The importance of internal loops within RNA substrates of ADAR11

The importance of internal loops within RNA substrates of ADAR11

Article No. jmbi.1999.2914 available online at http://www.idealibrary.com on J. Mol. Biol. (1999) 291, 1±13 The Importance of Internal Loops within ...

765KB Sizes 17 Downloads 54 Views

Article No. jmbi.1999.2914 available online at http://www.idealibrary.com on

J. Mol. Biol. (1999) 291, 1±13

The Importance of Internal Loops within RNA Substrates of ADAR1 Katrina A. Lehmann and Brenda L. Bass* Department of Biochemistry and HHMI, University of Utah, 50 N. Medical Drive Salt Lake City, UT 84132, USA

Adenosine deaminases that act on RNA (ADARs) are a family of RNA editing enzymes that convert adenosines to inosines within doublestranded RNA (dsRNA). Although ADARs deaminate perfectly basepaired dsRNA promiscuously, deamination is limited to a few, selected adenosines within dsRNA containing mismatches, bulges and internal loops. As a ®rst step in understanding how RNA structural features promote selectivity, we investigated the role of internal loops within ADAR substrates. We observed that a dsRNA helix is deaminated at the same sites whether it exists as a free molecule or is ¯anked by internal loops. Thus, internal loops delineate helix ends for ADAR1. Since ADAR1 deaminates short RNAs at fewer adenosines than long RNAs, loops decrease the number of deaminations within an RNA by dividing a long RNA into shorter substrates. For a series of symmetric internal loops related in sequence, larger loops (5six nucleotides) acted as helix ends, whereas smaller loops (4four nucleotides) did not. Our work provides the ®rst information about how secondary structure within ADAR substrates dictates selectivity, and suggests a rational approach for delineating minimal substrates for RNAs deaminated by ADARs in vivo. # 1999 Academic Press

*Corresponding author

Keywords: deaminase; double-stranded RNA; inosine; RNA editing

Introduction Adenosine deaminases that act on RNA (ADARs) convert adenosines to inosine residues within cellular and viral RNAs (see reviews by Bass, 1997; Maas et al., 1997; O'Connell, 1997). The enzyme activity has been detected in many metazoa, usually in the nucleus, and cDNAs for at least two distinct ADARs (ADAR1 and ADAR2) have been cloned. Like RNA splicing, adenosine deamination is a way to diversify the information encoded by a gene. For example, ADARs can deaminate adenosines within codons so that multiple protein isoforms can be synthesized from a single encoded mRNA. ADARs are involved in producing isoforms of mammalian serotonin receptors (Burns et al., 1997), several mammalian glutamate receptor subunits (Egebjerg & Heinemann, 1993; Higuchi et al., 1993; Lomeli et al., Abbreviations used: ADARs, adenosine deaminases that act on RNA; dsRNA, double-stranded RNA; RBP, RNA-binding proteins; dsRBM, dsRNA binding motif; ss, single-stranded. E-mail address of the corresponding author: [email protected] 0022-2836/99/310001±13 $30.00/0

1994), and the virally encoded hepatitis delta antigen (Polson et al., 1996). ADARs may have additional functions as well; in fact, other cellular and viral substrates have been identi®ed for which the function of deamination is unclear (Bass, 1997). ADAR substrates fall into two general categories based on how many of their adenosines are deaminated, i.e. how selectively they are deaminated (Bass, 1997). (The term selectivity has a de®ned meaning in the ADAR ®eld and refers to the percentage of the total number of adenosines in a base-paired region that are deaminated at complete reaction). Selectively modi®ed substrates are deaminated at less than 10 % of their adenosines, while substrates that are promiscuously, or non-selectively, modi®ed are deaminated at 50-60 % of their adenosines. Obviously, promiscuous deamination is ill-suited for generating protein isoforms of precise function, and in mRNAs where ADARs produce a functionally important codon change, deamination occurs selectively. Although it was initially speculated that ADARs would require accessory factors for selective deamination, many studies show that ADARs alone are capable of very selective deamination (Hurst et al., 1995; Dabiri et al., 1996; Melcher et al., # 1999 Academic Press

2

Internal Loops Promote Selective ADAR Deamination

1996; Polson et al., 1996). It is now generally accepted that features of the RNA substrate dictate selectivity. The speci®c features of ADAR substrates that determine selectivity are unclear. In vitro studies show that ADAR1 prefers to deaminate adenosines with a 50 neighbor of A, U, or C, over those with a 50 neighbor of G. In addition, adenosines very close to the 30 end of a strand are deaminated infrequently. These preferences are observed in both selectively and non-selectively deaminated substrates, but are not suf®cient to promote the exquisite selectivity observed within some biological substrates (see Polson et al., 1996). Although few studies have been performed, ADAR2 appears to have overlapping, but slightly different, speci®cities (Melcher et al., 1996). It has been proposed that the structure of the RNA substrate, rather than its sequence, might determine selectivity (Bass, 1997). ADARs are categorized as double-stranded RNA-binding proteins (dsRBPs) and like other dsRBPs (e.g. RNase III and PKR) require substrates that are completely, or largely, double-stranded. A survey of characterized ADAR substrates shows that RNAs that are completely double-stranded are deaminated promiscuously (non-selectively), while RNAs whose basepaired structures are periodically interrupted by mismatches, bulges and loops are deaminated

selectively. Although mutagenesis studies clearly show that base-pairs surrounding a given editing site are important for the precise and ef®cient deamination of that site (Higuchi et al., 1993; Lomeli et al., 1994; Yang et al., 1995; Polson et al., 1996), the effect of loops, mismatches and bulges on the total number of deamination sites within a given helix, i.e. selectivity, has not been investigated. As a ®rst step in understanding how RNA structure allows selective deamination, we have focused on the role of internal loops within ADAR substrates. Previous studies show that short duplexes are deaminated more selectively than long duplexes (Nishikura et al., 1991; Polson & Bass, 1994). Thus, we reasoned that internal loops within an ADAR substrate could uncouple adjacent helices to convert a long, promiscuously deaminated substrate into a series of short, selectivity deaminated substrates. Our experiments support this hypothesis, thus providing the ®rst information in regard to how RNA structure determines ADAR selectivity.

Results We examined the role of internal loops in promoting ADAR1 selectivity using six RNA molecules (Figure 1(a)). One molecule consisted of a completely base-paired dsRNA of 36 bp (36mer),

Figure 1 (Legend opposite)

Internal Loops Promote Selective ADAR Deamination

3

Figure 1. Nucleotide sequence and RNase T2 structure mapping of RNAs. (a) The same numbering scheme is used for the 36mer sequence of each RNA (blue), and purine (R) and pyrimidine (Y) rich strands are indicated. Internal loops, GC-rich helices, and the central helix of barbell molecules are labeled. (b) L0, L4, L6 and L12 molecules containing an R strand labeled at its 50 end with 32P were incubated in the presence or absence of RNase T2, under the same native conditions used in reactions with ADAR1 (see Materials and Methods). In separate reactions, RNA was digested with RNase T1 under denaturing conditions for sequence orientation purposes (data not shown). In contrast to a single R strand (ss), the barbells were protected from RNase T2 cleavage in regions predicted to be basepaired. Consistent with the predicted structures, the left loop, right loop and 30 overhang were digested with RNase T2. The L12, L6 and L4 loops show the expected number of RNase T2-accessible nucleotides, and one or two accessible nucleotides 30 of the predicted loop closure; moderate increases in solvent accessibility of nucleotides 30 of internal loops has been observed previously (Weeks & Crothers, 1993). The L6 and L4 loops show a degree of T2 accessibility intermediate to that of the L12 loops and the corresponding region of L0. All four RNAs show similar RNase T2 cleavages in the 30 overhang. The secondary structure of L8 was not determined, but subsequent studies showed that L8 was deaminated by ADAR1 with the same selectivity as barbell RNAs with larger (L12) and smaller (L6) loops (see Figure 3(a) and (b)) whose structures were con®rmed. OH- indicates alkaline hydrolysis lanes. Numbering is shown to the right of the gel, along with parentheses indicating the central helix, left loop, right loop, and 30 overhang.

formed between a purine (R) and a pyrimidine (Y) rich strand; previous studies show that this molecule is deaminated by ADAR1 and has six deamination sites at complete reaction (Polson & Bass, 1994). Four of the molecules were ``barbell'' molecules that contained the 36mer sequence bounded on each side by symmetric internal loops of 12 (L12), eight (L8), six (L6), or four (L4) nucleotides;

loops were closed with GC-rich helices at each terminus. Barbell molecules of decreasing loop sizes (L8, L6, L4) were created by changing the sequence of the L12 Y strand so as to extend the length of the GC-rich helix and decrease the loop size; thus, the R and Y strands had identical lengths from one barbell to the next. Importantly, the external GC-rich helices of all molecules were

4 predicted to be too short to serve as ADAR substrates if separated from the rest of the molecule. In the sixth molecule, L0, the GC-rich helices were extended completely to form a continuous helix with the 36mer sequence; this molecule was called L0 to convey the fact that it contained loops of zero nucleotides. All molecules except the 36mer contained a 12 nt 30 overhang on each strand to facilitate primer binding during primer extension analyses. The loops of the barbells consisted of adenosines and cytidines and were designed to minimize pairing and stacking interactions. We con®rmed that these sequences were unpaired internal loops by treating the molecules with ribonuclease T2 (RNase T2; Figure 1(b)), which cleaves 30 of unpaired nucleotides. Consistent with the predicted structures, pronounced T2 cleavage occurred in all loops, as well as the 30 overhangs, but not within base-paired regions. As expected, only the 30 overhang was cleaved in the L0 molecule. As mentioned, our starting hypothesis was that internal loops, by uncoupling adjacent helices, could convert one long ADAR substrate into a series of shorter substrates. If this hypothesis was correct, and each RNA helix was indeed uncoupled from adjacent helices by the loops, ADAR1 should treat internal loops as helix termini. That is, the deamination pattern within a particular sequence should be identical whether the sequence existed as a free dsRNA or was bounded by internal loops. To test this idea we reacted 36mer and L12 with ADAR1 and compared the reaction products. RNAs were internally labeled with [32P]ATP during transcription, and adenosine deamination was measured by determining the amount of [32P]AMP converted to [32P]IMP at various time points (Figure 2(a)). Previous studies show that dsRNAs greater than 50 bp are deaminated promiscuously by ADAR1, showing 50-60 % deamination at reaction completion (Polson & Bass, 1994). Since the L12 molecule contained a total of 54 canonical base-pairs, we expected at least 50 % of the adenosines would be deaminated at complete reaction if the internal loops were not acting as helix termini. However, we found that 36mer and L12 were deaminated to the same extent throughout the course of the experiment (zero to four hours, see the legend to Figure 2). A 5 fmol sample of each substrate was reacted, and both molecules showed between 21-23 fmol of inosine at complete reaction. These data showed that ADAR1 deaminated the 36mer and L12 with the same selectivity, which supported our idea that the L12 loops were acting to uncouple the 36 bp helix from its adjacent external helices. We next determined which adenosines were deaminated in each molecule using a primer extension assay (Figure 2(b)). Since inosine prefers to pair with cytidine, in this assay, inosines are evident by the loss of a ddTTP stop and the corre-

Internal Loops Promote Selective ADAR Deamination

sponding appearance of a ddCTP stop. Again, the results supported the idea that ADAR1 recognizes internal loops as helix termini. At complete reaction with ADAR1, 36mer and L12 were deaminated at the same six primary deamination sites (A5, A6, A9, A10, A14, A15; Figure 2(c)). The only difference between the two molecules occurred at A2, which was deaminated to a very minor extent in the L12 molecule (see Discussion). Thus, for the most part, ADAR1 appeared to recognize the L12 loops as helix ends, and the selectivity observed in the free 36mer molecule was maintained. The fact that the deamination sites were almost identical in the two molecules further emphasized that the loops were functioning as effective helix ends, since ADAR1 disfavors adenosines close to 30 termini. If the loops were not acting as effective helix ends, we would expect deamination of additional adenosines closer to the loop, such as those between A16 and A34. We next wanted to determine how short an internal loop could be and still function as a helix terminus. We extended our analysis to include barbells L8, L6, and L4, which have loops of eight, six and four nucleotides, respectively (Figure 1(a)). We also analyzed L0 (Figure 1(a)), the molecule in which the barbell loops had been closed to form a completely base-paired molecule of 66 bp. We reasoned that barbells with loops that could uncouple helices would be deaminated similarly to 36mer, while those with loops that could not uncouple adjacent helices would be deaminated promiscuously, and show a deamination pattern similar to that of L0. We performed a time-course experiment like that described in the legend to Figure 2(a), and analyzed the accumulation of inosine within all six RNAs over time (Figure 3(a)). At each time point, the amount of inosine observed for L12, L8, and L6 was almost identical with that of the 36mer, suggesting the loops in these molecules could uncouple adjacent helices. However, throughout the time-course the L4 barbell contained much more inosine, approaching that observed in L0; this suggested that, at least for the sequences we tested, loops 5six nucleotides, but

Internal Loops Promote Selective ADAR Deamination

5

Figure 2. Comparison of the reaction of 36mer and L12 RNAs. (a) RNAs were incubated with ADAR1, and at various times, aliquots were removed, digested to 50 nucleotide monophosphates (50 NMPs), and analyzed by TLC. The graph shows the amount of IMP (fmol) produced during the incubation of 36mer (squares) and L12 (circles) with ADAR1. Each point is the average of two to four experiments, with standard deviation bars shown. The amount of IMP observed at four hours (data not shown) veri®ed that the reaction was complete at two hours, and control experiments con®rmed that the enzyme was active throughout the incubation. (b) The sites of deamination within 36mer and L12 were mapped by primer extension sequencing using reverse transcriptase (RT). RNAs were incubated for two hours with (‡) or without (ÿ) ADAR1 and subsequently reverse-transcribed in the absence of dideoxynucleotides (lanes 2-3), or in the presence of ddGTP (lane 4), ddATP (lane 5), ddTTP (lanes 6-7) or ddCTP (lanes 8-9). Lane 1 was loaded with the primer only. Numbers to the right of each gel show adenosine number, and parentheses indicate the left and right loops of L12. (c) The diagram shows the location of inosines as determined in (b), and was con®rmed by mapping with a mutant (E46Q) RNase T1 (data not shown; see Materials and Methods). Based on comparisons with the RNase T1 mapping data, deaminations evident by the loss of a ddTTP stop, as well as the corresponding gain of a ddCTP stop, were considered to occur in the majority of the population of molecules (I), whereas deaminations evident only by the gain of a ddCTP stop were considered to be minor deaminations that existed in a minor fraction of the population (i). Italicized adenosines within L12 mark a region of possible local conformational heterogeneity. Deamination sites were not observed within the L12 loops.

6

Internal Loops Promote Selective ADAR Deamination

completion. In contrast, much more inosine was observed after the complete reaction of 5 fmol of L4 (47 fmol) or L0 (58 fmol). Although the range of values observed for the L4 and L0 samples were slightly overlapping at the two hour time point (see error bars, Figure 3(a)), within a given experiment the amount of inosine observed in L0 was consistently 20 % higher than that observed in L4. The overall time-course also emphasizes that L4 and L0 react with ADAR1 slightly differently. While the loops of four nucleotides are not acting as internal loops to uncouple adjacent helices, they are probably affecting the substrate in other important ways (see Discussion). We next mapped the inosines within L6 and L4 by primer extension (Figure 4(a)). L4 contained many more deamination sites than L6, and the location of each deamination site is shown in Figure 4(b). L6 showed exactly the same deamination pattern as 36mer, indicating that its loops were acting as helix termini. In contrast, L4 contained more than twice the number of deamination sites as L6, and had a pattern of deamination very similar to that seen in L0. Consistent with the idea that the external GC-rich helices are too short to act as ADAR substrates by themselves, none of the adenosines in the external helices of L6 were deaminated. In contrast, in L4 two adenosines in one of the external helices were deaminated (see small i's in Figure 4(b)); these two adenosines were also deaminated in L0, emphasizing that ADAR1 is recognizing L4 and L0 as similar molecules. Each time we decreased the length of a loop to make a different barbell RNA, we concurrently increased the length of the adjacent GC-rich helix. Thus, it was possible that the lower selectivity observed in L4 compared to L6 was due to the increase in the length of the external helices (and thus an increase in the overall stability of the duplex) rather than its smaller loops. To test this possibility, we created a chimeric barbell RNA that contained the L6 loops adjacent to L4 external helices (L6/L4 chimera). The L6/L4 chimera showed the same selectivity as L6 (25.9 (2.9) fmol inosine, n ˆ 5; data not shown) but half the total inosines as L4. Thus, the ability of the L6 barbell to be deaminated with the same selectivity and preferences as 36mer is dependent on its internal loops and not its external helices.

Figure 3. Determination of the minimal loop length required to uncouple helices in the barbell molecules. (a) All RNAs shown in Figure 1(a) were incubated with ADAR1 and the amount of IMP produced over time was determined as described in the legend to Figure 2. For each RNA, the reaction was complete at two hours, as veri®ed by analyses of four hour time points. The symbols used in the graphs are correlated with the appropriate RNAs to the right of the plot. The two hour points are the average of four (36mer, L12 and L8), ®ve (L6), or six (L4 and L0) experiments, whereas shorter time points are the average of two experiments; standard deviation (s.d.) bars are shown. (b) The TLC plate shows relative amounts of AMP and IMP at a representative two hour time point used for the graph shown in (a). The positions of AMP, IMP, and the origin (ori) are labeled to the right of the plate. RNAs were incubated with (‡) or without (ÿ) ADAR1 prior to digestion to 50 NMPs. The average amount of IMP observed for all two hour time points is given below the corresponding lane, with s.d.

Discussion

not those 4four nucleotides, were equivalent to helix termini. Figure 3(b) shows the primary data for the two hour time point (when the reaction is complete), as well as the amount of inosine observed. Reactions containing 5 fmol of 36mer, L12, L8 or L6 contained 21-23.5 fmol of inosine at reaction

Here we show that internal loops play an important role in determining the number of deamination events that occur within an ADAR substrate. In particular, we show that internal loops are equivalent to helix termini for ADAR1. Thus, a single, long dsRNA substrate can be converted into a series of short, independent substrates by the insertion of internal loops. Since ADARs deaminate short dsRNAs more selectively than long dsRNAs, internal loops can dramatically decrease the num-

Internal Loops Promote Selective ADAR Deamination

ber of deamination events that occur within an RNA. The restriction of adenosine deamination to a select number of sites is essential when ADARs act on mRNAs to alter speci®c codons. Our work provides the ®rst information on how secondary structure within ADAR substrates contributes to selective deamination and also adds to the general understanding of how internal loops function within RNAs. A selectivity model In our favorite model, all aspects of ADAR selectivity can be understood with respect to its nucleic acid binding properties. Simply put, ADARs bind to double-stranded RNA but not single-stranded RNA (Bass & Weintraub, 1987; Wagner & Nishikura, 1988). Since ADARs are not sequence speci®c (Polson & Bass, 1994), an increase in the length of a dsRNA, or the number of base-pairs it contains, should increase the number of ADAR binding sites; similar arguments have been made with regard to sequence-independent DNA-binding proteins (McGhee & von Hippel, 1974). According to our model, a long dsRNA would be deaminated at more adenosines because it has more binding sites. For example, if we assume that a single molecule of ADAR1 interacts with 20 bp, we would predict the 36mer has 17 different ADAR binding sites (36 ÿ 20 ‡ 1), while the L0 molecule, with 66 bp, would have 47 different binding sites. For the barbell molecules analyzed, we observed that loops 5six nucleotides uncoupled adjacent helices and increased selectivity, while loops 4four nucleotides did not (Figures 3(b) and 4(b)). According to our model, and as shown in Figure 5, this difference exists because the internal loops of L6 decrease the number of binding sites compared to L4. To understand why loops must be a certain length to uncouple adjacent helices, it is important to remember that a particular sequence of RNA is never completely paired or unpaired, but at equilibrium favors one state over the other. Since the unfavorable entropy associated with incorporating loop nucleotides into a helix increases with loop length, the population of L6 molecules will have fewer paired loops at equilibrium than the population of L4 molecules. Further, the addition of a dsRBP such as ADAR1 would shift the equilibrium toward the paired state. Although the binding energy incurred when an ADAR interacts with a substrate might be able to compensate for disruption of a helix by a single mismatch, successive unpairing of adjacent base-pairs will eventually create an internal loop whose free energy favors the unpaired state. From these considerations we expect that the minimal length necessary for a loop to function as a helix end will vary with the sequence of the loop and adjacent helices, and further, that certain asymmetric internal loops will also act as helix termini. Our future studies will be aimed at exploring these issues.

7 Our experiments with the L12 barbell suggest that some loops will be too large to maintain the ®delity of deamination within an adjacent helix. In contrast to the other barbell molecules studied, the deamination sites within L12 were not identical with those within the 36mer. In particular, one additional, very minor, deamination site occurred at A2 of L12 (Figure 2(c)). We suspect that the large loops of L12 are more destabilizing than the loops of other barbells, and may be exploring alternative conformers that involve nearby sequences. Possibly, an alternate pairing between the loops and bases within the helix promotes deamination at A2. Based on our observations with L12 and the other barbell molecules, we predict that loops that can both de®ne helix termini for ADARs and maintain the ®delity of deamination, will have an optimal size range, being neither too small to uncouple helices (like L4), nor too large to sample competing conformations with nearby sequences. Given that ADAR1 prefers to bind dsRNA over ssRNA, anything that shifts the equilibrium between the paired and unpaired state should effect ADAR binding. Thus, in addition to helical length, alterations of the thermodynamic stability of a duplex, such as the insertion of mismatches, should alter its selectivity. We propose that L4 is deaminated a bit more selectively than L0, because its internal loops, although too short to act as helix termini, increase the single-stranded character of the RNA. These ideas are related to those raised during early studies of ADAR to explain why deamination of completely base-paired RNAs stopped when 50-60 % of the adenosines were deaminated. Since modi®cation of AU base-pairs in dsRNA creates the considerably less stable IU mismatch (Bass & Weintraub, 1988; Stroebel et al., 1994), it was proposed that the reaction stopped when the molecule no longer had enough double-stranded character to be bound and deaminated by an ADAR (Bass & Weintraub, 1987; Polson & Bass, 1994).

What structural features allow internal loops to function as helix ends? The RNase T2 experiments were performed under the same conditions used for ADAR1 reactions, and indicate that under these conditions, the structures of the barbells are largely as we designed them. That is, sequences designed to exist as unpaired loops were digested with the singlestrand-speci®c ribonuclease, while regions designed to exist as double-stranded helices were resistant. However, the RNase T2 data do not give speci®c details about possible non-canonical features of the loops. Further, since RNase T2 is a single-stranded RNA binding protein, the addition of RNase T2 to the RNA solution could shift the equilibrium between the paired and unpaired states from that normally encountered by ADAR1.

8

Internal Loops Promote Selective ADAR Deamination

Figure 4 (Legend shown opposite)

Internal Loops Promote Selective ADAR Deamination

9

Figure 5. A model to explain why internal loops promote selectivity. According to our model, internal loops promote selective deamination by decreasing the number of ADAR1 binding sites within a substrate. ADAR1 molecules (green) are shown interacting with a completely base-paired dsRNA (c), or the same molecule with internal loops large enough to uncouple helices (a) or too short to uncouple helices (b). Since ADARs are not sequence speci®c, binding sites occur along the entire length of a completely base-paired dsRNA, and these molecules are deaminated promiscuously (c). Internal loops like those in (a) decrease the number of binding sites and allow more selective deamination; these loops may create a physical end to the helix by thermodynamically uncoupling adjacent helices. As depicted for the external helices of the molecule in (a), internal loops may sometimes increase selectivity by creating helices that are too short to support ADAR binding. The internal loops in the molecule shown in (b), while too short to uncouple helices, decrease the overall stability of the molecule. As previously proposed (Polson & Bass, 1994; Bass, 1997), mismatches and small internal loops may decrease the number of deamination events by shifting the conformational equilibrium towards the single-stranded state, thus decreasing the number of ADAR binding sites in the population. As shown for each example, the IU mismatches (red) that accumulate during the reaction ultimately lead to a molecule that is too unstable to be further modi®ed, and at this point the reaction terminates. See the text for further discussion. ADAR1 molecules are depicted according to previous observations regarding deamination preferences (Polson & Bass, 1994).

Previous studies show that internal loops within an RNA molecule can increase or decrease thermodynamic stability (Peritz et al., 1991; SantaLucia et al., 1991; Xia et al., 1997), promote conformation-

al ¯exibility such as bending (Zacharias & Hagerman, 1996), and widen the major groove of helical regions (Weeks & Crothers, 1993). Structural analyses show that the nucleotides of some loops

Figure 4. Location of inosines within L6 and L4. (a) The locations of inosines in L4 (lanes 1-4) and L6 (lanes 5-8) were mapped by primer extension sequencing as described in the legend to Figure 2(b). Numbering of adenosines within the central helix is shown to the left of the gel. The particular dideoxynucleotide included in each reaction, as well as whether the RNA was incubated in the presence (‡) or absence (ÿ) of ADAR1 is indicated at the top of the autoradiogram. RT sequencing products (cDNAs) produced from deaminated RNAs (‡ ADAR1; lanes 2, 4, 6, 8-10) have a faster gel mobility due to their altered nucleotide composition (cytidines instead of thymidines) relative to unmodi®ed samples (ÿADAR1; lanes 1, 3, 5, 7). Further, since cDNAs derived from deaminated RNAs are complex populations of molecules, bands in these lanes are less sharp in appearance. Inosines within L0 were mapped using the same method (data not shown). In some instances, sequence information near the primer annealing site was unclear, possibly due to abortive RT elongation of the primer. (b) The diagram compares the location of inosines within 36mer, L6, L4 and L0 RNAs. The continuous black line indicates the boundary between molecules whose loops are long enough to uncouple helices (5six nucleotides) and those whose loops are too short to uncouple helices (4four nucleotides). Inosines were not observed within the L6 and L4 loops. Barbell Y strands were not mapped for inosines, since previous studies (Polson & Bass, 1994) show that the three adenosines within the 36mer Y strand are not deaminated. While L0 may contain deaminated adenosines on its Y strand due to the increased stability of this duplex, the possible presence of such deaminations does not affect the conclusions drawn in this study.

10 incorporate into adjacent helices (Holbrook et al., 1991; Jang et al., 1998), while those of others disrupt the structure of an otherwise continuous A-form helix (Wimberly et al., 1993; Cai & Tinoco, 1996; Correll et al., 1997). At present we do not know which of these structural features, if any, are important for interactions of loops with ADARs. However, as suggested by the selectivity model proposed above, we prefer the idea that internal loops act as helix ends not because ADARs recognize and interact with their particular structural features, but because their particular structural features cannot be recognized as an RNA helix. This idea is consistent with the fact that editing sites in RNAs whose structures are proven occur in canonical base-pairs or single base mismatches. However, it is important to note that there is one report of deamination sites occurring within internal loops (Herb et al., 1996). At present it is not clear if this is a true anomaly or that the predicted structure is incorrect. Our work ®ts nicely in the context of previous studies that demonstrate the ability of internal loops to thermodynamically uncouple RNA helices (Weeks & Crothers, 1993). These previous studies indicate that internal loops are able to uncouple helices by decreasing enthalpically favorable stacking interactions so as to create an effective helical end similar to a blunt end. Although future studies will be required to understand the physical basis of our observations, recognition of RNAs by ADARs may provide an example of a protein:RNA interaction that makes use of the ability of internal loops to thermodynamically uncouple adjacent helices. Relationship of ADAR1 to other dsRBPs Interestingly, like ADAR1, other dsRNA-binding proteins have been observed to distinguish between perfect and irregular RNA duplexes, suggesting that internal loops may serve a similar function in these proteins. RNase III acts promiscuously on long, perfectly base-paired dsRNA, cleaving the RNA at numerous sites to yield 12-15 bp fragments (Robertson, 1982; Nicholson, 1996). In contrast, imperfect duplexes are cleaved at only one to two sites, and internal loops can further restrict cleavage to a single site in some substrates (Chelladurai et al., 1993, and see reviews by Robertson, 1982; Court, 1993; Nicholson, 1996). Likewise, the kinase activity of PKR is activated by 33-80 bp perfect dsRNAs, but not by irregular or shorter dsRNAs (see reviews by Mathews & Shenk, 1991; Clemens & Elia, 1997). All of these proteins (ADAR1, RNase III and PKR) bind dsRNA using an amino acid sequence known as the dsRNA binding motif (dsRBM), suggesting it is this shared sequence that mediates the discrimination between a perfectly base-paired duplex and one containing mismatches, bulges and loops. Consistent with our selectivity model, which proposes differences in secondary structures alter

Internal Loops Promote Selective ADAR Deamination

selectivity by affecting protein binding, both RNase III and PKR bind perfect dsRNAs with higher af®nity than irregular dsRNAs (Chelladurai et al., 1993; Bevilacqua & Cech, 1996). Studies of the effects of internal loops within substrates of other ADARs, such as ADAR2, have not yet been performed. However, the fact that all ADARs have dsRBMs, show promiscuous deamination on completely base-paired dsRNA and selective deamination on dsRNA containing periodic structural disruptions, suggests internal loops function to uncouple helices in substrates of all ADARs. Of course, future studies will need to be performed to con®rm this. Implications for biological substrates Our results help to explain the exquisite selectivity observed in some natural ADAR substrates. For instance, the production of the two isoforms of hepatitis delta antigen relies on highly selective deamination of the HDV antigenomic RNA (Polson et al., 1996). In this case, less than 1 % of the 340 adenosines of the antigenomic RNA are deaminated in vivo, or in vitro with puri®ed ADAR1. The HDV antigenome, while largely basepaired, is predicted to contain numerous internal loops. Based on our studies, we propose that these internal loops allow ADAR1 to recognize the long, 1700 nt HDV antigenome as a series of short segments. Due to the frequency of loops within the HDV antigenome, we predict that most of the helical segments are too short to act as ADAR substrates, thus limiting deamination to a select number of sites. Since ADAR1 deaminates helices ¯anked by appropriate internal loops with the same selectivity and preferences as those with blunt termini, it follows that a helical segment between adjacent loops could be removed from a biological substrate without loss of deamination speci®city. Thus, our studies suggest the ®rst rational approach for delineating minimal substrates for the RNAs that are deaminated by ADARs in vivo.

Materials and Methods Construction of transcription vectors Plasmids pKL36A and pKL36S were used as templates for transcription of the Y and R strands of L12, respectively, and were constructed from plasmid pGEM-9zf (Promega) after excision of the multiple-cloning region and phage T7 and SP6 promoter sequences (NotI to S®I). Excised sequences were replaced by an insert that included an optimal T7 promoter sequence (Milligan et al., 1987) and the template to transcribe one RNA strand of L12. Inserts were created by annealing complementary DNA oligos to create dsDNA inserts with NotI and S®I compatible ends, and are shown 50 to 30 below (36Acpy and 36Atpl anneal to create the pKL36A insert; 36Scpy and 36Stpl anneal to make the pKL36S insert): 36Acpy, CGGCCGTCGACTAGTAATACGACT CACTATAGGGCCCGGGAACACACCTGGTCCCTGTC

11

Internal Loops Promote Selective ADAR Deamination CTTGTTATTTTCCTTGGTTAATTACACAAGGCGCGC CCACACCACGGATCCATGGGCATGC; 36Atpl, GGC CGCATGCCCATGGATCCGTGGTGTGGGCGCGCCTT GTGTAATTAACCAAGGAAAATAACAAGGACAGGG ACCAGGTGTGTTCCCGGGCCCTATAGTGAGTCGTA TTACTAGTCGACGGCCGACT; 36Scpy, CGGCC GTCGACTAGTAATACGACTCACTATAGGGCGCGCC AACACAAATTAACCAAGGAAAATAACAAGGACAG GGACCAGGACACAACCCGGGCCCACGCACCAGAT CTAGAGCATGC; and 36Stpl, GGCCGCATGCTCTA GATCTGGTGCGTGGGCCCGGGTTGTGTCCTGGTCCC TGTCCTTGTTATTTTCCTTGGTTAATTTGTGTTGGCG CGCCCTATAGTGAGTCGTATTACTAGTCGACGGCC GACT. Gel-puri®ed oligo nucleotides were phosphorylated with phage T4 polynucleotide kinase (T4 PNK) and ATP prior to annealing in 10 mM Tris (pH 8.0), 1 mM EDTA (1  TE), and ligating with the 2818 bp fragment of the pGEM-9zf vector. Nucleic acid preparation Transcription vectors pKL36A and pKL36S were linearized at the BamH1 and BglII (NEB) sites, respectively, within the inserted sequence for use in run-off transcription. Transcription reactions (100 ml) contained 1 unit/ml phage T7 RNA polymerase (Life Technologies) and the manufacturer's buffer, and included 500 mM NTPs, 1.8 units/ml rRNasin (Promega), 100 mM DTT, 5 mg linearized plasmid. Other ssRNAs (Y and R strands of 36mer, Y and R strands of the L6/L4 chimera, and Y strands of L8, L6, L4 and L0) were synthesized from partially ssDNA templates that included an optimal T7 promoter sequence (Milligan et al., 1987) using the T7 MegaShortScript in vitro transcription kit (Ambion). RNAs were transcribed in the presence of either 0.2 mCi 3000 Ci/mmol [a-32P]ATP for TLC applications (in which case the non-radioactive ATP concentration was lowered to 250mM), or 0.1 mCi [5,6-3H]UTP to enable quanti®cation of the transcript. For RNase T2 and RNase T1 mapping applications, the RNA strand to be analyzed was gel puri®ed, then labeled at its 50 end using 6000 Ci/mmol [g-32P]ATP as described (Polson & Bass, 1994). The L12 R strand was annealed with other Y strands to create all the barbell RNAs shown in Figure 1(a). Gelpuri®ed Y and R strand RNAs were combined in equal molar amounts in a 15 ml volume in 10 mM Tris (pH 8.0) and annealed by heating to 95  C for three minutes, followed by slow cooling to room temperature for 45 minutes. Annealed RNAs were incubated brie¯y at ÿ20  C before the addition of 15 ml native gel buffer (2  ˆ 20 % (w/v) sucrose, 1  TBE, 0.5 % (w/v) SDS, 0.01 % (w/v) xylene cyanol and 0.01 % (w/v) bromophenol blue), and electrophoresis at 100 V for 12 hours on a 10 % (w/v) polyacrylamide native gel. dsRNAs were processed as described (Polson & Bass, 1994) prior to quanti®cation in triplicate by diethylaminoethyl cellulose (DEAE, Whatman) ®lter paper binding. Deamination reactions Adenosine deamination reactions (50 ml) contained 5 ml Xenopus laevis xADAR1 AF-Blue-650 M pool (Hough & Bass, 1994) and 5 fmol dsRNA in the standard assay buffer (1  AB) described by Hough & Bass (1994), except additional glycerol was excluded and EDTA was

5.1 mM. Both ADAR1 and dsRNA were pre-incubated separately in 1  AB for ten minutes on ice or at 25  C, respectively, prior to mixing of these components. RNAs were reacted to completion with ADAR1 (120 minutes, 25  C) unless otherwise noted. Reactions were stopped, and RNAs puri®ed as described (Polson & Bass, 1994). Thin-layer chromatography For TLC, RNAs were digested to 50 NMPs as described (Saccomanno & Bass, 1994), except nuclease P1 reactions were at 50  C in 40 ml of 1  TE. Puri®ed 50 NMPs were lyophilized, Cerenkov counted, and resuspended in water to the same cpm/ml. PEI-cellulose plates (Bodman) were pre-run in water and dried as described (Paul & Bass, 1998), prior to spotting nuclease P1 reaction products and 50 AMP and 50 IMP markers (prepared as described by Bass & Weintraub (1988)). Spotted plates were dried, then soaked in anhydrous methanol as described (Paul & Bass, 1998), prior to one-dimensional chromatography in a standard solvent (Saccomanno & Bass, 1994). Percentage deamination was calculated as described (Saccomanno & Bass, 1994). The molar amount of inosine was determined as the product of the percentage of deamination, the number of adenosines per molecule of unreacted duplex, and the amount of dsRNA reacted (5 fmol). Primer extension Oligonucleotides used as primers for primer extension were gel puri®ed, then 50 -end-labeled with phage T4 PNK (USB) and 6000 Ci/mmol [g-32P]ATP. Oligo-1 (50 CCTGGTCCCTGT 30 ) was used for primer extension of 36mer and L12, and SR14D (50 TCTGGTGCGTGGGC 30 ) was used for primer extension of barbell RNAs. Oligo-1 hybridizes to the last 12 nt of the R strand of the 36mer sequence. Mapping of inosines using a mutant (E46Q) RNase T1 (Steyaert et al., 1991) as described (Polson & Bass, 1994) demonstrated that the oligo-1 primer did not obscure deaminated adenosines (data not shown). SR14D hybridizes to nucleotides in the right GC-rich helix and 30 overhang of the R strand. Inosines were mapped by primer extension as described (Inoue & Cech, 1985) with the following modi®cations. RNA (50 fmol) was combined with 50 endlabeled primer in 1  hybridization buffer, incubated at 90  C for three minutes, and placed on dry ice to facilitate annealing of primer to the RNA. Reverse transcription reactions contained 2 ml of the above RNA-primer mixture, and were done as described (Inoue & Cech, 1985), except reactions were incubated at 42  C for 30 minutes with 1.5 units AMV-RT (Boehringer Mannheim). Reactions were performed as full extensions to verify RNA integrity, or with the addition of 40 mM ddNTP to obtain sequence information. An equal volume of 2  formamide loading buffer (Polson & Bass, 1994) was added prior to heating cDNAs (®ve minutes, 90  C). Electrophoresis (50 W, three hours) was on a 12 % (w/v, 19:1) polyacrylamide, 1  TBE, 8 M urea gel (43 cm long). Gels were dried and subjected to autoradiography and Phosphorimager analysis. RNase T2 structure mapping Approximately 30,000 cpm of (32P-50 )-end-labeled RNA was incubated in a 10 ml reaction containing 1  AB (see above), 10 mg Torula yeast RNA and RNase

12 T2 (Life Technologies). RNase T2 activity was inef®cient under our reaction conditions at 25  C, and so was used without dilution (23 units). Prior to addition of RNase T2, RNAs were incubated in 1  AB (ten minutes, 25  C). RNase T2 reactions were incubated for 20 minutes, then stopped by adding 10 ml of 2  formamide loading buffer and placing tubes on dry ice. Alkaline hydrolysis of 45,000 cpm RNA and 10 mg Torula yeast RNA was performed as described (Polson & Bass, 1994). Samples were heated (90  C, three minutes), then loaded onto a 12 % (w/v) polyacrylamide, 8 M urea, 1  TBE gel (43 cm long) prior to electrophoresis (50 W, one hour 45 minutes). Gels were dried and subjected to autoradiography and Phosphorimager analysis.

Acknowledgments We thank members of our research group for helpful advice and discussions, and Mark Macbeth and Brian Wimberly for critically reading this manuscript. We thank Ron Hough for providing puri®ed xADAR1, and Ed Meenan for the synthesis of DNA oligos. Oligonucleotides were synthesized by the Howard Hughes Medical Institute oligonucleotide synthesis facility at the University of Utah supported by the Department of Energy (Grant # DE-FG03-94ER61817). This work was supported by funds to B.L.B from the National Institute of General Medical Sciences (GM44073). B.L.B. is an HHMI Associate Investigator.

References Bass, B. L. (1997). RNA editing and hypermutation by adenosine deamination. Trends Biochem. Sci. 22, 157162. Bass, B. L. & Weintraub, H. (1987). A developmentally regulated activity that unwinds RNA duplexes. Cell, 48, 607-613. Bass, B. L. & Weintraub, H. (1988). An unwinding activity that covalently modi®es its double-stranded RNA substrate. Cell, 55, 1089-1098. Bevilacqua, P. C. & Cech, T. R. (1996). Minor-groove recognition of double-stranded RNA by the doublestranded RNA-binding domain from the RNA-activated protein kinase PKR. Biochemistry, 35, 99839994. Burns, C. M., Chu, H., Rueter, S. M., Hutchinson, L. K., Canton, H., Sanders-Bush, E. & Emeson, R. B. (1997). Regulation of serotonin-2C receptor G-protein coupling by RNA editing. Nature, 387, 303-308. Cai, Z. & Tinoco, I., Jr (1996). Solution structure of loop A from the hairpin ribozyme from tobacco ringspot virus satellite. Biochemistry, 35, 6026-6036. Chelladurai, B., Li, H., Zhang, K. & Nicholson, A. W. (1993). Mutational analysis of a ribonuclease III processing signal. Biochemistry, 32, 7549-7558. Clemens, M. J. & Elia, A. (1997). The double-stranded RNA-dependent protein kinase PKR: structure and function. J. Interferon Cytokine Res. 17, 503-524. Correll, C. C., Freeborn, B., Moore, P. B. & Steitz, T. A. (1997). Metals, motifs, and recognition in the crystal structure of a 5S rRNA domain. Cell, 91, 705-712. Court, D. (1993). RNA processing and degradation by RNase III. In Control of Messenger RNA Stability (Belasco, J. G. & Braverman, G., eds), pp. 71-116, Academic Press, San Diego.

Internal Loops Promote Selective ADAR Deamination Dabiri, G. A., Lai, F., Drakas, R. A. & Nishikura, K. (1996). Editing of the GluR-B ion channel RNA in vitro by recombinant double-stranded RNA adenosine deaminase. EMBO J. 15, 34-45. Egebjerg, J. & Heinemann, S. F. (1993). Ca2‡ permeability of unedited and edited versions of the kainate selective glutamate receptor GluR6. Proc. Natl Acad. Sci. USA, 90, 755-759. Herb, A., Higuchi, M., Sprengel, R. & Seeburg, P. H. (1996). Q/R site editing in kainate receptor GluR5 and GluR6 pre-mRNAs requires distant intronic sequences. Proc. Natl Acad. Sci. USA, 93, 1875-1880. Higuchi, M., Single, F. N., Kohler, M., Sommer, B., Sprengel, R. & Seeburg, P. H. (1993). RNA editing of AMPA receptor subunit GluR-B: a base-paired intron-exon structure determines position and ef®ciency. Cell, 75, 1361-1370. Holbrook, S. R., Cheong, C., Tinoco, I., Jr & Kim, S. H. (1991). Crystal structure of an RNA double helix incorporating a track of non-Watson-Crick base pairs. Nature, 353, 579-581. Hough, R. F. & Bass, B. L. (1994). Puri®cation of the Xenopus laevis double-stranded RNA adenosine deaminase. J. Biol. Chem. 269, 9933-9939. Hurst, S. R., Hough, R. F., Aruscavage, P. J. & Bass, B. L. (1995). Deamination of mammalian glutamate receptor RNA by Xenopus dsRNA adenosine deaminase: similarities to in vivo RNA editing. RNA, 1, 1051-1060. Inoue, T. & Cech, T. R. (1985). Secondary structure of the circular form of the Tetrahymena rRNA intervening sequence: a technique for RNA structure analysis using chemical probes and reverse transcriptase. Biochemistry, 82, 648-652. Jang, S. B., Hung, L. W., Chi, Y. I., Holbrook, E. L., Carter, R. J. & Holbrook, S. R. (1998). Structure of an RNA internal loop consisting of tandem C-A‡ base pairs. Biochemistry, 37, 11726-11731. Lomeli, H., Mosbacher, J., Melcher, T., Hoger, T., Geiger, J. R., Kuner, T., Monyer, H., Higuchi, M., Bach, A. & Seeburg, P. H. (1994). Control of kinetic properties of AMPA receptor channels by nuclear RNA editing. Science, 266, 1709-1713. Maas, S., Melcher, T. & Seeburg, P. H. (1997). Mammalian RNA-dependent deaminases and edited mRNAs. Curr. Opin. Cell Biol. 9, 343-349. Mathews, M. B. & Shenk, T. (1991). Adenovirus virusassociated RNA and translation control. J. Virol. 65, 5657-5662. McGhee, J. D. & von Hippel, P. H. (1974). Theoretical aspects of DNA-protein interactions: co-operative and non- co-operative binding of large ligands to a one-dimensional homogeneous lattice. J. Mol. Biol. 86, 469-489. Melcher, T., Maas, S., Herb, A., Sprengel, R., Seeburg, P. H. & Higuchi, M. (1996). A mammalian RNA editing enzyme. Nature, 379, 460-464. Milligan, J. F., Groebe, D. R., Witherell, G. W. & Uhlenbeck, O. C. (1987). Oligoribonucleotide synthesis using T7 RNA polymerase and synthetic DNA templates. Nucl. Acids Res. 15, 8783-8798. Nicholson, A. W. (1996). Structure, reactivity, and biology of double-stranded RNA. Prog. Nucl. Acid Res. Mol. Biol. 52, 1-65. Nishikura, K., Yoo, C., Kim, U., Murray, J. M., Estes, P. A., Cash, F. E. & Liebhaber, S. A. (1991). Substrate speci®city of the dsRNA unwinding/modifying activity. EMBO J. 10, 3523-3532.

Internal Loops Promote Selective ADAR Deamination O'Connell, M. A. (1997). RNA editing: rewriting receptors. Curr. Biol. 7, R437-439. Paul, M. S. & Bass, B. L. (1998). Inosine exists in mRNA at tissue-speci®c levels and is most abundant in brain mRNA. EMBO J. 17, 1120-1127. Peritz, A. E., Kierzek, R., Sugimoto, N. & Turner, D. H. (1991). Thermodynamic study of internal loops in oligoribonucleotides: symmetric loops are more stable than asymmetric loops. Biochemistry, 30, 64286436. Polson, A. G. & Bass, B. L. (1994). Preferential selection of adenosines for modi®cation by double-stranded RNA adenosine deaminase. EMBO J. 13, 5701-5711. Polson, A. G., Bass, B. L. & Casey, J. L. (1996). RNA editing of hepatitis delta virus antigenome by dsRNA-adenosine deaminase. Nature, 380, 454-456. Robertson, H. D. (1982). Escherichia coli ribonuclease III cleavage sites. Cell, 30, 669-672. Saccomanno, L. & Bass, B. L. (1994). The cytoplasm of Xenopus oocytes contains a factor that protects double-stranded RNA from adenosine-to-inosine modi®cation. Mol. Cell. Biol. 14, 5425-5432. SantaLucia, J., Jr, Kierzek, R. & Turner, D. H. (1991). Stabilities of consecutive A. C, C. C, G. G, U. C, and U. U mismatches in RNA internal loops: evidence for stable hydrogen-bonded U.U and C.C.‡ pairs. Biochemistry, 30, 8242-8251.

13 Steyaert, J., Opsomer, C., Wyns, L. & Stanssens, P. (1991). Quantitative analysis of the contribution of Glu46 and Asn98 to the guanosine speci®city of ribonuclease T1. Biochemistry, 30, 494-499. Stroebel, S. A., Cech, T. R., Usman, N. & Beigelman, L. (1994). The 2,6-diaminopurine riboside-5-methylisocytidine wobble base pair: an isoenergetic substitution for the study of G-U pairs in RNA. Biochemistry, 33, 13824-13835. Wagner, R. W. & Nishikura, K. (1988). Cell cycle expression of RNA duplex unwindase activity in mammalian cells. Mol. Cell. Biol. 8, 770-777. Weeks, K. M. & Crothers, D. M. (1993). Major groove accessibility of RNA. Science, 261, 1574-1577. Wimberly, B., Varani, G. & Tinoco, I., Jr (1993). The conformation of loop E of eukaryotic 5S ribosomal RNA. Biochemistry, 32, 1078-1087. Xia, T., McDowell, J. A. & Turner, D. H. (1997). Thermodynamics of nonsymmetric tandem mismatches adjacent to G.C base pairs in RNA. Biochemistry, 36, 12486-12497. Yang, J.-H., Sklar, P., Axel, R. & Maniatis, T. (1995). Editing of glutamate receptor subunit B pre-mRNA in vitro by site-speci®c deamination of adenosine. Nature 374, 77-81. Zacharias, M. & Hagerman, P. J. (1996). The in¯uence of symmetric internal loops on the ¯exibility of RNA. J. Mol. Biol. 257, 276-289.

Edited by D. E. Draper (Received 12 March 1999; received in revised form 25 May 1999; accepted 28 May 1999)