The RNA-binding Site of Bacteriophage Q (cid:98) Coat Protein*

The coat proteins of the RNA bacteriophages Q (cid:98) and MS2 are specific RNA binding proteins. Although they possess common tertiary structures, they bind different RNA stem loops and thus provide useful models of spe- cific protein-RNA recognition. Although the RNA-binding site of MS2 coat protein has been extensively char- acterized previously, little is known about Q (cid:98) . Here we describe the isolation of mutants that define the RNA- binding site of Q (cid:98) coat protein, showing that, as with MS2, it resides on the surface of a large (cid:98) -sheet. Mutations are also described that convert Q (cid:98) coat protein to the RNA binding specificity of MS2. The results of these and other studies indicate that, although they bind different RNAs, the binding sites of the two coat proteins are sufficiently similar that each is easily converted by mutation to the RNA binding specificity of the other. The coat proteins of the RNA bacteriophages play dual roles in the

The coat proteins of the RNA bacteriophages Q␤ and MS2 are specific RNA binding proteins. Although they possess common tertiary structures, they bind different RNA stem loops and thus provide useful models of specific protein-RNA recognition. Although the RNA-binding site of MS2 coat protein has been extensively characterized previously, little is known about Q␤. Here we describe the isolation of mutants that define the RNAbinding site of Q␤ coat protein, showing that, as with MS2, it resides on the surface of a large ␤-sheet. Mutations are also described that convert Q␤ coat protein to the RNA binding specificity of MS2. The results of these and other studies indicate that, although they bind different RNAs, the binding sites of the two coat proteins are sufficiently similar that each is easily converted by mutation to the RNA binding specificity of the other.
The coat proteins of the RNA bacteriophages play dual roles in the viral life cycle. In addition to serving as the major structural proteins of the virus particles, they act as translational repressors of viral replicase synthesis. This latter function is the result of coat protein interaction with an RNA stem loop which contains the replicase ribosome-binding site. The coat protein of bacteriophage MS2 is the most intensively studied of the RNA phage coat proteins. Its binding target on viral RNA has been thoroughly characterized (1), coat protein itself has been subjected to detailed genetic analysis of its RNA binding function (2)(3)(4)(5)(6), and x-ray structures of the coat protein in both the free and RNA-bound forms are available (7)(8)(9). The coat proteins of related phages are less well characterized, but, since some bind different RNAs, they provide opportunities to understand the basis of RNA binding specificity. The RNA binding targets of the coat proteins of MS2 and Q␤ are shown in Fig. 1. The two coat proteins are about 25% identical in amino acid sequence and possess highly similar tertiary structures. Thus they utilize a common structural framework to bind structurally distinct RNAs.
We previously reported genetic analyses of the MS2 coat protein RNA-binding site utilizing a two-plasmid system in which coat protein expressed from one plasmid (pCT119) translationally represses synthesis of a replicase-␤-galactosidase fusion protein from the second plasmid (pRZ5). We constructed an equivalent two-plasmid system for Q␤ coat protein in order to similarly dissect its RNA-binding site. Here we describe this system and its application in identifying amino acid residues important for the interaction of Q␤ coat protein with its RNA. We also describe the isolation and characterization of specific-ity mutations that confer to Q␤ coat protein the ability to bind the MS2 translational operator.

EXPERIMENTAL PROCEDURES
Plasmid Constructions-A two-plasmid system suitable for the isolation of MS2 coat mutants with altered translational repressor activities has been previously described in detail (2). In that system coat protein expressed from pCT119 represses the synthesis of a replicase-␤-galactosidase fusion protein from pRZ5. For this study we constructed an analogous genetic system for Q␤. A cDNA of the Q␤ viral genome had been previously cloned in plasmid pQ␤7 (10). We excised the the Q␤ coat sequence from pQ␤7 as a BglII-BamHI fragment and inserted it into the BamHI site of pUC119 (11). The resulting plasmid, pQCT119, places the coat sequence under transcriptional control of the lac promoter of Escherichia coli on a plasmid which contains a ColEI replication origin and confers resistance to ampicillin. The plasmids pRZQ and pRZQ5 contain Q␤ replicase-lacZ fusions and were constructed in a manner analogous to the previously described pRZ5 simply by inserting synthetic oligonucleotides containing different versions of the Q␤ translational operator sequence (see Fig. 1) in place of the MS2 operator of pRZ5. The resulting plasmids use the lac promoter to drive transcription of the replicase-lacZ fusion. In addition each possesses a P15A origin of replication and confers resistance to chloramphenicol. The synthetic translational operator sequences were also inserted into pT7-2 (U. S. Biochemical Corp.) to produce pT7ropQ and pT7ropQ5. We also cloned an operator variant lacking the bulged A into pT7-2, creating pT7ropQ-NB (see Fig. 1).
Construction of Mutational Libraries of the Q␤ Coat Sequence-Mutations were introduced into the Q␤ coat sequence using error-prone polymerase chain reaction with primers that flanked the coat gene of pQCT119. Three different mutagenesis protocols were used. 1) For some experiments we followed the method of Cadwell and Joyce (12) without modification. The PCR 1 reaction conditions were as follows: 20 fmol of DNA, 30 pmol of each primer, 50 mM KCl, 10 mM Tris-HCl, pH 8.3, 7 mM MgCl 2 , 0.5 mM MnCl 2 , 1 mM dTTP, 1 mM dCTP, 0.2 mM dGTP, 0.2 mM dATP. After addition of 5 units of Taq DNA polymerase the reaction was subjected to 30 cycles of 94°C for 1 min, 45°C for 1 min, and 72°C for 1 min. The mutant library produced by this method yielded about 50% repressor-defective clones, and sequence analysis showed that the mutations were highly random. However, the high mutation rate resulted in a predominance of multiple hit mutations. 2) To lower the mutagenesis frequency so as to favor single-hit mutations, conditions were altered to give 1.5 mM MgCl 2 , 0.065 mM MnCl 2 , and 1 mM each dNTP. These conditions resulted in libraries in which 10% of clones had the repressor-defective phenotype. However, DNA sequence analysis of the mutants showed that nucleotide substitutions were not as random as desired; a preponderance of transition mutations was obtained. 3) To reduce the mutagenesis frequency while retaining high randomness we made other modifications. First we altered the conditions of the mutagenesis reaction to include 7 mM MgCl 2 , 0.25 mM MnCl 2 , and 1 mM each dNTP. We also effectively reduced the size of the mutagenesis target. Taking advantage of the existence of internal NdeI and AlwNI sites we fragmented the PCR product into three pieces of approximately equal size, which were then used to replace the corresponding fragments in pQCT119. This resulted in three mutant libraries, each derived from mutagenesis directed to a different one-third of the coat sequence. About 10% of the clones obtained in this way produced defective translational repressors. Since it yielded the largest * This work was supported by a grant from the National Institutes of Health. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Isolation and Characterization of Repressor-defective Mutants-Mutated DNA was introduced into strain CSH41F Ϫ (pRZQ5). After about 24 h of growth on plates containing 5-bromo-4-chloro-3-indolyl-␤-Dgalactoside (X-gal), blue colonies were picked, transferred with sterile toothpicks into 1 ml of 2 ϫ YT medium, and grown to saturation. Cells were then collected by centrifugation, resuspended in 0.5 ml of 10 mM Tris-HCl, pH 7.4, 1 mM EDTA, and disrupted by sonication. After removing cellular debris by centrifugation, bromphenol blue and glycerol were added to the lysed samples. These were then applied to a 1% agarose gel (SeaKem Gold Agarose from FMC Corp. BioProducts) containing 50 mM sodium phosphate (pH 7.4), 1 mM EDTA, and electrophoresed until the dye had migrated 10 cm. Taking advantage of the presence of RNA in the capsids, they were visualized by staining the gel with ethidium bromide followed by UV illumination. Gels were also stained with Coomassie Brilliant Blue. Mutants that passed this capsid assembly screen were subjected to nucleotide sequence analysis and their protein products were purified (2). Filter binding analyses of their RNA binding functions were performed as described before (13) using 32 P-labeled RNA produced by run-off transcription of pT7rop plasmids cleaved with BamHI (14). The dissociation constants of the various proteins were determined in a protein-excess filter-binding assay (15) in which a low concentration of radioactive RNA (about 10 pM) was mixed with varying concentrations of coat protein (1 ϫ 10 Ϫ10 M to 1 ϫ 10 Ϫ6 M) in MMK buffer (0.1 M MES, pH 6, at 0°C, 80 mM KOAc, 10 mM Mg(OAc) 2 , and 0.01 mg/mL bovine serum albumin). Each protein preparation was also subjected to titration with excess 32 P-labeled RNA in a filter-binding experiment in order to determine the concentration of active protein in each preparation. Individual protein preparations  Table II are adjusted accordingly.
Isolation and Characterization of Specificity Mutants-Mutant library DNA was introduced into CSH41F Ϫ (pRZ5) and plated on solid medium containing X-gal. After 24 h of growth at 37°C, colonies displaying white to pale blue colony color phenotypes were picked into 1 ml of LB medium and grown to saturation. Since suppression of the operator-constitutive phenotype may sometimes be the result of an assembly defect rather than changes in RNA binding activity (5), the products of the mutant coat genes were assayed for capsid formation by agarose gel electrophoresis as described above. Those mutants passing the capsid-assembly test were selected for DNA sequence analysis. Solution assays of ␤-galactosidase were performed by the method of Miller (16). The wild-type and mutant coat proteins were purified by methods described elsewhere (2). Operator RNAs labeled with 32 P were produced by transcription in vitro of BamHI-cleaved pT7rop5 (3) or pT7ropQ5 (see above) as described by Draper et al. (14). Protein-excess nitrocellulose filter-binding assays of RNA binding activity were conducted as described by Witherell and Uhlenbeck (15).

Construction of a Two-plasmid Genetic System for Q␤-
The details of the two-plasmid genetic system used in the screening for MS2 translational repressor mutants have been elaborated before (2). For the present study we constructed an analogous system for Q␤ coat protein. Initially we constructed pRZQ, which contains a synthetic version of the wild-type Q␤ operator fused to lacZ. Unfortunately, this version of the operator was utilized poorly for translational repression by Q␤ coat protein.
Inspection of the operator sequence revealed the presence of two AUG triplets (at Ϫ9 and Ϫ4) in addition to the true replicase start site at ϩ1 (Fig. 1B). We wondered whether the existence of these potential translational start sites might increase ribosome binding at the expense of coat binding. Therefore, we constructed pRZQ5. It contains a mutant version of the Q␤ operator which eliminates the extra AUGs while retaining the elements required for recognition by coat protein (see Fig.  1A). Filter binding assays revealed that coat protein binds both operators in vitro with a K d of about 1 ϫ 10 Ϫ9 M (Fig. 2). Cells containing both pQCT119 and pRZQ5 plasmids resulted in white colonies, and ␤-galactosidase activity was repressed 44fold. Therefore, these modifications resulted in efficient translational repression in vivo while leaving the in vitro binding characteristics of the operator unaltered.
The Significance of the Bulged A Nucleotide in the Q␤ Operator-The sequence and structural requirements of the Q␤ operator have been previously defined (15). Those studies showed that binding activity for coat protein required the presence of an 8-base pair stem and a 3-nucleotide loop. Analysis of the binding properties of hairpin variants showed that the exact sequence of the stem was unimportant as long as its secondary structure was preserved. Moreover, the identity of only one of the three loop nucleotides was essential (the A at ϩ8). It was also found that the bulged A nucleotide could be deleted without substantially affecting coat protein binding. We found this surprising, since the activity of the MS2 operator is highly dependent on the presence and identity of the nucleotide at the bulged position (17), and because a bulged A residue is a conserved feature in the translational operators of a number of related RNA phages (18). We constructed pT7ropQ␤NB (no bulge) to test the role of the bulged A. Transcription in vitro produced the bulge-less operator variant shown in Fig. 1B. The results of filter binding studies are given in Fig. 3, where it is seen that the wild-type operator, in our hands, gives a K d of 1 ϫ 10 Ϫ9 M, whereas the bulge-less variant has a K d of about 5 ϫ 10 Ϫ9 M. This represents a 5-fold loss of binding affinity. Although there is a slight disagreement with Witherell and Uhlenbeck (15) over the exact magnitude of the effect of the bulge deletion (they found a 1.6-fold difference), our results confirm that the effects are much smaller than for the similar mutation of the MS2 operator (1). The consequences of bulge deletion on translational repression in vivo could not be tested, because the deletion of the extrahelical A destroys the replicase translational start site.
Random Mutagenesis and Isolation of Repressor-defective Coat Mutants-The two-plasmid system facilitates the isolation of coat mutants with altered translational repressor activities. Wild-type coat protein expressed from the pQCT119 plasmid efficiently represses the synthesis of the replicase-␤galactosidase fusion protein encoded on the plasmid pRZQ5. This results in white colonies on X-gal plates. Mutations that give rise to defective repressors yield blue colonies. After transformation of CSH41F Ϫ (pRZQ5) with a mutant library in pQCT119 and plating on LB medium containing X-gal, clones representing a spectrum of color-colony phenotypes were picked. In order to discard any mutants whose repressor defects might be secondary consequences of wholesale disruption of protein structure, each mutant coat protein was screened for the ability to assemble into virus-like particles using the electrophoretic assay described under "Experimental Procedures." Since capsid assembly requires multiple intersubunit contacts, we assume this to be a sensitive indicator of native folding. Capsids were easily visualized by ethidium bromide staining of host RNA within the particle (results not shown). We worried, however, that the repressor-defective mutants might fail to encapsidate RNA and would be unstainable with ethidium bromide. For this reason gels were also stained with Coomassie Blue. However, in no case did we find a mutant that stained with Coomassie Blue and not ethidium bromide.
Three different error-prone PCR mutagenesis methods were utilized in this study (see "Experimental Procedures" for details). The first used exactly the conditions of Cadwell and Joyce (12). This method is reported to introduce highly random mutations at a frequency of about 0.7%. A library of coat mutations produced in this way yielded about 50% blue colonies, but, of the 900 clones we screened, only 21 passed the capsid assembly test. Nucleotide sequence analysis showed that the high mutagenesis rate resulted in multiple mutations, increasing the likelihood that any given isolate would be defective for protein folding or stability. Consequently, we altered the conditions of PCR to lower the mutation rate to a point that favors single-hit mutations. However, we found that the resulting mutations were less random with regard to substitution type. We eventually found compromise conditions which lowered the mutation rate to an intermediate level while retaining high randomness. We further reduced the frequency of multiple hits by targeting mutations to three different restriction fragments of the gene. This strategy reduces the mutagenesis target size. Most of the mutants we characterized were from this library. To assess the spectrum of mutations present in this library, 21 repressor-defective mutants which also failed the capsid assembly test were subjected to nucleotide sequence analysis. The 32 nucleotide substitutions we identified in the 21 mutants were distributed throughout the coat coding sequence and were equally comprised of transition and transversion mutations. A disadvantage of targeting mutations to restriction fragments is that it precludes the isolation of mutations within the recognition sequences of the restriction endonucleases used to generate the fragments. The NdeI site contains parts of codons 88 -90, and the AlwNI site contains portions of codons 40 -43. Both sites contain sequences affecting potential binding site residues. However, this constraint does not apply to the 21 repressor-defective assembly-competent mutants derived from heavy mutagenesis of the whole gene (see above), and, indeed, one of these mutations alters codon 89 within the NdeI site. Nevertheless, mutations in these positions may be under-represented in our collection.
Characterization of the Mutant Repressors-Mutants that survived the screen for native structure were subjected to DNA sequence analysis. The nucleotide and amino acid substitutions we found are listed in Table I. Representative proteins were purified (2), their concentrations of active protein were determined (see "Experimental Procedures"), and each was tested for its ability to retain 32 P-labeled wild-type Q␤ operator RNA on nitrocellulose filters (15). At some positions in the amino acid sequence more than one substitution was found to result in the repressor-defective phenotype. In such cases only one of the mutant proteins was subjected to RNA binding studies. The  protein-excess binding curves are shown in Fig. 4, and the results are summarized in Table II. Each repressor-defective mutant with the possible exception of V32A corresponded to a defect in RNA binding. The locations of the affected amino acids in the structure of Q␤ coat protein are shown in Fig. 5A.
As was the case with MS2, the Q␤ RNA-binding site resides on the surface of the coat protein ␤-sheet.

Isolation of Q␤ Coat Mutants That
Bind MS2 RNA-Given the fact that the MS2 and Q␤ coat proteins share homologous tertiary structures, we wondered whether the Q␤ coat protein could be modified to bind the MS2 RNA operator. Wild-type Q␤ coat protein expressed from pQCT119 fails to repress synthesis of the MS2 replicase-␤-galactosidase fusion enzyme encoded on pRZ5 and results in the formation of blue colonies on X-gal plates. Random mutations within the Q␤ coat sequence were generated by the error-prone PCR method of Cadwell and Joyce (12), the mutant library was introduced into CSH41F Ϫ (pRZ5) and plated on solid medium containing X-gal. About 4000 colonies were screened for the white or pale blue phenotype and seven were picked for further analysis. Four of these produced coat proteins competent for assembly into virus-like particles as judged by electrophoretic mobility in agarose gels (results not shown). Sequence analysis of the other three revealed that they possessed multiple nucleotide substitutions, and that each contained mutations affecting the so-called FG loop (8). We showed previously that mutations within the FG loop of MS2 coat protein blocked assembly of coat protein dimers into capsids (5). This resulted in an elevated intracellular concentration of the active repressor species and increased translational repression without any improvement of RNA binding affinity. We assumed that these three Q␤ mutants were also in this category and did not characterize them further. Table III shows the nucleotide and amino acid substitutions of the four specificity mutants that passed the capsid-assembly test. The mutants we call suM1 and suM2 have two and three amino acid substitutions respectively, but have the D91N substitution in common. This led us to suspect that the D91N mutation alone was responsible for the translational repressor phenotypes of these mutants. This is also supported by the fact that, of the affected amino acids, only Asp 91 resides in a location in coat protein where it is likely to directly influence RNA binding. The D91N single mutant was constructed by isolating the EcoRI-NdeI fragment of the D91N,N22Y,T18S mutant and ligating it to pQCT119 cut with the same enzymes. For similar reasons we also suspected that the translational repressor phenotype of the suM4 mutant was conferred by the Q65H substitution. In this case, the appropriate single mutant had already been isolated (suM3). The translational repression efficiencies of the various mutants were determined and comparison of the ␤-galactosidase activities revealed that the two single mutants D91N and Q65H, repressed the Q␤ and MS2 operators similarly to the multiple mutants (see Table IV).
The RNA Binding Properties of the Specificity Mutants-The coat proteins produced by the Q65H and D91N mutants were purified and subjected to nitrocellulose filter-binding assays. In these experiments no attempt was made to correct for the fraction of active protein, but the binding activities measured in vitro are consistent with the translational repressor activities determined in vivo. Q65H and D91N exhibited K d values of 5 ϫ 10 Ϫ10 M and 2 ϫ 10 Ϫ9 M with the Q␤ operator, respectively, whereas the wild-type protein displayed a K d of 4 ϫ 10 Ϫ9 M ( Fig. 6 and Table V). On the other hand, the Q␤ coat protein interacted with MS2 RNA poorly, yielding a K d of 1 ϫ 10 Ϫ7 M. As predicted from their activities as translational repressors in vivo, both mutants possessed stronger affinities for the MS2 RNA operator, with dissociation constants of 3 ϫ 10 Ϫ9 M for Q65H, and 5 ϫ 10 Ϫ9 M for D91N. A schematic diagram of the Q␤ coat ␤-sheet structure depicts the positions of residues 65 and 91 within the RNA-binding site (Fig. 5A). DISCUSSION Our genetic approach to dissection of the Q␤ RNA-binding site relies on translationally repressing the synthesis of a hybrid replicase-␤-galactosidase protein. We initially constructed pRZQ for this purpose. It contains a synthetic version of the wild-type Q␤ operator. Unfortunately, pRZQ was not susceptible to efficient translational repression by Q␤ coat protein expressed from pQCT119. We did not know how to explain this behavior, but noticed the presence of two AUG triplets (at Ϫ9 and Ϫ4 in Fig. 1B) in addition to the replicase initiation codon (at ϩ1). Since translational repression is the result of competition between coat protein and ribosomes for binding of the operator RNA, we speculated that the extra AUGs might somehow tip the balance in favor of ribosome binding. This was the reasoning behind the construction of pRZQ5, which contains nucleotide substitutions that preserve the required RNA secondary structure, but destroy the two extra AUGs (see Fig. 1). These changes confer the desired capacity for translational repression. This is not the result of any difference in affinity of the two operator RNAs for coat protein, since their binding behaviors in vitro are indistinguishable (Fig. 2).
Do these observations imply that the extra AUGs are bound by ribosomes? N-terminal sequence analysis of the replicase protein suggests that the AUG at ϩ1 is the bona fide translation initiation codon (19). Although the AUG at Ϫ9 is in the replicase reading frame and could potentially lead to the synthesis of a protein three amino acids longer, we are unaware of any evidence for the existence of such an elongated product in Q␤-infected cells. Since pRZQ produces about 50% more ␤-ga- FIG. 5. A schematic view of the ␤-sheet surface of coat protein where the RNA-binding site resides. It is this surface of the protein which, in the assembled virus-particle, is oriented toward the interior of the capsid. Since the Q␤ and MS2 coat proteins are close structural homologues, we use the same basic illustration of ␤-strands connected by loops to represent them both. Shown in A are the locations of amino acids identified in this study whose substitution affects translational repression by Q␤ coat protein (Q␤ numbering). Blackened circles represent the repressor-defective mutations listed in Table I. Shaded circles correspond to the locations of the specificity mutations Q65H and D91N. For comparison B illustrates the locations of amino acids (MS2 numbering) in MS2 coat protein whose substitution resulted in the repressor-defective phenotype (3). In order to illustrate the extent of conservation of RNA-binding site residues in the Q␤ and MS2 coat proteins C shows a superposition of A and B. Positions which contain identical amino acids at structurally homologous positions are indicated as filled circles. Here the amino acids are identified by MS2 numbering. lactosidase than pRZQ5 2 it is possible that the Ϫ9 AUG is utilized for translation initiation in this system. On the other hand, the AUG at Ϫ4 cannot produce an elongated protein, since it resides in the Ϫ1 reading frame and is followed, after two codons, by a nonsense triplet. Since pRZQ5 simultaneously inactivates both of the extra AUGs, we cannot distinguish their relative contributions to inhibition of repression. Moreover, we do not know whether the behavior of pRZQ means that the operator is poorly repressed in intact Q␤ RNA in infected cells. Placement of the operator in the unusual context of pRZQ may have altered its function, perhaps somehow making the extra AUGs more susceptible to ribosome binding.
We previously reported the identification of amino acid residues making up the RNA binding site of MS2 coat protein.
Their side chains reside on one surface of a large ␤-sheet. The location and make-up of the RNA-binding site was confirmed by the x-ray structure of the MS2 RNA-protein complex (7). In the current study we applied a similar genetic strategy to identify the amino acid constituents of the Q␤ RNA-binding site. Given that MS2 and Q␤ coat proteins are evolutionary relatives and possess homologous amino acid sequences, we assumed that Q␤ coat protein would have high structural similarity to MS2, and would probably utilize a similar ␤-sheet surface for RNA binding. These expectations were confirmed by the mutational analyses reported here and by the x-ray structure of Q␤ coat protein which Lars Liljas and colleagues kindly made available to us while this manuscript was in preparation. Table I shows the nucleotide and amino acid substitutions that resulted in the repressor-defective phenotype. Table II and Fig.  4 show that these substitutions also caused defects in RNA binding in vitro. As with MS2, the RNA-binding site mainly resides on the surface of the coat protein ␤-sheet, although some loop residues are also implicated. Most of the essential Q␤ amino acids occupy positions that are structurally equivalent to important RNA-binding site residues of MS2 coat protein.
The degree of conservation of these residues is shown in Table  I which lists, in parentheses, the equivalent MS2 residues in those cases where a clear structural homologue is readily identified. Nearly all these amino acids are conserved between the two proteins. Fig. 5B shows the locations of the amino acid residues identified by genetic analysis as important for RNA binding by MS2 coat protein. A similar map of the Q␤-binding site, based on the studies reported here, is shown in Fig. 5A. Fig. 5C is a superposition of Fig. 5, A and B, and shows the degree of amino acid identity in residues required for RNA binding by the two proteins. The conserved aspects of the two sites may represent portions of coat protein structure required for binding of structurally similar parts of the operators.
It is possible, for several reasons, that the Q␤ RNA-binding site includes residues not identified by this analysis. For example, some types of mutations could be absent in our mutational library. However, sequence analysis of 32 mutations in 21 repressor-defective assembly-defective mutants revealed a wide spectrum of different substitution types (not shown). Therefore, it seems unlikely that any amino acid residue whose identity is crucial for RNA binding is missing from this analysis, but, of course, we cannot rule out the possible existence of mutational cold spots which would lead to under-representa-    tion of certain mutants. It is also possible that certain residues play a dual role, functioning both in RNA binding and in protein folding or stability. Mutants with substitutions at these sites will not have passed our screen for capsid formation. Although we obtained multiple isolates of all but one of the repressor-defective mutants, it is also possible that the mutant library contains repressor defects not yet isolated. For these reasons the RNA-binding site could be more extensive than is indicated by the present set of repressor-defective mutations. Unconserved amino acids within the two binding sites presumably account for their differing RNA binding specificities. We isolated four specificity mutants of Q␤ coat protein based on their abilities to repress the MS2 operator. Three of them contained multiple amino acid substitutions. This was a consequence of the high mutation rate of the error-prone PCR method we used in generating the mutant library. We restricted our analyses to residues 91 and 65. The D91N,A114G and T18S,N22Y,D91N mutants had only the D91N substitution in common, and, of the affected amino acids, Asp 91 was the only one present on the surface of the ␤-sheet where the RNAbinding site resides. The single mutant D91N was constructed to test this assertion, and we found that the translational repressor activity of both of the original mutants was conferred by the D91N substitution alone (Table IV). Similarly, Q65H possessed the translational repressor properties of the T29S,Q65H double mutant. For these reasons only the D91N and Q65H single substitutions were characterized further. Both mutants possess slightly increased affinities for the normal binding target of Q␤ coat protein, the Q␤ translational operator. This is evident both in the translational repression data presented in Table IV and in the in vitro binding affinities of the proteins for Q␤ operator RNA ( Fig. 6 and Table V). Thus, D91N binds the Q␤ operator 2-fold more tightly than does wild-type Q␤ coat protein. Meanwhile, Q65H shows an 8-fold improvement in binding of this RNA. The most dramatic effects, however, are found in the increased affinities for the MS2 operator. D91N and Q65H, respectively, bind the MS2 operator RNA 20-fold and 33-fold more tightly than does wild-type Q␤ coat protein.
The effects of the D91N substitution are easily rationalized. Asp 91 of Q␤ coat protein occupies a position which is structurally homologous to Asn 87 of MS2 coat protein. Genetic and structural studies identified Asn 87 of MS2 as an important site of interaction with RNA (6, 7). It forms a hydrogen bond with the Ϫ5 uridine residue in the MS2 operator. On the other hand, Asp 91 of Q␤ is apparently not required for binding of the Q␤ operator, since no substitutions at this site were found in our extensive collection of repressor-defective Q␤ coat mutants, and because the D91N substitution itself clearly does not impair binding to Q␤ RNA. It is striking that converting Asp 91 to its MS2 counterpart dramatically improves activity for the MS2 operator. In fact, Q␤-D91N binds MS2 RNA nearly as well as does MS2 coat protein itself. Clearly, the RNA binding surface of Q␤ coat protein is sufficiently similar to that of MS2 that it readily accommodates MS2 RNA, and a single amino acid substitution converts Q␤ coat protein to a good repressor of MS2.
The effects of the Q65H substitution are not as easily understood. Gln 65 in Q␤ coat protein occupies a position which is structurally homologous to Thr 59 of MS2. Substitutions like T59S or T59A result in repressor defects in MS2 coat protein (3,6). However, the MS2 mutant T59Q is not repressor-defective for the MS2 operator. 2 Although it is clear from the crystal structure of the MS2 coat protein-RNA complex that Thr 59 makes contact with RNA, the contact apparently does not in-volve the side chain (7). The effects of the Q65H substitution on RNA binding by Q␤ coat protein suggest that histidine in this position establishes new contacts with the two RNAs, substantially increasing affinity for both the Q␤ and MS2 operators.
We have shown that Q␤ coat protein easily acquires binding activity for MS2 RNA by mutation. In work being reported elsewhere we show that MS2 coat protein acquires the ability to bind the Q␤ operator with similar ease. Thus, the MS2 and Q␤ coat proteins are sufficiently similar that their RNA-binding specificities are readily interconverted. Perhaps this is not unexpected given the similarities in the amino acid sequences of their RNA-binding sites. However, some may find it surprising that these similar proteins bind operators with the large apparent structural differences illustrated in Fig. 1. Compared to MS2, the Q␤ operator requires a longer base paired stem, a smaller loop, and is relatively indifferent to deletion of the bulged A. In MS2 the critical nature of the bulge is easily reconciled with the structure of the protein-RNA complex where quasi-symmetric interactions between the two halves of the dimer and the As at Ϫ4 and Ϫ10 are observed. The amino acid residues (Val 29 , Thr 45 , Ser 47 , and Lys 61 ) of MS2 coat protein that form the binding sites for these two As are conserved in Q␤, thus satisfying one of the important requirements for interaction with MS2 RNA. However, given the relative dispensability of the Q␤ bulged A and the apparent difference in its spatial relationship to the loop A, the quasi-symmetry of these interactions is likely abolished in the Q␤ coat protein-RNA complex. We suspect that the interaction of coat protein with the loop A is conserved between the two phages, but the interaction with the bulged A probably is not. Other contacts must be formed to compensate for this loss. The differences in makeup of the RNA-binding site of the two coat proteins presumably reflect this fact. On the other hand, the similarities of the binding sites might not reflect any significant similarities in the precise modes of interaction of the two RNAs with their respective coat proteins. Experiments currently in progress should determine whether the nature of the interaction with the loop A is a conserved feature in the RNA-protein complexes of MS2 and Q␤.