Requirement of dimerization for RNA editing activity of adenosine deaminases acting on RNA.

Adenosine deaminases acting on RNA (ADAR) convert adenosine residues into inosines in double-stranded RNA. Three vertebrate ADAR gene family members, ADAR1, ADAR2, and ADAR3, have been identified. The catalytic domain of all three ADAR gene family members is very similar to that of Escherichia coli cytidine deaminase and APOBEC-1. Homodimerization is essential for the enzyme activity of those cytidine deaminases. In this study, we investigated the formation of complexes between differentially epitope-tagged ADAR monomers by sequential affinity chromatography and size exclusion column chromatography. Both ADAR1 and ADAR2 form a stable enzymatically active homodimer complex, whereas ADAR3 remains as a monomeric, enzymatically inactive form. No heterodimer complex formation among different ADAR gene family members was detected. Analysis of HeLa and mouse brain nuclear extracts suggested that endogenous ADAR1 and ADAR2 both form a homodimer complex. Interestingly, endogenous ADAR3 also appears to form a homodimer complex, indicating the presence of a brain-specific mechanism for ADAR3 dimerization. Homodimer formation may be necessary for ADAR to act as active deaminases. Analysis of dimer complexes consisting of one wild-type and one mutant monomer suggests functional interactions between the two subunits during site-selective RNA editing.

One type of RNA editing converts adenosine residues into inosine within the double-stranded RNA (dsRNA) 1 region of substrate RNAs (1)(2)(3). Because inosine is treated as guanosine by the translational machinery, this A-to-I editing could lead to functional alterations of the affected genes. For instance, A-to-I RNA editing results in the expression of editing isoforms of glutamate receptor (GluR) ion channel subunits (4,5) and serotonin 2C subtype receptors (5-HT 2C R) (6). Editing of the so-called "Q/R" site of the ␣-amino-3-hydroxy-5-methylisoxazole-4-propionic acid GluR-B subunit dramatically decreases the Ca 2ϩ permeability of the channel (7). Substantial reduction in G-protein coupling efficiency is noted with A-to-I editing of 5-HT 2C R RNA at five positions (A to E sites) located in the intracellular II loop region (6, 8 -10). A-to-I RNA editing also occurs in non-coding sequences. Editing of its own intron sequence by adenosine deaminase acting on RNA (ADAR) 2 creates an alternative splice acceptor site leading to synthesis of a truncated translation product, which may be a negative feedback mechanism to regulate the activity of ADAR2 (11). In all these examples, a dsRNA structure formed between the exonic sequences containing an editing site(s) and downstream or upstream intronic sequences has been proven to be essential for editing (4 -6, 12). Systematic search with a recently devised method for cloning of inosine-containing RNAs has led to identification of more than two dozen editing sites occurring in the intron and 3Ј-untranslated regions of new target genes. A-to-I RNA editing of these non-coding regions may affect the splicing rate, the translation efficacy, or stability of the edited mRNAs (13). Furthermore, the intronic and untranslated region sequences subjected to A-to-I RNA editing often contain common repetitive elements such as Alu and LINE1 repeats forming a long dsRNA structure, raising the possibility that A-to-I RNA editing may be involved in a mechanism regulating the very abundant repetitive sequences in mammalian genomes (2,3,13). Finally, A-to-I RNA editing of dsRNAs derived from transgenes appears to prevent silencing of the transgenes regulated by RNA interference, revealing the potential intersection of the two mechanisms, RNA editing and RNA interference both evolved to deal with dsRNA (14).
Members of the ADAR gene family have been implicated as the enzymes responsible for A-to-I RNA editing. Three separate mammalian gene family members (ADAR1 to ADAR3) have been identified (15)(16)(17)(18)(19)(20)(21)(22). Data base search has identified corresponding fish ADARs revealing the conservation of the complete set of ADAR gene family members in vertebrates through evolution (23,24). In invertebrates, a single Drosophila dADAR, very similar to mammalian ADAR2 (25), and two less conserved Caenorhabditis elegans c.e.ADAR1 and c.e.ADAR2 have been identified (15,26). Mammalian ADAR1 and ADAR2 are detected ubiquitously (15, 16, 18 -20), whereas the expression of mammalian ADAR3, Drosophila dADAR, and C. elegans c.e.ADAR1 is restricted mainly to nervous systems (21,22,25,27). Analysis of ADAR null mutation phenotypes has revealed the importance of A-to-I RNA editing. Flies with a null mutation of dADAR, although viable, display defective locomotion and behavior accompanied by various neurological and ana-tomical changes in the brain. This phenotype is most likely because of the lack of editing in the transcripts of several known target genes such as cac Ca 2ϩ channel and para Na ϩ channel (25). C. elegans strains containing homozygous deletions of both c.e.ADAR1 and c.e.ADAR2 genes show defects in chemotaxis, whereas aberrant development of the vulva is occasionally detected with worms lacking c.e.ADAR1 (27). Mice with a homozygous ADAR2 null mutation die several weeks after birth with repeated episodes of epileptic seizures because of underediting of GluR-B RNA at the Q/R site, a major target of ADAR2 (28). Chimeric mouse embryos derived from ADAR1 ϩ/Ϫ ES cells die at the midgestation stage with a phenotype indicative of dyserythropoietic defects (29). It has not yet been ruled out that antisense effects generated by transcripts derived from the ADAR1-targeted allele may contribute to the observed embryonic lethal phenotype (29).
Purified recombinant ADAR1 and ADAR2 proteins displayed in vitro their distinctive editing site selectivity with known RNA substrates (18 -20, 30, 31). For instance, ADAR2 edits almost exclusively the D site of 5-HT 2C R and the Q/R site of GluR-B RNA, whereas ADAR1 barely edits these sites. However, ADAR1 selectively edits the A and B sites of 5-HT 2C R RNA and the intronic hot spot (ϩ60 site) of GluR-B RNA. The result of in vitro editing studies indicate a significant difference among ADAR gene family members in their interaction with substrate RNA. Specific structural features of the dsRNA binding domains and their N-terminal regions may form the molecular basis of this editing site selectivity. There are only two dsRNA binding motif repeats in the RNA binding domain of ADAR2 and ADAR3, in contrast to three dsRNA binding motifs in ADAR1. ADAR2 lacks a part of the N terminus region of ADAR1, just upstream of its RNA binding domain, where ADAR1 contains two repeats of a Z-DNA binding motif (32). ADAR3 has a unique N-terminal region containing the arginine-rich R domain (21,22). Alternatively, the deaminase domains and relatively divergent C-terminal regions of ADAR gene family members may also contribute to the difference observed in their RNA editing site selectivity as indicated by the studies of domain exchange between ADAR1 and ADAR2 (21,33).
A longstanding question with regard to the enzymatic activities of ADARs is whether they act as monomeric or oligomeric forms and whether oligomerization plays a role in the siteselective editing mechanism. The catalytic domain of ADAR is very similar to that of the cytidine deaminase gene family (1,2,15). E. coli cytidine deaminase forms a homodimer (34), as does APOBEC-1, another cytidine deaminase involved in C-to-U RNA editing of apolipoprotein B mRNAs (35)(36)(37). In both cases homodimerization is required for enzymatic activity (34 -37). It is possible that the RNA editing site selectivity observed with ADAR1 and ADAR2 is dependent on their state of oligomerization. Curiously, the third member of the ADAR gene family, ADAR3, is incapable of editing all known sites of GluR-B and 5-HT 2C R RNAs. The lack of enzymatic activity may be related to its oligomerization state. In the present studies, we have investigated whether ADAR gene family members undergo oligomerization. In addition, we have examined the possible formation of heteromeric oligomers among different ADAR gene family members.
ADAR Expression Constructs-A 6His epitope tag sequence was introduced into the N termini of the existing FLAG epitope-tagged expression constructs, pBac-F-ADAR1, pBac-F-ADAR2a, and pBac-F-ADAR3 (20,22,38). Different regions of human ADAR1 (amino acids 2-72), ADAR2a (amino acids 2-35), or ADAR3 (amino acids 2-100) were prepared by PCR amplification of human ADAR1 (15), ADAR2a (20), and ADAR3 cDNA plasmids (22) using a set of oligonucleotide primers designed to create NotI and BamHI restriction sites. Not-H-ADAR1UP and H-ADAR1DW primers were used for PCR amplification of the ADAR1 sequence, Not-H-ADAR2UP and H-ADAR2DW primers for ADAR2a, and Bam-H-ADAR3UP and H-ADAR3DW primers for ADAR3. Restriction sites AflII, StuI, and NotI were utilized for ligation of the PCR products at their 3Ј ends into pBac-F-ADAR1, pBac-F-ADAR2a, and pBac-F-ADAR3, respectively. The resultant constructs, termed pBac-H-ADAR1, pBac-H-ADAR2a, and pBac-H-ADAR3, contain a Kozak sequence that is strongly preferred by baculovirus for protein translation initiation at the N terminus region (39). The region amplified by PCR was confirmed by sequencing. Baculovirus expression constructs were then transformed in DH10Bac for transposition into the bacmid and subjected to blue/white screening for identification of recombinant baculoviruses.
Expression of the ADAR Recombinant Baculovirus-Sf9 cells were grown to a density of 2 ϫ 10 6 cells/ml and infected with either a single or a combination of two ADAR recombinant viruses (1:1 ratio) at a multiplicity of infection of 10 -20. At 48 h post-infection, ϳ1 ϫ 10 9 cells were collected.
Extract Preparation-All procedures were carried out at 4°C. HeLa cell extract was prepared as described previously (40). Mouse brain nuclear extract was prepared by the Dignam method (41) with a minor modification as follows. Fresh mouse brains were minced using a pair of scissors, and further homogenized by a motor-driven Potter homogenizer in 3 times the packed cell volume of phosphate-buffered saline. The cell pellet was suspended in a buffer containing 10 mM Hepes (pH 7.9), 10 mM KCl, 1.5 mM MgCl 2 , 0.5 mM DTT, 1ϫ complete protease inhibitor mixture (Roche Diagnostics, Indianapolis, IN), and 0.5 mM phenylmethylsulfonyl fluoride, and kept on ice for 20 min. Cells were lysed by 10 -20 strokes with a glass Dounce homogenizer followed by centrifugation at 10,000 rpm for 15 min in a Type 65 Ti Beckman rotor. The nuclear pellet was suspended in 3 pellet volumes of a buffer containing 20 mM Hepes (pH 7.9), 0.42 M NaCl, 1.5 mM MgCl 2 , 0.2 mM EDTA, 25% glycerol, 1.0 mM DTT, 1ϫ Complete protease inhibitor mixture, and 0.5 mM phenylmethylsulfonyl fluoride. After five gentle strokes in a glass Dounce homogenizer, the protein extract was cleared of debris by centrifugation at 30,000 rpm for 30 min.
Sequential Epitope Tag Affinity Column Purification of ADAR Dimer Complexes-All column chromatography procedures were carried out at 4°C. Total cell extract was prepared from Sf9 cells infected with a single or combination of two recombinant baculoviruses (38). The cell extract, dialyzed against buffer A (0.05 M Tris, pH 7.0, 0.15 M NaCl, 5 mM EDTA, 1.0 mM DTT, 20% glycerol, 0.25 mM phenylmethylsulfonyl fluoride, 0.05% Nonidet P-40) was first passed through a 1.0-ml (1.0 ϫ 1.3 cm) anti-FLAG M2-monoclonal antibody (mAb)-agarose gel (Sigma) affinity column equilibrated with buffer A containing 0.15 M NaCl and 1 mM ␤-mercaptoethanol instead of 1 mM DTT. After washing the column with 10 ml each of buffer A containing 0.15 M NaCl, 0.75 M NaCl, and again 0.15 M NaCl, the complex was eluted with 5 ml of buffer A containing 0.15 M NaCl and 200 g/ml FLAG peptide. The pooled peak fractions were dialyzed against buffer B (10 mM Tris, pH 7.5, 0.3 M NaCl, 20% glycerol, 0.05% Nonidet P-40, 1 mM ␤-mercaptoethanol) and then applied to a TALON metal resin (BD Biosciences, Palo Alto, CA) affinity column. Following extensive washing with buffer B containing 10 mM imidazole, proteins were eluted with 150 mM imidazole. The yield of recombinant proteins during the sequential affinity chromatography was followed by Western blotting analysis using an anti-FLAG M2 mAb (Sigma) or anti-6His 6XHN mAb (BD Biosciences). The purity of recombinant proteins purified by the first ("1ϫ purified") and second ("2ϫ purified") affinity column chromatography were determined by electrophoresis on a 10% SDS-PAGE gel followed by silver staining.
In some experiments, the ADAR complex purified on a M2 mAbagarose column was treated with RNases prior to its application to the TALON affinity column. The recombinant ADAR proteins (1ϫ purified) were treated with single-stranded RNA (ssRNA) specific RNases A (0.5 units/ml) and T1 (10 unit/ml) obtained from Roche Diagnostics or with dsRNA-specific RNase V1 (1 unit/ml) obtained from Pierce. RNasedigested ADAR proteins were dialyzed against buffer B, and then subjected to TALON affinity column chromatography. The RNase digestion conditions were tested separately with uniformly [␣-32 P]ATPlabeled c-myc antisense ssRNA or dsRNA (40), confirming their complete digestion with the relevant RNase(s).
Size Exclusion Column Chromatography Analysis-Purified ADAR proteins (1 g) or crude nuclear extract (2 mg) was applied to a 24-ml (1 ϫ 30 cm) column of Superose 12 HR 10/30 (Amersham Biosciences) for size exclusion chromatography. The buffer system used was 0.05 M Tris (pH 7.0), 0.5 M NaCl, 5 mM EDTA, 1 mM DTT, 20% glycerol, and 0.1% Nonidet P-40. Purified recombinant ADAR proteins were concentrated to 100 l using Centricon (Amicon) before applying to the column. Fractions (0.5 ml) were collected at a flow rate of 0.4 ml/min using a fast protein liquid chromatography system. The molecular weight of ADAR (monomer or oligomer) was estimated by comparison with molecular weight standards obtained from Sigma; bovine thyroglobulin (669,000), horse spleen apoferritin (443,000), sweet potato ␤-amylase (200,000), yeast alcohol dehydrogenase (150,000), bovine serum albumin (66,000), and bovine carbonic anhydrase (29,000). The peak for the ADAR complex was confirmed by Western blotting analysis, and the peak position of the marker proteins was determined by measuring the optical absorption at 280 nm.
In Vitro RNA Editing Assay-Editing of a synthetic 5-HT 2C RNA C5 was assayed in vitro as described previously (22), using 1ϫ or 2ϫ purified recombinant homodimer complexes as well as 2ϫ purified heterodimer complexes consisting of one wild-type and another nonfunctional mutant ADAR monomer. The standard editing reaction contained 20 fmol of a synthetic C5 RNA substrate, 10 ng of recombinant ADAR proteins, 0.02 M Hepes (pH 7.0), 0.1 M NaCl, 10% glycerol, 5 mM EDTA, 1 mM DTT, and 250 units/ml RNasin (Promega). The reactions were incubated at 30°C for various times. Quantitation of editing efficiency at five sites of 5-HT 2C R RNA was carried out by dideoxyoligonucleotide/primer extension assay as described previously (10,22). The ratio of the edited and unedited RNAs was estimated by quantifying the radioactivity of the primer-extended products with a phosphorimaging system (Amersham Biosciences).
Western Immunoblot Analysis-Proteins were fractionated on an SDS-8% polyacrylamide gel and transferred to Immobilon TM -P nylon membrane (Millipore, Bedford, MA). Blots were blocked in a buffer containing phosphate-buffered saline and 3% nonfat dry milk. MAbs 15.8.6, 1.3.1, and 3.591 for detection of native and recombinant ADAR1, ADAR2, and ADAR3 proteins, respectively (22,42), and mAbs M2 and 6XHN for FLAG-and 6His epitope-tagged recombinant ADAR proteins, respectively, were used. ADAR-specific protein bands were detected by peroxidase-conjugated goat antibodies directed against mouse immunoglobulins (Kirkegaard and Perry Lab., Gaithersburg, MD) and chemiluminescense staining using Renaissance TM (PerkinElmer Life Sciences).

Recombinant ADAR1 and ADAR2 but Not ADAR3 Proteins
Form Stable Homodimeric Complexes-A set of baculovirus constructs for ectopic expression of ADAR1, ADAR2, and ADAR3 with either a FLAG or a 6His epitope tag at the N terminus were prepared. Two different sizes of ADAR1 protein, a full-length 150 kDa and a shorter 110 kDa form (p150 and p110) are synthesized because of differential usage of two Met initiation codons (17). The full-length ADAR1 (p150), and ADAR2a among the four known splicing isoforms of ADAR2 (19,20), were investigated in the present studies. FLAG-and 6His epitope-tagged ADAR1, ADAR2, or ADAR3 proteins were coexpressed in Sf9 cells infected with approximately a 1:1 ratio of two different recombinant baculoviruses, and purified by sequential affinity chromatography, first on M2 anti-FLAG mAb-agarose gel and then TALON metal resin as schematically shown in Fig. 1. Each purification step was monitored by Western analysis using anti-FLAG or anti-6His antibody. Both ADAR1 and ADAR2 were purified as oligomeric complexes containing both FLAG-and 6His-tagged protein (Fig. 2, lanes 3 and 8 for ADAR1 and lanes 4 and 9 for ADAR2). The binding of FLAG-tagged ADAR1 or ADAR2 protein to the first affinity column (FLAG mAb column) was nearly complete, whereas a substantial amount (30 to 50%) of the 6His-tagged ADAR1 or ADAR2 protein was detected in this first flow-through fraction as expected. The unbound 6His-tagged ADAR protein represents, most likely, the oligomeric complex consisting of 6Histagged monomers only (see Fig. 1). In contrast, only FLAGtagged ADAR (30 to 50% of the total FLAG-tagged protein present in the original extracts, again representing complexes composed entirely of FLAG-tagged monomers) but almost no 6His-tagged ADAR was detected in the flow-through of the second TALON affinity column, indicating complete binding of the 6His-tagged ADAR that had been preselected by FLAG mAb-agarose gel chromatography (i.e. the oligomeric complex consisting of both FLAG-and 6His-tagged monomers). Overall, the yield of the 2ϫ purified oligomeric complex was 30 to 50% of the ADAR present in the extracts, consistent with the amount expected on the basis of monomers sorting randomly into oligomeric complexes without regard to the FLAG or 6His tag. This essentially complete recovery establishes that the oligomer represents the major form of the complex and shows also that our sequential affinity column purification scheme did not selectively enrich a rare form of the complex. The copurification of FLAG with 6His epitope-tagged ADAR1 as well as ADAR2 were confirmed through a similar sequential affinity column chromatography, but in the reverse order, i.e. TALON metal resin first and then M2 anti-FLAG mAb-agarose gel (data not shown). In contrast, recombinant ADAR3 protein was detected in Western analysis only by using the antibody corresponding to the type of affinity chromatography applied first (Fig. 2, lane 5) but not by the reciprocal antibody corresponding to the second affinity chromatography (Fig. 2, lane 10). These results clearly indicate that recombinant ADAR1 and ADAR2 but not ADAR3 form oligomers.
The apparent molecular mass of ADAR1 (full-length p150 form), ADAR2a, and ADAR3 have been estimated to be 150, 90, and 80 kDa, respectively, by SDS-PAGE (18 -22, 30, 31). To determine the size(s) of ADAR1 and ADAR2 recombinant proteins purified through sequential affinity chromatography, the oligomeric complexes eluted from the second affinity column were fractionated on a Superose 12 size exclusion column (Fig.  3, A and B, top panels). Based on the standard size markers, the sizes of ADAR1 and ADAR2 oligomeric forms were estimated to be 300 and 180 kDa, respectively, indicating that they are both homodimers.
The sequential affinity column chromatography purification procedure as designed precludes detection of ADAR monomer in the 2ϫ purified fraction (Fig. 1). Therefore, to check for free monomers we also examined the apparent sizes of ADAR1 and ADAR2 when expressed as single epitope-tagged proteins and purified by a single affinity column (1ϫ purified). We found that the fractionation profiles for both 1ϫ purified ADAR1 and ADAR2 were identical to those of the 2ϫ purified ADAR proteins. Distinctive elution peaks anticipated for the monomeric forms of ADAR1 (150 kDa) and ADAR2 (90 kDa) were not detected (data not shown). In addition, we carried out the sequential affinity chromatography of two proteins, each separately tagged with FLAG and 6His epitope, and purified on a single affinity column (1ϫ purified) following in vitro mixing and incubation. We found that there was no significant in vitro exchange of two differentially epitope-tagged ADAR1 or ADAR2 proteins, at least during a 2-h incubation at 30°C (data not shown). Taken together, our results suggest that both ADAR1 and ADAR2 recombinant proteins ectopically expressed in Sf9 cells form predominantly a stable homodimer complex.
Superose 12 column chromatography conducted with FLAG epitope-tagged recombinant ADAR3 proteins purified by a single affinity column chromatography (1ϫ purified) revealed a complex elution pattern significantly different from one expected for its monomeric state (Fig. 3C, upper panel). Because the silver staining of the recombinant ADAR3 proteins indicated that they were more than 90% homogeneous, the results were surprising. Although the presence of a minor peak at the position expected for the monomer form was clearly detected (80 kDa, indicated by an arrow), the majority of recombinant ADAR3 proteins eluted as a broad smear covering a range from ϳ400 kDa to much smaller than the monomeric form, suggesting possible nonspecific interaction with the Superose 12 matrix. Using high salt (2 M NaCl) and different pH buffers did not change the elution pattern.
We previously have reported the presence of a ssRNA binding domain located within the arginine-rich R domain (22). Thus, we reasoned that binding of RNA molecules from the insect cells might be responsible for the unusual migration of ADAR3. However, size fractionation of the ADAR3 after treating extensively with RNases specific for both ssRNA and dsRNA (see below) was identical to that of untreated protein (data not shown). In conclusion, the results of sequential affinity column purification strongly suggest a monomeric state for recombinant ADAR3, but we currently do not understand the reason for its unusual migration on Superose 12.
Interestingly, FLAG epitope-tagged ADAR1 and ADAR2 expressed in Sf9 cells and purified on M2 anti-FLAG mAb-agarose gels have been demonstrated to be enzymatically active in deamination of adenosines on a long synthetic dsRNA substrate or in the site-selective in vitro RNA editing of GluR or 5-HT 2C R substrate RNAs (20,30), whereas recombinant ADAR3 proteins are inactive in these assays (22). We now realize that the ADAR1 and ADAR2 recombinant proteins used for our previous studies were predominantly homodimeric forms, and thus it may be that the enzymatic activity of these two recombinant ADARs are related to a dimer formation.
RNA Independent Homodimerization of ADAR1 and ADAR2-The size exclusion chromatography with doubly purified fractions revealed no large oligomeric complexes that would form via binding of multiple ADAR monomers to a long dsRNA. However, homodimerization of ADAR1 and ADAR2 recombinant protein could be dependent on the presence of a short dsRNA substrate of Sf9 cell origin serving to bridge two monomers of ADAR1 or ADAR2. Both RNA-dependent and -independent homodimerization of the dsRNA-activated protein kinase PKR has been reported (43)(44)(45) 3 and 8), ADAR2 (lanes 4 and 9), or ADAR3 (lanes 5 and 10), purified by sequential column chromatography based on both epitope tags, were identified by Western analysis using mAbs specific to the epitope tag. Anti-FLAG mAb was used for Western analysis following the first M2 anti-FLAG mAb affinity column chromatography purification (lanes 3-5), whereas anti-6His mAb was used after the second TALON affinity column chromatography purification (lanes 8 -10). Recombinant ADAR2 tagged with one epitope only, FLAG or 6His (F-ADAR2 and H-ADAR2), were included as controls to show the specificity of the two mAbs used for Western analysis (lanes 1, 2, 6,  and 7). ADAR1 or ADAR2 oligomeric complexes containing both FLAG and 6His epitope tags (F/H-ADAR1 and F/H-ADAR2) were identified with anti-6His mAb (lanes 8 -10) and also with anti-FLAG mAb (not shown). In contrast, ADAR3 was detected only with mAb matching the epitope used for the first affinity chromatography, i.e. FLAG in this set of experiments (lane 5), but not with mAb matching the epitope used for the second affinity chromatography, i.e. 6His (lane 10). When the order of the sequential affinity chromatography was reversed, i.e. TALON first and M2-FLAG mAb-agarose second, ADAR3 (6His-tagged) was detected again only after the first affinity chromatography with anti-6His mAb (not shown). Proteins purified by single (or first) affinity chromatography are indicated as 1X, whereas those purified by two sequential affinity column are designated as 2X. Approximately 10 ng each of 1ϫ purified and 20 ng each of 2ϫ purified proteins were loaded onto 8% SDS-polyacrylamide gels. somewhat similar to ADAR. We therefore tested the homodimer complex copurified by sequential affinity chromatography for their sensitivity to single-strand and double-strand specific ribonuclease treatments (Fig. 4). FLAG and 6His epitope-tagged ADAR1 or ADAR2 recombinant proteins remained together as a dimer regardless of RNase A and T1 (ssRNA specific) or RNase V1 (dsRNA specific) treatment. Thus, the association of two different epitope-tagged monomers is unlikely to be mediated through an RNA molecule(s) (Fig.  4B). On the other hand, it is still possible that an RNA molecule directly involved in formation of the ADAR homodimer may be resistant to the ribonuclease digestion because of its close contact with ADAR proteins. To eliminate the possibility, we looked for RNA molecules bound to the ADAR homodimer complex by 32 P labeling and PAGE analysis of the labeled RNAs. Approximately 2 g of 2ϫ purified ADAR1 or ADAR2 was subjected to proteinase K digestion and subsequent RNA extraction. The total RNA extracted was then labeled using 32 P-labeled pCp and T4 RNA ligase and analyzed by 7 M urea-PAGE. By including known amounts of a synthetic 21nucleotide RNA molecule as an internal control, we concluded that no significant level of RNA is present (less than 0.1 ng of RNA for 2 g of the doubly purified ADAR homodimer). This is less than one RNA base per dimer, clearly insufficient to act as a bridge between monomeric proteins (data not shown).
Homodimerization of Native ADARs-To confirm homodimerization of native ADAR1 and ADAR2, HeLa cell and mouse brain nuclear extracts were subjected to Superose 12 size exclusion column chromatography. In HeLa cells, both p150 and p110 forms of ADAR1 were detected (Fig. 3A, middle  panel). The peak of native ADAR1 p150 coincided with that of the recombinant p150 homodimer complex (300 kDa), and ADAR1 p110 proteins also eluted in the fractions expected for its homodimer complex (220 kDa). In mouse brain, only the p110 form of ADAR1 was detected in nuclear extracts, and it appeared in the fractions expected for the homodimer (Fig. 3A,  lower panel). Native ADAR2 proteins appear also to exist mainly as a complex of homodimer size (Fig. 3B, middle and lower panels). Both the ADAR2a and ADAR2b splicing isoforms of native human ADAR2 differing in size by 40 amino acid residues (19,20) were detected in the HeLa nuclear extract. In mouse, the size difference, only 10 amino acid residues, was not sufficient for detection of these two isoforms separately (18). We noted a slightly broader elution peak of native ADAR1 and ADAR2 especially with a tailing toward smaller molecular weight regions in comparison with recombinant homodimer complexes, possibly indicating the presence of some monomer form (Fig. 3, A and B, middle and lower panels). As with the recombinant protein, fractionation of the native mouse brain ADAR3 resulted in a complex elution pattern, but with a minor peak that coincides with the size for the anticipated ADAR3 homodimer (160 kDa), suggesting the possibility that native ADAR3 may also form a homodimeric complex in brain (Fig.  3C, lower panel). Unlike the recombinant protein, smearing  (panels A and B). Positions of ADAR3 monomer (80 kDa) and homodimer (160 kDa) are also indicated by arrows (panel C). Recombinant ADAR3 proteins were detected also in all fractions collected beyond the 30 fractions shown, up to fraction 50.
FIG. 3. Analysis of ADAR oligomeric complexes by Superose 12 gel filtration column chromatography. Oligomeric forms of ADAR proteins fractionated by Superose 12 gel filtration column chromatography were analyzed by Western blotting analysis using mAb specific to into fractions corresponding to small molecules was not observed with native ADAR3. More importantly, no obvious peak of monomer was detected for native ADAR3 (Fig. 3C, lower  panel), in contrast to recombinant ADAR3 (Fig. 3C, upper panel). It should be pointed that a fraction of native ADAR1 (detected for certain extract preparations but not seen clearly in Fig. 3A, middle and lower panels) and ADAR3 proteins (Fig.  3C, lower panel) migrated with an apparent molecular weight of Ͼ600,000, whereas no larger complex of native ADAR2 was detected other than the homodimer size complex in the extracts of HeLa cells and mouse brain. Interestingly, digestion of the extracts with RNase A and T1 prior to Superose 12 size exclusion column chromatography resulted in shifting of at least a part of the larger ADAR1 or ADAR3 complexes to complexes of homodimer size. Most importantly, however, RNase digestion (ssRNA and dsRNA specific) did not affect the size of the homodimer-like complex of ADAR1, ADAR2, or ADAR3, indicating RNA independent homodimerization of native ADARs (data not shown). We have recently reported the association of both ADAR1 and ADAR2 with large nuclear ribonucleoprotein particles, consisting of four splicesomal subunits that assemble together with the pre-mRNA. However, the size of the complexes observed in the present studies are far smaller than the large nuclear RNP particles (200 S), which can be detected with the HeLa nuclear extract prepared through a specific gentle extraction procedure in the presence of 2 mM vanadyl ribonucleoside RNase inhibitor (42). Thus, the nature of the larger complexes of native ADAR1 and ADAR3 detected in the present studies (both RNase digestion sensitive and resistant) is not clear at this time.
Different Members of the ADAR Gene Family Do Not Associate as Heterodimers-We next investigated the possibility of heterodimer complex formation between two different members of the ADAR gene family (Fig. 5). Two differentially epitope-tagged ADAR gene family members, i.e. ADAR1 and ADAR2 (lanes 1 and 7), ADAR1 and ADAR3 (lanes 2 and 8), or ADAR2 and ADAR3 (lanes 3 and 9), were coexpressed in the same cell (Sf9) and purified by sequential affinity chromatography. The presence of both ADAR proteins tagged with FLAG  1 and 7), transfection II, F-ADAR3 and H-ADAR1 (lanes 2 and 8), and transfection III, F-ADAR3 and H-ADAR2 (lanes 6 and 12), were purified first by M2 anti-FLAG mAb (lanes 4 -6) and then by TALON affinity column chromatography (lanes 10 -12). Two recombinant proteins are detected in the extracts with anti-FLAG (lanes 1-3) and anti-6His mAb (lanes 7-9). FLAGtagged proteins were detected with anti-FLAG mAb (lanes 4 -6) following the first affinity chromatography (1X), but no 6His-tagged proteins (i.e. no heterodimer) were detected with anti-6His mAb (lanes 10 -12) after the second affinity chromatography (2X).  6 -10), were tested at 20°C. Both ssRNA and dsRNA were digested completely within 30 min. B, F/H-ADAR1 and F/H-ADAR2 dimer complexes purified by affinity chromatography on anti-FLAG M2 mAb-agarose beads (1ϫ purified) were subjected to RNase digestion using either A and T1 or V1, at 20°C for 30 min, further purified by TALON affinity chromatography, and then tested with Western analysis using anti-FLAG mAb. Controls (C) were done exactly the same but without RNase treatments. and 6His was confirmed in the extracts, but no formation of heterodimer complexes between any combination of two different ADAR gene family members was detected (lanes [10][11][12]. Although these experiments with the recombinant ADAR proteins suggest no formation of heterodimer complexes, it is possible that a heterodimer complex may form in vivo, i.e. between ADAR2 and ADAR3 in brain. We therefore conducted coimmunoprecipitation experiments using nuclear extracts of mouse brain in an attempt to detect heterodimer complex formation between two different ADAR native proteins. Mouse nuclear extract immunoprecipitated with a mAb specific to an ADAR gene family member was examined by Western blotting analysis with mAbs specific to the remaining two ADAR gene family members. All the ADAR gene family member-specific mAbs successfully immunoprecipitated the anticipated cognate ADAR protein but did not coimmunoprecipitate any of the remaining ADAR gene family members (data not shown). These results suggest that formation of heteromeric complexes among different ADAR gene family members is unlikely.
Functional Interaction between the Two Monomers of the ADAR Homodimer-The two monomers within an ADAR1 or ADAR2 homodimer may bind to two separate substrate RNAs and deaminate adenosine residues at each catalytic center independently. Alternatively, the two monomers may act cooperatively for conversion of an adenosine residue of a single substrate RNA. To obtain some insights into possible functional interaction between the two monomers, we examined the enzymatic activities of a heterodimer complex made between wild-type and a non-functional ADAR mutant in comparison to those of mutant or wild-type homodimers (Fig. 6). ADAR1 E912A with a Glu 912 3 Ala substitution and ADAR2a E396A with a Glu 396 3 Ala substitution were used as the mutant monomers. The glutamate residues Glu 912 of ADAR1 and Glu 396 of ADAR2 are located within the tripeptide sequences HAE and PCG, which are highly conserved among ADAR gene family members as well as cytidine and deoxycytidylate deaminase gene family members. These residues are believed to play a critical role in proton transfer functions required for the hydrolytic deamination reaction (15,38). We have previously shown that site-directed mutagenesis of Glu 912 of ADAR1 (E912A) results in complete abolishment of the deaminase activity without affecting substrate RNA binding capability (38). Formation of the heterodimer between one wild-type and one mutant monomer was first confirmed by their copurification through sequential affinity column chromatography as above (data not shown). Non-selective ADAR activity, which converts multiple adenosines to inosines in a sequence-independent manner, was determined on a long 575-bp synthetic c-myc dsRNA (Fig. 6A). Site selective A-to-I RNA editing activity was monitored by determining the editing of 5-HT 2C R RNA at the A site by ADAR1 and the D site by ADAR2. Preferential editing of the A and D sites by ADAR1 and ADAR2, respectively, has been demonstrated previously in vitro using recombinant proteins (6,22,46). Preliminary time course experiments were conducted separately to choose conditions under which the enzymatic reaction remains first-order in enzyme concentration, so that the results can be compared quantitatively. The enzymatic activities of heterodimers consisting of one wild-type and one non-functional mutant monomer were found to be approximately half (55% for ADAR1 and 52% for ADAR2) of the wild-type homodimer activity when tested with the long c-myc dsRNA substrate (Fig. 6A). In contrast, siteselective RNA editing activity by the heterodimer, determined on 5-HT 2C R RNA, decreased to ϳ30% of the wild-type homodimer activity (Fig. 6B). These results may indicate that natural substrate RNAs induce cooperative interactions be-tween the two monomers in the wild-type homodimer complex. Presumably, the three-dimensional structure of natural substrates such as 5-HT 2C R RNA, which includes short dsRNA regions, loops, and bulges, facilitate simultaneous interactions with both monomers in the dimer complex.

DISCUSSION
Homodimerization of ADAR1 and ADAR2-In this study, we have demonstrated that recombinant ADAR1 and ADAR2 both exist predominantly as stable homodimers. Purification of the complexes via sequential affinity chromatography with two FIG. 6. Functional interaction between two subunits of the dimer complex in A-to-I RNA editing. Enzymatic activities of wildtype (both 1ϫ and 2ϫ purified), non-functional mutants (ADAR1 E912A and ADAR2 E396A), and heterodimer complexes consisting of 6His-tagged wild-type and FLAG-tagged mutant subunits were tested. By conducting a preliminary time course experiment, an appropriate incubation time within the linear reaction range was selected. ADAR1, open bars. ADAR2, filled bars. Three independent experiments were conducted (n ϭ 3), and standard errors are indicated. A, A-to-I base modification activities were determined with a synthetic long c-myc dsRNA (575 bp) by using 10 ng each of recombinant enzyme for 15 min at 37°C. The results were normalized relative to the values obtained with wild-type 1ϫ purified enzymes. A-to-I base modification by ADAR1 and ADAR2 wild-type enzymes (1ϫ purified) was 11.5 and 21.9%, respectively. B, site-specific editing of 5-HT 2C R RNA was monitored. In vitro editing of 5-HT 2C R RNA at the A site by ADAR1 and at the D site by ADAR2 was carried out for 30 and 10 min, respectively, at 30°C. A site editing by ADAR1 wild-type (1ϫ purified) was 25.2%, whereas D site editing by ADAR2 wild-type (1ϫ purified) was 51.6%. different epitope tags revealed for the first time the presence of oligomeric complexes, whereas size exclusion column chromatography identified the complexes as homodimers. The homodimer formation is mediated by protein-protein interaction between two monomers and is independent of binding to RNA. Although we originally assumed that both oligomeric as well as monomeric forms of ADAR proteins might be present in equilibrium, our results suggest that recombinant ADAR1 or ADAR2 ectopically expressed in Sf9 cells appears to form predominantly a homodimer, possibly immediately after translation or even during translation as occurs for some dimeric proteins (e.g. tubulin). Furthermore, homodimers once formed appear to be very stable without detectable exchange of their monomer components under physiological conditions. Native ADAR1 biochemically purified from various sources (bovine liver, calf thymus, and Xenopus oocytes) has been reported to have various sizes (80 to 120 kDa), smaller than the full-length p150 form, probably because of translation initiation at the internal methionine codon (equivalent to our p110 form) or truncation of the N terminus as a result of nonspecific proteolysis (47)(48)(49). In contrast to our findings in this study with recombinant ADAR1, biochemically purified native ADAR1 was found to be a monomer by size exclusion column chromatography and by glycerol gradient sedimentation analysis (47)(48)(49). In addition to the differences in protein size and possible species differences, another possible explanation for the discrepancy is that native ADAR1, predominantly in the homodimer state, might become dissociated into monomers because of the rather vigorous biochemical purification procedures applied. Interestingly, it has been reported that the monomeric form of Xenopus ADAR1, once biochemically purified, does not re-dimerize even under low ionic strength conditions (49). In all previous studies, the purification of native ADAR1 was assayed by enzymatic activity, suggesting that the biochemically purified monomeric ADAR1 was capable of deaminating the dsRNA substrate used in their assay (46 -48). An alternative interpretation consistent with that data is that ADAR1 protein dissociated into monomers because of the nonphysiological conditions applied during purification, but was reconstituted into homodimers upon binding to dsRNA during the A-to-I base modification assay, which restored its enzymatic activity. Recently, homodimerization of ADAR2 on a dsRNA substrate has been proposed based on the results of kinetic analysis (50). A-to-I RNA editing was observed only under conditions allowing ternary complex formation between the ADAR2 homodimer and GluR-B RNA substrate, indicating a requirement for ADAR2 dimer formation for its site-selective editing activity (50). Binding of a substrate RNA and acceleration of homodimerization of PKR has been reported also (43)(44)(45).
Our results on recombinant ADAR1 and ADAR2 are supported by size exclusion column chromatography analysis of HeLa and mouse brain nuclear extracts, which detected the anticipated homodimer size complex of native ADAR1 and ADAR2. The native complexes were detected by Western blotting analysis of the crude extracts using ADAR1-or ADAR2specific antibodies, and thus we cannot exclude the possibility that they represent the monomer associated with some currently unknown molecule (protein or RNA). However, the identical sizes of the native complexes and recombinant homodimers suggest that a large fraction of native ADAR1 or ADAR2 most likely also exist as homodimers.
Possible Dimerization of Native ADAR3 in Brain-In contrast to the results with ADAR1 and ADAR2, we could not detect the homodimer of recombinant ADAR3. Recombinant ADAR3 ectopically expressed in Sf9 cells apparently remains as a monomer, which may explain its lack of enzymatic activity (22). The behavior of the recombinant ADAR3 during size exclusion column chromatography was also aberrant, eluting from a Superose 12 column as a broad smear, including fractions corresponding to physically impossible sizes. It may be that a majority of recombinant ADAR3, incapable of homodimerizing, does not have a uniform structure and migrates through the column without forming a distinctive fractionation peak. Alternatively, monomeric recombinant ADAR3 may interact non-specifically with the matrix material of the sizing column used, possibly because of lack of a required post-translational modification that takes place in brain but not in Sf9 insect cells. This hypothesis then predicts that native ADAR3 protein may form a homodimer and behave differently from the recombinant proteins during size exclusion column chromatography. Indeed, size fractionation analysis of mouse brain extracts on a Superose 12 column revealed an elution peak corresponding to a complex of ϳ160 kDa, the anticipated size of the ADAR3 homodimer, in addition to a separate peak corresponding to a much larger molecular mass complex (Ͼ600 kDa). The size of the 160-kDa complex also corresponds to the size of a potential ADAR2/ADAR3 heterodimer (Fig. 3C, lower panel).
Heterodimerization among proteins related to the human ADAR family has been observed in other organisms. The ADAT gene family (tRNA-specific A-to-I editing enzymes) has been identified in yeast because of their deaminase domain sequence homology to ADAR (1). ADAT1 edits A 37 of tRNA Ala as a monomer (51), whereas A 34 at the first anticodon position of tRNA Ala is edited specifically by a heterodimer formed by ADAT2 and ADAT3 (52). ADAT2 is the catalytically active subunit, whereas ADAT3 is the regulatory subunit not directly involved in the A-to-I deamination mechanism (52). More recently, the possibility of heterodimer formation between c.e.ADAR1 and c.e.ADAR2 has also been indicated (27). We were intrigued by the possibility that enzymatically inactive ADAR3 might form a heterodimer complex with an enzymatically active ADAR member (e.g. ADAR2), and therefore investigated possible heterodimer formation among three different ADARs. We found no indication of any heterodimer formation between two different ADAR gene family members, including the heterodimer consisting of ADAR2 and ADAR3, at least among recombinant proteins co-expressed in the same Sf9 insect cell. Taken together with our preliminary results from coimmunoprecipitation experiments of mouse brain extracts, we conclude that the complex eluting around 160 kDa is likely to represent a ADAR3 homodimer that forms only in mouse brain.
Interaction of Two Monomers-The structural basis of the interactions between two monomer subunits of ADAR1 or ADAR2 is currently unknown. The deaminase domain structure of ADAR is predicted to have a similarity to E. coli cytidine deaminase (1,15). In the x-ray crystal structure of E. coli cytidine deaminase the homodimer has a 2-fold symmetry axis, indicating that the two monomers are structurally indistinguishable and predicted to be functionally equivalent (34). Molecular modeling of another cytidine deaminase APOBEC-1, ApoB RNA editing enzyme, predicts a structural configuration of the homodimer very similar to that of E. coli cytidine deaminase (37). Extensive interactions widely spread over many regions are involved in the formation of the monomer-monomer interface of E. coli cytidine deaminase and APOBEC-1 homodimers (34,37). The active site in each monomer is completed only with contributions from the other partner subunit (34,37). We have recently conducted studies to map the regions required for formation of the homodimer using the ADAR1 mutant baculovirus constructs (38) coexpressed with the wildtype construct and sequential affinity chromatography purification of the dimer complex. 2 Although the conclusions of those studies remain preliminary because of the unstable nature of certain deletion constructs, it appears that the interface interactions of the two monomers occurs over a widespread region including the deaminase domain as well as the dsRNA binding domains. In contrast, the N-terminal region containing the Z-DNA binding domain (amino acids 1 to 295) is not required, because formation of the heterodimer between p150 and p110 of ADAR1 can be detected. The regions critical for nuclear import or export of ADAR1 have been mapped recently (53,54). It would be interesting to know if formation of the homodimer is involved also in the nuclear-cytoplasmic shuttling mechanism. The stable nature of the recombinant ADAR1 and ADAR2 homodimers predict its de novo formation during translation and transport as a homodimer unit. However, it is also possible that the monomer is used as a transport form (nuclear import or export), whereas homodimer formation may be a part of a mechanism to concentrate the active complex in one compartment of the cell.
Dimerization is known to affect the enzymatic activity as well as substrate specificity (45,52). Each dsRNA binding domain of ADAR is independently capable of binding to a dsRNA region as short as 15 to 20 bp. A number of ADAR proteins bind to multiple sites of a long completely complementary dsRNA (55), but also to a discrete site of a specific hairpin RNA (56). Thus, each monomer of ADAR1 or ADAR2 may bind to a separate dsRNA molecule through its own dsRNA binding domains. However, it is also possible that one homodimer binds to a single substrate RNA whereas the dsRNA binding domains of the two monomers make cooperative interactions. In an attempt to understand the functional interactions of the two monomers, enzymatic activity of a heterodimer consisting of one wild-type and one inactive mutant monomer was compared with the homodimers consisting of only wild-type functional monomers. The point mutation was in a position that is equivalent to the residue in APOBEC-1 (E63) predicted to play a critical role in both the catalysis of the deamination reaction and in monomer interface interactions (37). Interestingly, the site-selective editing activity of ADAR1 or ADAR2 heterodimer complexes revealed substantial cooperation of the two monomers. Our results hint that the glutamate residue, Glu 912 of ADAR1 or Glu 396 of ADAR2, may contribute to the formation of the active site for the partner monomer in addition to the catalysis mechanism. Surprisingly, the enzymatic activity of the heterodimers suggested that each monomer can independently catalyze non-selective deamination of a long dsRNA substrate. We do not presently have enough information to distinguish several possible models for the interactions of the two monomers or each monomer with a substrate RNA. A model for the interaction of two monomer subunits of APOBEC-1 with a single substrate RNA (ApoB RNA) has been proposed (37). In this model, the targeted cytidine residue is deaminated at the catalytic center of one monomer, whereas a specific downstream uridine residue of the same RNA is bound by the partner subunit. In contrast, each monomer of the E. coli cytidine deaminase homodimer catalyzes deamination of one cytidine independently (34). On the analogy of the model for APOBEC-1 and ApoB RNA interaction, one possible interpretation of our results is that the number of substrate RNAs interacting with each ADAR homodimer may vary depending on the tertiary structure of the RNA. A long uninterrupted and completely complementary dsRNA (A-form double helix, and most likely rod shaped) may be independently bound and deaminated by each monomer. In contrast, a substrate RNA with relatively short dsRNA regions, loops, and bulges such as 5-HT 2C R may be bound by the dsRNA binding domains of both monomers, and deaminated after formation of an active complex in which interactions between the two monomers brings about alignment of a select adenosine residue with one of the two catalytic centers. Our future studies will define interactions of the two subunits of ADAR1 and ADAR2 homodimers and address their significance for enzymatic activity as well as intracellular localization.