A-to-I RNA Editing: Recent News and Residual Mysteries*

Adenosine to inosine modification in pre-mRNA, with inosine acting as guanosine during translation, was the most recent type of RNA editing to be discovered (1). At present, it appears to be the most widespread type of nuclear pre-mRNA editing in higher eukaryotes (6, 14, 25). Adenosine deaminases acting on RNA (ADARs), the enzymes responsible for conversion of A-to-I in double-stranded (ds) RNA, were first noticed as cellular RNA unwinding activity (2, 3), as they lead to destabilization of RNA duplexes by introducing I U mismatches. Since the initial cloning of the first RNA-specific adenosine deaminase, ADAR1 (4, 5), a family of A-to-I editing enzymes (ADAR1–3) has emerged (Fig. 1) (6–10). Both ADAR1 and ADAR2 are detected in many tissues, whereas ADAR3 is expressed only in restricted regions of the brain (4, 5, 7–10). Members of the ADAR gene family share common structural features such as two or three repeats of a dsRNA binding motif and a separate deaminase or catalytic domain (4, 5). Certain structural features are unique to particular ADAR members. For instance, ADAR1 contains two Z-DNA binding motifs (11), whereas ADAR3 includes an arginine-rich single-stranded RNA binding domain at the N terminus (9, 10) (Fig. 1). In vitro RNA editing studies have revealed a significant difference in site selectivity displayed by ADAR1 and ADAR2 (7, 8), whereas no RNA editing activity has been demonstrated yet for ADAR3 (9, 10). Orthologues of the mammalian ADARs have also been characterized and cloned from fruit fly (12) and worms (6) (Fig. 1), and related genes have been identified in fish genomes (13). Subsequently, the subfamily of tRNAspecific A-to-I editing enzymes (ADAT1–3) was uncovered based on their sequence homologies to ADARs. ADAT3 is an adenosine deaminase in yeast shown to form an enzymatically active heterodimer with ADAT2 (14). ADAT1 and ADAT2 are conserved from yeast to humans (15), and it is believed that an ADAT1-like adenosine deaminase is the ancestor of the current A-to-I editing enzymes (14).

Adenosine to inosine modification in pre-mRNA, with inosine acting as guanosine during translation, was the most recent type of RNA editing to be discovered (1). At present, it appears to be the most widespread type of nuclear pre-mRNA editing in higher eukaryotes (6,14,25). Adenosine deaminases acting on RNA (ADARs), 1 the enzymes responsible for conversion of A-to-I in double-stranded (ds) RNA, were first noticed as cellular RNA unwinding activity (2,3), as they lead to destabilization of RNA duplexes by introducing I⅐U mismatches. Since the initial cloning of the first RNA-specific adenosine deaminase, ADAR1 (4,5), a family of A-to-I editing enzymes (ADAR1-3) has emerged ( Fig. 1) (6 -10). Both ADAR1 and ADAR2 are detected in many tissues, whereas ADAR3 is expressed only in restricted regions of the brain (4,5,(7)(8)(9)(10). Members of the ADAR gene family share common structural features such as two or three repeats of a dsRNA binding motif and a separate deaminase or catalytic domain (4,5). Certain structural features are unique to particular ADAR members. For instance, ADAR1 contains two Z-DNA binding motifs (11), whereas ADAR3 includes an arginine-rich single-stranded RNA binding domain at the N terminus (9, 10) ( Fig. 1). In vitro RNA editing studies have revealed a significant difference in site selectivity displayed by ADAR1 and ADAR2 (7,8), whereas no RNA editing activity has been demonstrated yet for ADAR3 (9,10). Orthologues of the mammalian ADARs have also been characterized and cloned from fruit fly (12) and worms (6) (Fig. 1), and related genes have been identified in fish genomes (13). Subsequently, the subfamily of tRNAspecific A-to-I editing enzymes (ADAT1-3) was uncovered based on their sequence homologies to ADARs. ADAT3 is an adenosine deaminase in yeast shown to form an enzymatically active heterodimer with ADAT2 (14). ADAT1 and ADAT2 are conserved from yeast to humans (15), and it is believed that an ADAT1-like adenosine deaminase is the ancestor of the current A-to-I editing enzymes (14).

ADAR Substrates
A partially double-stranded RNA structure involving exonic and intronic sequences seems essential for editing to take place (16 -18). By far, most of the genes found to undergo A-to-I RNA editing in mammals, Drosophila melanogaster, Caenorhabditis elegans, and squid, are expressed in the nervous system. Prominent examples are transcripts of the mammalian glutamate receptors (GluRs) and the serotonin receptor subunit 2C (5-HT 2C R), where deamination of exonic adenosines leads to single amino acid changes in the resulting proteins often with profound consequences for receptor function (see Fig. 2). A-to-I editing was also found to occur in non-coding regions of pre-mRNAs (17,19), most notably in the transcripts of the editing enzyme ADAR2 (19). By editing its own transcripts, an alternative splice acceptor site is created in intron 1, leading to alternative splicing resulting in a nonfunctional protein (19). Interestingly, the only ADAR-like gene in Drosophila, DADAR, which is most closely related to the mammalian ADAR2 protein, is also self-edited but within the coding region (12).
A-to-I RNA editing of the antigenome RNA of hepatitis delta virus, which replaces a translational stop signal with a tryptophan codon (UAG to UGG) (20) is an obligatory step in the hepatitis delta virus life cycle. In addition, an intronic branch site adenosine of SH-PTP1 tyrosine phosphatase has been proposed to be edited by ADAR (21). No examples have yet been identified of A-to-I editing creating a translational start codon (ATA to ATG) or changing a consensus polyadenylation signal (AAUAAA). However, such discoveries are likely.
In contrast to its exquisite selectivity for adenosine residues deaminated in a natural RNA substrate with a characteristic hairpin structure including loops and bulges, ADARs almost randomly modify many adenosines (ϳ50%) within a long completely complementary dsRNA. One source of such extended dsRNA molecules is viral RNAs that are synthesized during replication and also transcription of viral genomes. These viruses, such as measles, seem to encounter ADARs and under certain circumstances are converted to hypermodified dsRNAs with many I⅐U mismatched base pairs, denoted I-dsRNAs (6,22).
The physiological importance of A-to-I RNA editing has been confirmed by analysis of ADAR null mutation phenotypes. For instance, ADAR2 Ϫ/Ϫ mice died young after repeated episodes of epileptic seizures caused by underediting of GluR-B RNA at the Q/R site, which controls Ca 2ϩ permeability of the resulting ion channel (23). Lethal phenotypes including dyserythropoietic defects were observed in chimeric mouse embryos derived from ADAR1 ϩ/Ϫ embryonic stem cells. However, the possibility that antisense transcripts derived from the ADAR1 targeted allele may contribute to the observed phenotypes has not been ruled out (24). The genetic inactivation of DADAR in Drosophila resulted in flies with extreme behavioral deficits and neurological symptoms such as paralysis, locomotor uncoordination, and tremors, which increased in severity with age (25). The observed phenotype is likely to be the result of the total lack of A-to-I RNA editing in the brains of the mutant flies (25). In light of these findings it can be speculated that dysfunction of A-to-I RNA editing in humans could be cause for or contribute to certain pathophysiological processes and diseases.

A-to-I RNA Editing of Repetitive Sequences within
Introns and 3-UTRs The measured amount of inosine, especially in the brain, is substantial and suggests that many genes undergo A-to-I editing (26). However, since the initial discovery of A-to-I RNA editing (1), only a few edited genes have been identified from discrepancies between the mRNA (cDNA) and genomic sequences, and their discovery was entirely serendipitous. A recently devised method for the enrichment and cloning of inosine-containing RNAs provides the first comprehensive search tool and has lead to the isolation of new edited RNA sequences; 10 from C. elegans and 19 from human brain (27). They include transcripts from diverse genes such as tankyrase, NADH-ubiquinone oxidoreductase, and a snoRNA pre-cursor (27). All of the identified RNAs have A-to-I conversions in non-coding regions, mainly in 3Ј-UTR and intronic sequences, often involving repetitive elements such as Alu and LINE1 repeats (27). Potentially, the modifications in 3Ј-UTR secondary structures could alter the stability, transport, or translation of the mRNA, and intronic editing might influence the kinetics or efficiency of splicing (Fig. 2).
Because no novel or known genes with editing sites in coding regions were isolated during the unbiased screening (27), it might be that the editing events that were discovered previously within the coding regions represent exceptional cases where, due to the beneficial function of the edited gene variant, positive selection has led to an increase in editing rates. It will be interesting to see if exhaustive screening of the enriched RNA pools will eventually lead to an accurate estimate of the total number of edited genes in the mammalian genome and the ratio of coding to non-coding editing sites. However, two dozen or so newly identified genes do not explain the discrepancy between significant amounts of inosine detectable within the poly(A) fraction of cellular RNA on one hand and the paucity of newly identified targets on the other. What could explain this seeming contradiction? One possibility is that there are a small number of genes that get edited extensively, contributing most of the measurable inosine. Such a mode of modification by ADARs, termed hyperediting, has previously been reported for the host-induced editing of viral RNAs (6). Indeed, some of the substrates in C. elegans and human are modified at multiple positions (27). At the same time, the observed amount of inosine in mRNA (up to 1 inosine in every 17,000 nucleotides (26)) could also be explained if many genes are edited at few positions but with low efficiency. This would make it much more difficult to identify individual editing events, because only a small fraction of the gene's transcripts carry the modification. Also, it is possible that the RNA editing machinery continuously probes nascent transcripts for new editing sites. In light of the few constraints on non-coding sequences (i.e. intronic, 5Ј-UTR, and 3Ј-UTR), novel, editable RNA structures could continue to appear at a significant rate within the pool of primary transcripts. Because such probing is a non-directional and initially low rate phenomenon, it would be difficult to detect these editing events and to distinguish them from reverse transcription or sequencing errors. One can imagine a scenario where opportunistic editing of newly formed RNA structures generates a constant "inosine background" that could account for most of the inosine measured. This is supported by the fact that the abundance of repetitive sequences in the mammalian genomes, some still active mobile elements, leads to a high number of inverted repeats embedded within expressed sequences. The presence of repetitive elements in most of the newly identified editing substrates certainly supports the latter scenario. It is tempting to speculate that ADARs might be involved in the regulation of transposons and repetitive elements or possibly in the phenomena of gene imprinting and X chromosome inactivation (Fig. 3). Repetitive sequences and transcription of sense and antisense strand RNA, both with the potential to form dsRNA, have been proposed to play a critical role in these epigenetic modification processes (28).

More Recoded Transcripts and Consequences
Several new A-to-I RNA editing sites occurring within the coding regions of neurotransmitter receptors and channels have been reported recently. In D. melanogaster, transcripts for the putative nicotinic acetylcholine receptor D␣6 were found to be edited at several positions within the ligand binding region of the receptor channel (29). This finding adds another neurotransmitter receptor to the growing list of genes targeted for A-to-I editing that reside in the nervous system. Furthermore, a novel case of invertebrate A-to-I editing characterized in squid also involved a gene important for neurotransmission: the delayed rectifier K ϩ channel (SqKv1.1A) (30). Thirteen editing sites have been identified, which are all positioned in membrane-spanning segments with influence on the channel's gating behavior. In addition, one editing site located in a domain important for subunit assembly was shown to regulate tetramerization of the subunits (30). Curiously, the fact that eight valine residues are introduced by editing is reminiscent of the cluster of adaptive mutations occurring in certain fish genes to maintain their function at low temperature, suggesting a possible temperature-dependent regulation of editing (30,31).
An important function for channel subunit trafficking and assembly has recently been traced to the same editing position in GluR-B, which dominantly regulates channel gating (32) and also is important for efficient further processing of its pre-mRNA (23). The edited GluR-B subunits (harboring an arginine at the Q/R site) are retained within the endoplasmic reticulum until incorporation of the protein into heteromeric receptors (32). Thus, the Q/R site of GluR-B is the molecular determinant for a set of critical receptor functions, and all of them rely on the quantitative recoding of the glutamine codon by RNA editing (31,32).

Deregulation of A-to-I RNA Editing
Could local or systemic A-to-I editing activity be the cause of phenotypes that correspond to known diseases in humans, in particular, with respect to brain disorders and neurodegenerative processes? Editing of 5-HT 2C receptor pre-mRNA at a total of five sites leads to replacement of three amino acid residues in the second intracellular domain, resulting in dramatic alterations in G-protein coupling functions of the receptor (16). Among all RNA editing isoforms detected, those carrying Gly at position 158 displayed the most dramatically decreased G-protein coupling activities (33). Interestingly, substantial alterations in the 5-HT 2C receptor RNA editing levels and concomitant increase in the expression of the receptor isoform carrying Gly at position 158 were observed in the prefrontal cortex of suicide victims (34). The measured changes in the expression pattern of the receptor isoforms are anticipated to have resulted in dampening of the overall efficacy of serotonergic neurotransmission in these depressed individuals (34). Interestingly, the treatment of mice with an antidepressant Prozac TM (serotonin re-uptake inhibitor) induced changes in editing levels that were converse to those observed in suicide victims (34). These preliminary results, together with two related reports (35,36), raise the possibility that changes in RNA editing of the 5-HT 2C R may be involved in the etiology of neuropsychiatric disorders and also that dynamic changes in editing levels could be induced by various psychoactive drugs and antidepressants. Due to the high variance in 5-HT 2C R editing within different cell types of the brain and our present ignorance about how RNA editing is regulated, larger data sets will be needed to assess the significance of these findings for those disease states and to unravel the causal relationships between changes in serotonin levels and alteration in editing (31,34).
With respect to the editing of glutamate receptors in the central nervous system, mounting evidence points to a link between abnormal editing levels and seizure vulnerability. Epileptic seizures are the major phenotypic features of transgenic mice with underediting at the GluR-B Q/R site, mainly due to the increased macroscopic AMPA-R conductance in these mice (23,37,38). Also, when eliminating Q/R site editing in the GluR-6 kainate receptor subunit, the mutant mice exhibit an increased susceptibility to kainate-induced seizures (39). Underediting of the GluR-B Q/R site might, among other reasons, also be an explanation for the occurrence of epileptic seizures in patients with glioblastomas (malignant tumors of glia cells) (40). This is suggested from findings that in these tumors the RNA editing activity of ADAR2, the enzyme crucial for GluR-B Q/R site editing (23), is significantly depressed (40). Tumor-intrinsic Q/R site editing was decreased to levels comparable with those measured in the mouse models. As in the other cases described, it will be important to determine whether the observed changes in RNA editing patterns are consequences of the disease state or are involved in driving disease progression. In either case, the clinical symptoms of depression (5-HT 2C R) or epilepsy (GluR) might be treated more effectively if account is taken of the concomitant deregulation in RNA editing.

Determinants for A-to-I RNA Editing Efficacy
What are the determinants that could cause a significant change in the rate of editing at a specific site? Currently, we can neither extract that information from known ADAR substrates nor predict whether a given RNA sequence will be edited in vivo. Apart from the requirement for double-stranded regions and certain 5Ј and 3Ј neighbor preferences that allow for the assessment of editing probability in vitro (6), additional features of the substrates and additional protein factors appear to be involved in vivo. A major consideration for what happens in vivo will be how accessible the RNAs actually are for binding and modification by these enzymes.
Because of the involvement of intronic sequences in the dsRNA structures essential for editing of 5-HT 2C R and GluR pre-mRNA, A-to-I RNA editing must occur before or simultaneously with splicing. Therefore, it is likely that RNA editing and splicing machineries interact with each other (Fig. 4). In the brains of ADAR2 Ϫ/Ϫ mice the almost complete absence of GluR-B RNA editing at the Q/R site due to the total loss of the editing enzyme ADAR2 indeed changed the kinetics of splicing (10-fold reduction) (23), confirming the close relationship between the processes of RNA editing and splicing. In fact, the minute amounts of edited GluR-B primary transcripts undergo preferential splicing as judged from the level of editing measured in mature GluR-B mRNA (40% Arg compared with 10% in GluR-B pre-mRNA (23)). Preferential splicing of edited transcripts has also been observed in the case of ADAR2 selfediting (40).
A growing number of snoRNAs are found that guide the methylation and uridylation of rRNAs, tRNAs, snRNAs, and probably mRNAs through the formation of a short RNA duplex with their target sequences (reviewed in Ref. 41). One such mammalian snoRNA, specifically expressed in the brain and paternally imprinted, harbors a guide sequence specific for 5-HT 2C R mRNA (42). Intriguingly, it is predicted to guide the 2ЈO-methylation of the adenosine residue, which also undergoes A-to-I editing (C site). The modification is expected to interfere with the ADAR deamination reaction (Fig. 4). Again, the relative localization of the methylation and editing machineries may be the most important factor in determining the kind and rate of modifications introduced.
The stability of a dsRNA structure, especially one with many mismatched base pairs and bulges, may be sensitive to subtle change of body temperature (31); a feature of particular importance for poikilotherms. Opening of the dsRNA structure involved in A-to-I RNA editing, possibly regulated by various dsRNA helicases and annealing activities, appears to be another critical step (Fig. 4). Mutation in RNA helicase A, a specific ATP-dependent dsRNA helicase, blocks resolution of the dsRNA structure and results in the occurrence of a "splicing catastrophe" or aberrant splicing and skipping of exons of the Drosophila para-Na ϩ channel transcripts at the region surrounding the editing sites (43). This indicates that the overall editing efficiency of a given substrate RNA may significantly change depending on the stability of the dsRNA structure and/or its splicing rate (43). Furthermore, it has been reported recently that both ADAR1 and ADAR2 proteins are complexed with large nuclear ribonucleoprotein particles that constitute all known factors required for pre-mRNA splicing (44). This suggests an interesting possibility that certain physiological conditions (e.g. fever or hypothermia) or pharmacotherapies may affect directly or indirectly the availability and abundance of RNA helicase A or various splicing factors as well as ADAR expression levels, which all in turn regulate RNA editing levels (33,43) (Fig. 4).

ADAR1 Functions in Processes Other Than Site-selective
Pre-mRNA Editing A number of unique structural and functional features set ADAR1 apart from the other known ADAR gene family members (Fig. 1). Although in vitro ADAR1 has been shown to be capable of carrying out editing of certain sites such as the A and B sites in 5-HT 2C R transcripts (10,45) and the intronic hotspot1 in GluR-B (7, 23), its involvement in vivo in site-selective editing of nuclear pre-mRNA has not yet been demonstrated. Its unique domain architecture, expression, and localization indicate that ADAR1 might in fact have in vivo functions other than nuclear pre-mRNA editing. Two major ADAR1 variants, differing in the lengths of their N-terminal sequence, are expressed due to the use of different promoters. The interferon-inducible, full-length ADAR1 150-kDa protein is present in both the cytoplasm and nucleus, whereas the constitutively expressed 110-kDa isoform is exclusively nuclear (46) (Fig. 3). A role for the 150-kDa ADAR1 in the cellular, interferon-mediated antiviral response has been postulated (6,46). The 150-kDa ADAR1 includes a unique N-terminal domain that has been shown to bind DNA in the Z-conformation with high affinity. One supposition has been that Z-DNA binding helps to localize ADAR1 to actively transcribed target genes (11), and Z-binding proteins have recently been implicated in the modulation of chromatin remodeling (47). Within the same N-terminal region of the 150-kDa ADAR1 protein a nuclear export signal was found, which leads to nucleocytoplasmic shuttling of the enzyme (48).
Any RNA that is at least partially double-stranded represents a potential substrate for A-to-I editing. This raises the question of whether other cellular processes that rely on double-stranded RNA molecules as functional entities, such as RNAi, will be influenced by A-to-I RNA editing (Fig. 3). RNAi is a post-transcriptional process whereby dsRNA induces the homology-dependent degradation of cognate mRNA in the cytoplasm. RNAi may be involved in other processes such as chromatin remodeling in the nucleus. For the RNAi mechanism operating in the cytoplasm, the most significant member of the ADAR gene family is the 150-kDa form of ADAR1. The potency of the trigger dsRNAs mediating RNAi has been reported to be significantly reduced following A-to-I editing in vitro by ADAR proteins (49). Thus it is possible that ADAR1 may interact with precursors of the recently identified class of short RNAs such as micro-RNAs and small transient RNAs and affect their synthesis (Fig. 3).

Cellular Activities Specific for Inosine-containing RNAs
The studies described above may indicate the presence of a mechanism to eliminate or deal with dsRNA, either viral or cellular, after hypermodification by ADARs. Several cellular activities that specifically act on inosine-containing RNA (I-RNA or I-dsRNA) have been identified (Fig. 3). A cytoplasmic ribonuclease activity that specifically cleaves I-dsRNA (I-RNase) has been reported recently. Cleavage occurs at specific sites with dsRNA consisting of alternate I⅐U and U⅐I base pairs, presumably introduced by p150 ADAR1 (50). In addition, a nuclear-localized protein p54 nrb capable of binding to hypermodified I-RNAs has also been reported (22). The biological function of p54 nrb , which forms a protein complex with two other proteins, PSF (splicing factor) and matrix 3 (nuclear matrix protein), is currently not understood (22). Taken together, these cellular activities specific for inosine-containing RNAs represent the "smoking gun" of frequent encounters between ADARs and endogenous or viral dsRNA and provide evidence for the existence of cellular mechanism(s) to process hypermodified RNA molecules.