways in which bacteriophages manipulate their bacterial hosts, it would not be surprising if this is found in the future, paralleling observations in eukaryotes.

Animal microRNA turnover

Animal miRNAs

In animals, multiple classes of small RNAs are generated by diverse biogenesis mechanisms and loaded into Argonaute proteins. After loading into Argonaute, the bases of the small RNA are displayed, guiding the complex to bind via base pairing to complementary RNAs (for review, see Nakanishi 2016). The small RNA–Argonaute complex (in most cases dubbed RNA-induced silencing complex [RISC]) carries out diverse regulatory functions upon target binding. Most RISCs silence expression of target RNAs post-transcriptionally, though some also act at a transcriptional level and/or promote target gene expression (Youngman and Claycomb 2014; Gutbrod and Martienssen 2020). Outside of animals, Argonautes have an even greater diversity of binding partners, including DNA guides or targets (Lisitskaya et al. 2018).

The most ubiquitously expressed class of small RNAs in animals is microRNAs (miRNAs) (Bartel 2018). MiRNA biogenesis depends on characteristic sequence and structural features of primary miRNA transcripts, most notably an ∼35 bp hairpin that is recognized by the Microprocessor complex (Lee et al. 2003; Denli et al. 2004; Gregory et al. 2004; Han et al. 2004, 2006; Auyeung et al. 2013; Fang and Bartel 2015). The Microprocessor cleaves both strands of this hairpin to yield an ∼55–70 nt precursor hairpin that is exported to the cytoplasm, where its loop is cleaved off by Dicer (Bernstein et al. 2001; Grishok et al. 2001; Hutvágner et al. 2001; Ketting et al. 2001; Knight and Bass 2001). This yields an RNA duplex that is loaded into an Argonaute protein, after which one strand (the “star” or “passenger” strand) is ejected and degraded; the remaining complex, consisting of the miRNA guide strand loaded snugly into Argonaute, forms the core of the miRNA-induced silencing complex (miRISC) (Hutvágner and Zamore 2002; Mourelatos et al. 2002). In miRISCs, the miRNA's “seed” sequence (nucleotides 2–8) is well ordered and displayed by its interaction with Argonaute, leading to a dominant role of the seed in target RNA recognition and binding; other bases can support target interaction, especially when seed pairing interactions are weak (Bartel 2018). Binding to target RNAs leads to translational repression and decay of target mRNA through recruitment of GW182 proteins (TNRC6 in mammals) that function as scaffolds for recruitment of the deadenylation and decapping machinery (PAN2/3 and CCR4–NOT) (Braun et al. 2013). Below, we discuss the fate of miRNAs after loading into Argonaute and our current understanding of how they are targeted for decay. Select cases of parallels or contrasts in the decay of other animal small RNA classes are also highlighted.

Homeostatic relationship between microRNA and Argonaute stability

In eukaryotic systems, small RNAs are highly protected by their mode of binding to Argonaute proteins. The 5′ end is tightly bound by Argonaute's MID domain, and the 3′ end is bound in the PAZ domain through a less stable interaction (Sheu-Gruttadauria and MacRae 2017). Thus, similar to Hfq in bacteria, Argonaute shields its resident small RNA from cellular nucleases (Diederichs and Haber 2007; Winter and Diederichs 2011).

Unlike Hfq, whose stability is not modified by sRNA interaction, the loading of a small RNA into Argonaute is reciprocally stabilizing for some Argonaute proteins. For example, the stabilities of miRNA Argonautes like Drosophila AGO1, mouse Ago2, and Arabidopsis AGO1 and AGO2 are promoted by their association with miRNAs (Derrien et al. 2012; Martinez and Gregory 2013; Smibert et al. 2013). In the absence of miRNA biogenesis, the unloaded forms of these proteins are quickly degraded. Because these unloaded Argonautes are unstable, conditions that impair miRNA biogenesis or loading of Argonaute (e.g., HSP90 inhibition) decrease overall Argonaute abundance (Suzuki et al. 2009; Johnston et al. 2010; Connerty et al. 2016).

Whether the miRNA or Argonaute is limiting for the formation of RISCs appears to depend on cellular context. In the context of miRNA overexpression, increasing the level of Argonaute allows for higher levels of miRNA accumulation (Diederichs and Haber 2007), but experiments in which only the protein or the small RNA is overexpressed are more informative about the component whose endogenous expression is limiting in vivo. In some contexts (fly tissues and mouse and human cell cultures), the pool of miRNAs is limiting to the formation of RISCs; increasing the expression of miRNAs boosts the level of Argonaute, but overexpressing Argonaute alone does not increase miRNA levels (Martinez and Gregory 2013; Smibert et al. 2013; Kretov et al. 2020). In other cases (Drosophila S2 cells and mouse T cells), the situation is reversed, where Argonaute is limiting such that not all miRNAs can be loaded (Bronevetsky et al. 2013; Reichholf et al. 2019). In S2 cells, the limiting amount of Ago1 aids in specificity of the loaded miRNA pool (Reichholf et al. 2019); perhaps this level of quality control is dispensable in other contexts in which Argonaute is made in excess.

The pathways that degrade these unloaded Argonautes differ depending on the cellular context examined. This decay relies on the ubiquitin-dependent proteasome in Dicer^−/− MEFs (Smibert et al. 2013) but instead requires autophagy in Drosha RNAi Drosphila S2 cells, DGCR8^−/− mESCs, or hen1(lf) seedlings (Derrien et al. 2012; Martinez and Gregory 2013; Kobayashi et al. 2019a). While these apparent differences may be due to the cellular context, future mechanistic work will more clearly delineate commonalities and divergences among these pathways. Such analysis has been performed in the context of Drosophila S2 cells, where an unloadable mutant version of Ago1 was used to identify factors involved in degradation of unloaded Argonaute (Kobayashi et al. 2019a). In the S2 cell context, an E3 ubiquitin ligase, Iruka, ubiquitinates unloaded Ago1 and directs it to a selective autophagy pathway via VCP (valosin-containing protein) and two of its substrate recognition adapters (Ufd1 and Npl4) (Fig. 2; Kobayashi et al. 2019a,b). Like miRNA-binding Argonautes, piRNA-associated Argonautes also depend on small RNA loading for stability, at least in the case of Drosophila Piwi and Caenorhabditis elegans PRG-1 (Olivieri et al. 2010; Weick et al. 2014).

Figure 2.

Assembly and decay of a RISC. Upon loading, the miRNA is stabilized by interactions of its 5′ end with Argonaute's MID domain and of its 3′ end with the PAZ domain. Unloaded miRNA Argonautes are targeted for degradation; in Drosophila, the E3 ubiquitin ligase Iruka recognizes the unloaded form. The loaded RISC represses target mRNAs through base pairing predominantly within the miRNA seed. Specialized RNAs with extensive base pairing to the miRNA trigger an alternative regulatory interaction known as target-directed miRNA degradation (TDMD) via recruitment of E3 ubiquitin ligase ZSWIM8 (Dora in Drosophila and EBAX-1 in C. elegans). Extensive base pairing also drives miRNA 3′ end display and target-directed tailing and trimming (TDTT), which is dispensable for TDMD.

Interestingly, not all Argonaute proteins are similarly labile in the unloaded state. In the case of Drosophila Ago2, which binds small interfering RNAs (siRNAs), the protein's stability is also modulated by the ubiquitin proteasome pathway (Chinen and Lei 2017), but the protein is equally stable in the presence or absence of its cognate small RNAs (Smibert et al. 2013). This stability of the unloaded Drosophila Ago2 likely aids in rapid response to antiviral threats by maintaining an available pool of the Argonaute to load exogenously derived siRNAs (van Rij et al. 2006). Thus, the difference in the biological roles between siRNA Argonautes like Drosophila Ago2 and miRNA Argonautes like Drosophila Ago1 and mammalian Ago1–4 may have driven the differential stabilities of the unloaded proteins. This model would predict that these properties would hold true in other systems with dedicated miRNA and siRNA Argonautes, such as C. elegans (Tabara et al. 1999; Grishok et al. 2001; Seroussi et al. 2023); this remains to be determined.

What underlies these differences in unloaded Argonaute stability at the molecular level? Are there simply Argonaute-specific degrons that are recognized by the E3 ligases only in the unloaded or loaded state? Does rigidity between Drosophila Ago2's two lobes also prevent the structural changes that distinguish the loaded from the unloaded state? Structure-guided domain swaps between Argonaute paralogs (for instance, in the hinge region between the two lobes or in putative degrons) will be informative to determine which regions of the protein encode differential stability upon loading.

Genome-wide profiling of miRNA decay

When loaded into Argonaute, miRNAs are generally very stable. Profiling of miRNAs following acute inactivation of either transcription or the miRNA biogenesis machinery enabled the earliest examinations of half-lives (Bail et al. 2010; Gantier et al. 2011; Lehrbach et al. 2012; Miki et al. 2014; Guo et al. 2015; Vieux et al. 2021). Because of the long half-lives of miRNAs, these measurements suffer from the possibility of secondary effects of long-term inhibition of transcription or miRNA biogenesis. Metabolic labeling has allowed more precise kinetic analysis of miRNA turnover at steady state without these confounding effects (Duffy et al. 2015; Marzi et al. 2016; Kingston and Bartel 2019; Reichholf et al. 2019). These measures roughly agreed with earlier profiling, demonstrating that half-lives are generally long (median half-life 11–34 h, depending on cell type), in contrast to mRNAs (median half-life 2.2–4.3 h) (Herzog et al. 2017; Kingston and Bartel 2019; Reichholf et al. 2019; Eisen et al. 2020). In developmental models that do not reach a steady state, thus impairing the application of metabolic labeling methods, mathematical modeling of miRNA production and decay using the unstable star strand as a proxy for production rates has allowed for refined estimates of miRNA decay (e.g., in developing C. elegans) (Nahar et al. 2024). These and other studies further showed that various miRNA sequences display characteristic differences in stability, with a subset of miRNAs undergoing rapid decay. Clear early examples of sequence-specific rapid miRNA decay include miR-382 in HEK293 and the extended miR-16 family in NIH3T3 cells (Bail et al. 2010; Rissland et al. 2011).

miRNA 3′ end remodeling

Deep sequencing studies also revealed a large number of miRNA sequence isoforms (“isomirs”). Some isomirs arise from alternative cleavage during biogenesis, resulting in shifts in position of the 5′ or 3′ end of the miRNA, and the 5′ isomers are especially impactful on miRNA function due to their shifting of the functional seed sequence (Fernandez-Valverde et al. 2010; Fukunaga et al. 2012; Lee and Doudna 2012; Zhou et al. 2012; Tan et al. 2014; Bofill-De Ros et al. 2019; Panzade et al. 2022). Tailing by nucleotidyl transferases and trimming by exonucleases during and after biogenesis also modify the sequences of miRNA precursors and mature miRNAs (Burroughs et al. 2010; Berezikov et al. 2011; Han et al. 2011; Liu et al. 2011; Wyman et al. 2011; Westholm et al. 2012; Lee et al. 2019; Vieux et al. 2021). These modifications influence the stability and processing efficiency of canonical and noncanonical miRNA precursors (Heo et al. 2008, 2009, 2012; Hagan et al. 2009; Lehrbach et al. 2009; Thornton et al. 2012; Chang et al. 2013; Ustianenko et al. 2013; Yoda et al. 2013; Bortolamiol-Becet et al. 2015; Kim et al. 2015, 2020; Reimão-Pinto et al. 2015). The function of tailing and trimming of mature miRNAs is less clear and is context-dependent.

The finding that tailing and trimming are promoted by miRNA binding to highly complementary targets (target-directed tailing and trimming [TDTT]) drove much of the interest in the relationship between 3′ remodeling and decay (Ameres et al. 2010; Baccarini et al. 2011; Xie et al. 2012). However, recent work further elucidating the mechanism of target-mediated miRNA degradation (TDMD) shows that TDTT is dispensable for TDMD (see more on TDMD below) (Kleaveland et al. 2018; Han et al. 2020; Shi et al. 2020). Furthermore, 3′ end modifications that prevent TDTT, such as 2′-O-methylation, do not prevent TDMD (Ameres et al. 2010; Han et al. 2020; Kingston and Bartel 2021). Thus, TDMD is clearly separable from TDTT, though nucleases involved in TDTT may act redundantly with other nucleases to drive TDMD.

Still, even when highly complementary targets are not overexpressed, the prevalence of tailed and trimmed isoforms increases over time after biogenesis (Kingston and Bartel 2019; Reichholf et al. 2019; Vieux et al. 2021). Functional studies examining the impact of depletion of terminal nucleotidyl transferases show context-dependent effects of tailing. While uridylation plays important roles in regulating miRNA precursors (see above), uridylation of mature miRNAs has very limited or no impact on miRNA stability (Jones et al. 2009, 2012; Knouf et al. 2013; Thornton et al. 2014; Gutiérrez-Vázquez et al. 2017; Yang et al. 2019, 2022a; Vieux et al. 2021). Adenylation of miR-122 by the cytoplasmic poly(A) polymerase GLD2 stabilizes the miRNA in some contexts (Katoh et al. 2009; Burns et al. 2011; D'Ambrogio et al. 2012; Hojo et al. 2020). In other cases, miRNA adenylation has no effect on miRNA stability (Mansur et al. 2016; Vieux et al. 2021; Yang et al. 2022a), whereas in others, adenylation promotes miRNA decay (Boele et al. 2014; Shukla et al. 2019). In at least two contexts, adenylation drives global miRNA decay: Maternally deposited miRNAs are adenylated by a GLD2 paralog, Wispy, resulting in miRNA clearance at the Drosophila maternal-to-zygotic transition, and a viral poly(A) polymerase drives decay of host miRNAs in poxvirus-infected cells (Backes et al. 2012; Lee et al. 2014). What determines the differential outcome in the few known cases of regulatory tailing and trimming in these and other contexts is not known and remains of great interest. In other small RNA classes, the relationship between 3′ end modifications and stability is also emerging. In C. elegans, small RNAs bound to the germline Argonaute CSR-1 are uridylated, which appears to dampen their levels (van Wolfswinkel et al. 2009). In animal piRNAs, 3′ end trimming followed by 3′ end 2′-O-methylation are the final steps in 3′ end formation, and both steps are required for piRNA stability (Bronkhorst and Ketting 2018; Gainetdinov et al. 2021; Pastore et al. 2021). Some classes of siRNAs are also 2′-O-methylated, dependent on their resident Argonaute, including Drosophila siRNAs, which are Ago2-bound and ERGO-1-bound 26G RNAs in C. elegans (Horwich et al. 2007; Ameres et al. 2010; Billi et al. 2012; Kamminga et al. 2012; Montgomery et al. 2012). These siRNAs are also stabilized by 2′-O-methylation, yet other classes of siRNAs in C. elegans and siRNAs in vertebrates function in the absence of methylation (Tam et al. 2008; Watanabe et al. 2008; Billi et al. 2012; Kamminga et al. 2012; Montgomery et al. 2012). These differential requirements for methylation may have arisen due to different properties of the siRNAs’ resident Argonautes and/or differences in how the siRNAs engage and act on their targets. Notably, in plants and cnidarian animals (exemplified by the sea anemone Nematostella vectensis), miRNAs bind and cleave fully complementary targets, thus acting like siRNAs. In these organisms, both miRNAs and siRNAs are protected by 2′-O-methylation of 3′ ends (Li et al. 2005; Yu et al. 2005; Yang et al. 2006; Grimson et al. 2008; Moran et al. 2014; Modepalli et al. 2018). Thus, extensively complementary target engagement by a small RNA cohort may promote the evolutionary retention of its Argonaute's ability to recruit the methyltransferase Hen1, though this is clearly not strictly required.

Broad regulators of miRNA stability

Through efforts to determine the mechanisms that underlie differential miRNA stability, multiple drivers of miRNA decay have been identified. Early studies demonstrated that a multiprotein nuclease complex called the RNA exosome was required for the rapid decay of miR-382 in a reconstituted in vitro system (Bail et al. 2010). In C. elegans, the decapping scavenger enzyme DCS-1 cooperates with the 5′-to-3′ exoribonucleases XRN-1/2 and the RNA exosome to degrade a subset of miRNAs (Chatterjee et al. 2009; Bossé et al. 2013). Similarly, its ortholog, DcpS, modulates miRNA stability in human cells (Meziane et al. 2015), but what determines whether a miRNA is targeted by this pathway is so far unclear. A phenomenon known as target-mediated miRNA protection (TMPP) prevents miRNAs from being routed to this XRN-2-mediated decay pathway, but this mechanism remains broadly unresolved (Chatterjee et al. 2011). PQN-59, a protein with a prion-like domain, was also recently described as a modulator of miRNA stability in C. elegans, but little is known about this mechanism and whether it relates to DCS-1/XRN-1/2 (Carlston et al. 2021). Another endonuclease, Tudor–staphylococcal nuclease (TSN), is responsible for instability of a subset of miRNAs (Elbarbary et al. 2017a,b). Degradation by TSN-mediated miRNA decay (TumiD) requires an AC dinucleotide within the terminal four nucleotides of the miRNA (Elbarbary et al. 2017a,b). So far, how TumiD is regulated or whether it is preferentially deployed in any specific contexts to degrade its substrates is unknown.

In other scenarios, wholesale miRNA decay is observed in specific developmental contexts. In many types of retinal neurons, miRNAs display rapid turnover via an unknown mechanism (Krol et al. 2010). As noted above, in the Drosophila maternal-to-zygotic transition, maternal miRNAs are cleared; this process requires the terminal adenylation of miRNAs by a GLD2 ortholog, Wispy, and the nuclease that subsequently degrades the tailed miRNAs is currently unknown (Lee et al. 2014). Similarly, in the context of poxvirus infection, a viral poly(A) polymerase drives host miRNA clearance by an unknown nuclease (Backes et al. 2012).

Regulated decay of small RNA populations is achieved in some cases by targeting the loaded Argonautes for proteasomal decay. Such a mechanism drives wholesale miRNA clearance during T-cell maturation (Bronevetsky et al. 2013). Similarly, the piRNA-binding Argonaute MIWI is degraded in an APC-dependent manner at the end of mouse spermatogenesis, resulting in clearance of piRNAs (Zhao et al. 2013). In this case of proteolysis-driven small RNA degradation, piRNAs are required for APC recruitment but presumably in a sequence-independent manner because the whole pool of MIWI is targeted (Zhao et al. 2013). This therefore represents an interesting case where the loaded (rather than unloaded) conformation of Argonaute is specifically targeted for decay. In the arms race between cricket paralysis virus and its hosts, the virus encodes a suppressor of RNAi (called 1A) that binds to Drosophila Ago2, both directly inhibiting its activity and targeting it to the proteasome via recruitment of host ubiquitin ligase components Elongin B and Elongin C (Nayak et al. 2018).

Target-directed microRNA degradation

In addition to wholesale small RNA clearance, viral infection and developmental transitions also trigger selective decay of specific miRNAs. In recent years, a major proportion of observed cases of sequence-specific miRNA decay have been attributed to a process dubbed target-directed miRNA degradation (TDMD) (de la Mata et al. 2015). First described in 2010, TDMD occurs when a specialized RNA (“TDMD trigger”) base pairs with a miRNA, driving its decay, echoing the action of decoy RNAs in bacteria (Ameres et al. 2010; Cazalla et al. 2010). TDMD triggers generally base pair with the targeted miRNA in two regions: the 5′ seed region and a major portion (≥6 nt) of the 3′ half of the miRNA (Ameres et al. 2010; Cazalla et al. 2010; Baccarini et al. 2011; Libri et al. 2012; Marcinowski et al. 2012; Lee et al. 2013; de la Mata et al. 2015; Haas et al. 2016; Sheu-Gruttadauria et al. 2019). An intervening bulge between these two duplexes prevents Argonaute-mediated cleavage of the target, prolonging the lifetime of the complex bound to the TDMD trigger. This unique bipartite basepairing interaction results in a conformational change that affects both the miRNA and its Argonaute (Fig. 2; Sheu-Gruttadauria et al. 2019). The miRNA 3′ end is pulled out of its binding pocket in the Argonaute PAZ domain; this untethers the two lobes of Argonaute, causing an opening of the central cleft (Sheu-Gruttadauria et al. 2019).

In this configuration, the 3′ end of the microRNA is exposed to the cytoplasm. Thus, exonucleases and nucleotidyl transferases heavily modify the 3′ ends of miRNAs undergoing TDMD by trimming and tailing, respectively. This target-directed tailing and trimming (TDTT) was one of the most salient features of TDMD in early studies (Ameres et al. 2010; Baccarini et al. 2011; Xie et al. 2012; Haas et al. 2016). Recent work has uncoupled TDTT from TDMD, demonstrating via depletion of nucleotidyl transferases or protective modification of the miRNA 3′ end that TDTT is dispensable for TDMD (Kleaveland et al. 2018; Han et al. 2020; Shi et al. 2020). Rather, TDTT is likely a consequence of the TDMD conformation of miRISCs (Sheu-Gruttadauria et al. 2019). Studies of miRNA isomers in cells expressing Argonaute containing PAZ domain mutations that constitutively release the miRNA 3′ end support this interpretation; a high proportion of trimmed and tailed isoforms are observed in these scenarios (Sheu-Gruttadauria et al. 2019; Yang et al. 2020).

How is the TDMD conformation recognized and targeted for degradation? Recent genome-wide screens have identified an Elongin BC-containing E3 ubiquitin ligase in which the substrate recognition module ZSWIM8 targets the Argonaute of miRNAs undergoing TDMD for ubiquitination and proteasomal degradation, ultimately resulting in decay of the exposed miRNA (Wang et al. 2013; Han et al. 2020; Shi et al. 2020).

Similar to the rapid degradation of unloaded Argonautes, some Argonaute proteins are susceptible to ZSWIM8-dependent target-mediated degradation, whereas others are resistant. So far, this has been best characterized in Drosophila, where miRNA–Ago1 RISCs are susceptible to target-mediated degradation dependent on the ZSWIM8 ortholog Dora, whereas siRNA–Ago2 RISCs are not (Kingston and Bartel 2021). As proposed by Ameres et al. (2010), this differential susceptibility to extensively complementary targets may reflect requirements imposed by the Argonautes’ respective target repertoires: Ago1-bound miRNAs engage with targets primarily through limited base pairing involving the miRNA seed, whereas Ago2-bound siRNAs engage fully complementary targets and thus frequently experience release of the sRNA 3′ end from the Argonaute PAZ domain. Examining the susceptibility to TDMD of an animal miRNA Argonaute that engages highly complementary targets (e.g., in N. vectensis) may inform whether these two properties indeed constrain each other (Grimson et al. 2008; Moran et al. 2014; Modepalli et al. 2018).

Many key mechanistic questions remain. Although the conformational change in Argonaute driven by TDMD triggers is well documented, whether and how this conformation can specifically recruit ZSWIM8 are unclear. Where ZSWIM8 binds Argonaute is not yet known; perhaps this binding is dependent on a degron that is exposed by the TDMD conformation. Testing the contribution of the TDMD conformation to ZSWIM8 recruitment is difficult because base pairing to TDMD triggers induces both the conformational change and all downstream degradation events. Three hints suggest that the TDMD conformation may not be sufficient (or may even be dispensable) for TDMD. First, TDMD triggers designed de novo do not always induce decay, and regions outside the miRNA binding site in endogenous TDMD triggers are conserved and, in some cases, required for TDMD (Bitetti et al. 2018; Kleaveland et al. 2018; Li et al. 2021). This suggests that the requirements for a TDMD trigger are more extensive than the regions that base pair to the miRNA. Second, PAZ domain mutations with constitutive miRNA 3′ end release do not drive strong and consistent Argonaute and miRNA decay (Sheu-Gruttadauria et al. 2019; Yang et al. 2020). Third, miRNA decay driven by the C. elegans ZSWIM8 ortholog EBAX-1 has different base pairing requirements; the developmentally timed decay of the mir-35 family depends only on the miRNA's seed sequence, suggesting little to no contribution from 3′ end pairing (Donnelly et al. 2022). Thus, relaxed base pairing configurations may be able to trigger TDMD in the absence of the conformation change or 3′ end display. While specialized pipelines for identifying TDMD triggers are likely to accelerate their discovery (Simeone et al. 2022), such methods should be supplemented with more agnostic empirical approaches like CLASH in light of the potential for relaxed TDMD base pairing requirements (Li et al. 2021; Donnelly et al. 2022). The overall topology of TDMD trigger RNAs also has an emerging role in TDMD efficiency (Pawlica et al. 2016; Gorbea et al. 2022; Fuchs Wightman et al. 2024); understanding how the topology and conserved sequence blocks outside the miRNA binding site contribute to the potency of TDMD will also be important future areas of investigation (Pawlica et al. 2016; Bitetti et al. 2018; Kleaveland et al. 2018; Li et al. 2021; Fuchs Wightman et al. 2024). In the coming years, the identification of new TDMD triggers and functional dissection of these triggers and the broader TDMD machinery will address these open questions.

Biological roles of TDMD

How does TDMD contribute to physiology and development? The first identified TDMD triggers were viral noncoding RNAs that target host microRNAs for destruction, suggesting that viruses have hijacked this endogenous regulatory pathway for their own benefit (Cazalla et al. 2010; Libri et al. 2012; Marcinowski et al. 2012; Lee et al. 2013). More recently, endogenous TDMD triggers have been uncovered. Interestingly, the first trigger discovered was a vertebrate long noncoding RNA (lncRNA; libra in zebrafish) that gained coding potential in part of the vertebrate lineage including mammals (NREP in mice) (Bitetti et al. 2018). Nrep induces TDTT and TDMD of miR-27b, driving its decay in the cerebellum. Loss of the binding site that triggers TDMD results in behavioral abnormalities in libra mutant zebrafish and Nrep mutant mice (Taylor et al. 2008; Bitetti et al. 2018). A second prominent TDMD trigger is the lncRNA Cyrano, which targets the miRNA miR-7 for decay in neurons and differentiating myoblasts, though Cyrano knockout does not perturb normal fish or mouse development (Kleaveland et al. 2018; Goudarzi et al. 2019; Yang et al. 2022b). Interestingly, Cyrano is a more “potent” TDMD trigger than others; its action is multiple turnover, destroying ∼17 miR-7s per Cyrano molecule (Kleaveland et al. 2018). Most other TDMD triggers act much less efficiently, generally degrading less than one miRNA molecule per molecule of TDMD trigger (Baccarini et al. 2011; de la Mata et al. 2015; Ghini et al. 2018), despite the fact that TDMD triggers are not degraded during TDMD (de la Mata et al. 2015; Kleaveland et al. 2018; Han et al. 2020).

More recently, additional mRNAs that induce TDMD have been identified in mammals, including Serpine and BCL2L11 (Ghini et al. 2018; Li et al. 2021). Whether these two cases are ZSWIM8-dependent is not yet known, and future work will determine whether they act through the same mechanism as NREP and Cyrano (Ghini et al. 2018; Li et al. 2021; Shi et al. 2023). Nonetheless, the miRNA binding sites in these mRNAs are required for wild-type cellular behavior and appropriate response to apoptotic stimuli.

ZSWIM8 loss-of-function studies have hinted at the roles for TDMD in organismal biology. In Drosophila, knockout of the ZSWIM8 ortholog Dora/Pelado is lethal (Kingston and Bartel 2021; Molina-Pelayo et al. 2022). Using S2 cell culture, TDMD triggers corresponding to most of the miRNAs stabilized in Dora knockouts were identified (residing in both mRNA 3′ UTRs and lncRNAs) (Kingston et al. 2022; Sheng et al. 2023). One of the lncRNA TDMD triggers is responsible for cuticle integrity (Kingston et al. 2022), whereas one of the mRNA TDMD triggers ensures robustness in response to oxidative stress (Sheng et al. 2023). Reducing the dosage of one of the miRNAs stabilized in the Dora mutant, miR-3, partially rescues the Dora mutant lethality, supporting the model that aberrantly stabilized miRNAs contribute to this deleterious phenotype (Kingston et al. 2022). Other functions of Dora may also contribute to this phenotype, such as its role in actin cytoskeleton polarization, which appears to be ubiquitin ligase-independent (Molina-Pelayo et al. 2022).

In C. elegans, the ZSWIM8 homolog EBAX-1 was previously shown to drive proteasomal decay of the neuronal pathfinding Robo receptor SAX-3 (Wang et al. 2013). Knockout of ebax-1 results in neuronal pathfinding defects and other poorly characterized mild developmental phenotypes (Wang et al. 2013). Although miRNA stabilization has been observed in ebax-1-null animals (Shi et al. 2020; Donnelly et al. 2022; Kotagama et al. 2024; Nahar et al. 2024), whether any of the ebax-1^−/− phenotypes are due to RISC stabilization rather than that of SAX-3 or another substrate is as yet unknown.

ZSWIM8 knockout mice have allowed for the assessment of the contribution of TDMD to mammalian development (Jones et al. 2023; Shi et al. 2023). Knockout in the whole animal results in developmental defects in the lungs (incomplete sacculation) and heart (globose morphology) (Jones et al. 2023; Shi et al. 2023). Pups are born underweight and die postnatally (Jones et al. 2023; Shi et al. 2023). These studies unveiled that numerous miRNAs are sensitive to ZSWIM8-mediated decay in an array of tissue-specific patterns and suggest that many (at least ∼55) additional TDMD triggers remain to be identified (Jones et al. 2023; Shi et al. 2023). Notably, knockout of two miRNAs, miR-322 and miR-503, suppressed the low-weight phenotype of the ZSWIM8 knockout mouse, suggesting that the size phenotype is due to aberrant stabilization of miR-322 and miR-503 (Jones et al. 2023). The lung phenotype was not suppressed by knockout of multiple candidate miRNAs stabilized in the ZSWIM8 mutant (Shi et al. 2023). Whether the lung sacculation and other phenotypes of null mice are due to miRNA stabilization, stabilization of other ZSWIM8 substrates (such as Dab1 in the developing nervous system) (Wang et al. 2023), or loss of proposed ubiquitin ligase-independent roles of ZSWIM8 (Okumura et al. 2021; Molina-Pelayo et al. 2022) remains an open question. In mouse myoblasts, ZSWIM8 knockdown accelerates differentiation into myotubes despite the fact that Cyrano knockdown attenuates this process in human myoblasts (Okumura et al. 2021; Yang et al. 2022b). Assuming conservation between these two systems, these results suggest that ZSWIM8 has additional substrates or functions that are epistatic to its role in TDMD in this context, again highlighting the need for caution in attributing ZSWIM8-null phenotypes to abrogated TDMD.

General conclusions for miRNA decay

1. Argonaute is protective of bound small RNAs. In many cases, this stabilizing relationship is reciprocal, as in the case of miRNA- and piRNA-binding Argonautes, which are labile in the absence of their cognate small RNAs.

2. Argonaute-bound miRNAs are constitutively relatively stable. Mechanisms that target the Argonaute for decay are a prevalent means of destabilizing small RNAs in either a sequence-independent or sequence-dependent manner.

3. Target-directed miRNA degradation (TDMD) drives sequence-specific miRNA decay via recruitment of ZSWIM8, the substrate recognition module of an E3 ubiquitin ligase.

4. TDMD generally occurs via engagement of specialized target RNAs (“TDMD triggers”) that extensively base pair to the miRNA, though recent evidence suggests that extensive base pairing is dispensable in some cases.

5. Although suspected instances of TDMD inferred from ZSWIM8 loss-of-function studies continue to rise, most TDMD triggers remain unidentified. Identification of additional TDMD triggers will aid elucidation of both the mechanism and emerging biological roles of TDMD.

Common themes and future perspectives in understanding bacterial and animal small RNA decay

In reviewing bacterial and animal small RNA decay mechanisms, multiple parallels and contrasts emerge. In both cases, the stability of small RNAs generally greatly exceeds that of mRNAs. The overarching common principle governing small RNA stability in both systems is the need for association with their protein cofactor. Production of either component of these complexes can be limiting for accumulation. In bacteria, Hfq availability seems to be limiting for complex formation with small RNAs; in animals, the limiting factor for miRNA abundance depends on cell type. A point of contrast arises in the impact of the small RNA on stability of the binding protein. While Hfq is stable in the absence of its cognate small RNAs, miRNA-binding Argonautes are labile in the unloaded state. Perhaps this difference is a reflection of the fact that Hfq is limiting for sRNA complex formation, whereas Argonautes are often generated in excess of miRNAs; the lack of an sRNA-less pool of Hfq obviates the need for any mechanism targeting its decay. When Argonaute is limiting in animals (e.g., Drosophila S2 cells), this provides a layer of quality control to the loaded small RNA pool. Whether limited Hfq availability also plays such a role is an intriguing possibility. Consistent with the importance of maintaining proper Hfq levels is negative feedback control of Hfq translation by active Hfq; abrogating the feedback—and thus increasing Hfq levels—leads to a growth defect (Morita and Aiba 2019).

In rare contexts, wholesale decay of small RNAs occurs, but sequence-specific small RNA decay is more prevalent. A commonality for sequence-specific decay between Hfq-bound sRNAs and miRNAs is destabilization by base pairing to a highly complementary target RNA. In the case of Hfq-bound sRNAs, these targets are generally the same RNA that is regulated by the sRNA, also known as “use and lose” sRNA-mediated regulation followed by sRNA decay. In other cases, expression of decoy noncoding RNAs drives the destruction of complementary small RNAs (as in the case of ChiX), or, in some cases, sponge RNAs inhibit sRNA function without eliciting decay. The case of miRNAs bears similarity to the bacterial decoy RNAs, with a dedicated set of specialized targets—“TDMD triggers”—inducing decay of the miRNA without being repressed themselves (de la Mata et al. 2015; Kleaveland et al. 2018; Han et al. 2020); like many bacterial decoys, many of the TDMD triggers so far identified are noncoding RNAs whose sole purpose may be regulation of their cognate miRNA.

In the cases where bacteria use decoys in analogy to TDMD triggers, we have observed several commonalities between the scenarios in which these ad hoc decay drivers are used. The first described cases of natural TDMD triggers were virally encoded noncoding RNAs, and likewise, prophages express decoy RNAs to degrade select bacterial small RNAs (Tree et al. 2014). Thus, in both cases, small RNA decay is hijacked in the pathogen–host arms race. In terms of host RNAs that drive small RNA decay, the clearest use of TDMD is to degrade specific miRNAs in a tissue-specific and developmentally regulated manner (Bitetti et al. 2018; Kleaveland et al. 2018); in bacteria, the identification of a sponge that affects the transition from individual to community behaviors can be considered a developmental switch (Huber et al. 2022). Finally, decoy RNAs in bacteria most commonly drive specific sRNA decay, making the sRNA-dependent regulation more responsive to their inducing stresses and helping drive a return to homeostasis after the stress passes. Although stress-induced TDMD has yet to be described, this is likely to be unveiled by studies of TDMD in additional settings in the near future.

The broader biological implications of these pathways are still not fully understood and will be the focus of future work. In the case of animal TDMD, Dora mutant flies and Zswim8 knockout mice have unveiled important consequences for loss of TDMD for early animal development; conditional knockouts will further unravel postembryonic roles of TDMD in diverse tissues. Importantly, these studies should continue to be interpreted with caution—and bolstered with miRNA knockout rescue experiments—because our understanding of non-Argonaute substrates of Zswim8 and potential ubiquitin ligase-independent roles of Zswim8 is limited. It has been more of a challenge to document the consequences of the “use and lose” mode of bacterial sRNA decay, given that it is not yet clear whether one or many mRNA targets drive decay.

In conclusion, we have known for almost 30 years that miRNAs and bacterial sRNAs are important regulators of growth and development. Thus, it should be no surprise that regulation of their destruction will play as important a role as regulation of their synthesis, and it is exciting to see the recent growing understanding of the mechanisms of regulatory RNA turnover.

Competing interest statement

The authors declare no competing interests.