2-5-Mediated Decay (2-5AMD): from antiviral defense to control of host RNA

Abstract

Mammalian cells are exquisitely sensitive to the presence of double-stranded RNA (dsRNA), a molecule that they interpret as a signal of viral presence requiring immediate attention. Upon sensing dsRNA cells activate the innate immune response, which involves transcriptional mechanisms driving inflammation and secretion of interferons (IFNs) and interferon-stimulated genes (ISGs), as well as synthesis of RNA-like signaling molecules comprised of three or more 2’−5’-linked adenylates (2–5As). 2–5As were discovered some forty years ago as IFN-induced inhibitors of protein synthesis. The efforts of many laboratories, aimed at elucidating the molecular mechanism and function of these mysterious RNA-like signaling oligonucleotides, revealed that 2–5A is a specific ligand for the kinase-family endonuclease RNase L, which decays single-stranded RNA (ssRNA) from viruses and mRNAs (as well as other RNAs) from hosts in a process called 2–5A-mediated decay (2–5AMD). During recent years it has become increasingly recognized that 2–5AMD is more than a blunt tool of viral RNA destruction, but a pathway deeply integrated in sensing and regulation of endogenous RNAs. Here we present an overview of recently emerged roles of 2–5AMD in host RNA regulation.

Graphical abstract

Introduction

Double-stranded RNA (dsRNA) is a widespread viral material that triggers global RNA decay, and one of the fastest and deepest blocks of global translation in mammalian cells described to date (Donovan et al. 2017; Chitrakar et al. 2019). Arrest of protein synthesis during the dsRNA response is mediated by a diffusible signaling molecule containing three or more 2’,5’-linked adenylate residues (2–5A), that resembles a short stretch of RNA (Hovanessian and Kerr 1978; Chitrakar et al. 2019). 2–5As are produced by the interferon (IFN)-inducible cytosolic dsRNA sensors oligoadenylate synthetase 1 (OAS1), OAS2, and OAS3 (Rebouillat and Hovanessian 1999; Donovan et al. 2013; Ibsen et al. 2014; Donovan et al. 2015), and exert their effect through specific activation of the ubiquitously expressed mammalian endonuclease RNase L (Clemens and Williams 1978; Zhou 1993). Activated RNase L cleaves Un^N consensus motifs (n=A~U>>G,C; N=A,G,C,U; ^ is the scissile phosphodiester bond) within various single-stranded RNAs (ssRNAs) in the cell. This action can deplete all translationally active mRNAs within tens of minutes and cause a deep translational arrest escaped only by antiviral proteins (Burke et al. 2019; Rath et al. 2019). This pathway, termed 2–5A-mediated decay (2–5AMD) (Rath et al. 2019), serves as a mechanism for translational control by the innate immune system in the cells of all higher vertebrates. In this review we discuss the role of 2–5AMD in shaping the innate immune response to both pathogen-derived and endogenous dsRNA (endo-dsRNA), and describe an emerging role for this pathway in sensing leaking introns and intergenic RNAs.

Structure of the 2–5AMD effector enzyme: RNase L

RNase L contains a regulated endoribonuclease domain belonging to the kinase-extension nuclease (KEN) family, fused to a kinase-homology domain (KH) (Fig. 1A) (Manning 2002). This domain combination is shared only by one additional protein, a key receptor in the unfolded protein response, Ire1 (Lee et al. 2008). The KH domain structurally resembles protein kinases such as CDK2, but has no identifiable kinase activity (Silverman et al. 1988; Han et al. 2014). At the N-terminus, the KH domain bears an ankyrin-repeat domain (ANK) that regulates nuclease activity by directly sensing 2–5A. [Figure 1 near here]

Figure 1.

Structure of RNase L with bound activator and substrate mimic. A) Structure of the RNase L dimer with individual protomers shown as surface (chain A) or ribbon (chain B) representation. B) Close-up view of the active site with bound RNA substrate mimic. This image illustrates the structural mechanism of Un^N consensus motif recognition by RNase L. The orientation of the active site corresponds to the bottom view of panel (A). Black arrows show the direction of the RNA molecule bound at the dimeric active site. Nucleotide numberings are relative to the cleavage site (position 0). PDB accession entry 4OAV was used, adapted from Han et al. 2014. A color version of this figure is available online.

Upon activation, RNase L forms a crossed homodimer held together by the binding of two 2–5A molecules (Fig. 1A) (Han et al. 2012; Han et al. 2014). Each 2–5A copy resides in a composite pocket created by two ANK domains and one KH domain. The dimer of RNase L is additionally stabilized by two homotypic interfaces, one formed by two KH domains and another formed by two KEN domains. The latter creates a dimeric RNase that binds a single RNA substrate. The structure reveals the conserved U nucleotide is recognized by a single KEN protomer, while the other KEN protomer is poised for performing RNA cleavage (Fig. 1B) (Han et al. 2014). Curiously, while the structure shows 2–5A activates RNase L by holding the dimer together, deletion of the ANK domain partially activates RNase L in the absence of 2–5A. Thus, the ANK domain could have an additional, dimerization-suppressing function that is released by 2–5A binding (Dong and Silverman 1997). Structural analysis of the hypothetical auto-repressed state of RNase L remains to be performed. Of further note, the RNase L paralog Ire1 assembles into higher-order oligomers (Korennykh et al. 2009). For RNase L, high-order assembly is also expected based on cooperativity with Hill coefficient ~ 5, as well as crosslinking studies (Han et al. 2012). The structural basis underlying cooperative activation of RNase L also awaits future work. Functionally, high-order assembly of RNase L should enhance sensitivity to 2–5A and dsRNA, and allow complete shutdown of activity at low levels of 2–5A, such as under physiological conditions that do not require the action of RNase L.

RNase L is a broad-spectrum endonuclease that cleaves any ssRNA preferentially at Un^N consensus sites (Fig. 1B lower panel) (D H. Wreschner et al. 1981; Iordanov et al. 2000; Washenberger et al. 2007; Huang et al. 2014; Han et al. 2014; Rath et al. 2015). In human cell extracts as well as live human cells under conditions of relatively low RNase L activation, this preference stands out, and mRNAs containing AU-rich regions are preferentially degraded (Rath et al. 2015). Under conditions of strong RNase L activation leading to global translational shutdown in A549 cells (strong 2–5AMD), the preference for Un^N in degraded mRNAs paradoxically disappears and all mRNAs appear to be cleaved at the same rate regardless of Un^N sequence content (Rath et al. 2019). This loss of sequence preference can be explained if one postulates that during strong 2–5AMD in A549 cells RNase L operates under Briggs-Haldane conditions, perhaps with assistance from mRNA-docked ribosomes (Rath et al. 2019). In this “perfect” regime, every RNase L-mRNA binding event should be catalytically productive, leading to uniform decay of all transcripts. Further validation and in-depth analysis of this Briggs-Haldane hypothesis would be invaluable for understanding precisely how 2–5AMD can produce global and uniform decay of all human mRNAs irrespective of mRNA length or Un^N content.

2–5A synthetases that activate 2–5AMD: dsRNA receptors OAS1, OAS2 and OAS3

RNase L is constitutively present in almost all types of mammalian tissues, and its expression does not respond strongly to interferons or dsRNA. Therefore, the activity of 2–5AMD is controlled almost exclusively by the cellular level of the second messenger 2–5A. 2–5A molecules are produced by three closely related dsRNA receptors with polymerase beta (pol-β)-like fold, OAS1, OAS2, and OAS3. In OAS1, dsRNA docking induces a conformational change that narrows the active site, bringing together the catalytic residues required for 2–5A synthesis (Fig. 2, left column) (Donovan et al. 2013). Whereas OAS1 has a single domain, OAS2 and OAS3 contain duplicated OAS1-like domains that are critical to their activity (Fig. 2, middle column) (Donovan et al. 2015; Koul et al. 2020). These duplicated domains are pseudoenzymes that probably emerged from gene duplication followed by an adaptive shift in function from nucleotidyl transferase activity to strong RNA binding (Donovan et al. 2015). [Figure 2 near here]

Figure 2.

Structure and characteristics of OAS proteins. Left column, top to bottom) Structure of human OAS1 bound to dsRNA, human OAS3 domain 1 bound to dsRNA, and human OASL (PDB accession numbers: 4IG8, 4S3N, 4XQ7). The structure of OAS2 is not yet known. Middle and right columns) Domain architecture and characteristics of the OASs. Adapted from Donovan et al. 2015. A color version of this figure is available online.

Indeed, biochemistry and X-ray crystallography experiments determined that OAS3 employs the non-enzymatic domains to increase its affinity for dsRNA. The increase is achieved by adopting a constitutive dsRNA-compatible conformation in the duplicated domains, which eliminates conformational energetic barriers to dsRNA binding. However, the loss of the conformational change should inevitably result in uncontrollable 2–5A production. This situation creates an evolutionary conflict: better dsRNA binding versus well-regulated 2–5A synthesis. It was proposed that domain duplication provides the means to escape from this conflict by retaining regulated enzymatic activity in just one domain, while strengthening dsRNA binding in the duplicated domains, which evolved into pseudoenzymes to prevent uncontrollable 2–5A production. Therefore, domain duplications in OAS3 (and presumably also in OAS2) serve to improve binding and selectivity to longer dsRNAs (Fig. 2) (Donovan et al. 2015).

The number of additional domains correlates with experimentally determined dsRNA length preferences, with the shortest dsRNAs activating OAS1 and the longest activating OAS3 (Fig. 2) (Donovan et al. 2013; Donovan et al. 2015; Koul et al. 2020). Thus domain duplication enables differential dsRNA sensing in the OAS family. Indeed, these proteins serve non-redundant functions, and vary in their response to different viral and endogenous dsRNA sources (Sadler and Williams 2008; Chitrakar et al. 2019; Chitrakar et al. 2021).

The fourth member of the OAS family, OASL, is a functional outlier. Although OASL has the same fold as OAS1 (Fig. 2, left column), it does not synthesize 2–5A. Although OASL exhibits antiviral activity, this activity is independent of 2–5AMD (Zhu et al. 2014; Ibsen et al. 2015), and its expression is induced by viral infection as well as IFNs (Melchjorsen et al. 2009; Chitrakar et al. 2019). OASL binds to the dsRNA receptor helicase RIG-I via ubiquitin-like and dsRNA-binding domains, helping RIG-I to detect viral dsRNA (Zhu et al. 2014; Ibsen et al. 2015).

Pathogen defense connections of 2–5AMD

RNase L blocks replication of many viruses as well as some bacteria, indicating that 2–5AMD is a component of anti-pathogen immunity (Silverman 2007; Li et al. 2008). The importance of this mechanism for defense is increasingly clear for viruses, owing to a rapidly growing list of viral proteins, such as 2’,5’-specific nucleases, which are key for viral evasion of 2–5AMD and antiviral immunity (Xiang et al. 2002; Min and Krug 2006; Sanchez and Mohr 2007; Zhao et al. 2012; Zhang et al. 2013; Silverman and Weiss 2014; Sánchez-Tacuba et al. 2015; Goldstein et al. 2017; Drappier et al. 2018; Ancar et al. 2020; Asthana et al. 2021). Additional evidence comes from the observation that RNase L cleaves the RNA of various viruses (Washenberger et al. 2007; Cooper, Jha, et al. 2014; Cooper, Banerjee, et al. 2014). Thus, direct destruction of viral RNAs by 2–5AMD likely contributes to its antiviral activity (Nilsen and Baglioni 1979; Han et al. 2004; Silverman 2007; Cooper, Banerjee, et al. 2014). For hepatitis C virus (HCV) and poliovirus RNA genomes it has been shown that RNase L preferentially cleaves Un^N motifs (Han et al. 2004; Cooper, Jha, et al. 2014), and the majority of mapped cuts occur at a small number of sites. For the segmented influenza A virus genome, RNase L more frequently targets segments of greater length and abundance, with a few discrete hotspots of activity (Cooper, Banerjee, et al. 2014). Recent findings show that cleavage of viral RNA during 2–5AMD does not necessarily restrict replication. For example, Zika virus genomes are degraded by RNase L, but virus production is not diminished (Whelan et al. 2019). On the contrary, RNase L appears to support Zika virus replication factories, with knockout cells generating reduced viral titer (Whelan et al. 2021). Thus the direct activity of RNase L on viral genomes, as well as its effect on replication outcome, varies greatly between pathogens. Indeed, viral RNA destruction is only one facet of the role of 2–5AMD in the innate immune response.

OAS1 was shown to be a key sensor that activates 2–5AMD during SARS-CoV-2 infection (Wickenhagen et al. 2021). Two genome-wide association studies reported a correlation between individual differences in COVID-19 disease severity with genetic variation at the locus containing OAS1, OAS2 and OAS3 (Pairo-Castineira et al. 2021; Zeberg and Pääbo 2021). However, due to linkage disequilibrium in this region it is difficult to assign a causality to a single mutation, particularly within study groups of homogeneous ancestry (Sams et al. 2016; Huffman et al. 2022).

OAS1 emerged as a determinant of COVID-19 infection outcome upon analysis of a cluster of sites within the OAS1 locus. These sites are significantly variable between critically ill COVID-19 patients and the general population (Zhou et al. 2021; Wickenhagen et al. 2021), and lower OAS1 expression correlates with increased disease severity (Zhou et al. 2021). One OAS1 allele in particular (rs10774671) appears to confer protection from hospitalization with COVID-19 infection, independent of patient ancestry, and is a likely causative variant (Huffman et al. 2022). This intronic single-nucleotide polymorphism (SNP) controls splicing of the OAS1 transcript, and consequent prenylation and membrane-association the protein (Bonnevie-Nielsen et al. 2005; Noguchi et al. 2013; H. Li et al. 2017; Skrivergaard et al. 2019; Soveg et al. 2021; Wickenhagen et al. 2021). Of note, the same SNP was linked also to West Nile virus infection risk (Lim et al. 2009), and responsiveness of chronic HCV patients to IFN therapy (El Awady et al. 2011).

For OAS2, an intriguing immune function as a blocker of milk secretion has been described. According to this work, viral infection of a mother can activate OAS2 and arrest lactation to prevent passing the virus to offspring via milk (Oakes et al. 2017).

2–5AMD integration in cell physiology

Innate immunity functions as a rapid, broad-spectrum defense mechanism that slows down infections during the critical early period before antibody production is active. In higher vertebrates this is enacted primarily by the IFN system. IFN synthesis is triggered by the recognition of pathogen-associated molecular patterns, and signals the production of IFN-stimulated genes (ISGs) that execute defensive functions. The sensor proteins RIG-I and MDA5 for example, recognize dsRNA and induce the transcription of IFN/ISGs (Fig. 3) [Figure 3 near here]. Among the ISGs are all the OASs, particularly OAS2. Nevertheless, basal levels of OAS3 are readily detectable in uninfected cells and are sufficient for robust 2–5A synthesis even before the onset of RIG-I/MDA5 signaling (Birdwell et al. 2016; Chitrakar et al. 2019). 2–5AMD can therefore act as the first line of response to dsRNA that precedes IFNs and ISGs. Once activated, RNase L cleaves all accessible cytosolic RNAs, leading to strong mRNA depletion and production of RNA fragments with 2’,3’-cyclic phosphates (Cooper, Jha, et al. 2014; Donovan et al. 2017; Rath et al. 2019).

Figure 3.

Role of RNase L within dsRNA-sensing innate immune pathways. The dsRNA-binding RNA helicases RIG-I and MDA5 activate transcriptional IFN responses, which includes the production of antiviral genes and all three OAS enzymes: OAS1, OAS2 and OAS3. The OAS enzymes synthesize 2–5A and activate RNase L. Notably, although RIG-I/MDA5 upregulate the OAS enzymes, substantial quantities of OAS1 and OAS3 are constitutively basally expressed (Zhao et al. 2013). This basal expression can be sufficient to allow RNase L activation by dsRNA rapidly and independently from RIG-I/MDA5 (Chitrakar et al. 2021). A color version of this figure is available online.

These RNA fragments were proposed to perpetuate IFN synthesis by signaling through RIG-I/MDA5 (Malathi et al. 2007; Luthra et al. 2011; Jung et al. 2020; Steinberg et al. 2021), as well as activate the NLRP3 inflammasome to potentiate IL-1β production (Chakrabarti et al. 2015), and promote apoptosis (Siddiqui et al. 2015). RNase L activity is strongly pro-apoptotic in both cultured cells and mice (Castelli et al. 1997; Zhou et al. 1997; Rusch et al. 2000; Siddiqui et al. 2015; Y. Li et al. 2017; Banerjee et al. 2019; Boehmer et al. 2021). The mechanism underlying the pro-apoptotic activity is unclear, although the global arrest of translation by 2–5AMD is likely a factor (Rath et al. 2019). It has been also proposed that apoptosis in tumor cells depends on depletion of the mRNA for anti-apoptotic factor Mcl-1 (Boehmer et al. 2021). Reduction of Mcl-1 protein levels is generally sufficient for apoptosis (Subramaniam et al. 2008; Zhou et al. 2013; Goodall et al. 2016). The action of RNase L was further associated with autophagy(Chakrabarti et al. 2012; Siddiqui and Malathi 2012; Siddiqui et al. 2015; Ramnani et al. 2021).

Beyond viruses: 2–5AMD in non-infectious diseases

In mammals RNase L is broadly expressed throughout the body, and is enriched in the lungs, gastrointestinal tract, bone marrow and lymphoid tissues, consistent with its immune function (Zhou et al. 1997; Zhou et al. 2005). Mice lacking RNase L show defects in the antiviral activity of IFNα, as well as increased susceptibility to infection by numerous viruses (Zhou et al. 1997; Silverman 2007). Mouse studies further show a pro-inflammatory function for RNase L in experimental skin allograft rejection (Silverman et al. 2002), as well as colitis and colitis-associated cancer models (Long et al. 2013) demonstrating a greater role for RNase L in the immune system outside of antiviral defense. Genetic variation in OAS1 is also implicated in immune dysregulation in humans. Allelic variation at a particular SNP (rs10774671, described above) is linked to differential susceptibility to type-I diabetes (Field et al. 2005; Zeng et al. 2014; Pedersen et al. 2021), the autoimmune condition Sjögrens syndrome (H. Li et al. 2017), and lethal polymorphic autoinflammatory immunodeficiency (Magg et al. 2021). It is clear 2–5AMD has a broader function in immune regulation aside from its antiviral activity.

Chronic fatigue syndrome (CFS)

Alterations in the 2–5A pathway have been observed in people with Chronic fatigue syndrome (CFS), a condition characterized by profound exhaustion that is not relived by rest. Increased RNase L activity and 2–5A levels are seen in CFS patients compared to controls (Suhadolnik et al. 1994; Suhadolnik et al. 1997; Suhadolnik et al. 1999; De Meirleir et al. 2000; Ikuta et al. 2003). Deregulation of the RNase L pathway in CFS patients is correlated with reduced number of natural killer cells (Nijs et al. 2003) and increased apoptotic activity (Frémont et al. 2002). A cleaved, low molecular weight form of RNase L is also present in CFS patients, and was a proposed biomarker for the condition (Suhadolnik et al. 1997; De Meirleir et al. 2000; Demettre et al. 2002). However, the CFS connections do not establish causality and remain correlative. Efforts to attribute clinical importance were inconclusive (Nijs and Meirleir 2005). Therefore, 2–5AMD is not presently accepted as a contributing factor in CFS.

Hereditary prostate cancer

Early investigations into the genetic causes of familial prostate cancer identified several loci associated with the condition. Hereditary prostate cancer locus 1 (HPC1) is one such region that tends to segregate with families who exhibit more severe disease (Goode et al. 2001). In 2002 HPC1 was mapped to RNASEL, with multiple different germline mutations implicated (Carpten et al. 2002; Rökman et al. 2002; Rennert et al. 2002; Casey et al. 2002). However, the role of HPC1 mutations in prostate cancer etiology was questioned by subsequent studies of incidence in familial and sporadic cases that found no statistical significance (Wang et al. 2002; Downing et al. 2003; Wiklund et al. 2004; Maier et al. 2005). Thus, RNase L mutations identified to date do not appear to drive prostate cancer. However, they could be an aggravating factor in cancer progression in combination with additional causes (Wang et al. 2002; Silverman 2003; Madsen et al. 2008). Mutations in RNASEL were independently linked to pancreatic cancer, both sporadic and familial (Bartsch et al. 2005), supporting a broader role of 2–5AMD deficiencies in cancer development (Madsen et al. 2008).

Breast and prostate cancers: migration and metastasis inhibition

Among strong cellular phenotypes of 2–5AMD is enhanced adhesion and motility of cells that lack the 2–5AMD nuclease, RNase L (Banerjee et al. 2015; Rath et al. 2015). It has been shown that RNase L restricts human cell migration and adhesion as robustly as microRNA miR-200, a known key suppressor of metastasis (Rath et al. 2015). Simultaneously, strong, miR-200-like inhibitory action of RNase L on metastasis was also documented in mice. When implanted into mice, RNase L-depleted human prostate cancer cells cause aggressive metastases, while wild-type cells remain primarily at the site of implantation (Banerjee et al. 2015). Based on these results, RNase L is a candidate metastasis suppressor gene that could become a drug target. Supporting this idea, examination of a panel of breast cancer microarrays in our laboratory (unpublished) shows that aggressively metastatic breast cancer cells have markedly low RNase L expression (Fig. 4) (Kang et al. 2003). [Figure 4 near here]

Figure 4.

RNase L expression is low in aggressively metastatic human breast cancer cells. Observation of RNase L expression in human breast cancer cell lines, strongly metastatic versus weekly metastatic. RNase L expression is statistically repressed in highly metastatic cell lines (Kang et al. 2003). A color version of this figure is available online.

Compendium of host RNA targets of 2–5AMD

Advances in RNA-seq methodology are revealing that, in addition to ribosomal RNAs and mRNAs, a variety of host RNAs are targets of RNase L. Although the acronym 2–5AMD was originally introduced specifically to describe RNase L-mediated cleavage of mRNAs (by analogy with nonsense-mediated decay; NMD) (Rath et al. 2019), here we refer to 2–5AMD as a combined effect of RNase L on all host RNAs. The timing and function of RNase L cleavage varies by RNA species.

Decay of mRNAs that have large 3’-UTRs with many micro-RNA binding sites

Some mRNAs are destabilized by low endogenous activity of RNase L in the absence of viruses (Rath et al. 2015; Rath et al. 2019). Studies using 2–5A to activate RNase L in cell extracts and live human cells found overlap between homeostatic mRNA targets destabilized by 2–5AMD and microRNA-regulated transcripts, particularly those enriched in miR-21, miR-30, miR-155, miR-192, and miR-200 binding sites (Rath et al. 2015). These microRNAs control cell cycle, apoptosis, carcinogenesis, and adipogenesis (Valastyan and Weinberg 2011; Zaragosi et al. 2011; Bracken et al. 2014; Filios et al. 2014), which recapitulate many recognized effects of RNase L. Thus, the homeostatic activity of RNase L resembles the effects of micro-RNAs (albeit achieved without relying on any microRNAs). Transcripts that are not targets of RNase L predominantly encode ribosomal and mitochondrial proteins (Rath et al. 2015). Reported RNase L targets were stabilized in knockout cells relative to wild-type, indicating constitutive activation under homeostatic conditions. Moreover, inverse enrichments were seen in introns compared to exons, which points to transcriptional compensation for RNase L-mediated depletion. Unlike other RNA species, individual mRNAs do not have specific site preferences, and can be cut by RNase L in many positions (Rath et al. 2019).

Global mRNA decay by 2–5AMD during acute dsRNA stress

Upon activation by high dsRNA loads, 2–5AMD rapidly depletes the vast majority of translating mRNAs in the cell, leading to translational shutdown (Clemens and Williams 1978; Burke et al. 2019; Rath et al. 2019). This activity precipitates polysome collapse and concomitant accumulation of empty 80S and 40S initiation complexes (Rath et al. 2019). Although dsRNA also inhibits global translation by activating the eIF2α kinase PKR, this mechanism is delayed compared to 2–5AMD and has a milder affect on global protein synthesis (Donovan et al. 2017).

2–5AMD rapidly depletes all translationally active basal mRNAs in the cell. The cytoplasm is subsequently repopulated with mRNAs of IFNs and ISGs (Fig. 3). In this way, 2–5AMD facilitates a translational switch from basal needs to defense protein production. Activation of sweeping mRNA destruction amidst innate immune response presents a paradox for translation of IFNs and other antiviral proteins. This paradox is reinforced by the finding that defense mRNAs are not inherently resistant to 2–5AMD (Rath et al. 2019). We found that IFNs evade 2–5AMD via a positive feedback mechanism (Rath et al. 2019). The IFN response involves positive feedback with 2–5AMD-resistatnt intermediates (IFN proteins, phospho-STAT etc). These stable components dynamically protect mRNAs encoding IFNs and ISGs from 2–5AMD (Rath et al. 2019). This mechanism is somewhat reminiscent of the superinduction phenomenon, in which cytokine expression is regulated at the transcriptional level during bacteria-induced translation block (Barry et al. 2017).

Not all transcripts are accessible to RNase L. Approximately 11% of basal mRNAs remain untouched during 2–5AMD (Burke et al. 2019; Rath et al. 2019). These transcripts contribute little to basal translation, with the ~89% of mRNAs that are decayed accounting for a disproportionate 99.7% of protein synthesis (Rath et al. 2019). Most likely the 2–5AMD-resistant, non-translating mRNAs are pre-mRNAs and mRNAs not yet exported from the nucleus (Burke et al. 2019). Another subset of resistant RNAs are non-coding RNAs, which also do not participate in translation and reside largely in the nucleus (Rath et al. 2019). Thus, 2–5AMD targets the most translationally active transcripts. It is tempting to speculate that ribosomes may assist in RNase L targeting of mRNAs, perhaps by providing the strong binding surface that could explain the Briggs-Haldane kinetics describing mRNA decay in live cells (Rath et al. 2019).

Global wipeout of mRNAs likely provides a mechanism to focus translation on defense needs during the dsRNA response. In this way 2–5AMD resembles release of translational resources by other stress responses. However, stress response pathways typically rely on special sequences, such as uORFs encoded in 5’-UTRs of privileged mRNAs (Sidrauski et al. 2015). 2–5AMD does not require such licensing mechanisms and achieves translational reprogramming using the transcriptional dynamics of mRNAs induced during the IFN response.

18S and 28S ribosomal RNAs

For decades it has been accepted that RNase L-mediated cleavage of rRNA renders ribosomes non-functional and explains the global loss of protein synthesis in 2–5A-treated extracts (Goswami and Sharma 1984; Iordanov et al. 2000; Cooper, Banerjee, et al. 2014). The high reproducibility and consistency of rRNA cleavage was adapted as the most common readout for RNase L activation (D H. Wreschner et al. 1981; D. H. Wreschner et al. 1981; Silverman et al. 1983; Iordanov et al. 2000; Rath et al. 2019). However, the progression of rRNA cleavage does not correlate with that of translation loss (Toots et al. 1988; Donovan et al. 2017), and cleaved ribosomes are fully translationally competent (Donovan et al. 2017; Rath et al. 2019). When mapped to the ribosome structure, the 18S and 28S rRNA cut sites are found in peripheral surface loops, further supporting opportunistic rather than biologically relevant cleavage (Rath et al. 2019). Thus, the ribosome is a marker, but not a biological target of 2–5AMD.

Small RNAs: tRNAs and Y-RNAs

Among the most abundant cleaved substrates of RNase L are tRNAs and Y-RNAs (Donovan et al. 2017; Rath et al. 2019). Analogous to rRNAs, tRNAs and Y-RNAs are cut at a few preferred sites. RNase L specifically cuts within the anticodon stem-loops of tRNAs, primarily targeting tRNA-His, tRNA-Pro, and tRNA-Gln (Donovan et al. 2017). Although this activity might produce signaling fragments implicated in cancer proliferation (Ivanov et al. 2011; Dhahbi et al. 2014; Honda et al. 2015), to the best of our knowledge evidence of such a signaling role for tRNA products of 2–5AMD is unavailable. Fragments are readily detectable by the RtcB-qPCR method, although the level of tRNAs remain unchanged, except under conditions of very strong and prolonged RNase L activation (Donovan et al. 2017). It is therefore unlikely that tRNA depletion significantly contributes to translational arrest. Cleavage of Y-RNAs is similarly site-specific (Donovan et al. 2017). The function of these sites and short Y-RNA fragments are unknown, although similar cleaved Y-RNAs have been detected in cancer patients and in apoptotic cells (Rutjes et al. 1999; Dhahbi et al. 2014; Kowalski and Krude 2015). In summary, current data do not attribute a biological function to tRNA and Y-RNA cleavage.

Host dsRNA sensing by 2–5AMD

Loss of ADAR1 or DNA methylation activates 2–5AMD

Although dsRNA has historically been associated with viruses, deleterious inflammatory effects have recently been linked to accumulation of endogenous dsRNAs upon disruption of DNA or RNA maintenance mechanisms (White et al. 2014; Ahmad et al. 2018). Low basal levels of endo-dsRNAs can arise from genetically encoded abundant retroelements such as Alu repeats. These dsRNAs are cleared by the dsRNA-editing enzyme ADAR1, which disrupts Alu RNA duplexes (Rice et al. 2012). Loss of ADAR1 activity causes cell death, and leads to embryonic lethality in mice or the autoimmune Aicardi-Goutières syndrome in humans (Rice et al. 2012; Liddicoat et al. 2015; Y. Li et al. 2017). The lethality of ADAR1 knockouts can be prevented by deleting either the IFN-inducing dsRNA sensor MDA5 or its downstream adapter MAVS, indicating a key role of IFN signaling in sensing endo-dsRNAs (Mannion et al. 2014; Liddicoat et al. 2015; Pestal et al. 2015). Independently, the cell-lethal phenotype of ADAR1 deficiency can be rescued by deletion of RNase L (Y. Li et al. 2017). RNase L is located downstream of MDA5/MAVS signaling via IFN dependence of OAS1, OAS2 and OAS3 expression. Thus, all lethality-rescuing knockouts described above belong to the same axis of endo-dsRNA sensing. These studies suggest that 2–5AMD is the executor of cell death upon loss of ADAR1 and point to endo-dsRNAs as cell-intrinsic physiological activators of 2–5AMD.

Separate from ADAR1 activity, abnormal accumulation of endo-dsRNAs can arise from derepression of genetic repeats caused by DNA demethylating drugs. One such agent, 5-Aza-CdR, is a US Food and Drug Administration (FDA)-approved drug that clears cancerous cells presumably via IFN-mediated apoptosis (Leonova et al. 2013; Chiappinelli et al. 2015; Roulois et al. 2015). The cytotoxic mechanism of this compound was recently linked to activation of RNase L (Banerjee et al. 2019), which supports the importance of 2–5AMD in endo-dsRNA surveillance and provides another reminder of the therapeutic potential of this pathway.

2–5AMD is a surveillance system for paired retroelements in introns and intergenic RNAs

The full spectrum of molecular mechanisms responsible for endo-dsRNA biogenesis is unknown and is currently actively explored. Our lab discovered that introns and intergenic transcripts provide a major source of endo-dsRNA formed by widespread inverted L1, L2 and Alu repeats (Chitrakar et al. 2021). This finding was made serendipitously while investigating the hypothesis that the dsRNA-cleaving endonuclease DICER1, like ADAR1, is a suppressor of endo-dsRNAs. To our surprise, depletion of DICER1 by multiple methods, including CRISPR knockouts, did not cause endo-dsRNA accumulation. However, phosphorothioate (PS) DNA used as antisense oligonucleotides during investigation of DICER1 were strong triggers of 2–5AMD (Chitrakar et al. 2021).

We determined that PS DNAs block nuclear decay of introns and intergenic RNAs, which subsequently spillover into the cytoplasm. RNA-seq analysis of these RNAs revealed an abundance of intronic and intergenic inverted retroelements (IIIRs), which form ≥50bp dsRNA duplexes sensed by OAS3 (Fig. 5). [Figure 5 near here] We noticed that while IIIRs activate 2–5AMD and PKR, they do not activate RIG-I/MDA5 sensing and IFN production (Chitrakar et al. 2021). Thus, 2–5AMD and RIG-I/MDA5/IFN axes of innate immunity appear to be tuned toward different functions. Whereas RIG-I/MDA5/IFN signaling is triggered primarily by viral dsRNAs, 2–5AMD is activated by dsRNA of both viruses and host. This separation of functions can be expected considering that the RIG-I/MDA5/IFN arm serves to send IFNs from infected cells to alert naïve cells of the risk of viral infection. Introns and intergenic dsRNAs (IIIRs) do not pose an infection risk to surrounding cells. Thus, IIIRs activate the cell-intrinsic defense mechanism of 2–5AMD. Within this hypothesis, the insensitivity of the RIG-I/MDA5/IFN axis to endo-dsRNAs could be the result of evolution, which eliminated unnecessary paracrine innate immune responses to host dsRNA.

Figure 5.

2–5AMD senses and degrades human introns and intergenic RNAs that encode an abundance of base-paired retroelements (Intronic and Intergenic Inverted Retroelements; IIIRs). Disrupted homeostasis of introns and intergenic RNAs in the nucleus causes IIIR leakage into the cytoplasm. These RNAs carry abundant endo-dsRNAs, which activate 2–5AMD through OAS3 (Chitrakar et al. 2021). A color version of this figure is available online.

In addition to demonstrating that 2–5AMD is a sensor of IIIRs, this work found that 2–5AMD is also a cleavage and decay system for IIIR-carrying host RNAs. IIIRs contain both double-stranded and single-stranded regions, and are thus perfectly matched for the recognition and destruction arms of 2–5AMD. Of note, discovery of the IIIR/RNase L connection caused us to analyze the nucleotide composition of cleaved introns, in which we found they are strongly enriched for a single nucleotide: U. Based on this finding, it has been proposed the Un^N consensus sequence of RNase L may signify 2–5AMD evolution specifically towards recognition of IIIR-carrying host RNAs.

Conclusion

As a conserved pathway of dsRNA surveillance present in the cells of all mammals, 2–5AMD is important for both antiviral innate immunity and host RNA regulation. These two facets are closely interlinked; antiviral action of 2–5AMD almost certainly arises from degradation of viral RNAs, as well as reprogramming of host cells, particularly from reorganization of the translational landscape toward synthesis of antiviral proteins.

The roles of 2–5AMD in regulation of host RNAs have remained elusive for decades largely due to a lack of tools needed for these studies. Advances in RNA-seq, such as pioneering work on cyclic phosphate sequencing by others (Cooper, Jha, et al. 2014) and by our laboratory (Donovan et al. 2017; Rath et al. 2019) began to uncover the landscape of human RNA targets (Fig. 6). [Figure 6 near here] Some of the identified targets, such as rRNAs and small RNAs appear to be “bystander hits” that do not bear any function. In contrast, almost all human mRNAs, introns that carry IIIRs and intergenic RNAs that carry IIIRs are emerging as biologically relevant substrates of 2–5AMD (and in the case of IIIRs, also as activators).

Figure 6.

Spectrum of currently known RNA targets of RNase L and 2–5AMD. A color version of this figure is available online.

IIIRs do not activate the dsRNA sensing arms of innate immunity that are responsible for synthesis of IFNs. Thus, 2–5AMD is an unusual axis of dsRNA sensing, which serves not only for antiviral defense, but also for surveillance and clearance of rogue endo-dsRNAs that escape their normal degradation fate in the nucleus. We posit that leakage of these IIIR-carrying RNAs provides a signal of potentially harmful cellular stresses from disrupted RNA homeostasis. 2–5AMD provides a corrective response that clears these RNAs when possible, or executes apoptosis as a terminal measure.