Analysis of the Functional Role of a G·A Sheared Base Pair by in Vitro Genetics*

A classical genetic strategy has been combined with an in vitro selection method to search for functional interactions between the two domains of the hairpin ribozyme. G21 is located within internal loop B; it is proposed to form a sheared base pair with A43 across loop B and to bind a Mg2+ ion. Both nucleotides are important for ribozyme function, and G·A sheared base pairs are a very widespread motif in structured RNA. We took advantage of its presence in the hairpin ribozyme to study its functional role. Pseudorevertants, in which the loss of G21 was compensated by mutations at other positions, were isolated by in vitro selection. The vast majority of G21 revertants contained substitutions within domain A, pointing to functional communication between specific sites within the two domains of the hairpin ribozyme. The possibility of a direct or redundant contacts is supported by electrophoretic mobility shift studies showing that a complex formed between domain B of the ribozyme and the substrate was disrupted and restored by base substitutions that have analogous effects on catalytic activity. The functional significance of this complex, the role of the nucleotides involved, and the basis for magnesium ion requirement is discussed.

Four natural small catalytic RNA motifs are known to mediate cleavage and ligation, reactions central to RNA processing events associated with replication of the genomes of certain RNA viruses or virus-like RNA replicons. These include the hairpin (1-3) hammerhead (4,5), hepatitis ␦ (6), and Varkud satellite (7) ribozymes (also see Ref. 8). Divalent cations including Mg 2ϩ are known to support catalysis. However, recent evidence indicates that divalent metal ions function to facilitate ribozyme folding but do not play a direct obligatory role in catalysis for the hairpin, hammerhead, and VS ribozymes (9 -12). For each of these three ribozymes, it appears that catalytic function is resident in the folded structure of the RNA itself.
The 21-kDa complex between the 50-nucleotide hairpin ribozyme and its 14-nucleotide substrate consists of four short helical elements and two internal loops (Fig. 1). The complex is organized into two domains. Domain A consists of the substrate and substrate-binding sequence, forming helix 1, internal loop A, and helix 2. Domain B is located entirely within the ri-bozyme and consists of helix 3, internal loop B, and helix 4. The two domains of the ribozyme-substrate complex interact through tertiary contacts to generate a complex necessary for catalytic activity. This was initially established through linker insertion studies (13,14) and domain separation and reconstitution experiments (15). Covalent cross-linking studies were used to develop a preliminary tertiary structure model for the ribozyme-substrate complex (16). Subsequently, fluorescence resonance energy transfer analysis (17,18), and hydroxyl radical protection experiments (19) were used to establish the requirements for docking of the two domains and to map the interface between the domains in the docked complex. G 21 is located at a particularly interesting location within the hairpin ribozyme. Ultraviolet irradiation at low intensities induces a specific photo-cross-link between G 21 and U 42 (20). Although the cross-linked RNA has little or no catalytic activity (20), the cross-linkable conformation is likely to represent a ground state structure of the active ribozyme species, because 1) 90% cross-linking can be achieved with molecules that fold homogeneously and 2) domain B modifications that increase cross-linking yield concomitantly increased catalytic activity (21). 1 The region of loop B in which G 21 is embedded represents a common folding motif, also found within loop E of eukaryotic 5 S rRNA, the conserved central domain of viroids, and the sarcin-ricin loop of large rRNA (20,(22)(23)(24). Within this motif, a local reversal of strand direction results in cross-strand stacking of the two cross-linkable bases. In the context of the hairpin ribozyme, G 21 forms a sheared base pair with A 43 and stacks on U 42 (Fig. 2). A remarkable feature of this structure is that both the major and minor grooves are wide, and the Watson-Crick base pairing faces of some of the bases are left unpaired and are projected into the grooves. This structural model has recently been confirmed by the NMR-derived structure of the isolated loop B (22). It has been proposed that this motif is well suited for specific binding of protein or RNA species (25). Particularly noteworthy is the observation that N-7, N-1, and N-2 of G 21 are protected from chemical modification upon Mg 2ϩ addition (22). This could be related to the recent finding that a Mn 2ϩ ion can bind in between A 20 and G 21 (26).
The identification of specific tertiary contacts between the two domains presents a very challenging problem. Several biochemical and biophysical strategies have been used successfully to probe the tertiary structure of small ribozymes. Recently, an interaction between the nucleotide G ϩ1 of the substrate and C 25 has been biochemically identified (27).
Here, we describe a fundamentally different strategy, based on a classical genetic approach, to search for functional interactions between the two domains of the hairpin ribozyme. The method is in vitro selection of pseudorevertants of loss of function mutations. In genetically tractable biological systems, the selection and analysis of second site revertants has been widely applied to examine functional interactions between molecules and within different parts of the same molecules. Advantages of the in vitro adaptation of this strategy to RNA structure are, first, that the investigator has complete control over the location, distribution, and frequency of mutations and, second, that the selection can be performed with very many more variants than can be accommodated in an in vivo selection. Our results clearly demonstrate functional compensation between specific sites within the two domains of the hairpin ribozyme. A detailed study allowed us to assess the function of the G⅐A sheared base pair and to understand the role of one of the magnesium ions required for activity.

MATERIALS AND METHODS
RNA Preparation-RNA was transcribed with T7 RNA polymerase using synthetic oligonucleotide templates (28). Partially doublestranded oligonucleotides (200 pmol) were transcribed by T7 RNA polymerase at 37°C for 3 h in 1-ml reactions containing 40 mM Tris-HCl (pH 8), 25 mM MgCl 2 , 2 mM NTPs, 1 mM spermidine, and 5 mM dithiothreitol. Transcripts were purified by electrophoresis through 20% polyacrylamide, 7 M urea gels, eluted by diffusion, precipitated, and quantitated by UV absorbance. Body-labeled substrate RNA was prepared as described above, except that 100 pmol of template was used in a 100-l reaction containing 50 pmol of [␣-32 P]CTP and 0.5 mM unlabeled CTP. Following gel purification, substrates were quantified by measuring radioisotope incorporation.
Mixed RNA-DNA oligonucleotides were synthesized by solid-phase synthesis as described (29). End-labeling of RNA molecules was accomplished by dephosphorylation of 5Ј-ends using calf intestinal alkaline phosphatase (1 unit/50 pmol of RNA) for 1 h at 50°C, followed by phosphorylation with T4 polynucleotide kinase (2 units/10 pmol of RNA) using 25 pmol of [␥-32 P]ATP and purified as described above.
Activity Assays-Cleavage assays were conducted using hairpin ribozymes synthesized in two segments, as described (30). The 5Ј and 3Ј ribozyme segments were mixed at a concentration of 100 nM, denatured for 1 min at 90°C, and then renatured for 20 min on ice in cleavage buffer (5 or 12 mM MgCl 2 (as indicated), 40 mM Tris-HCl (pH 7.5)). To assay a large number of mutants under standardized conditions, we examined cleavage time courses under the following multiple turnover conditions, which allows us to have a reasonable estimate of the cleavage efficiency over a wide range of activity. In these experiments, 20 nM ribozyme and 200 nM substrate were incubated at 37°C in cleavage buffer (5 or 12 mM MgCl 2 (as indicated), 40 mM Tris-HCl (pH 7.5)), reactions were initiated by mixing an equal volume of ribozyme and substrate, and aliquots were removed and quenched at 0, 5, 10, 15, 30, and 60 min. Reaction products were separated using a 20% denaturing polyacrylamide gel, and dried gels were quantified using a Bio-Rad GS-250 Molecular Imager. To determine initial reaction rates, results were fitted by linear regression to the equation ln F ϭ Ϫkt ϩ b, where F is the unreacted fraction and t is the reaction time. Each mutant was assayed side by side with the wild type construct and relative initial rate constants (k rel ϭ k mutant /k wt ) were calculated. Each construct was assayed at least twice. Absolute initial rates varied by up to 30% from experiment to experiment, but the reported relative values were reproducible to within a 10% error, usually less. Mg 2ϩ dependence was determined using the cleavage protocol described above but using single turnover conditions (200 nM ribozyme, 20 nM substrate); the Mg 2ϩ concentrations used were 2, 5, 10, 20, 50, and 100 mM, and depending on the efficiency of the considered ribozyme, aliquots were taken at 0, 2, 5, 10, and 15 min or at 0, 5, 10, 15, 30, and 60 min.
Photo-cross-linking-RNA cross-linking was performed as described in Ref. 20. Gel-purified radiolabeled RNA was denatured at 70°C for 2 min and then renatured for 20 min on ice in cleavage buffer (12 mM MgCl 2 ). Twenty-microliter aliquots were irradiated at a distance of 1 cm with a 254-nm hand-held ultraviolet light source. Irradiated samples were then analyzed by electrophoresis through 15% denaturing polyacrylamide-urea gels, and reaction products were quantitated as described above. Cross-linked nucleotides were mapped using end-labeled RNA and partial alkaline hydrolysis and identified by electrophoresis alongside enzymatic RNA sequencing ladders (20).
In Vitro Selection-The general method was carried out as described (31,32). The transcriptional template was generated by annealing and extension of two overlapping oligonucleotides, a 74-nucleotide segment containing the coding sequences for a T7 promoter, a primer binding site, part of the ribozyme sequences (5Ј-TAATACGACTCACTATAGG-ACGCACGTGAGCTCGAGTAAACAGAGAANNCACCCAGA(ATC)AA-ACACACGTTGTGGT-3Ј), and a 63-nucleotide segment representing the antisense sequences for the remainder of the ribozyme, a linker, the substrate, and a second primer binding site (5Ј-CGACGTCGGCTCTA-GAGAAACAGGACTNNCAGAGATCTGTACCAGGTAATATACCACA-ACGTG-3Ј). Overlapping sequences are in boldface type. Underlined nucleotides were mutagenized to a total frequency of 12% (i.e. each mutant base was present at a frequency of 4%). The double-stranded template (400 pmol) was transcribed with T7 RNA polymerase for 3 h; these conditions are permissive for self-cleavage of the wild type molecule and active variants. Cleaved (active) and uncleaved (inactive) molecules were separated on a 8% polyacrylamide, 7 M urea gel and eluted. For the active pool, ligation reactions were carried out by incubating the active fraction with 200 pmol of a substrate for ribozyme ligation in cleavage buffer (12 mM MgCl 2 ) at 0°C for 30 min. The oligonucleotide substrate for ligation was a DNA-RNA hybrid containing a binding site for primer 2 (P2, 5Ј-GUCCUGUUUTCTAGAGACGC-CTGCTG-3Ј; ribonucleotides in boldface type). The ligation product was isolated using a 8% polyacrylamide, 7 M urea gel, reverse-transcribed, and amplified by polymerase chain reaction using the primers P1 (5Ј-CACGAGGCGTCTCTAGA-3Ј) and P2 (5Ј-TAATACGACTCACTAG-GACGCACGTGAGCTCG-3Ј). Inactive molecules were reverse transcribed and amplified using primers P2 and P3 (5Ј-CGACGTCGGCTC-TAGAG-3Ј).
Gel Shift Assay-Substrate and loop B domain interactions were analyzed using a gel shift assay essentially as described previously (33). Trace amounts of 5Ј-radiolabeled substrate (Ͻ1 nM) were incubated in siliconized tubes with increasing amounts of unlabeled loop B domain (0, 0.1, 0.5, 1, 2, 5, 10, and 20 M). The mix was incubated for 20 min at 37°C, and binding was allowed to proceed for 2 h on ice. The 37°C 20-min incubation allows the loop B domain to reach the same folding state as a "standard" denaturation renaturation (90°C for 1 min followed by 20 min on ice), but avoids dimer formation as monitored with a cross-link assay (20). 2 After the addition of 5% glycerol, samples were run on a 15% nondenaturing polyacrylamide gel containing 25 mM magnesium acetate and 40 mM Tris acetate, pH 7, for 16 h at 11 watts and 4°C. The results were quantitated using a Bio-Rad GS 250 molecular imager. The proportion of bound substrate as a function of concentration of loop B domain was plotted on a semilogarithmic scale, and the K D value was estimated as the observed concentration corresponding to half of the binding at saturation. 2 S. Butcher, personal communication.

G 21 Mutants Strongly Inhibit Cleavage-In vitro
selection experiments showed that G at ribozyme position 21 is important for cleavage (34). However, selection experiments generally are of limited utility in understanding defects associated with specific mutations. Typically, only a small fraction of inactive clones are recovered, and inactive variants usually contain multiple base substitutions. To begin analysis of the functional importance of G 21 , we synthesized ribozymes containing each of the four possible bases at position 21, and conducted parallel trans-cleavage assays. G 21 substitutions reduced cleavage efficiency by 10 -100-fold in a cleavage buffer containing 5 mM Mg 2ϩ , with U being the most highly inhibitory (Table I;  cleavage deficiencies was observed at 12 mM Mg 2ϩ for all three mutants. To better evaluate the effect of ions, we carried out a study to compare the magnesium dependence of cleavage for the 21G3 U mutant and the WT. 3 The apparent K Mg 2ϩ for the WT ribozyme is 15 mM, and the saturation is obtained by 50 mM. These values are in good agreement with the values found in other studies using the same sequences (17,37,38). The requirement is clearly increased for the 21G3 U mutant for which the K Mg 2ϩ is about 30 mM, although a level of activity comparable with the WT molecule is reached over 100 mM.
When analyzed on a non denaturing polyacrylamide gel, the WT ribozyme-substrate complex run as two bands. The fastest migrating band corresponds to the active "docked" ribozyme, while the low mobility complex was shown to be an inactive complex in which domain A is stacked over domain B (39,40). When conducted with the 21G3 U ribozyme, the gel shift mobility assay in the presence of 5 mM Mg 2ϩ showed that most of the complex is undocked (data not shown).
Mutations at A 43 , the pairing partner of G 21 in the loop B structure, indicated that a purine is required at this position. 43A3 G shows only a 2-3-fold reduction in cleavage activity. However, no cleavage was detectable during analysis of the 43A3 C and 43A3 U mutation (see also . It is striking that the most inhibitory substitutions at positions 21 and 43 are those with the potential to generate canonical base pairs (21G3 U, 43A3 C, and 43A3 U). These results could reflect that the formation of a 21⅐43 Watson-Crick base pair induces an inactive conformation, possibly by collapsing the helix 3 proximal segment of internal loop B into an A-form extension of helix 3.
In Vitro Selection of Second Site Revertants of G 21 Mutations-In vitro selection was used to search for second site revertants that could compensate for mutation of G 21 to A, U, or C (Fig. 1). The synthetic DNA template encoded a complex population of sequence variants containing equimolar quantities of A, U, and C at position 21, while G was absent. In addition, we mutagenized the 19 functionally important bases within internal loops A and B to a frequency of two additional mutations per molecule and completely randomized the two base pairs of helix 2 that are proximal to internal loop A (see "Materials and Methods"). The mutation 39U3 C had previously been isolated as a nonspecific suppressor of several ribozyme mutations (31,34,41). To avoid the potential recovery of this previously characterized suppressor, we did not mutagenize U 39 into the population before initiating selection. This design yields a pool of variant molecules with a complexity of approximately 10 6 unique sequences (42). To prevent recovery of a true revertant (G 21 ) due to polymerase misincorporation, a single round of selection was performed. Previous experience shows that a single round selection for active variants provides a very high level of enrichment for DNA encoding active ribozymes and that the selected population consists of both highly active and a variety of suboptimal species. 4 It should be noted that a round of selection provides two selective steps for RNA catalysis, cleavage followed by ligation.
The DNA template pool was transcribed for 3 h under conditions favorable for self-cleavage; then cleaved (active) and uncleaved (inactive) species were separated by preparative gel electrophoresis. To empirically assess the sites and frequencies of sequence variation in the pool, cDNA from inactive transcripts was synthesized, cloned, and sequenced. The frequency and position of mutations were consistent with those predicted by design of the variant pool. To recover the active populations, molecules that had undergone cleavage were allowed to carry out RNA-catalyzed ligation, and then cDNA copies of active ribozymes were synthesized, amplified, cloned, and sequenced.
Cloned cDNA molecules from the active pool were characterized by sequencing and by monitoring the amount of selfcleavage during in vitro transcription (reaction including 20 mM MgCl 2 and 2 mM spermidine). Representative clones derived from the active pool are shown in Table II. As expected, all recovered clones contained G 21 substitutions. A small number of clones containing only G 21 mutations were identified and showed the very low levels of self-cleavage activity expected from the trans-cleavage assays described above. A number of pseudorevertants were identified that displayed self-cleavage activities much higher than those with G 21 mutations alone. Most of these clones had dual base substitutions at the Ϫ3⅐12 base pair within helix 2, previously shown to be essential for substrate binding and cleavage (43). In these clones, G 21 mutations to U, C, and A were accompanied by changes of the a Ϫ3 ⅐U 12 base pair to g⅐C, c⅐G, and u⅐G. In addition, a single pseudorevertant was identified, which contained a transversion of the base immediately 5Ј of G 21 , 20A3 C.
Second Site Mutations in Helix 2 Base Pair (Ϫ3⅐12) Rescue G 21 Mutations-The functional compensation of G 21 mutants by substitutions at the Ϫ3⅐12 base pair is particularly interesting because the compensatory mutations lie in the loop A domain of the ribozyme, while the original mutation is in the loop B domain. Trans-cleavage assays using synthetic ribozymes and oligonucleotide substrates were carried out in order to examine more closely the efficiency and specificity of this compensatory effect. Using the two-piece ribozyme construct previously described and characterized (30,37), we generated ribozymes and substrates to test all possible combinations of bases at position 21 and base pairs at Ϫ3⅐12. Results are shown in Fig. 3 and Table III (please note that helix 4 sequence is 5Ј-C 27 ACC-3Ј and 5Ј-GGUG 35 -3Ј except for the 3 The abbreviation used is: WT, wild type. 4 B. Sargueil, A. Berzal-Herranz, and J. Burke, unpublished results. simple G 21 mutants that have a 2-G⅐C-base pair extended helix 4, which slightly enhances their activity (21) and therefore slightly undermines their mutant phenotype).
Because it is the most strongly inhibited, we will first discuss the 21G3 U mutant. The autoradiograms of Fig. 3 indicate that changing the a Ϫ3 ⅐U 12 base pair to g⅐C dramatically increases the activity of the 21G3 U mutant. The increase in the initial cleavage rate is over 100-fold in the presence of 5 mM MgCl 2 (Table III). This same base pair substitution also increases the initial cleavage rate of the G 21 ribozyme, but by a factor of 3 only. Interestingly, all base pair substitutions at Ϫ3⅐12 enhance the activity of the 21G3 U mutant ribozyme. Rate enhancement varies from 7-to 110-fold. A similar pattern is observed when G 21 is changed to A or to C. For each G 21 mutation, the naturally occurring base pair (a Ϫ3 ⅐U 12 ) is the least favorable for cleavage. For 21G3 A and 21G3 C, base pair substitutions at Ϫ3⅐12 increase initial cleavage rates by factors of 2-16 relative to a Ϫ3 ⅐U 12 . In every case, g Ϫ3 ⅐C 12 has the most significant effect.
Regardless of the identity of the base at position 21, cleavage is strongly inhibited by mutational disruption of the Ϫ3⅐12 base pair (data not shown). This result is consistent with the selection experiment described above, in which all active variants recovered could form Ϫ3⅐12 base pairs, and is also fully consistent with previous selection and mutational studies (43,44).
To screen for potential backbone interactions involving nucleotides G 21 , a Ϫ3 , and U 12 , the effects of substitutions with 2Ј-deoxyribonucleotides at these three sites were examined. None had a strong effect; deoxyribose substitution at positions G 21 , a Ϫ3 , and U 12 reduced initial cleavage rates by 2-fold, no reduction, and 4-fold, respectively (data not shown, and see Refs. 45 and 46). We conclude that only U 12 2Ј-OH could potentially be involved in the rescue phenomenon observed.
Another suppressor of G 21 mutations has previously been isolated (31,34,41). To analyze the possibility of an interaction between U 39 and the Ϫ3⅐12 base pair, we combined mutations at these positions together with the 21G3 U mutation. 39U3 C appeared to have a very marginal effect compared with g Ϫ3 ⅐C 12 . We concluded that these two suppressors act in different ways.
Finally, the magnesium requirement for cleavage was determined for the 12U3 C and 21G3 U/12U3 C ribozymes (Fig. 4). The K Mg 2ϩ(app) for the 12U3 C construct is 3-4 mM, and the saturation is obtained at about 10 mM. Once more, this is in good agreement with the data obtained previously by others using the 12U3 C variant of the WT ribozyme (47,48). The revertant shows a similar behavior except that the K Mg 2ϩ(app) value is slightly higher (about 6 mM).
Partial Suppression of 21G3 U by 20A3 C-Independent of the Ϫ3⅐12 substitutions, the mutation 20A3 C was recovered as a second site revertant of the 21G3 U mutation and showed a significant increase in self-cleavage activity relative to 21G3 U. Cleavage assays using trans-acting ribozymes were conducted and confirmed that 20A3 C is indeed a second site suppressor of 21G3 U, increasing activity by a factor of approximately 15 (Table IV, top).
To explore the specificity of this suppression, we synthesized and analyzed a panel of ribozymes with mutations within loop B at positions 20 and 21. In the context of the wild type ribozyme, all mutations of A 20 are slightly inhibitory (Table IV, top). In the 21G3 U background, the only significant compensatory effect observed was that of the original isolate, 21G3 U/20A3 C.
To investigate the possibility of a direct contact between A 20 and the Ϫ3⅐12 base pair, we examined the effects of combinations of A 20 and helix2 mutants on the suppression of 21G3 U mutations (Table IV, bottom). The g Ϫ3 ⅐C 12 substitution strongly enhanced the activity of the 21G3 U/20A3 C and 21G3 U/20A3 U ribozymes but had little effect on the inactive 21G3 U/20A3 G ribozyme. The effect of the Ϫ3⅐12 base pair mutation is independent of the nucleotide present at position 20 except for the inactive 20A3 G mutation (see below). The rescue by a Ϫ3 ⅐U 12 substitution may in some cases obscure the A 20 effect.
Two further results point to the likely importance of noncanonical base pairing between A 20 ⅐C 44 and G 21 ⅐A 43 . First, mutations that would be expected to extend Watson-Crick base pairing of helix 3 into loop B are severely inhibitory (Table IV,  top). Second, the loss of activity induced by the 20A3 G transition can be partially rescued by the 44C3 U mutation (data not shown). A G 20 ⅐U 44 wobble pairing is nearly isosteric to the one seen for A 20 ⅐C 44 in the NMR-derived structure (22).  Table I. The suppressive effect of 20A3 C is likely to be a consequence of a local conformational effect within loop B; this could, for example, reflect the presence of a Mg 2ϩ binding site in between A 20 and G 21 (26).
Effects of A 43 Mutations on the Suppression of G 21 Mutants-In the model of loop B structure, G 21 forms a sheared base pair with A 43 (16,20,22,49). Therefore, the functional effects of A 43 mutations were examined in the wild type molecule (Table V). Since G 21 is also linked genetically to the Ϫ3⅐12 base pair, we also examined the effects of base pair changes at Ϫ3⅐12 on the A 43 mutants. The simultaneous mutation of G 21 and A 43 is very detrimental in any context (Table V). The effect of the mutations of the Ϫ3⅐12 base pair is only seen with G 43 and is really significant for the g Ϫ3 ⅐C 12 substitution, which partially restores activity in the 21G3 U/43A3 G and 21G3 C/ 43A3 G contexts (at least a 40-fold improvement). Together with the data on G 21 , this strongly suggests that the purine requirement at position 43 is not only constrained by the G⅐A sheared base pair.

UV-Cross-linking Assays to Monitor Folding of Loop B-
The structure-induced photo-cross-linking of G 21 to U 42 (20) can be used as an activity-independent assay to monitor folding of the RNA within loop B. We examined the time course and reaction sites of photo-cross-linking for several of the mutants that are under investigation in this study.
Interestingly, all variants at position 21 retain their ability to form a UV-induced cross-link (Fig. 5A), although the yield of cross-linked product is significantly reduced relative to the wild type molecule (Fig. 5B). Mapping of the 5Ј cross-linking site for the 21G3 U mutant shows that the cross-link occurs at or in the immediate vicinity of U 21 (data not shown). The 3Ј crosslinking site maps unambiguously to U 42 , the same site used by the wild type molecule (note that in Ref. 20 the 3Ј cross-linking site was incorrectly interpreted as U 41 ). Together, these results indicate that we observe the same cross-link, reflecting the same local conformation present in all of the molecules tested, although a smaller proportion of the mutant molecules are folded into the photoreactive structure.
In contrast, A 43 mutants show a slightly different pattern. Cross-links to the correct sites are observed for each mutant but with different efficiencies (Fig. 5B). 43A3 G and 43A3 U mutants retain a significant amount of cross-linking, while 43A3 C almost abolishes the photoreactivity.
Although the helix 2 base pair substitution g Ϫ3 ⅐C 12 rescues the cleavage deficiency of the 21G3 U mutant, it has no effect on the reduction in cross-linking efficiency induced by the same mutation. As expected, the Ϫ3a3g⅐12U3 C substitution has no effect on the cross-linking rate of the wild type molecules.
These results indicate that the photoreactive structure containing cross-strand stacking is robust enough to withstand base substitutions that result in severe loss of catalytic activity. The observation that the g Ϫ3 ⅐C 12 substitution partially rescues activity without increasing the cross-linking efficiency confirms that the helix 2 substitution does not affect folding of loop B.
A Complex between the Substrate and Domain B-Gel mobility shift assays were used to examine possible interactions between the substrate (S; from u Ϫ5 to u ϩ9 ), substrate-binding strand (SBS; from A 1 to A 13 ), and domain B (from A 15 to A 50 ). When labeled substrate-binding strand was used, the addition of substrate caused a decrease in mobility consistent with formation of the substrate-substrate-binding strand duplex (Fig.  6A). However, no such shift was observed in the presence of domain B, whether or not the substrate was present. These results are consistent with previously published observations, which demonstrate formation of a stable substrate-strand binding site duplex (40,50,51). In contrast, a complex was observed when labeled substrate was incubated with domain B (Fig. 6B). Two bands with mobilities lower than that of unbound substrate were observed, with the lower mobility species predominating at higher loop B domain concentrations. These two species may reflect differences in conformation or differences in composition. The titration experiment indicates that the substrate-domain B complex has an apparent K D of approximately 0.6 M (Fig. 6B). To better evaluate the significance of the complex, we repeated the experiment with the "docking" mutant ϩ1g3a substrate. Our results are fully consistent with what has been observed with the complete ribozyme using fluorescence resonance energy transfer experiments: the ϩ1g3a substrate shows a decreased affinity (apparent K D of 3.5 M) for the loop B (18).
To examine the possibility of fortuitous base pairing between the substrate and the loop B domain, we used a DNA analog of the substrate as a control and found that no complex could be detected (data not shown). We tested the effect of combinations of mutations at ribozyme position 21 and substrate position Ϫ3. Strikingly, the 21G3 U substitution in domain B eliminated the formation of the substrate-domain B complex, while the Ϫ3a3g substrate substitution restored complex formation (Fig. 6, C and D). The apparent dissociation constant for the complex between the mutant substrate and mutant domain B is 0.1 M, slightly lower than that of the wild type molecules (Fig. 6, C and D). It also runs more homogeneously than the wild type, consistent with the slight increase of activity (Fig.  6C). Disruption and restoration of the substrate-domain B in-  teraction by the same mutations that inhibit and rescue catalytic activity suggest that this complex could be relevant to normal ribozyme function. However, it is important to note that we have made numerous attempts to identify catalytic activity in the substrate-domain B complex under a variety of conditions, and no evidence for catalytic function has been obtained.

Isolation of Suppressor Mutations by in Vitro Selection-In
vitro selection technology has provided new tools for investigating the structure-function relationships of biologically important nucleic acids and for the isolation of RNA molecules with novel properties from random sequence pools. Here, we report an in vitro selection strategy to search for second-site revertants of strongly inactivating point mutations. This strategy has been adapted from organismal genetics, where both intragenic and extragenic suppressors have long been invaluable for identifying functional interactions between sites within and between gene products. Our results show clearly that the same strategy can be successfully conducted entirely in vitro. Biochemical studies of synthetic trans-acting ribozymes confirm that the base substitutions isolated in the selection experiments do, indeed, result in the restoration of RNA cleavage activity to the G 21 mutant ribozymes. While the reversion of the G 21 mutant phenotype is clearly established, these studies do not themselves tell us the exact nature of the functional compensation of G 21 mutations (discussed below).
Structural Importance of the G 21 ⅐A 43 Sheared Base Pair-Because mutations at G 21 retain some activity, the presence of the G 21 ⅐A 43 sheared base pair in the catalytic structure is still a matter of discussion (37,45). Below we discuss the possibility for the local geometry to withstand G 21 mutations. The G 21 ⅐A 43 sheared base pair exposes N-7 and O-6 of G 21 on the outside of the helix (22)(23)(24)(52)(53)(54). Stacking of G 21 and U 42 is likely to be responsible for the observed photo-cross-linking reaction (22). In the phylogenetically derived and NMR-derived model, this base pairing appears as a keystone of the peculiar loop B structure. Strikingly, our findings show that some G 21 and A 43 mutations retain a significant cross-linking activity and, even more surprising, that the hairpin ribozyme can recover a full activity despite the presence of this expected important structural defect.
Several lines of evidence suggest that the overall geometry of a G⅐A sheared base pair may be retained in the presence of mutations. First, the G⅐A pair contains two hydrogen bonds to the backbone (23,24), and the geometry of the pair may be maintained with only one base-specific hydrogen bond when a third strand interacts with the sheared pair (53,54). Second, the base pairs that flank G⅐A pairs are known to constrain backbone orientation and stacking (55). Finally, hydrogen bonds are formed with the adjacent backbone (54,56). Congruent with our finding that some mutations can be tolerated, functional group substitutions of a G⅐A pair in two different sequence contexts can be introduced without dramatic loss of stability (52,57). Similarly, the two A nucleotides participating in the sheared base pairs of the hammerhead ribozyme can also be substituted with ribopurine without dramatic activity loss (58,59). These data together with the retention of a substantial photosensitivity suggest that a significant fraction of the mutant molecules still adopt the correct overall geometry. A recent study combining phylogenetic comparison and molecular modeling of the 5 S rRNA loops shows that all the mutations of the G of a G⅐A sheared base pair can be accommodated with a similar geometry. Our results are in good agreement with Leontis and Westhof modeling, except for the G 21 ⅐U 43 and G 21 ⅐G 43 mutants that retain some photoreactivity and for which they could not model an isosteric pair (60). Interestingly, the U 43 mutant is as photoreactive as the G 43 mutant but in contrast is completely inactive and not rescued by the Ϫ3a3g/ 12U3 C substitution; this suggests another role for A 43 .
In conclusion, it is likely that most of the G 21 ⅐A 43 mutants are not so much altered in the overall structure of loop B but rather in another function, which could be the interdomain interaction.
Possible of 0.6 M and sufficient stability to be visualized by electrophoretic methods. No complexes between loop B and the substrate binding strand could be observed, either in the presence or absence of substrate. Particularly striking is the finding that formation of the complex is inhibited by the 21G3 U substitution and rescued by the Ϫ3a3g change, in correlation with the loss of cleavage activity that results from 21G3 U and its restoration by the Ϫ3a3g change. These results suggest that the mechanism through which 21G3 U inhibits cleavage activity could involve destabilization of an interaction between domain B and substrate.
The nature of the complex between substrate and domain B is unknown. Examination of the sequences for potential Watson-Crick base pairing revealed the potential to form several short duplexes. The longest of them consists of five contiguous base pairs between the 3Ј-end of the substrate (u ϩ5 through u ϩ9 ) and loop B positions A 22 through A 26 . However, four observations indicate that these potential short duplexes may not be responsible for the slowly migrating complex. First, no complex was observed when a DNA analog of the substrate was used. Second, modification of the sequence of helices 3 and 4 did not eliminate complex formation. Third, none of the identified potential interactions would be directly affected by the 21G3 U or the Ϫ3c3g substitutions. Fourth, Pinard et al. (27) recently demonstrated that the g ϩ1 position of the substrate interacts with C 25 of loop B; consistent with this finding, we found that ϩ1g3a mutation destabilizes the observed complex.
The complex between substrate and domain B has no detectable cleavage activity and is therefore unlikely to represent an intermediate on the normal cleavage pathway. However, there are several observations suggesting that the physical interaction that we observe may have some relevance to the catalytic structure. First, formation of the complex is disrupted and restored by the same mutations in the two domains that have been identified as inactivating and compensating mutations on the basis of activity-based in vitro selection experiments. Second, covalent cross-linking experiments indicate that hairpin ribozymes derivatized with an azidophenacyl group at the A 20 -G 21 linkage form specific adducts with the nucleotides at positions 11 and 12, the latter being the base pairing partner of nucleotide Ϫ3 (61). Third, a preliminary structure model of the hairpin ribozyme-substrate complex aligns the two domains such that the Ϫ3⅐12 base pair is closely opposed to the G 21 ⅐A 43 pair (16). Finally, hydroxyl radical footprinting shows that the ribozyme portion of helix 2 and A 43 are internalized upon docking of the two domains (19).
Nature of the Functional Compensation and Role of the G 21 ⅐A 43 Sheared Base Pair-The results summarized above indicate that G 21 is important, but not essential, in maintaining the active configuration of loop B. The g Ϫ3 ⅐C 12 substitution renders the activity relatively independent of the nucleotide present at the position 21. Despite the lack of an obvious specificity, this substitution cannot be considered as a trivial general up-mutation, since it does not have a significant effect on most A 43 mutants and cannot restore the phenotype of the G 8 , 5 g ϩ1 , and G 11 mutants (data not shown). The increase of the wild type molecule upon the g Ϫ3 ⅐C 12 substitution could be due to an improvement in loop A-loop B interaction or simply to the stabilization of a weak helix 2. We do not favor the latter hypothesis, since the rate-limiting step of the hairpin ribozyme seems to be a structural rearrangement upon cleavage or the chemical reaction itself (discussed in Ref. 62). Furthermore, the reverse base pair (c Ϫ3 ⅐G 12 ) does not have such a drastic effect.
When run on a nondenaturing gel, the 21G3 U mutant runs mainly as undocked molecules. Furthermore, our results suggest that both the G 21 ⅐A 43 and the Ϫ3⅐12 base pair are involved in interdomain interactions. Finally, we report here the existence of a loop B-substrate complex that is disrupted upon G 21 mutation and restored by the Ϫ3a3g substitution. Although 5 B. Sargueil and J. Burke, manuscript in preparation. we cannot make conclusions regarding the nature and the functional role of this complex (see above), it clearly indicates that some interactions mediated by G 21 and c Ϫ3 can be made between the substrate and the loop B domain. This led us to the conclusion that G 21 variants are "docking" mutants that can be restored by high Mg 2ϩ concentration. The lack of interpretable specificity of the mutation suppression pattern does not argue in favor of a direct contact between the G 21 and the Ϫ3⅐12 base pairs. In contrast, the identity of A 43 seems constrained beyond the requirement for a sheared base pair. Furthermore, two lines of evidence suggest that A 43 and U 12 are implicated in interdomain interactions. First, using NAIM, Ryder and Strobel (45) identified the Rp phosphate oxygen at A 43 and the 2Ј-hydroxyl at U 12 to be required for optimal activity (45). Second, these two positions are internalized upon docking of the two domains as monitored by hydroxyradical footprinting. We propose that the A 43 shallow groove face interacts with the a Ϫ3 ⅐U 12 minor groove. In the most straightforward scenario, A 43 interacts with a Ϫ3 ⅐U 12 through a Mg 2ϩ ion, which would be superfluous in the presence of g Ϫ3 ⅐C 12 . Nevertheless, this hypothesis does not take into account the newly identified metal binding site between A 20 and G 21 . We are therefore proposing that the interdomain interaction is sensitive to the positioning of A 43 by the G 21 ⅐A 43 base pairing, which in turn is likely to be influenced by the binding of a Mg 2ϩ ion (see below). G 21 mutations may alter the configuration of A 43 , such as it that can no longer interact efficiently with a Ϫ3 ⅐U 12 . The substitution Ϫ3a3g/12U3 C restores the interaction by providing a new dispatch of interacting functional group in the shallow groove and the additional g Ϫ3 exocyclic amine. In any construct, this interaction may need a structural rearrangement to make A 43 more accessible (reviewed in Ref. 2). Although to date we have been unable to develop a compelling structural model for any of these contacts, our hypothesis seem to be a reasonable explanation for our results and the data accumulated by other authors (19,45,47,63). Such a contact could be incorporated in the model of Earnshaw et al. (16) with minor rearrangements. 6 Our data cannot rule out that a Ϫ3 ⅐U 12 and G 21 ⅐A 43 base pairs are independently involved in two functionally redundant interdomain contacts, although in this case, we would have expect the G 21 ⅐A 43 partner to appear in the selection experiment. The existence of an ion binding site at G 21 is already supported by two independent sets of experiments. First, in previous results of chemical modification experiments, G 21 shows protection from kethoxal (N-1, N-2) and NiCr (N-7) modification upon folding Mg 2ϩ addition (49). Second, Butcher et al. (26) have recently identified a Mg 2ϩ binding site between A 20 and G 21 phosphates. In their model, the ion bridges the 2-phosphate and is therefore likely to constrain the geometry of the backbone (see, for example, Ref. 64). Interestingly enough, we isolated 20A3 C as a suppressor of G 21 mutations. In this case, it seems likely that a local conformation effect partially restores the Mg 2ϩ binding site altered by G 21 mutations.
The presence of one or two Mg 2ϩ binding sites in the hairpin ribozyme is still a matter of discussion. Walter et al. (65) hypothesize that two ion binding sites were necessary for a ribozyme constituted of a four-way junction, while only one was required for the two-way junction ribozyme (65). A careful examination of the sequences used in the different studies showed that the Mg 2ϩ requirement is correlated with the identity of the Ϫ3⅐12 base pair independently of the nature of the junction (17,18,37,38,47,62,66). Interpreted in the light of the works cited above, our data show that the WT ribozyme has a requirement for two Mg 2ϩ ions, with the g Ϫ3 ⅐C 12 mutation exempting the ribozyme from binding the second ion. Our results strongly suggest that the second Mg 2ϩ is bound by A 20 -G 21 and that the g Ϫ3 ⅐C 12 mutation allows the ribozyme to make an interdomain contact without having to bind the second Mg 2ϩ .
It is significant to note that the Ϫ3a3g/12U3 C substitutions appear to increase the activity of the wild type ribozyme in low Mg 2ϩ . In addition, it is present in the majority of the molecules derived through in vitro selection (43) as well as in the two other known natural occurrences of the ribozyme (67,68). The reason for the presence of a functional but suboptimal base pair in the sTRSV version of the hairpin ribozyme is not known, and it could be imposed by other viral functions.