Limitations of Translational Accuracy
Chapter
65
C. G. KURLAND, DIARMAID HUGHES, AND MÅNS EHRENBERG
Ribosomes most often make polypeptides that are faithful translations of the codon sequences in mRNAs. Occasionally, translation fails to produce a canonical product. Instead, a protein may contain an amino acid substitution at some position, or part of the polypeptide may have been produced after a shift in the reading frame, or the polypeptide may be abortively terminated. The first of these failures, a missense error, is what is usually called to mind as an error of translation. Nevertheless, the other sorts of errors are more costly to the economy of cells, and avoiding them probably represents a deeper mechanistic challenge to the evolution of ribosomes than does avoiding missense errors.
It is easy to understand why frameshift or abortive termination events are more costly to cells than are missense errors. As we will see below, a protein containing an amino acid substitution may be somewhat less active than the canonical version of the same protein, but a frameshift product or an abortively terminated protein is in most instances completely inactive. Likewise, when we review the mechanisms of translation, it becomes evident that the measured translocation of RNA molecules from one ribosomal binding site to another is largely what translation is all about. This suggests, in turn, that maintaining acceptable levels of the movement accidents that lead to abortive termination or loss of reading frame must represent a major challenge to the translation mechanism.
The elongation of a polypeptide is a stepwise process in which all components of the system except the amino acids are reused in successive cycles of peptide bond formation (Fig. 1). It is convenient to think of the elongation cycle as consisting of two functional compartments, each associated with a different elongation factor subcycle; one involves the selection of an aminoacyl-tRNA (aa-tRNA) to match a codon, and the other involves the translocation of the polypeptide and mRNA that returns the ribosome to the beginning point of the next cycle.
In the first compartment, an aa-tRNA is bound in a complex with elongation factor Tu (EF-Tu) and GTP, the so-called ternary complex. The ternary complex is tested at the decoding site of the ribosome for a proper match with the codon that is being translated at this ribosomal site. The ternary complex with aa-tRNA may dissociate from the ribosome during this testing period, or it may be accepted. In the latter case, GTP is hydrolyzed, EF-Tu and GDP dissociate from the ribosome, and the aa-tRNA is bound at the ribosomal decoding site. The EF-Tu subcycle is completed by the regeneration of the ternary complex from EF-Tu, a recharged aa-tRNA, and GTP that has itself been regenerated from GDP and Pi. The elongation factor EF-Ts is instrumental in the regeneration of the ternary complex from the EF-Tu·GDP complex.
The elongation of the polypeptide chain by one amino acid is effected by the transfer of the nascent polypeptide from the tRNA that has translated the previous codon to the aa-tRNA that is matched with the current codon being translated. This transfer provides the transition to the second elongation factor subcycle.
In the second compartment, the advance of the ribosome along the mRNA is mediated by elongation factor G (EF-G) in complex with GTP. Here, the complex of the peptidyl-tRNA fixed with the mRNA at the codon that it translated is translocated to a new position on the working surface of the ribosome. This leaves the next codon in line at the decoding site ready to initiate a new cycle of elongation. The peptidyl-tRNA is now positioned at a site that can carry out the transfer reaction corresponding to peptide bond formation after the next aa-tRNA molecule has been selected. The EF-G and GDP that are released during the translocation subcycle are regenerated as an EF-G·GTP complex in preparation for future translocation cycles. Until recently, these steps of the elongation cycle had been visualized in terms of a very specific ribosome model, the two-site model.
Although it has been superseded recently by a more detailed model, the Watson two-site model (276) goes a long way toward displaying the problems that must be solved by a translating ribosome. One involves the proper matching of tRNA to codon, and the other involves the coordinate movements of tRNA and mRNA through separate ribosomal A- and P-sites associated in this model with the acceptor (A) and donor (peptidyl [P]) functions of peptide bond formation.
Two important refinements have been added to the Watson model (276) in recent years. One is the identification of a third functional site for tRNA binding, the so-called exit site (E-site). Hints that tRNA can bind to more than two distinguishable ribosomal sites have been in the literature for decades (246, 287). However, it took the persistent work of Nierhaus and collaborators (185, 212, 213) to lend credence to the idea that a third ribosomal site (E-site) site is the residence of the deacylated tRNA after it has left the P-site and before it exits from the ribosome.
The second refinement is more complex because it concerns the multiplicity of ways in which the 30S and 50S ribosomal subunits bind tRNA during the intermediate states of the elongation cycle. The two ribosomal binding sites of Watson’s model (276) were defined originally by the two states of susceptibility of peptidyl-tRNA to the action of puromycin. In contrast, attempts to visualize an elongation cycle that makes use of the subunit structure of ribosomes tend to generate a higher order of binding states for tRNA during the elongation cycle. In particular, it was suggested that an interplay of partial contributions from A- and P-sites would generate hybrid binding states that could mediate the movements of peptidyl-tRNA (35, 239). Indeed, it was suggested that at least three states of tRNA binding including one hybrid intermediate are a minimum mechanistic requirement for the translocation of the peptidyl-tRNA–mRNA complex (141).
The complexities of these speculations were exceeded by the complexity of the results from experiments in which the binding sites of tRNA on ribosomes were identified by a chemical masking protocol applied in vitro to ribosomes in different stages of translation (175, 176, 190). No fewer than six different tRNA-binding states were identified in the four distinguishable states of the tRNA-mRNA-ribosome complex studied in these experiments. In the scheme of Noller et al. (190), a cycle of peptide bond formation requires an ordered sequence of movements of the tRNA and/or the ribosomal subunits so that a tRNA progresses from the initial 30S A-site through the A/A-site, to the hybrid A/P-site, to the P/P-site, to the hybrid P/E-site, and finally to the 50S E-site before exiting the system (Fig. 1). This complex flow pattern illustrates the mechanical complexities of a peptide bond cycle, and it provides the basis for our expectation that measured movement is an important aspect of the translational accuracy problem.
The classification of aberrant translation products is simple: there are essentially two sorts. The first contains one or more unorthodox substitutions within an otherwise complete amino acid sequence. The second involves a polypeptide produced by the failure to completely translate the mRNA sequence in the correct reading frame; typically this sort of product has a nonstandard sequence length. It may be missing a longer or a shorter part of the canonical sequence, or it may have new sequences appended to the canonical sequence. The first is a missense error, and the second is a processivity error. The classification of the events leading to these two sorts of variants is much more complex. Indeed, there are at least two missense events that can create processivity errrors.
The reason for referring to abnormally long or short products as processivity errors may require some explanation. At all stages of translation, the nascent polypeptide is thought to be anchored to the same ribosome via the successive tRNA species that elongate it one amino acid at a time (45, 86, 276); that is to say, translation is processive. Since the code is punctuated, misreading the mRNA sequences in particular ways can compromise the processivity of translation. For example, if a sense codon is misread as a stop codon, the nascent polypeptide will be terminated in an incomplete form. This sort of event is called a false stop (122). Alternatively, the stop codon at the end of the translation sequence may be misread as a sense codon, or a prior frameshift may bring it out of play. In either event, a somewhat longer product may be produced; this sort of event is called termination readthrough. However, this is just the beginning of the list of processivity errors.
Translational frameshift events come in two varieties: slipping and hopping (16, 283, 284). Slipping occurs when the translational machinery advances along the mRNA in an unorthodox way that leads to a loss of reading frame; here, the mRNA is read as though one nucleotide were added or subtracted from the sequence after the shift (14). The shifted ribosome will with high probability encounter an out-of-phase stop codon within a short run of codons, and the translation of the shifted polypeptide will terminate in this case (157). Hopping occurs when the translation apparatus appears to leave the mRNA sequence at some codon, omits to translate a stretch of the mRNA after this codon, but resumes the elongation of the nascent polypeptide either in phase or out of phase at a similar codon further downstream in the mRNA sequence. The gap created by the hopping ribosome may be anywhere from 1 to 20 amino acids long (283, 284).
Processivity may also be interrupted by an event that has been called drop-off (164, 165, 166). Here, the nascent polypeptidyl-tRNA dissociates from the mRNA-programmed ribosome, and the tRNA is removed by an enzyme with that sole function, peptidyl-tRNA hydrolase. If peptidyl-tRNA hydrolase is inactivated, the bacterium will eventually die, which indicates that the peptidyl-tRNA is toxic (166).
Finally, any obstruction of the processivity of mRNA transcription will be reflected in a proportionally truncated polypeptide product. Little is known about transcriptional frameshift events, but abortive transcriptional termination is known to be influenced by ribosome phenotypes (44, 123, 128) as well as by the idiosyncracies of individual mRNA sequences (4, 77).
Quantitation of the frequencies with which these diverse accidents occur is complicated by the fact that bacteria possess a battery of proteins that can rehabilitate some aberrant proteins (71, 72, 92, 119, 272) and destroy others (43, 71, 72, 92, 99, 119, 233, 244, 272). Thus, the competition between proteolytic enzymes and chaperonin proteins determines the stability of proteins with missense substitutions as well as the rates at which truncated proteins are destroyed. This means that the steady-state levels of aberrant proteins are influenced by the intensity of such posttranslational events. Since the activities of both chaperonin proteins and proteolytic enzymes are sensitive to physiological conditions, their influence on the steady-state levels of aberrant proteins may vary according to the physiological state of the cell (90, 91, 203, 263, 270). Accordingly, there will be a physiologically regulated discrepancy between the frequencies of accidents on ribosomes and the frequencies of disabled proteins found in the cell. In particular, measurements of steady-state levels of errors will always underestimate the frequencies of accidents that occur on ribosomes.
At this point, it would be helpful to present a table of average error frequencies to support a discussion of the impact of the different sorts of aberrations on bacteria. Unfortunately, the space of error events has not been sampled with anything like the thoroughness needed to generate reliable representative frequencies. What we have instead is a very spotty sampling of the error space. As a consequence, a few relevant numbers must be discussed rather than simply presented.
The earliest attempts to estimate missense error frequencies yielded frequencies close to 3 × 10–4 per codon (65, 70, 152, 153). A later series of direct chemical measurements at a few codons surprisingly yielded error rates closer to or higher than 10–3 (32, 199, 200). The most recent critical summaries of missense data report frequencies between 5 × 10–5 and 5 × 10–3 per codon in laboratory strains of Escherichia coli growing under normal conditions (197, 198). Following Parker (198), we take as a global average 5 × 10–4 missense events per codon. If we assume that a representative protein is assembled from 500 amino acids, such an error rate would leave 78% of the protein copies missense error free. On the other hand, for an object such as a ribosome with a protein complement that corresponds to roughly 104 amino acids, virtually no two ribosomes in a bacterium would be identical, and the average ribosome would have roughly five missense substitutions.
There are to our knowledge two sorts of global estimates of processivity errors. The first is that of Manley (157), who used antibodies to identify incomplete β-galactosidase sequences that were fractionated electrophoretically in gels. The quantities of these incomplete polypeptides suggest that 31% of all ribosome starts on a lacZ mRNA normally are aborted before completion of the β-galactosidase subunit. The second sort of measurement consists of measuring the difference in the recovery of monomer and dimer copies of β-galactosidase on gels; this difference is equivalent to the loss of processivity that occurs in translating a monomer (123). This sort of measurement suggests that close to 25% of the ribosome starts are aborted before the monomers are completed. Given the experimental uncertainties and the possibilities of strain differences, these two measurements are in excellent agreement. They suggest that the global processivity error rate is close to 2.5 × 10–4, or 1 in 4,000 codons.
Nevertheless, there are at least two caveats. As we will document below, ribosome mutants tend to generate higher processivity losses (62, 123, 264), which indicates that there may be significant differences between the processivity errors of different bacterial strains. Furthermore, the Manley (157) experiments revealed that the processivity losses associated with the translation of β-galactosidase are dominated by only nine different aborted polypeptide chains. One interpretation of this observation is that there are only a limited number of frameshift windows in any given gene, rather than that there are a very few hot spots for frameshifting or drop-off. Accordingly, it seems likely that the frequency of processivity losses is sequence dependent and that it may vary from one gene product to another. Currently, we have data only for lacZ.
As recounted above, the processivity losses can result from a number of different events: however, it seems that for wild-type cells, one of these events dominates. The contribution of abortive termination of transcription of lacZ, also referred to as polarity, has been estimated at one-third of the total processivity loss in laboratory wild-type bacteria (62, 123). False stops within lacZ mRNA are estimated at less than 2 × 10–5 per sense codon (122). Spontaneous frameshift events for wild-type ribosomes have been estimated at less than 3 × 10–5 per codon for the translation of lacZ mRNA (137). Finally, drop-off has been estimated biochemically for the entire polypeptide population in wild-type E. coli to occur with a frequency close to 4 × 10–4 per codon (164). Again, with due allowance for experimental uncertainties and strain differences, this figure suggests that drop-off is the dominant contribution to the processivity losses of gene expression in wild-type E. coli.
We have inherited from earlier generations of scientists the expectation that amino acid substitutions, like mutations, are so dangerous that cells will go to any extreme to be rid of them. An illustration of this prejudice is the ribosome editing model, which mobilizes the drop-off event to edit missense events on the translating ribosome (46, 47, 165, 167). In this model, the erasure of missense substitutions is realized by drop-off events which take place preferentially at codons that are translated by a mismatched tRNA. It is clear that with sufficiently high rates of drop-off events preferentially attending missense events on the ribosome, the missense error rates could be reduced significantly. Nevertheless, it is useful to ask in what way a trade-off between missense and drop-off events would benefit the cell.
Clearly, if an average missense event leads to the production of a toxic protein, and if the cost of removing the toxic product by drop-off is negligible, the trade-off that is postulated in the ribosome editing model would be advantageous. On the other hand, if the negative influence of the average misssense event on protein function is small and the cost of a drop-off event is relatively high, then the mechanistic coupling of the two events would amplify the negative consequences of missense errors rather than reduce them. Consequently, we need a way of comparing the relative costs of different sorts of translation aberrations.
Assessing the impact of processivity errors is rather straightforward. We expect frameshifted or abortively terminated gene products to be most often completely inactive (see, for example, reference 53). Hence, the translation of a polypeptide that is not correctly processed will have a cost that is associated with the effort wasted in synthesizing the terminated product. In this analysis, it is assumed that there is an average cost per amino acid incorporated into peptide that is made up of terms representing the time wasted in the synthesis of the dud protein by ribosomes, protein factors, tRNA molecules, and so on. Hence, an index of cost is simply the average chain length of the terminated product, which for sufficiently low probabilities of abortive termination corresponds to one-half the chain length of the canonical protein. This means that if a representative protein with one missense substitution is less than 50% as active as the canonical version, it pays to abortively terminate missense events by drop-off. On the other hand, if missense events lead on average to less than 50% debility, it does not pay to edit them by drop-off (141).
Assessing the impact of events in which a mismatch has occurred between the codon and the aa-tRNA that has been matched with it on the ribosome is somewhat more complex. First, we need to recall that not all mismatched codon-anticodon interactions on the ribosome lead to a dysfunctional amino acid substitution. As a consequence, if there is a high probability of drop-off at such mismatch events, this will destroy a significant fraction of proteins that otherwise would be fully functional. For example, because of the constraints of the genetic code, a random codon change by one nucleotide will lead to the choice of a synonymous codon with a probability close to 0.29 (182). The remaining mismatch events that lead to amino acid changes tend to be conservative because of the way the genetic code is arranged (237, 289). For instance, the probability of generating a charge change in the amino acid side chain by a random mismatch event is less than 0.3 (158, 183). This means that the chance of generating a disruptive amino acid substitution in a protein by a random codon-anticodon mismatch a priori is significantly lower than 0.5.
In fact, in a statistical comparison, the relative likelihood of generating by random mutations in a gene (lacZ) an inactive gene product (β-galactosidase) as the result of a mutation that leads to abortive termination compared to that leading to a missense substitution was found to be approximately 1 in 300 (145). Even in one of the most highly conserved genes, tufB, this chance increases only to 1 in 10 (3). In other words, for most genes it will be exceedingly difficult to identify single missense substitutions that completely inactivate an enzyme. Indeed, proteins with single amino acid substitutions are most often quite active compared with the canonical homologs (33, 73, 106, 107, 159, 174, 196, 206, 209, 228, 235). Recent summaries of results obtained with mutant proteins show that of the thousands of random amino acid substitutions studied, much less than half inhibit the function of the affected proteins by more than 50% (138, 160). These data suggest that even if all codon-anticodon mismatch events were to lead to an amino acid substitution, the expected effect on protein function, for all except the most highly conserved proteins, would be relatively small.
In summary, the arrangements of the genetic code taken together with the robust character of protein structure suggest that the functional consequences of a codon-anticodon mismatch during translation will lead most often to much less than a 50% loss of protein activity. Accordingly, ribosome editing is not a highly recommended way to control missense substitutions because in such a scheme the cure is worse than the disease (141).
We have introduced translation errors in the context of a generalized cycle of polypeptide elongation, and we have viewed the consequences of errors in this cycle from the perspective of their average frequencies as well as of their impact on individual proteins. Although this perspective is a useful one, it provides too narrow a base to allow us to identify the selective factors that determine the error levels in bacteria. For example, in our review of the ribosome editing model, the reference point for assessing the impact of translation errors was the loss of 50% of normal activity for an aberrant protein. We have stressed the fact that relative to this reference point, the structures and functions of proteins are robust to missense substitutions. Accordingly, one might conclude that to a good first approximation, missense errors do not create significant functional problems for proteins. In contrast, from the perspective of the population biology of bacteria, a 50% loss of a growth-limiting activity is tantamount to a catastrophe. So our first injunction is that over the time span of evolution, or even for the intervals that influence the microevolution of bacterial populations, small differences in growth rates are bound to influence the character of competing systems. This means that the impact of a missense substitution can be considerably less than 50% and still be scored as biologically significant.
Second, larger proteins or protein aggregates are potentially able to accumulate a large number of missense substitutions at error frequencies at which an average-size protein is virtually error free (Fig. 2). For example, the β and β' subunits of RNA polymerase have chain lengths that are three times longer than that of the 500-amino-acid-long representative protein that we discussed above. Likewise, DNA replication, transcription, and translation are dependent on large protein assemblies that consist of many tens of thousands of amino acids. This means that at codon-normalized missense frequencies that would have a barely perceptible effect on the functions of average-size proteins, the accumulation of multiple substitutions could produce a more telling effect on these central functions. As a consequence, we expect small, incremental changes in the accuracy phenotypes to be relevant to the evolution of ribosomes as well as to the stability of bacterial populations.
Bacterial mutants of wild-type laboratory strains have provided a unique and rich source of information about the different sorts of incremental changes in accuracy phenotypes that can be tolerated by bacteria under laboratory growth conditions. Unfortunately, the spectrum of mutant phenotypes is biased and limited largely to those that alter the response of the translation system to a few antibiotics. This situation probably will change in response to the practical problems of dealing with a growing frequency of natural antibiotic-resistant bacterial populations as well as to the worsening epidemic profile in countries where health service has broken down or where there are large numbers of AIDs patients. In the meantime, the particularities of the available mutants have led to very general ideas concerning connections between missense and processivity errors as well as the relationship of translational fidelity to bacterial growth rates. We will make use of these patterns later, but first we need to review the details of the available phenotypes in order to document their general properties.
This chapter is concerned primarily with the accuracy of ribosomal reactions and the consequences of errors in ribosome function. We have little to say about the errors of aa-tRNA synthesis because they are so infrequent compared to those of the ribosome reactions that they scarcely contribute to the errors of translation (reviewed by McClain [161]). In contrast, missense, nonsense, and frameshift errors are promoted at high frequency by tRNA mutants with alterations in their anticodons or at bases outside the anticodon. These are discussed in chapter 60. Likewise, tRNA structures most often contain nucleotides that are highly modified, and mutations in several of the modification enzyme genes are associated with alterations in translational accuracy. These are reviewed by Björk in chapter 57.
The story of ribosomal accuracy mutants is closely linked to the story of bacterial mutants resistant to the aminoglycoside antibiotic streptomycin. Streptomycin was shown to inhibit protein synthesis (74), and both streptomycin-dependent (Smd) (172) and streptomycin-resistant (Smr) (184) mutants of E. coli had been identified before the existence of the ribosome had been established. Spotts and Stanier (243) were the first to suggest that the ribosome was the target for streptomycin, and this suggestion was confirmed by in vitro translation studies (75, 156, 238). Mixing experiments in vitro showed that it was the 30S ribosomal subunit which responded to streptomycin (52, 56) and that it was the seat of the Smd phenotype (148). Finally, reconstitution experiments showed that protein S12 on the 30S subunit, coded by the gene rpsL, was responsible for the Smd (23) and Smr (195) phenotypes.
The primary binding site of streptomycin was identified as 16S rRNA (25). The precise location of this site has been the subject of many studies, often with conflicting results. The most recent data obtained by using concentric photoaffinity labeling locate streptomycin in the interface of the two ribosomal subunits close to S5 and L11 (1). Thus, both rRNA and ribosomal proteins are involved in the responses of ribosomes to streptomycin.
It was shown early on that streptomycin could cause phenotypic suppression of some auxotrophic mutants (18, 84, 95, 97, 146, 194). This finding suggested that the antibiotic induced misreading in the translation of the genetic code. It was found subsequently that many Smr (rpsL) mutants restrict the natural low-level readthrough of some nonsense codons (91). In some cases, the error-restrictive effect of Smr mutations on nonsense readthrough could be reversed by the addition of streptomycin. The error-restrictive effects of Smr mutants appeared to be fairly general, reducing both missense and nonsense reading by mutant tRNA suppressors in bacteria and phage (25, 51, 194, 245). In addition, the low-level spontaneous suppression of many frameshift mutations in lacZ was also restricted by rpsL mutations (14).
The misreading effects of streptomycin and the restrictive effects of rpsL mutants could also be expressed in vitro with translation systems that mimicked to some degree the phenotypic effects expressed in vivo. For example, a poly(U)-directed translation system with wild-type components occasionally makes the error of incorporating leucine instead of phenylalanine (36). Streptomycin increases the incorporation of leucine, isoleucine, serine, and tyrosine (57, 58, 124, 201), but ribosomes from a restrictive Smr (rpsL) strain misread poly(U) much less often than do ribosomes from a wild-type strain (8). Not all Smr mutations in rpsL are error restrictive; the phenotypes range from nonrestrictive (or weakly error prone) through semirestrictive to restrictive (29, 34).
Amino acid sequencing of some of the mutant S12 proteins (81, 117, 269) identified a number of substitutions at positions 42 and 87 that are associated with resistance to streptomycin and one at position 42 associated with streptomycin dependence. More recently, DNA sequencing of newly isolated mutants of Salmonella typhimurium (official designation, Salmonella enterica serovar Typhimurium) (265) and E. coli (260, 261; D. Hughes, unpublished data) has revealed that at least 19 different substitutions or three-base deletions at 11 codon sites in S12 can promote an Smr, pseudo-Smd (Smp), or Smd phenotype. The mutations are clustered in two areas, amino acid residues 41 to 45 and residues 87, 89, 90, 91, and 93, with one exceptional mutation at position 53. Smr mutations have been isolated only at residues 42, 53, and 87, whereas Smd mutations have been isolated at all of these positions except 53. Only a single mutation, Lys-42→Arg, is known to have a nonrestrictive Smr phenotype. Interestingly, a significant number (close to 20%) of the Smd isolates carry double mutations in rpsL (261). It is unclear how many of the observed phenotypes require double mutations and how many of the secondary mutations are silent, their occurrence possibly reflecting the mechanism behind the mutational process (260). These "ancillary mutations" all result in amino acid substitutions (260), and they occur at 10 different sites in rpsL, only 4 of which have been found to give rise to "primary" mutations. In several cases, identical single-site mutations in rpsL, isolated spontaneously, were seen to confer different phenotypes, either Smr or Smd (261). This finding suggests that extragenic ancillary mutations arise frequently, resulting in these cases in the probable conversion of an Smd to an Smr phenotype.
Two mutations deserve particular mention. First, there is the Smp mutation rpsL1204 (293), which is intermediate in phenotype between a restrictive Smr and an Smd mutation. It has proved useful experimentally because some of its phenotypic effects can be varied in response to changes in the level of streptomycin present. The Smp mutant carries two mutations in rpsL, at positions 85 and 90 (Hughes, unpublished data). One of the characterized Smr mutations, rpsL224 (Lys-87→Arg), is also responsive to streptomycin. The addition of streptomycin to the growth media converts this mutant phenotype from moderately restrictive to strongly error prone (29). A second interesting mutation is the error-restrictive mutation rpsL500 (265), which is the only Smr mutation not to be isolated in a selection for resistance to streptomycin. This exceptional mutation was isolated as a kirromycin-resistant mutation in a strain carrying one wild-type EF-Tu (tuf) gene and one kirromycin-resistant tuf gene. Kirromycin resistance normally results from mutations in both of the two tuf genes (111). The exceptional mutation is the most error-restrictive rpsL mutation known; it confers a relatively low level of resistance to streptomycin, and it maps at an unusual position (residue 53) in the protein (265, 267).
A deeper understanding of the phenotypic significance of each of these mutant positions will have to await detailed structural information on S12 and its neighborhood in the ribosome. The influence of the antibiotic streptomycin, and of the various mutations in rpsL, on the kinetics and accuracy of translation is discussed below. For the moment, we note that the existence of rpsL mutations leading to an error-restrictive (i.e, a hyperaccurate) phenotype provides an important clue to how the selective pressures in the evolution of ribosomes influence the accuracy of translation. It is obvious that translational accuracy is not maximized in wild-type bacteria. Why aren’t wild-type ribosomes more accurate, or at least as accurate as those of restrictive Smr mutants? Is there something negative about too much accuracy? We return to this issue later.
Neamine-resistant mutations arise in the S17 protein coded by rpsQ, and these mutations frequently are associated with altered electrophoretic mobility for the protein S17 (30). One of these mutations, His-30→Pro, restricts tRNA suppressor misreading of nonsense codons (291). The antibiotic neamine causes misreading during translationin vitro. Neamine-resistant S17 mutant ribosomes are less sensitive than wild-type ribosomes to neamine-induced misreading, and they are also restrictive in the absence of the antibiotic (30, 262).
Gentamicin is another antibiotic which induces misreading in poly(U)-directed translationin vitro. Gentamicin-resistant mutants with alterations in the large ribosomal subunit protein L6 encoded by rplF have been isolated (40) and shown to have a weak error-restrictive effect in vitro as well as in vivo (135). Compared with our extensive knowledge of rpsL mutant phenotypes, very little is known about the neamine- and gentamicin-resistant mutants.
Smd mutants, with their conditional-lethal phenotype, allow the relatively easy selection of new nonallelic mutations which relieve the dependent phenotype (38, 104). Mutations relieving the Smd phenotype have been identified in the small subunit ribosomal proteins S4 (24, 59, 103, 134) and S5 (103) and in 16S rRNA (5). A striking feature of these secondary mutations is that by themselves they cause translational errors and are thus ribosomal ambiguity, or Ram, mutations. Ram mutations are generalized misreaders of nonsense, missense, and frameshift mutations, and they also reverse the restriction on misreading imposed by rpsL mutations (9, 10, 14, 204, 219). Indeed, the very first Ram mutations isolated were selected as suppressors which relieved arginine auxotrophy caused by the restriction of nonsense (UAG) readthrough by the strA1 mutation (219). These mutations were altered in protein S4 (294).
Several different alterations associated with the Ram phenotype have been identified in S4 and S5 proteins. In protein S4, Ram mutations include short additions or deletions at the C terminus (63, 79) and amino acid substitutions at position 49 or 53 (269). Duplications or triplications of 31 or 41 nucleotides at the 3' end of the gene also cause a Ram phenotype (230). In protein S5, single amino acid substitutions at position 103 or 111 cause a Ram phenotype (117).
Mutants of S5 have also been isolated by other means. Selecting for resistance to the antibiotic spectinomycin provided S5 mutants with single amino acid substitutions at position 19, 20, or 21 (31, 60, 79, 80). Revertants of a temperature-sensitive alanyl-tRNA synthetase mutant (41, 42) carried mutant protein S5 (288). However, neither of these two classes of S5 mutants has a Ram phenotype (9, 204). A protein crystal structure of S5 has been determined, and the placements of the various mutations in it show that those associated with a Ram phenotype map on one face of the protein whereas those causing spectinomycin resistance map on another (208). It is proposed that these two S5 regions bind 16S rRNA close to their respective antibiotic binding sites, streptomycin with the central domain of 16S in the vicinity of C-912 and helix 27 (1, 37) and spectinomycin with helix 34 close to residue C-1192 (37). There is currently no crystallographic structure for S4. However, the available topographic information strongly suggests that proteins S4, S5, and S12 are grouped together on the small ribosomal subunit (37).
Mutations of protein L7/L12 on the large subunit that increase translational errors have been isolated (130, 131). Of four proteins with such mutations tested, one was isolated along with a tRNA mutant during a selection for suppression of a missense mutation, while the other three were isolated in the classical way as phenotypic revertants of the Smd phenotype. All four mutations increase the level of readthrough of some nonsense codons in vivo, but only one (isolated as an Smd suppressor) causes any significant enhancement of missense errors in vitro. One possible explanation for this difference is that this one mutant may have a reduced discrimination of EF-Tu ternary complexes, whereas the other three may be more specifically defective in their interaction with release factors.
The unusual severity of the growth and elongation rate depressions of the L7/L12 mutants compared with their modest effects on accuracy is apparently due to their perturbed interactions with both EF-Tu and EF-G (22). A further complicating factor in understanding their phenotypes is that L7/L12 is unique in being present in four copies on the ribosome; all other proteins and rRNAs are present in one copy. This implies that the phenotype of any L7/L12 mutant may depend on the influence of one to four proteins in the mature ribosomes. The crystal structure of L7/L12 is partially solved (147), and the positions of the mutants in the 120-amino-acid residue protein are known (130). The mutant with the Ram-like phenotype has a deletion of five amino acids (residues 38 to 42) in the flexible hinge region of the protein. Of the other three mutants, the one that was isolated in the selection for missense suppression has a complex alteration in the same region (deletion of residues 43 to 48 and duplication of residues 38 to 42, giving a loss of one amino acid), while the other two have single amino acid substitutions at positions 74 and 82.
In summary, the similarities between the Ram phenotypes and the effects of streptomycin on translational accuracy may be an important clue to the origin of the growth-inhibiting defect of Smd ribosomes. Thus, it has been inferred that Smd strains, rather than being dependent on streptomycin specifically, are instead dependent on a decreased level of translational accuracy (26). Here too it is worth emphasizing the parallel effects of Ram mutations and streptomycin on missense and frameshift errors. The relevance of these different correlations to the origins of the Smd phenotype is discussed later.
Antibiotic-resistant variants of bacterial rRNA have been selected as spontaneous mutants as well as constructed by site-directed mutagenesis in vitro. Here too, resistance to streptomycin has been a favorite theme. All of the Smr mutations identified so far in rRNA alter the small subunit 16S molecule in one of three different areas: position 13 (102), the 530 loop and stem (78, 85, 102, 162, 207), and positions 912 to 915 (78, 102, 180). All of these mutations reduce the binding of streptomycin to the ribosome, thereby reducing the error-enhancing effects of the antibiotic on translation, and some of them have also been shown to be error restrictive in the absence of the antibiotic (205).
An error-restrictive mutation has been selected in the small subunit 15S rRNA of yeast mitochondria. This mutation, parR-454, is close to the 3' end of the rRNA and restricts the naturally occurring high level of spontaneous frameshifting by these ribosomes (286).
Two constructed mutations in the large subunit 23S rRNA reduce misreading in vitro and nonsense readthrough and frameshifting in vivo; these are substitutions of G-2661 by C and U (G2661C and G2661U) in the universally conserved α-sarcin loop, which interacts with the elongation factors (163). The G2661C mutation causes loss of viability in combination with a hyperaccurate rpsL mutation (252). Viability can be restored to the double mutant by the introduction of a tufA mutation (250) or by the addition of streptomycin to the medium (21). The loss of viability in the double mutant and the hyperaccuracy of the single mutant are associated with a reduced efficiency of ternary complex association to the ribosome (21).
All three mutations have been constructed at position G2583 in 23S rRNA in what is believed to be part of the peptidyltransferase center of the ribosome. All displayed some degree of increased accuracy in translation of poly(U) in vitro, possibly because of an altered interaction between the 3' end of the tRNA and the peptidyltransferase center (227). U-2584, the immediately adjacent position, is protected in chemical footprinting experiments by the CCA terminus of P-site-bound N-acetyl-tRNAPhe and A-site-bound aa-tRNA (177).
Several mutations which increase the levels of a variety of translational errors have now been isolated or constructed in 16S rRNA (or its equivalent in yeast cells). A mutation at position 1054 in helix 34 of 16S rRNA was isolated as a suppressor of a UGA mutation (181). The initial analysis supported the view that this mutation specifically suppresses UGA, leading to the proposal of a model in which one of two UCA triplets (residues 1199 to 1204 on the opposite strand of helix 34) interacts directly with the UGA stop codon as an initial event in peptide chain termination. Several site-directed mutations in the 1200 region also confer a UGA suppressor phenotype, supporting this model (95). However, more recent studies suggest that these and other mutations in this region of helix 34 may have more general effects on accuracy such as causing nonspecific readthrough of all three nonsense codons and suppression of +1 and –1 frameshift mutations (179). The discrepancies in observed effects of mutations in helix 34 may be due to the differences in the quantitative resolution of the assays used by the different groups.
A mutation in yeast mitochondrial 15S rRNA, in the highly conserved region forming the stem of the 530 loop, corresponding to E. coli G517A causes suppression of ochre mutations (232). Apparently the same mutation in yeast nuclear 18S rRNA causes error restriction (48). In E. coli, it has been shown that the substitution of G-517 in 16S rRNA by A, U, or C, or indeed its deletion, leads to a Ram phenotype, with suppression of UAA, UAG, and UGA nonsense mutations and of –1 and +1 frame-shift mutations (193).
An attempt has been made to isolate Ram-type mutations in rRNA on the basis of the principles used to isolate the classic Ram mutations of S4 and S5, namely, by suppression of the Smd phenotype. One mutation, C1469U in 16S rRNA, was isolated. This mutation not only suppresses streptomycin dependence but also causes increased misreading in vitro (5). Current structural information suggests that position 1469 is not close to the ribosomal proteins normally associated with translational accuracy (S4, S5, and S12). This, in turn, suggests either that many of the mutations exert their effects indirectly on distant parts of the structure or that a significant number of ribosome components that are involved in accuracy remain to be identified.
Several of the sites in the peptidyltransferase region of 23S rRNA which have been implicated in rRNA-tRNA interactions by chemical footprinting experiments (175) have also been systematically altered and tested for their effects on translational accuracy; G-2252 and G-2253 are protected by the 3'-terminal CCA of P-site-bound tRNA, and five of the six possible changes at these positions decrease translational accuracy, causing increased readthrough of all three nonsense codons as well as of +1 and –1 frameshift mutations (100). These results are provocative because they are a hint that rRNA-tRNA interactions at the ribosomal P-site may influence the accuracy of decoding in the A-site.
A spontaneously occurring error-enhancing mutation has recently been isolated in the peptidyltransferase region of 23S rRNA by selection for suppression of a –1 frameshift mutation (192). The mutation, U2555A, is a general Ram mutation enhancing readthrough of all three nonsense codons as well as –1 and +1 frameshifting. This site also is protected in chemical footprinting experiments by P-site-bound N-acetyl-Phe-tRNAPhe and A-site-bound aa-tRNA (175). The aa-tRNA bound at the A-site also protects C-2254, which is immediately adjacent to the two P-site-protected residues studied by Gregory et al. (100). In addition, another 10 spontaneously occurring mutations have now been isolated in 23S rRNA. These were obtained by selecting for suppressors of the –1 frameshift mutation trpE91 (M. O’Connor, personal communication). The positions of these mutations include one in the α-sarcin loop at position C-2666, several in a region cross-linked to protein L6 and 5S rRNA, and several in a region believed to be at the interface of the ribosomal subunits. Although selected as –1 frameshift suppressors, each of these mutations causes suppression of all three stop codons and +1 frameshift mutations as well. Thus, all of these suppressor mutations should be considered generalized Ram mutations with a tendency to increase both missense and processivity error rates.
After EF-Tu, in ternary complex with aa-tRNA and GTP, has deposited the aa-tRNA at the A-site, it leaves the elongating ribosome in complex with GDP (173). The antibiotic kirromycin inhibits the release of EF-Tu·GDP from the ribosome, thus blocking further polypeptide elongation (290). Mutant forms of EF-Tu resistant to kirromycin have been isolated; among these are some with a Ram phenotype. The mutation Val-375→Thr enhances the rate of nonsense codon readthrough (112, 274), of frameshifting (112, 273), and of missense errors in vitro (114, 251). Two genes, tufA and tufB, code for EF-Tu in E. coli as well as in S. typhimurium. The degree of error enhancement depends on whether one or both genes carry this mutation (112). Recently a number of new kirromycin-resistant tuf mutations have been isolated and identified (2), some of which have an error enhancing phenotype (F. Abdulkarim and D. Hughes, unpublished data). These include mutations at amino acid residues 120, 160, and 375.
Reducing the normally large amount of EF-Tu in the cell (about 9% of total cell protein) by inactivation of either of the tuf genes causes a reduction in the level of readthrough of nonsense codons (266). This is because the release factors compete more successfully for interaction at the nonsense codon-programmed sites against a reduced concentration of near-cognate ternary complex. Similar effects of the competition between EF-Tu and release factors are seen on +1 and –1 frameshifting, indicating that a significant amount of the observed frameshifting occurs at the nonsense codon terminating the frameshift window (226).
It has been reported that in E. coli, Val-375→Thr in tufA and Gly-222→Asp in tufB act synergistically to increase the level of translational errors in vivo (273, 274). Individually, this same tufA mutation is error prone (112, 114, 251), while the tufB mutation is defective in interacting with the ribosome (247). The basis of this synergism is unclear because it was not observed for missense errors in vitro (251), and in vivo it seems to depend on the particular strain background used (249). However, a cooperative interaction between these two mutant EF-Tu species was found for their dissociation from GTP in the presence of kirromycin in vitro(6). It remains to be determined whether the observed synergism and cooperation are related to each other or to observations indicating that a functional ternary complex may contain two copies of EF-Tu-GTP associated with aa-tRNA (68, 278).
A kirromycin-resistant EF-Tu mutant, carrying tufAa, with an antisuppression phenotype that is expressed against suppressor tRNA species (250) is severely defective in binding aa-tRNA (248). Abdulkarim et al. (2) have identified the tufA mutation as a change at residue 378 in domain III of EF-Tu. They also have identified several new kirromycin-resistant mutations, at residues 120, 124, and 160 in domain I of EF-Tu, each with severe defects in the interaction with aa-tRNA (Abulkarim and Hughes, unpublished data). In addition, two histidine residues in domain I of EF-Tu, His-66 and His-118, have been identified as the sites of cross-links to tRNA (168). These particular positions have been altered by site-directed mutagenesis: His-66 to Ala (7) and His-118 to Ala, Glu, and Gly (7, 121). In each case, the tRNA affinity for EF-Tu is severely reduced. Such results suggest that the affinity of aa-tRNA for EF-Tu·GTP is sensitive to several specific, single amino acid substitutions in EF-Tu. In summary, much of the mutational and cross-linking evidence points strongly to an intimate interplay between domain I of EF-Tu and aa-tRNA.
The function of EF-G is to promote translocation in some as yet unknown way (127, 149, 240). It is evident in the Noller model that this step involves a reassociation of tRNA and mRNA with the functional sites of the ribosome that at a minimum corresponds to the movement of the A-site and P-site tRNAs into the P-site and E-site, respectively. Unfortunately, the coupled movements of mRNA during this rearrangement are not as well documented (28, 150, 241). Nevertheless, if we consider translocation as an orderly movement of tRNA and mRNA, mutants of EF-G might be expected to influence the accuracy of these movements. The most recent data concerning mutants of EF-G do not support this expectation.
Mutants of EF-G have been isolated as resistant to fusidic acid or to low levels of kanamycin. Two kanamycin-resistant EF-G mutations (110) have been found to reduce elongation and growth rates, but their influence on accuracy has not yet been reported. One fusidic acid-resistant mutation has been reported to reduce the level of spontaneous frameshifting at a few sites in lacI, without having an effect on nonsense readthrough (215). This finding suggests that EF-G may affect the accuracy of reading frame maintenance. In contrast, a more recent study of 18 new fusidic acid-resistant mutants of EF-G in S. typhimurium provides little support for this conclusion.
These new mutations cluster in three regions of the fus gene (120), corresponding to three interacting domains of the protein structure (4a, 55). Many of these mutations have adverse effects on elongation and growth rates. Indeed, there is a good correlation between the degree of growth impairment and an associated reduction in translation elongation rate (U. Johanson and D. Hughes, unpublished data). The mutant described previously by Richter Dahlfors and Kurland (215) is also kinetically impaired to an extent that places it in the middle of the range of variation described by the new Salmonella mutants.
A similar relationship between exponential growth rate and translation elongation rate has previously been seen for restrictive rpsL mutations alone (12, 29) and in combination with error-prone tuf mutations (268). In the case of the rpsL and tuf mutations, the reduced rates are associated with a reduced efficiency of interaction between the ribosome and ternary complex correlated with varying degrees of an error-restrictive phenotype (268). Although the analysis is still incomplete, it appears that the growth and elongation rate phenotypes of the fus mutations also reflect, at least in part, the efficiency with which EF-G interacts with the ribosome (Johanson and Hughes, unpublished data). Furthermore, the depressed kinetics of some of the mutant EF-Gs can be reversed by some mutations in 16S rRNA (Johanson and Hughes, unpublished data), which also is reminiscent of the cooperative interactions described for some combinations of rpsL and tuf mutations.
To determine whether the new fus mutations have any associated accuracy phenotypes, frameshift suppression rates and nonsense readthrough rates for all 18 mutations were measured (Johanson and Hughes, unpublished data). For 17 of the mutations there is a weak positive correlation between the level of nonsense readthrough and the level of frameshift suppression. Readthrough and suppression are weakly restricted in some cases but not significantly different from the wild-type level in the remainder. We know of no a priori reason to expect that EF-G mutations should have any direct influence on nonsense readthrough. Therefore, we suspect that the small degree of restriction observed in both nonsense readthrough and frameshift suppression is a secondary effect of the reduced elongation rates characteristic of the mutations.
Only 1 of the 18 mutations (fusA3) offers any real support for the idea that EF-G may have a specific effect on reading frame maintenance. The fusA3 mutant has a wild-type level of nonsense readthrough and is moderately restricted in frameshift suppression (by about 40%), which is reminiscent of the phenotype of the mutant described previously by Richter Dahlfors and Kurland (215). However, the low magnitudes of the restrictive effects of most of these mutations, which are independent of the magnitude of the effects on elongation rate and growth rate, hint that EF-G may have no direct influence on the accuracy of reading frame maintenance during translation. Alternatively, different selection procedures may be required to identify EF-G mutations with pronounced effects on frameshifting.
Thiostrepton is an antibiotic which binds to the 50S ribosomal subunit, interacting with the A-1067 region of 23S rRNA (253, 254). This antibiotic-ribosome interaction prevents both EF-G (27, 279) and EF-Tu (54, 178) from binding to the ribosome. Conversely, ribosomes with bound EF-G are unable to bind thiostrepton (108) or to bind EF-Tu in a ternary complex (214). These results suggest that the ribosome binding sites for both elongation factors and for thiostrepton overlap and include the 50S ribosomal subunit. Attempts have been made to locate more precisely the ribosome binding site of EF-Tu by fluorescence (225) and by immunoelectron microscopy with (87) or without (144) cross-linking. The data (reviewed by Spirin and Vasiliev [242]) are consistent with EF-Tu binding to both the 30S and 50S subunits, in the region of the L7/L12 stalk on the 50S subunit. This is also the area of the ribosome at which similar techniques identify an interaction with EF-G (88). EF-G has also been cross-linked to a short fragment of 23S rRNA containing the A-1067 thiostrepton site (236). Chemical protection of rRNA in ribosomes with EF-Tu bound by kirromycin, or EF-G bound by fusidic acid, suggests that specific bases in the 23S rRNA (at and around G-2661 in the α-sarcin loop by both factors and around A-1067 by EF-G alone) are protected by the drug-bound factor (177). Cleavage at G-2661 in the α-sarcin-sensitive loop of 23S rRNA in 70S ribosomes specifically blocks the binding of both elongation factors, leaving other ribosomal interactions unimpaired (105). Protection data do not distinguish between a direct protection by EF-Tu of these bases in rRNA as opposed to an indirect protection that is the result of EF-Tu-induced structural rearrangements of the ribosome. However, the direct interpretation is supported by genetic and biochemical data (21, 252) showing that the mutation G2661C in 23S rRNA interferes with ternary complex binding to the ribosome and that the defect can be reversed by a mutant EF-Tu (250). Intriguingly these effects are enhanced in the presence of an error-restrictive S12 mutation on the small ribosomal subunit. This would suggest that both the α-sarcin-sensitive loop of 23S rRNA on the large ribosomal subunit and the S12 region on the small ribosomal subunit are involved in the ternary complex interaction with the ribosome. The possibility of an interaction between EF-Tu and S12 is also suggested by the isolation of rpsL500 in a selection for kirromycin resistance (265). This mutation is discussed more thoroughly below.
Two mutants of EF-Tu that have a major, specific defect in their interaction with the ribosome have been independently identified. Each involves the mutation of a conserved glycine on the surface of domain II of EF-Tu; one is at position 220 (247), and the other is at position 280 (267). In each case, the mutant EF-Tu can form a ternary complex with aa-tRNA and GTP, but it subsequently fails to interact productively with the ribosome. At this stage, one can only speculate that this part of the surface of domain II of EF-Tu may interact with the ribosome.
The recent solution of the crystal structure of EF-Tu in its GTP-bound form (19, 132) and a refinement of the GDP-bound form (133) have led to renewed interest in the structure of the complex of EF-Tu with aa-tRNA. At the time of this writing, there are two contradictory published models (76, 129). One model, that of Kinzy et al. (129), is based on cross-linking and protease mapping studies of an EF-1α·GTP·aa-tRNA complex, with the results mapped onto the EF-Tu·GDP structure (133). They propose that the aminoacyl end of aa-tRNA is in contact with domain II of EF-Tu, that the D and TΨC arms are close to domain III, and that the anticodon arm is in solution near domain I. In contrast, the second model, that of Förster et al. (76), has the aminoacyl stem of aa-tRNA inserted into the cleft between domains I and II of EF-Tu·GTP, with the anticodon stem and loop protruding into solution over domain II.
Evidently, there is an asymmetry in the effects of mutations on the rates of growth and translational elongation: without exception they decrease these rates. So far no mutants have been observed to grow or translate significantly faster than laboratory wild-type strains. We take this as an indication that the wild-type laboratory strains are maximized for growth and translational elongation rates.
There are a multiplicity of ways of obtaining either a restrictive or a Ram variant in the sense that either sort of mutation can be expressed in ribosomal protein, RNA, or EF-Tu. Nevertheless, when mutations of opposite tendency are combined, they invariably tend to cancel or suppress each other. It would seem that there is a simple arithmetic for the sums of the effects of accuracy mutants on both the elongation rates and the growth rates.
Finally, all of the mutants that influence the missense and nonsense error rates also influence frameshifting error rates. This hints at an overlap between the mechanisms of aa-tRNA selection and mRNA movement.
Since mutants with altered translational accuracy can be either error prone or hyperaccurate, we may infer that the intermediate accuracy of wild-type translation must be some sort of compromise. It is easy to guess that the compromise might concern a trade-off between the higher efficiency of error-free proteins and the costs of maintaining the accuracy at a certain level. The conceptual problem is to formulate the nature of this trade-off in a way that can be exploited experimentally. This has been done by using the growth rate of bacteria as the indicator of the efficiency of translation as well as of the costs of accuracy (67).
The general formulation of this model, the growth optimization model (67), begins by demonstrating that there is an optimal arrangement of the components of a bacterium that will lead to a maximum growth rate in a particular medium. The optimum partitioning of bacterial mass is in general different for maximum growth in different media. For example, it can be shown that maximum growth rates are supported in richer media by bacteria with a high density of translation components, while a lower density of these components is required for maximum growth rates in poor media. This expectation is reflected in the observations of Maaløe (154) and Maaløe amd Kjeldgaard (155), who described earlier the variations in ribosome content for bacteria under different growth conditions. A detail that the growth optimization model predicts, which was not recognized by Maaløe and Kjeldgaard, is the inverse variation on the one hand for the ratio of the aa-tRNA·EF-Tu ternary complex to ribosomes and on the other hand for the concentrations of ribosomes at different growth rates. Likewise, the model suggests, in contrast to the expectations of Maaløe and Kjeldgaard, that rates of translation per ribosome will increase with increasing growth rates of the bacteria. Both of the expectations of the model are in agreement with experimental data (49, 50).
The growth optimization model is a kinetic optimization, in which the steady-state flows through each compartment correspond to the flow required to "feed" other compartments plus the flow corresponding to the growth expansion of the system. A key point is that all the components of each functional pathway must work at a mass-normalized rate that is a maximum. As a consequence, a mutation that changes the kinetics of the ribosome may be evaluated not simply by what it does to the rate of protein synthesis but according to how the rate normalized to the new mass of translation apparatus is changed. Here, a mutation that slows down ribosomes might be advantageous in principle if it reduces the size of the translation system proportionately more than it retards it. It is this mass normalization that permits the different functional compartments to be optimized with respect to each other. As we will see below, the mass normalization also provides a natural connection between the growth optimization and Michaelis-Menten kinetics (67). Accuracy optimization has a natural and simple formulation within the context of this more general model.
To simplify the presentation, we can consider first mutations that do not change the mass of the system but do affect the kinetics of aa-tRNA·EF-Tu ternary complex selection at the codon-programmed ribosome in a minimal kinetic scheme (140):
↓ kd
(1)
Here, a ternary complex, T, associates with the ribosomal binding site, Ra, in a diffusion-limited process governed by a rate constant, ka. The intermediate complex (Ri) so formed can either dissociate at a rate determined by kd or form a peptide bond through an enzymatic process whose rate is determined by kc. The ratio ka/kd is obviously the equilibrium constant for the association of T at the codon-programmed site but is shown as above to illustrate a branch point. The probability that the peptide will be formed from the intermediate Ri is for a cognate ternary complex determined by the ratio kd/kc = a and is 1/(1 + a). For a noncognate ternary complex, the equivalent probability will be 1/(1 + Da). We refer to a as the discard parameter and to D as the discrimination ratio, which describes how much more quickly the noncognate ternary complex is lost from the ribosomal intermediate than is the cognate species. Accordingly, the cognate rate of product formation is retarded by a factor of 1/(1 + a) and the noncognate rate is retarded by a factor of 1/(1 + Da). It is easily shown that at equal concentrations of a cognate and a noncognate ternary complex, the error rate (E) with which noncognate peptide bonds are incorporated according to this scheme is given by:
E = 1/[1+ (1 + Da)/(1 + a)] (2)
The parameter D can be thought of as a measure of the maximum difference in binding energy for cognate and noncognate species at the ribosome-bound codon.
This minimum formal apparatus allows us to simulate the response of the system to mutations that affect the accuracy of ternary complex acquisition by systematically perturbing only one parameter. Thus, D is taken as an invariant physical parameter that describes codon-anticodon interactions, so we can calculate the influence of mutational change of the discard parameter a on the rate and accuracy of peptide bond formation. We do this by calculating at each value of a both the rate of elongation relative to an arbitrary standard rate and the error frequency. We then correct the rate of peptide bond formation for the influence of the errors which we assume cause an average decrease in the efficiency of function of the proteins produced at any value of a. In this way, we can simulate the influence of variations in the discard parameter a on the effective rate of protein synthesis.
Figures 3 and 4 describe such a simulation for the system described in scheme 1. At low values of a, the rates of translation are high but the error rates are also high; consequently, the effective rate of protein synthesis is low. As the values of a increase, the rate of protein synthesis decreases but the error rate decreases disproportionately faster, so that the effective rate of protein synthesis increases until a maximum value is reached. Beyond this maximum, a further increase of a leads to a progressive decrease in the rate of peptide bond formation, but the corresponding improvement of the accuracy of translation has a negligible influence on the effective rate of protein synthesis. Evidently, it is possible to obtain different maxima for the effective rate of protein synthesis for different degrees of missense error impact (Fig. 5) as well as for different values of D (Fig. 6). Similar types of maxima are also obtained for more complex systems, such as multistep selections (Fig. 6). However, the point is that maxima describing the trade-off between accuracy and rate of translation can be obtained for the simplest as well as for more complex flow schemes. This means that we have a precise way to evaluate the relationship of accuracy to the growth optimization of bacteria. Thus, for a given set of assumptions, the maximum represents the fastest effective rate of translation, and therefore it represents the fastest rate of growth for the bacterium. Discussion of one more formal point is required before we review the relevant data.
We note that for fixed values of D, the only way to increase accuracy in this scheme is to increase the discard parameter a. In other words, accuracy is obtained by less efficient acquisition of the competing substrates, both cognate and noncognate. In a sense, the system pays for improved accuracy by a lower level of substrate saturation or, equivalently, by a larger Michaelis-Menten constant (Km). We also note that the potential rate loss due to greater values of a can be compensated for by increased substrate concentrations. However, in the context of growth rate optimization, this rate compensation will still have a negative effect on the growth rate because the mass of the system increases along with the substrate concentration, and consequently, the mass-normalized rate of the translation system will decrease.
This simple view of the kinetics of translation and of the impact of accuracy variants on these kinetics has been verified in a number of different ways. These will be documented in the discussion that follows.
Under most growth conditions, EF-Tu, in terms of mass, is the single most abundant protein in S. typhimurium and E. coli. Tubulekas and Hughes (266) have determined that EF-Tu is about 9% of total protein in S. typhimurium growing exponentially in glucose minimal medium or in rich medium. These values are similar to those determined for E. coli (271). Although seemingly counterintuitive, the growth optimization model predicts that even with these high concentrations of EF-Tu and comparable concentrations of aa-tRNA, ternary complexes should not saturate kinetically the translating ribosomes under most growth conditions. In fact, the prediction, which is in contrast to that of Maaløe (154) and Maaløe amd Kjeldgaard (155), is that the kinetic saturation of ribosomes should vary proportionately to the growth rate. In this variation, the degree of saturation increases with the demand for efficient translation kinetics, and it might approach saturation only at the very fastest growth rates sustained by the bacteria.
This possibility has the important consequence that ribosome mutants with enhanced accuracy of aa-tRNA selection would be expected to have ribosomes with larger Km values for the reaction with the ternary complex in vitro. Furthermore, the mutant bacteria themselves should have lower translation rates associated with slower growth rates. These predictions have been verified in some detail for a range of hyperaccurate mutants including Smr and Smp variants (9, 12, 29, 222). In summary, the growth optimization model can account for the cost of accuracy as a kinetic penalty that must be weighed against the kinetic value of error-free proteins that are maximally effective. Proteins with many errors are, kinetically speaking, a bargain, but their functions are hampered. The trade-off between these costs, the costs of accuracy and the costs of errors, leads to an optimum at which the maximum growth rate is supported by a kinetically efficient translation system with the most acceptable error rate: the wild-type bacterium.
It turns out that mutants of EF-Tu are potentially more suitable than ribosome mutants for quantitative studies of the accuracy optimization because the potential complications generated by the self-assembly of ribosomes are circumvented. The tufA and tufB genes code for identical EF-Tu proteins in S. typhimurium and each can be inactivated by Mud insertions (113). Strains with either tuf gene inactivated by Mud insertion contain about 65% of the normal amount of EF-Tu (266). The tufB gene normally contributes only one-third of the total, so that upon inactivation of tufA, the accumulation of tufB is apparently up-regulated. Translation elongation rates and exponential growth rates in these strains are each reduced to about 83% of the normal wild-type level. From the magnitude of the reduction in elongation rate relative to the reduction in the amount of EF-Tu, it was calculated, by using the Michaelis-Menten equation, that in vivo the saturation level of wild-type ribosomes by the ternary complex in glucose minimal medium is 0.63. In the strains in which one tuf gene is inactivated, the saturation level was calculated to be 0.50 (266). These saturation levels are sufficiently low that both translation rates and growth rates will be sensitive to changes in the kinetic saturation of ribosomes by the ternary complex in vivo.
In general, we expect that accuracy mutations affecting the ability of the ternary complex to interact with the ribosome will influence the kinetic saturation level. An extreme example of this is provided by the tufB mutation Gly-280→Val, which forms a ternary complex but fails to interact productively with the ribosome (267). Strains carrying this mutation have the same EF-Tu content as wild-type strains but have growth and elongation rates similar to those of strains with one tuf gene inactivated by Mud. Another example is provided by the phenotype of strains carrying a single active tuf gene and a ribosomal protein S12 mutation (rpsL500). Such double-mutant strains are inviable if the active EF-Tu is wild type but are viable if it is an error-prone mutant, i.e., carries a mutation which improves the interaction (i.e., lowers the Km) of the ternary complex interaction with the ribosome (265, 266). In each case of such double mutants, the actual amount of EF-Tu is the same, but the ability of the corresponding ternary complex to saturate the ribosome has been altered. The calculated saturation level for the ribosomes by the ternary complex in the inviable combination is 0.25 to 0.30. It may be that at this low level of saturation, bacterial viability breaks down because of secondary effects, for example, loss of attenuation control or other types of translational regulation of gene expression.
It is well established that error-restrictive S12 mutations reduce the kinetic efficiency of the interaction between the ternary complex and the ribosome (12, 20, 29, 265, 268). Furthermore, it was found for one extreme S12 mutation (rpsL500) that kinetic efficiency could be at least partially restored in vitro by the error-prone EF-Tu mutation Val-375→Thr (265). The interpretation is that the restrictive S12 mutations reduce the saturation level of the ribosome, while the error-prone EF-Tu mutations increase the saturation level, as has been seen previously with combinations of rpsL (restrictive) and rpsD (Ram) mutants (12).
A series of strains was constructed to test the effects of combining several restrictive rpsL mutations with the wild-type tuf gene or with one or two error-prone tuf mutations. The observed pattern is that each of the restrictive rpsL mutations causes a characteristic reduction both in translation elongation rate and in exponential growth rate (268). A similar linear correlation between elongation rate and growth rate had also been seen with a set of restrictive rpsL mutants from E. coli (12). When these measurements were made on the strains carrying both error-prone tuf mutations and the restrictive rpsL mutations, direct reversals of the slow elongation and growth rate phenotypes characteristic of the restrictive mutants were observed (268). Thus, major effects of these rpsL and tuf alleles are to reduce and increase, respectively, the kinetic saturation level of the ribosome by the ternary complex. Formally, it seems that the different mutations behave as though they have independent effects on the discard parameter a and that the sum of these effects determines the accuracy as well as the kinetic efficiency of the tRNA selection process.
As an aside, we should note that the phenotypic reversals of the elongation and growth rate defects of the stringent rpsL mutations by the Ram tuf mutations are incomplete in each case, even when the translation error levels are restored to the wild-type level (268). This is likely to be due to a second effect of rpsL mutations, namely, a reduced rate of GTP hydrolysis on EF-Tu (20) which may not be reversed by the increased affinity of EF-Tu for the ribosome.
What happens if the amount of EF-Tu in the cell increases? Overproduction of EF-Tu from plasmids does not enhance elongation rate (266), which shows that when the EF-Tu concentration is raised, translation elongation rates are limited by other factors. These may include the availability of aa-tRNA to form ternary complexes with any excess EF-Tu.
These data along with the similar behavior of EF-G mutants (Johanson and Hughes, unpublished data) described above support the expectation that a reduction in the kinetic saturation level of the ribosome by the elongation factors will reduce translation rates as well as growth rates. In other words, the bacterial ribosomes are normally not kinetically saturated by elongation factors. Therefore, the reduced growth rate of a hyperaccurate translation mutant is an index of the kinetic cost of the corresponding degree of accuracy enhancement. However, the situation is more complicated than this, as we show when we discuss the relationship of processivity errors to missense errors.
Direct support for the idea that ribosomes of so-called wild-type strains represent the growth-maximized phenotype under normal laboratory growth conditions comes from an unexpected quarter (169, 170). A group of 65 different strains, mostly from the ECOR collection (191), were grown in batch cultures on glucose minimal medium in order to measure their maximum growth rates. It was observed that the doubling times varied from 42 to 124 min. A few of the natural isolates had growth rates comparable to those of wild-type laboratory strains, with doubling times in the range 42 to 47 min. However, the surprising observation was that the vast majority grew with doubling times significantly longer than 50 min, with a mean value of 71 min (169).
Clearly, the natural isolates were not selected uniformly for one particular growth phenotype, and they certainly were not selected for maximum growth performance in glucose minimal medium. However, after eight of the natural isolates, including the slowest representatives, were cultivated for 280 generations in glucose-limited chemostats, their descendants grew at rates comparable to those of the wild-type laboratory strains, with a mean doubling time of 49 min and extremes of 44 and 52 min (170). In other words, growth in glucose minimal medium tends to select growth-maximized bacteria. It may be inferred from these observations that the long-term cultivation of bacteria under routine conditions tends to select a "laboratory phenotype" with maximum growth rates in batch cultures. We return below to the nature of the selective forces on natural strains. Before doing so, we need to discuss the performance characteristics of the ribosomes from the natural isolates.
Ribosomes prepared from the original natural isolates were analyzed in vitro. It was found that their kinetic properties were as broadly distributed as the growth rates of the bacteria themselves (169). Indeed, the kinetic efficiency with which the purified ribosomes interact with the aa-tRNA·EF-Tu ternary complex is linearly correlated with the growth rate of the parent bacterium in glucose minimal medium. We note that this is the same sort of correlation that is obtained with mutant and wild-type laboratory strains. In contrast, the missense error rates of the isolated ribosomes in vitro are randomly distributed with respect to both growth rates and ribosome kinetics. This is a pattern distinctly different from the one described above for laboratory strains, for which there is a tight correlation between missense error rates and translation kinetics.
Evidently the ribosome phenotypes of the natural isolates have evolved in response to forces different from those selecting the phenotypes of the laboratory strains. Nevertheless, ribosomes isolated from the descendants of the natural isolates that had been cultivated for 280 generations in chemostats have ribosome kinetics and missense error rates that are uniformly indistinguishable from those of wild-type laboratory strains (170). In other words, the ribosomes of bacteria selected for maximum growth rates in the laboratory have well-defined growth-optimized performance characteristics that may be identified with those of the ribosomes from wild-type laboratory strains.
It has been suggested that the phenotypes of the natural isolates are different from those of laboratory strains because natural isolates are selected in part for their ability to survive starvation conditions, not just to grow well when nutrients are available (171). Indeed, it was observed that natural isolates with slow growth rates and sluggish ribosomes survive carbon starvation better than do their peppier relatives. In addition, descendants of the slow growers that have been adapted to glucose minimal media in chemostats lose the superior capacity to survive carbon starvation. These observations suggest that there is some aspect of the growth-optimized bacterium that is antagonistic to efficient survival of carbon starvation and, in equal measure, that there is some selected aspect of efficient carbon starvation survival that is inimical to maximum growth rates (171). In this view, the growth characteristics and ribosome phenotypes of individual natural isolate clones reflect the recent nutritional history of the clone: those that have had access to nutrients are fast growers, and those that have starved much of the time are slow growers.
Finally, the selection of the maximum growth rate phenotype from the natural isolates is associated with a very marked selection of a fairly uniform missense error frequency that is characteristic also of the ribosomes from laboratory wild-type strains (170). The estimated missense error frequencies of the chemostat-selected ribosomes are low; for example, the error of leucine substitution for phenylalanine in a poly(U)-programmed system has a mean close to 4 × 10–4 and a range of 3.2 × 10–4 to 4.3 × 10–4 (170). Nevertheless, the very narrow breadth of this distribution suggests that even small deviations from this error rate are culled during the growth selections. This is in stark contrast to the apparently random distributions of the missense error rates in the natural isolates.
It is possible to relate the growth rate depression of some error-prone mutants to their missense error rates, but the interpretations of the relationship are not straightforward.
One example of error-prone phenotypes is provided by the EF-Tu mutation Ala-375→Thr, which reverses the restrictive rpsL phenotypes and is by itself error prone (112, 114, 115). Likewise, the Ram mutations in rpsD are error prone (10), and they reverse restrictive rpsL phenotypes (12). In the case of EF-Tu, the mutation primarily affects its initial interaction with the ribosome (114, 251) whereas the Ram ribosomes primarily affect proofreading of noncognate aa-tRNAs after GTP hydrolysis on EF-Tu (9, 11). The important point is that there is for these mutations, regardless of their origin, a strong correlation between increased error levels and decreased growth rates (114, 116). The correlation is linear over a 10-fold increase in missense errors measured in vitro, and over this range, the growth rate is halved. Thus, each successive 100% increase in missense errors adds approximately 11% to the time required to complete one cell generation.
Because the error-prone tuf and ribosome mutations enhance all types of errors (missense, nonsense, and frameshift), we cannot rule outa priori the possibility that the growth rate reductions are related to an unacceptably high level of frameshifting on some vital message. However, when β-galactosidase is used as a test protein in these assays, it is found that its specific activity rather than the amount of protein decreases progressively as the number of missense errors increases (114, 116). This finding suggests that processivity errors associated with the enhanced missense errors are not primarily responsible for the loss of activity or the decrease in growth rates with these error-prone mutants. Indeed, the reductions in β-galactosidase specific activity are dramatic: a fourfold increase in missense errors caused by mutant EF-Tu reduces specific activity at least twofold. However, the β-galactosidase subunit is at least twice the size of an average protein and is functional as a tetramer. Accordingly, we suggest that β-galactosidase is a good analog for one or more large protein complexes (e.g., RNA polymerase) that would be limiting for growth and, because of their size, more than usually sensitive to the impact of missense errors.
In what follows, we discuss only so-called spontaneous frameshifts; discussions of programmed frameshifts are found in papers by Atkins et al. (16, 17).
In previous sections it was noted that both restrictive and error-prone mutations tend to influence the frequencies of frameshift errors and missense errors in the same direction; that is, they simultaneously increase or decrease occurrences of both missense and frameshift mutations. These error correlations began to appear at a time when mutations that suppress frameshift errors were first analyzed and shown to change the structures of the anticodons of particular tRNA species (217, 218, 220, 221, 292). The natural inference from these data is that the codon-anticodon interaction normally guides the phased movement of the mRNA during an elongation cycle. A corollary of this inference is that distorted codon-anticodon interactions, as for example when a missense error has occurred, would be expected to lead with abnormally high frequencies to a frameshift error (137).
This expectation is confirmed in general by the observed phenotypes of translational accuracy mutants. Another connection between missense events and frameshift events is observed under conditions in which the aminoacylation of individual tRNA families is experimentally inhibited while protein synthesis continues at a reduced rate. Under these conditions, noncognate tRNA species will replace the uncharged tRNA species at some of its cognate codons. The expectation is that the resulting mismatches at starved codons will generate frameshift events at an abnormally high rate (137). A speculative mechanism for this effect was thought to be that mismatched tRNA species might slip out of phase and interact with overlapping, out-of-phase triplets. Such a mechanism accounts for the fact that the first generation of tRNA suppressors was found to favor runs of identical nucleotides in the mRNA strings at which they generated compensating frameshifts (217, 218, 220, 292). It turns out that data obtained from quite different systems suggest that a general mechanism for frameshifting includes slippage into a cognate or quasicognate phase mismatch.
For example, it is observed that both the deprivation of particular aa-tRNA species (82, 280, 281) and the accumulation of abnormally high concentrations of other tRNA species (15, 39) will raise frameshift frequencies. Further studies by Gallant and his coworkers defined the peculiarities of these frameshift events for AAG codons that are starved for their cognate Lys-tRNA (83, 151, 202, 282).
It turns out that all of the frameshift events associated with the starved AAG codons could be associated in one or another of two groups (Fig. 7): the leftwingers that produce –1 nucleotide frameshifts, which suppress the – mutations in the Crick et al. collection (53), and the rightwingers that produce +1 nucleotide frameshifts, which suppress the + mutations in the same collection. In addition, each of the classes is steered by a characteristic heptanucleotide sequence. For the rightwingers, the heptanucleotide contains a triplet to the immediate left of the AAG codon and one nucleotide to the immediate right of this codon. For the leftwingers, the four nucleotides to the immediate left of the AAG codon complete the heptanucleotide.
Lindsley and Gallant (151) suggest on the basis of amino acid sequences of the frameshifted products that for the rightwingers, the left-positioned triplet calls for an Ala-tRNAGCC which at the time of the shift is positioned in P-site, while the right-side nucleotide provides a wobble position at which a Ser-tRNA mismatched at AAG can slip into the overlapping AGC codon to accomplish the shift. Similarly, for the leftwingers, the four nucleotides to the left permit slippage leftward of a peptidyl-tRNAPhe from its cognate UUC to an overlapping CUU triplet simultaneously with the shift leftward of a Gln-tRNACAA from the mismatched AAG codon to accomplish the shift.
In summary, the data of Gallant and workers (83, 151, 202, 282) are consistent with the interpretation that frameshifts resulting from deprivation for charged tRNA species occur at the starved codons and that these are translated by a mismatched tRNA that has an overlapping triplet to the right or to the left with which to make a nearly cognate match, i.e., a quasicognate phase mismatch. This conclusion is consistent with the idea that tRNA suppressors of frameshift mutations are themselves shifty owing to alterations of their anticodon loops. In such cases, an extra base in an anticodon loop seems to permit the mutant tRNA to read overlapping triplets within mRNA strings of four or more repetitive nucleotides (137). Accordingly, we suggest by analogy that the majority of spontaneous frameshift events are associated with missense events. In particular, we expect frameshifts to occur at those codons for which a mismatched tRNA has an opportunity to shift into a match or near match with an overlapping, out-of-phase triplet. This will account for the parallel influences of mutations in ribosomes and EF-Tu on missense and frameshift frequencies.
We have earlier suggested that drop-off is the dominant processivity error under normal laboratory growth conditions. Furthermore, the available data suggest that drop-off events may occur at frequencies at least 1 order of magnitude higher than that for frameshift events in laboratory strains. Therefore, drop-off is potentially more of a problem for the growing bacterium than is frameshifting. In the following discussion, we suggest that drop-off events and frameshift events are distributed asymmetrically around the growth rate optimum of translational accuracy. Furthermore, we suggest that this arrangement sets constraints on the accuracy with which tRNA can be matched to codon during translation.
The report that streptomycin increases in parallel the missense error rates and the drop-off frequencies (47, 167) initially suggested that drop-off frequencies might be correlated with frameshifting (141). However, it has been observed more recently that streptomycin reduces dramatically the processivity error rate for an Smp mutant at the same time that it increases missense frequencies (62, 123). In addition, it is observed that a series of Ram mutants abortively terminate translation at moderately reduced frequencies that are not correlated with the more extreme variations of their missense rates (62). Therefore, we now believe that there is no strong correlation between missense and drop-off events.
In contrast, restrictive ribosome mutants of E. coli translate with processivity losses that are strongly correlated to their degree of missense restriction (62, 123). The observation that missense error-restrictive mutants are also frameshift restricted (14) suggests that the enhanced processivity losses of restrictive mutants are due to augmented drop-off frequencies. Accordingly, it seems that frameshifting events occur preferentially in association with missense, while drop-off frequencies become more prevalent as the restriction of missense is strengthened.
That increased missense restriction would be associated with increased drop-off was anticipated on the grounds of a simple extension of the mechanism described in scheme 1. In this scheme, for tRNA selection we require the more accurate ribosome to release tRNA species from its codon-programmed binding site more easily than the more error-prone ribosome. Indeed, accuracy of tRNA selection is in our formulation determined in part by the magnitude of this tendency to release tRNA from the decoding site, which is summarized in the discard parameter a. Now, we have seen in the analysis of the frameshift error mechanism that phase errors seem to occur when peptidyl-tRNA wanders away from an in-phase interaction with mRNA when noncognate pairing weakens the codon-anticodon interaction. Similarly, we may imagine that an enhanced tendency to release tRNA from the decoding site prior to peptide bond formation is retained after the peptide bond is formed. If there is an enhanced tendency for restrictive mutants to release peptidyl-tRNA from the decoding site either before or during translocation, this might be reflected in augmented drop-off frequencies for such mutants (141). In other words, an enhancement of a which is required for greater missense restriction would in this view lead to an enhanced drop-off rate.
More direct experimentation with ribosomes in vitro gives some support to this view. Karimi and Ehrenberg (126) have compared stabilities of dipeptidyl-tRNAs at the so-called A-sites of wild-type and mutant ribosomes. They find that the dissociation rates of two restrictive mutants are about 50% greater than that of the wild type, while the dissociation rate for an rpsD Ram mutant is fourfold lower than that of the wild type. Furthermore, they observe a systematic stabilization of dipeptidyl-tRNA at the A-site for all ribosome variants in the presence of streptomycin. In other words, the interpretation that drop-off results from an increased tendency to release tRNA from the ribosomes of restrictive mutants, and that streptomycin reduces the drop-off frequencies by enhancing the stability of peptidyl-tRNA binding at the ribosomes, is supported by the in vitro studies. There are two implications of these results that need to be expanded upon.
First, there is the question of what streptomycin does to stimulate or to support the growth of certain restrictive mutants such as the Smp and Smd variants. The suggestion is that very restrictive mutants such as the Smp and Smd variants are restrained by their inability to process long proteins such as the β and β' subunits of RNA polymerase and that streptomycin relieves this restraint by reducing the drop-off frequencies to more favorable levels (123). Hence, the hypothesis is that streptomycin stimulates the growth of these mutants by reducing processivity error rates at the expense of raising missense error rates. For the hyperrestrictive mutants, this trade-off is favorable.
Second, there is the question of how a positive coupling between missense restriction and drop-off rates influences the growth rate optimization of ribosome functions. The initial optimizations depended on the maximization of only two opposing tendencies: the loss of protein efficiency due to missense substitutions and the loss of translation efficiency due to the costs of restricting missense. Now, we have seen in the previous section that missense events are correlated with frameshift errors. This will compound the negative influences of missense by raising the average cost per missense event. There are a number of ways to simulate this effect, but the simplest is to assume that a small subset of all missense events always results in a frameshift (Fig. 5). It is evident that this sort of error coupling can sharpen the approach to the optimum value of a.
A similar but antisymmetrical effect of drop-off sharpens the decline from the optimum. Thus, the data discussed above suggest that it would be reasonable, at least for the sake of illustration, to assume that drop off increases as a function of a. For example, the processivity might be taken as proportional to the term 1/(1 + a). An optimization that takes into account the drop-off error is illustrated in Fig. 5.
The sharper optimization of ribosome performance that results from taking into account four parameters that influence the effective rate of translation provides an explanation of yet another observation that has been discussed above. We have emphasized that there is a remarkable convergence of ribosome performance characteristics following relatively short selection times for the descendants of natural isolates grown in chemostats. The simulation in Fig. 5 suggests that the combined influence of selections to optimize the kinetic efficiency of the ribosome and those to minimize the losses due to errors, especially those due to processivity errors, may account for the rapid convergence of ribosome performance characteristics in the chemostat selections. Likewise, it also seems clear that routine passage in laboratories can create a relatively uniform laboratory phenotype from the most diverse spectrum of natural isolates.
So far our exposition has been focused primarily on the generalities of the accuracy/error dualism and its relationship to the growth physiology of bacteria. To extend this discussion to the molecular domain, and in particular to help explicate the mechanisms of proofreading, we need to discuss the functions of nucleoside triphosphates in translation. By so doing, we hope to be able to present a general description of the molecular problems confronting a translation "machine" and to put into a proper perspective the role of the high-energy nucleoside triphosphates.
The elongation of a polymer such as a protein is a unidirectional process; therefore, one molecular problem is to identify the nature of the forces that drive it in one direction. What is pushing or pulling the translational machinery along the mRNA? There is an analogous directionality, or irreversibility, in the selection of the correct aa-tRNA at each codon in the mRNA (136). It is obvious from the very notion of chemical equilibrium that the process of translation takes place far away from chemical equilibrium. Here, it is assumed that the small free energy differences between peptide bonds for different amino acids are not relevant to their positions in proteins or to their codon specificities. In other words, for the purposes of our discussion, at equilibrium the standard free energy change for the formation of any peptide bond is taken to be equivalent to that for any other one. Therefore, it follows that at equilibrium, it will not be possible to select a particular sequence of amino acids on the basis of any codon-specific interactions. This conclusion is rigorously derivable, and it rests on a simple observation (136). Whatever the differences are in, say, the binding energy of a correctly matched aa-tRNA at a codon-programmed site compared with that of a mismatched aa-tRNA, they will be matched by differences in opposite sign for the release of the aa-tRNAs from the codon-programmed site. Therefore, close to equilibrium, any amino acid can make a peptide product as well as any other one, regardless of the strength of the binding of its specific tRNA to the codon-programmed site on the ribosome.
To obtain a situation in which the binding specificity of the codon-anticodon interaction can be expressed, the symmetry of binding and dissociation at equilibrium must be broken, and this can be done only by driving the reaction far from equilibrium (136). For example, scheme 1 (see above) has been written without all of the reactions in all directions that would obtain at equilibrium. Instead, it has been written with the assumption that kc is a rate-limiting step and that all other reactions following it are negligible. Therefore, this scheme describes only a reaction pathway far from equilibrium and for which the symmetrical dissociation reactions that neutralize the binding specificity of the association pathway have been removed.
The circumstances which make a reaction scheme such as scheme 1 valid are easy to describe. As long as the products of the reaction governed by kc are removed so quickly from the system that the back reaction to Ri is a negligible flow, scheme 1 is valid. Indeed, scheme 1 is a minimum, irreversible scheme for the selection of a substrate on the basis of binding specificity. The point is that an irreversible reaction scheme is a sine qua non for the expression of the binding specificity of the codon in the selection of aa-tRNA during translation in precisely the same way that it is a sine qua non for the movement of the ribosome unidirectionally, from one codon to the next (136). A variation on this theme involves coupling the flow of the selected substrate, e.g., aa-tRNA in our case, to the flow of a nonselected substrate, GTP in our case. In such a coupled reaction, the displacement from equilibrium of the nonselected substrate can drive the flow of the selected substrate, and the specificity of the selection can be supported by the displacement from equilibrium of the nonselected substrate (136). This is presumably one of the functions of the ternary complex containing EF-Tu, GTP, and aa-tRNA, namely, to couple the flow of amino acids, in a codon-specific manner, to the flow of GTP. In this case, the displacement from equilibrium is generated by the efficient regeneration of GTP from GDP, so that virtually no back reactions involving GDP are supported.
It is worth emphasizing that the GTP per se does not provide the driving force in its coupled reactions. The driving force is the displacement of GTP from equilibrium with GDP, which is ultimately provided by the metabolism of the cell.
However, the high-energy character of the phosphodiester bond does have a different function (69). The high-energy character of the nucleoside triphosphates is expressed as the large standard free energy decrease in going from, say, GTP plus water to GDP plus a phosphate ion. Inspection of the rate equations for schemes with and without the coupled reaction with GTP reveals that the free energy contribution of the nonselected substrate augments the kinetics with which the selected substrate is processed. The rate enhancement is expressed in the lower concentrations of selected substrate that are required in the coupled reaction compared with those required in the uncoupled reaction to maintain a certain flow of product. In a sense, the high free energy contribution of the nonselected substrate "pumps up" the concentration of the selected substrate. Since, as discussed above, the concentrations of substrates are a significant cost of accurate translation, this influence of the nucleoside triphosphates is very important, particularly for costly substrates such as ternary complexes.
The starting point of the models of Hopfield (109) and Ninio (188) was that the standard free energy difference between cognate and noncognate codon-anticodon pairs is not enough to explain the low frequency of translation errors observed in vivo (152, 153). In terms of schemes 1 and 2, we would say that Hopfield and Ninio were convinced that the discrimination factor D in scheme 2 is too small to account for the observed accuracy of translation. Their solution to this problem was to postulate that the selection of the aa-tRNA at a codon-programmed site proceeded in a succession of steps that compounded the accuracy of a single selection. This sort of multistep process was called kinetic proofreading or simply proofreading. In such a multistep process, the preferential selection of cognate aa-tRNA in the first step, depicted in scheme 1, would be enhanced by the repetition of the selection in subsequent steps that also could be depicted as in scheme 1. The consequence of this type of repetition is that the error of the first step is multiplied by that of the second step to yield the error of a two-step process. For example, if the error is 0.01 in the first step and is 0.01 in the second step, the overall missense error after two steps is 0.0001, and so on and so forth. The key is to provide a mechanism whereby stepwise reductions of the error can be driven on the ribosome in such a way that the accuracy enhancement is multiplicative.
It is here that our discussion of driving forces is relevant. Hopfield (109) and Ninio (188) pointed out that if there is an accuracy-enhancement over and above the standard free energy differences for cognate and noncognate codon-anticodon interactions (or equivalently for D), a special displacement from equilibrium is required. This displacement from equilibrium was identified with the GTP reaction in the EF-Tu cycle. If the displacement from equilibrium for this GTP reaction drives accuracy-enhancing steps, then the excess GTP hydrolysis rates associated with cognate and noncognate aa-tRNAs should be correlated with these proofreading flows. In particular, a proofreading scheme would be verified by experimental determinations of two inequalities. First, the ratio between the number of GTPs hydrolyzed at a particular codon per noncognate peptide bond formation and the number of GTPs hydrolyzed per cognate peptide bond formation must be greater than 1; this ratio corresponds to the proofreading amplification of accuracy. Second, the ratio of peptide bond formation for cognate compared to noncognate amino acid should be significantly greater than the proofreading ratio. Indeed, the ratio of these two ratios corresponds to the accuracy of the nonproofreading or initial selection step.
The first experimental indications that proofreading is used by translating E. coli ribosomes were obtained by Thompson and Stone (259). Using a partial poly(U) translation system, i.e., one lacking EF-G, they found that more GTPs were hydrolyzed per noncognate dipeptide (f nc) than per cognate dipeptide (f c). Their results were compatible with the existence of proofreading. Though not definitive, the data provided a rough indication of how much accuracy might be amplified by a second selection step in a single cycle of peptide bond formation. Subsequent experiments confirmed these results in the single-cycle system (257), and Ruusala et al. (223) demonstrated the existence of proofreading in a steady-state translation system that elongated polypeptides in the presence of EF-G. The latter poly(Phe) system was optimized for high accuracy and rate (118, 275), making the accuracy parameters relevant for the situation in vivo. Ruusala et al. (223) obtained missense error frequencies in the range of 10–4 for poly(Phe) synthesis when noncognate ternary complexes containing Leu-
competed with Phe-tRNAPhe ternary complexes. They also found that EF-Tu must make 100 times more cycles for every peptidyl transfer to the noncognate Leu-than for a transfer to the cognate Phe-tRNAPhe. This finding implies that the accuracy enhancement by proofreading is a factor of a 100. Such results also implied that the accuracy of E. coli ribosomes is equally partitioned between initial selection and proofreading. In such a case, the parameters of schemes 1 and 2 are identical in the initial selection step and in the proofreading step.
We recall that the intrinsic selectivity D summarizes how much more tightly a cognate ternary complex is bound to a particular codon than is a noncognate one. This parameter may have an upper bound which, of course, would be relevant to the limits of accuracy in a single-step ribosomal selection. Indeed, model experiments performed with tRNAs containing fully or nearly complementary anticodons indicate D values of 100 or even lower (101). Corresponding estimates for D values for near-cognate and cognate ternary complex pairs, obtained from direct experimental measurements, are still missing. Indirect experimental evidence suggests D values of around 400 (21) or somewhat lower when Phe-tRNAPhe competes with Leu-
ternary complexes. At the same time the accuracy of the corresponding reactions is about 3,000 for wild-type ribosomes and close to 25,000 for hyperaccurate ribosomes. This would mean that the accuracy is much higher than the D value, which in turn would exclude the possibility that translational accuracy is maintained by single-step selection.
Considerable effort was spent by Thompson and collaborators to determine this standard free energy difference between Phe-tRNAPhe and Leu-interacting with UUU triplets in the A-site (255, 258). They reported a maximum possible accuracy in the initial selection of 40,000 (D = 40,000). At the same time only a factor of 12 out of this large number was actually expressed in the initial selectivity, according to their measurements. However, a closer inspection of how the experimental data were interpreted by Thompson and Dix (255) shows that these experiments only allow the more conservative conclusion that D is 2,000 or larger at 0°C (142, 258). There is current interest in the determination of the D value because of the claims by Nierhaus (185, 186) that proofreading is unnecessary for high accuracy in translation and hence does not exist (see below).
The more indirect approach by Bilgin and Ehrenberg (21), mentioned above, gave a D value of about 400 for the somewhat more error-prone isoacceptor (Leu-
), indicating that the maximal, initial selectivity of the poly(U)-programmed ribosome in discriminating against Leu-may be around 1,000 (D = 1,000). Initial selection has been measured to be about 100 (223). This, in turn, would imply that the probability that a cognate ternary complex dissociates from the A-site without GTP hydrolysis is about 10% and that only 10% of the maximal accuracy is used in initial selection; that is to say, a = 0.1 and D = 1,000, since according to Ruusala et al. (223), the missense error is reduced at each step by a factor of 100.
The picture that emerges from in vitro experiments is that about 10% of the maximal accuracy is used in the initial selection step and that there is a corresponding 10% reduction in the effective association rate (k cat/Km) for a cognate ternary complex interacting with ribosomes. If there is a single proofreading step and if the D value is the same in proofreading as in initial selection, then about 10% of the highest possible accuracy would be used also here. This would imply that the proofreading branch is operating with an extra GTP hydrolysis of 10%, and the overall reduction in the rate factor (k cat/Km) for peptide bond formation, in relation to the association rate constant for ternary complex binding to the A-site, would be about 20%.
There are two ways of viewing the virtues of proofreading schemes. First, for a given D value, the attainable accuracy is higher than in a single-step selection. Second, for a given accuracy, the kinetic efficiency (k cat/Km) of cognate ternary complex selection can be significantly greater than with a single-step selection. In other words, the maximum effective elongation rate can, for a given D value, be significantly larger in proofreading schemes than in a single-step selection (Fig. 6). The more general optimization for n-step processes is described by Ehrenberg and Blomberg (66).
So far we have discussed the functions of the GTP cycle in terms of kinetic efficiency and the accuracy enhancement of proofreading mechanisms. That there may be other functions for GTP in the EF-Tu cycle is suggested by recent work. For example, Ehrenberg et al. (68) reported, first, that two GTPs are hydrolyzed in EF-Tu function for every peptide bond that is formed during poly(Phe) synthesis on poly(U)-programmed ribosomes. Such results contradict the classical picture that one molecule of GTP is hydrolyzed for every EF-Tu-catalyzed binding of a cognate aa-tRNA to the A-site (94, 125, 127). Second, they also found that the stoichiometry for the complex between EF-Tu, GTP, and aa-tRNA is 2:2:1 instead of 1:1:1 as in the classical description (93, 234, 285). These data suggest that extended ternary complexes with 2:1 stoichiometry between EF-Tu and aa-tRNA may be the real substrates for mRNA-programmed ribosomes in vivo.
Evidence confirming that two GTPs are used in EF-Tu function per peptide bond has come from experiments with an EF-Tu that is structurally altered (Asp-138→Asn) and which can bind XTP but not GTP (277, 278). These experiments demonstrated that there are always two XTPs hydrolyzed per peptide bond in poly(U) translation (277, 278), while only one GTP is hydrolyzed on EF-G per translocation (216, 278). If this XTP-specific EF-Tu variant has the same functional properties as wild-type EF-Tu, apart from a switch in specificity from GTP to XTP, then the experiments by Weijland and Parmeggiani (278) lend strong, independent support to the notion that two GTPs are minimally required in EF-Tu function for every peptide bond (68, 231).
From these data, one may tentatively conclude that there are indeed two GTPs hydrolyzed per cognate peptide bond in full (68, 231, 277, 278) as well as in partial (20, 64, 254, 256, 257, 277, 278) poly(U) systems. Experiments with messengers other than poly(U) are now necessary to exclude the possibility that these results are due to the idiosyncrasies of poly(U)-dependent poly(Phe) synthesis. The reason behind such an unexpected stoichiometry for the ternary complex is presently unknown.
The interaction surface between ribosome and ternary complex could be extensive. This means that potentially the ternary complex-ribosome interaction could be stabilized by a very large binding energy. This reflection has led Nierhaus (186) to suggest that some sort of allosteric rearrangement of ribosomes is required to allow release of the ternary complex during tRNA selection. Neither the identification of this problem nor the allosteric solution is completely unmotivated. However, we also recognize that large interacting surfaces do not necessarily involve exceedingly large binding energies because it is also possible for the ribosome to develop a weak interaction with its ternary complexes by including a sufficient amount of repulsive forces at the interacting interface.
Nevertheless, Nierhaus (186) has suggested that a negative allosteric interaction between the aa-tRNA in the A-site and the deacylated tRNA in the E-site (89, 185, 186, 210, 211) provides the required precision for the selection of aa-tRNA rather than proofreading (185). In addition, Nierhaus (185) argues against proofreading because "not one mechanism has been proposed up to now that satisfactorily explains a repeated melting and rejoining of codon-anticodon interactions." This criticism is based on a misunderstanding of proofreading: no such "melting and rejoining" has been postulated for proofreading, and it was never suggested to be part of the mechanism (66, 109, 136, 140, 142, 188, 223, 257). Much more could be said about the arguments for the allosteric model, but the essential fact is that it provides no alternative explanation to the excess hydrolysis of GTP observed for noncognate peptide bond formation that has been verified in great detail in several laboratories. However, some sort of ribosomal allostery is compatible with a proofreading mechanism. The only question is whether there is any experimental evidence for a role of allostery in tRNA selection.
The experimental base for the allosteric effects on accuracy does not seem compelling. According to the model, when a cognate or near-cognate ternary complex is allowed to develop anticodon interactions with the mRNA codon, the E-site-bound tRNA dissociates and the affinity of the A-site-bound ternary complex becomes very high (186). However, the supporting measurement to identify the stable complex in the A-site (229) were made in experiments performed at high concentrations of Mg2+. It is well known that the accuracy of tRNA selection under such conditions is drastically reduced (e.g., reference 222). In addition, there were no error measurements performed in these binding stability experiments (229), which greatly weakens the support that these measurements offer for the hypothesis (185, 186). The point is that other data (118) suggest that the error level is extremely high under the conditions used by Schilling-Bartezko et al. (229), while the allosteric model requires that the error level be very low.
Nierhaus is addressing a real problem in his efforts to rationalize the E-site. This problem has been introduced in our discussion of the opposition of the accuracy of tRNA selection and the processivity of translation. In order for initial selection and proofreading to operate with high accuracy, the ternary complex as well as the aa-tRNA must have relatively weak interactions with the ribosome. At the same time, peptidyl-tRNA should be stably bound to the A-site as well as the P-site to minimize the probability of drop-off. Accordingly, there might be a switch in affinity for the tRNA-ribosome interaction, after proofreading and before or simultaneously with peptide bond formation. One solution to this dilemma may be found in the hybrid state model (176, 189). According to this model, aa-tRNA interacts only with the small ribosomal subunit during both initial selection of the ternary complex and proofreading of aa-tRNA. After the proofreading is over, the CCA part of aa-tRNA binds to 23S rRNA in the large subunit, and this additional interaction gives the required stabilization for the binding of the peptidyl-tRNA (189). We have estimated that a cognate ternary complex dissociates from the the A-site, prior to GTP hydrolysis, with a rate of about 10 s–1 (see above). Under similar experimental conditions, a dipeptidyl-tRNA in the putative A/P state with the postulated 23S rRNA stabilization (189) dissociates with a rate of 8 × 10–3 s–1, i.e., more than 1,000-fold more slowly (126). Thus, kinetic experiments carried out under conditions supporting accurate translation suggest that the stability of the interaction between tRNA and ribosome changes in accordance with the hybrid state model.
Streptomycin causes pronounced misreading of the genetic code, and a host of ribosome mutants can be selected as variants that tolerate or are dependent on the antibiotic (see the earlier discussion for references). The effects of streptomycin on accuracy could in principle arise because the drug stabilizes aa-tRNA in the A-site, thereby reducing the discard parameter a in scheme 2 above. Streptomycin may also reduce the discrimination ratio D in scheme 2. Recently Karimi and Ehrenberg (126) demonstrated that streptomycin stabilizes the binding of a cognate dipeptidyl-tRNA (Phe) about threefold in the A/P state of wild-type ribosomes (Fig. 1). At the same time a noncognate dipeptidyl-tRNA () was stabilized more than 100-fold (R. Karimi and M. Ehrenberg, unpublished data). This finding indicates that streptomycin reduces moderately the discard parameter a but that the major effect of the drug is to reduce the discrimination ratio D. This may reflect a distortion of the codon-anticodon structure away from a configuration that "maximizes the energy difference between cognate and noncognate codon-anticodon interactions" as postulated by Kurland et al. (143) for the unperturbed, wild-type ribosome.
The experiments of Karimi and Ehrenberg (126; unpublished data) lead to the expectation that streptomycin should affect both the initial selection and proofreading steps of cognate tRNA selection by perturbing the interaction between the anticodon loop of the tRNA with the decoding region of the 30S subunit. In support of this conclusion, Bilgin and Ehrenberg (21) found an approximately 30-fold reduction in accuracy in both selection steps. We note that in the original experiment by Ruusala and Kurland (224), the drug had a larger effect on proofreading than on initial selection, which at the time led to an emphasis on its effects on proofreading rather than on initial selection (139, 142). The cause of this discrepancy is not known, but genetic differences between the strains from which the studied ribosomes were obtained may provide part of an explanation. Here it is worth stressing that the drug influences both selection steps. This means, for one thing, that streptomycin cannot be used to map a proofreading center on rRNA, as suggested by Noller (189).
Hyperaccurate Smr, Smp, and Smd ribosome variants were previously shown to have reduced kinetic efficiency for cognate tRNA selection (20, 21, 29, 140, 142, 222, 293). This loss in kinetic efficiency for translation, associated with reduced missense and nonsense errors, is in line with Ninio’s kinetic analysis of nonsense suppression data (140, 142, 187). All of these findings are consistent with the notion that the increased accuracy of these ribosome variants is caused by enhanced values of the discard parameter a, which reduces the association rate of cognate aa-tRNAs to the A-site. Intriguingly, in the early experiments it was found that the reduction in kinetic efficiency always seemed to correlate only with more aggressive proofreading of cognate aa-tRNAs, not with their initial selection (140, 142, 222). Another surprising finding was that the enhanced missense errors of three Ram mutants, all with ribosomal protein S4 changes, were also shown to be due to decreased proofreading accuracy, while their nonproofreading (initial) selectivity was unchanged (11). It therefore seemed that the effects of the S12 changes that generate Smr, Smp, and Smd phenotypes and, finally, the S4 changes that generate the Ram phenotype all had something to do with proofreading but that they had little if anything to do with the initial selection steps of translation (11, 140, 142, 222).
More recent, fast kinetic experiments in a partial translation system (20, 21, 61) indicate that the reduced efficiency of cognate substrate processing comes from the initial selection step and that it is not caused by aggressive proofreading. Thus, the number of EF-Tu-associated GTPs hydrolyzed per peptide bond is now found to be close to two for all ribosome types, which is the wild-type stoichiometry. This means that the reduced kinetic efficiency for cognate reading of hyperaccurate ribosomes with alterations in S12 is not due to aggressive proofreading of cognate tRNAs, as originally thought (13, 29, 222). Instead, these new experiments identify the loss of kinetic efficiency with initial selection specifically with a lowered rate of GTP hydrolysis on the ternary complex, possibly in combination with an increased dissociation rate of the ternary complex from the ribosomal A-site (20).
A new phenotype associated with these hyperaccurate ribosomes may explain the discrepancy between partial translation and full translation with EF-G present. Bilgin et al. (20) observed that the hyperaccurate ribosomes have a strong, ternary complex-dependent GTPase activity. This so-called idling reaction continues even when all active ribosomes have completed dipeptide synthesis. It is strongest for Smd ribosomes and significantly stronger than the wild-type reaction for Smp ribosomes. Addition of streptomycin to the Smd and Smp ribosomes reduces the idling to the low wild-type level. We suggest that this idling reaction may interfere with the estimation of GTP stoichiometry of Smr, Smp, and Smd ribosomes under certain conditions and that this may account for the apparent discrepancy between results in full and partial translation systems.
Accuracy is tuned also by RNA. A base change from G to C at position 2661 of 23S rRNA (252) leads to an almost threefold enhancement in the accuracy of tRNA selection (21). This accuracy enhancement is associated with a 30% reduction in k cat/Km for cognate ternary complex interaction with the A-site. The accuracy enhancement and the reduction in cognate code reading efficiency are both due to a change in the initial selection step and not in the proofreading branch, in contrast to what was guessed by Melançon et al. (163). This particular change in rRNA gives a phenotype which is quite close to that of the S12 Smp variant, at least in the absence of streptomycin. A combination of Smp and C-2661C leads to a new Smd phenotype (21).
In summary, the most recent data suggest that the influence of streptomycin on accuracy is distributed between the initial and the proofreading selection of tRNA at the ribosomal A-site. In contrast, mutations that alter the kinetic response of ribosomes to the antibiotic and cause hyperaccuracy specifically reduce the initial selection efficiency of cognate tRNAs. One consequence of the revision is that for these ribosome variants the proofreading costs for the accuracy enhancement have been reduced. Thus, these costs no longer include excessive GTP hydrolysis by aggressive proofreading of correctly matched tRNA species, which previously was thought to be a major cost (142). It now seems that the loss of kinetic efficiency in the initial selection of cognate ternary complexes is the principal cost of accuracy enhancement in this scheme.
The theoretical papers of Hopfield (109) and Ninio (188) introduced proofreading, and they generated a compelling problem at the same time that they solved another one. In effect, the notion of proofreading transformed what was initially thought to be a physical problem into a biological one. The general problem solved concerned the origin of the accuracy of biosynthetic reactions such as translation. A salient feature of proofreading schemes is that they can in principle support unlimited accuracy. In particular, as long as the displacements from equilibrium of nonselected substrates such as ATP or GTP are adequate, missense rates can be driven to exceedingly low levels. Displacements from equilibrium for nucleoside triphosphates are greater than 1010 (136). This being so, why are the missense error rates of translation as high as 10–4 to 10–3?
Much of this chapter has been devoted to providing a background to this second problem, though we have put off stating it until the end. In this chapter, we have systematically enumerated and described the different constraints that set upper bounds on the biological selection of the missense error rates under conditions of growth rate competition. These include the loss of kinetic efficiency of translation as well as the loss of processivity associated with augmented accuracy of tRNA selection. The device that we have used to relate what we call the costs of accuracy and the costs of inaccuracy is the optimization of effective translation. With the aid of this device we have been able to illustrate how different constraints shape the biological optimization of translational accuracy.
Nevertheless, we take seriously data suggesting that bacteria in nature have very different constraints on their translation system. We would be quite remiss if we did not conclude in all candor that the processes that select the ribosome phenotypes in nature are a complete mystery at present.
We thank Karin Olson for her expert help and generosity with overtime. Our work is supported by the Swedish Cancer Society, Natural Sciences Research Council, and Technical Sciences Research Council.
References
1. Abad, J. P., and R. Amils. 1994. Location of the streptomycin ribosomal binding site explains its pleiotrophic effects on protein biosynthesis. J. Mol. Biol. 235:1251–1260.
2. Abdulkarim, F., L. Liljas, and D. Hughes. 1994. Mutations to kirromycin resistance occur in the interface of domains I and III of EF-Tu·GTP. FEBS Lett. 352:118–122.
3. Abdulkarim, F., T. M. F. Tuohy, R. H. Buckingham, and D. Hughes. 1991. Missense substitutions lethal to essential functions of EF-Tu. Biochimie 73:1457–1464.
4. Adhya, S., and M. Gottesman. 1978. Control of transcription termination. Annu. Rev. Biochem. 47:967–996.
4a. Ævarsson, A., E. Braznikov, M. Garber, J. Zheltonosova, Y. Chirgadze, S. Al-Karadaghi, L. A. Svensson, and A. Liljas. 1994. Three-dimensional structure of the ribosomal translocase: elongation factor G from Thermus thermophilus. EMBO J. 13:3669–3677.
5. Allen, P. N., and H. F. Noller. 1991. A single base substitution in 16S ribosomal RNA suppresses streptomycin dependence and increases the frequency of translational errors. Cell 66:141–148.
6. Anborgh, P. H., G. W. M. Swart, and A. Parmeggiani. 1991. Kirromycin-induced modifications facilitate the separation of EF-Tu species and reveal intermolecular interactions. FEBS Lett. 292:232–236.
7. Anderson, C., and O. Wiborg. 1994. Escherichia coli elongation-factor-Tu mutants with decreased affinity for aminoacyl-tRNA. Eur. J. Biochem. 220:739–744.
8. Anderson, W. F., L. Gorini, and L. Breckenridge. 1965. Ribosomes in streptomycin-activated suppression. Proc. Natl. Acad. Sci. USA 54:1076–1083.
9. Andersson, D. I., S. G. E. Andersson, and C. G. Kurland. 1986. Functional interactions between mutated forms of ribosomal proteins S4, S5 and S12. Biochimie 68:705–713.
10. Andersson, D. I., K. Bohman, L. A. Isaksson, and C.G. Kurland. 1982. Translation rates and misreading characteristics of rpsD mutants in Escherichia coli. Mol. Gen. Genet. 187:467–472.
11. Andersson, D. I., and C. G. Kurland. 1983. Ram ribosomes are defective proofreaders. Mol. Gen. Genet. 191:378–381.
12. Andersson, D. I., H. W. van Verseveld, A. H. Stouthamer, and C. G. Kurland. 1986. Suboptimal growth with hyper-accurate ribosomes. Arch. Microbiol. 144:96–101.
13. Andersson, S. G. E., R. H. Buckingham, and C. G. Kurland. 1984. Does codon composition influence ribosome function? EMBO J. 3:91–94.
14. Atkins, J. F., D. Elseviers, and L. Gorini. 1972. Low activity of β-galactosidase in frameshift mutants of Escherichia coli. Proc. Natl. Acad. Sci. USA 69:1192–1195.
15. Atkins, J. F., R. F. Gesteland, B. R. Reid, and C. W. Anderson. 1979. Normal tRNA’s promote ribosomal frameshifting. Cell 18:1119–1131.
16. Atkins, J. F., R. B. Weiss, and R. F. Gesteland. 1990. Ribosome gymnastics—degree of difficulty 9.5, style 10.0. Cell 62:413–423.
17. Atkins, J. F., R. B. Weiss, S. Thompson, and R. F. Gesteland. 1991. Towards a genetic dissection of the basis of triplet decoding, and its natural subversion: programmed reading frame shifts and hops. Annu. Rev. Genet. 25:201–228.
18. Benzer, S., and S. P. Champe. 1962. Change from nonsense to sense in the genetic code. Proc. Natl. Acad. Sci. USA 48:1114–1121.
19. Berchtold, H., L. Reshetnikova, C. O. A. Reiser, N. K. Schirmer, M. Sprinzl, and R. Hilgenfeld. 1993. Crystal structure of active elongation factor Tu reveals major domain rearrangements. Nature (London) 365:126–132.
20. Bilgin, N., F. Claesens, H. Pahverk, and M. Ehrenberg. 1992. Kinetic properties of E. coli ribosomes with altered forms of S12. J. Mol. Biol. 224:1011–1027.
21. Bilgin, N., and M. Ehrenberg. 1994. Mutations in 23S ribosomal RNA perturb transfer RNA selection and can lead to streptomycin dependence. J. Mol. Biol. 235:813–824.
22. Bilgin, N., L. A. Kirsebom, M. Ehrenberg, and C. G. Kurland. 1988. Mutations in ribosomal proteins L7/L12 perturb EF-G and EF-Tu functions. Biochimie 70:611–618.
23. Birge, E. A., and C. G. Kurland. 1969. Altered ribosomal protein in streptomycin-dependent E. coli. Science 166:1282–1284.
24. Birge, E. A., and C.G. Kurland. 1970. Reversion of a streptomycin-dependent strain of Escherichia coli. Mol. Gen. Genet. 109:356–369.
25. Biswas, D. K., and L. Gorini. 1972. The attachment site of streptomycin to the 30S ribosomal subunit. Proc. Natl. Acad. Sci. USA 69:2141–2144.
26. Bjare, U., and L. Gorini. 1971. Drug dependence reversed by a ribosomal ambiguity, ram, in Escherichia coli. J. Mol. Biol. 57:423–435.
27. Bodley, J. W., L. Lin, and J. H. Highland. 1970. Studies on translocation. VI. Thiostrepton prevents the formation of a complex of a ribosome-G factor-guanine nucleotide complex. Biochem. Biophys. Res. Commun. 41:1406–1411.
28. Bogdanov, A. A., O. A. Dontsova, and R. Brimacombe. 1992. Messenger RNA path through the procaryotic ribosome p. 421–431. In K. H. Nierhaus, F. Franceschi, A. R. Subramanian, V. A. Erdmann, and B. Wittmann-Liebold (ed.), The Translational Apparatus. Plenum Publishing Corp., New York.
29. Bohman, K. T., T. Ruusala, P. C. Jelenc, and C. G. Kurland. 1984. Kinetic impairment of restrictive streptomycin-resistant ribosomes. Mol. Gen. Genet. 198:90–99.
30. Bollen, A., T. Cabezón, M. de Wilde, R. Villarroel, and A. Herzog. 1975. Alteration of ribosomal protein S17 by mutation linked to neamine resistance in Escherichia coli. J. Mol. Biol. 99:795–806.
31. Bollen, A., J. Davies, M. Ozaki, and S. Mizushima. 1969. Ribosomal protein conferring sensitivity to the antibiotic spectinomycin in Escherichia coli. Science 165:85–86.
32. Bouadloun, F., D. Donner, and C. G. Kurland. 1983. Codon-specific missense errors in vivo. EMBO J. 2:1351–1356.
33. Bowie, J. U., J. F. Reidhaar-Olson, W. A. Lim, and R. T. Sauer. 1990. Deciphering the message in protein sequences: tolerance to amino acid substitutions. Science 247:1306–1310.
34. Breckenridge, L., and L. Gorini. 1970. Genetic analysis of streptomycin resistance in Escherichia coli. Genetics 65:9–25.
35. Bretscher, M. S. 1968. Translocation in protein synthesis: a hybrid structure model. Nature (London) 218:675–677.
36. Bretscher, M. S., and M. Grunberg-Manago. 1962. Polyribonucleotide-directed protein synthesis using Escherichia coli cell-free system. Nature (London) 195:283–284.
37. Brimacombe, R. 1991. RNA-protein interactions in the Escherichia coli ribosome. Biochimie 73:927–936.
38. Brownstein, B. L., and L. J. Lewandowski. 1967. A mutation suppressing streptomycin dependence. I. An effect on ribosome function. J. Mol. Biol. 25:99–109.
39. Bruce, A. G., J. F. Atkins, and R. F. Gesteland. 1986. tRNA anticodon replacement experiments show that ribosomal frameshifting can be caused by doublet decoding. Proc. Natl. Acad. Sci. USA 83:5062–5066.
40. Bückel, P., A. Büchberger, A. Böck, and H. G. Wittmann. 1977. Alteration in ribosomal protein L6 in mutants of Escherichia coli resistant to gentamicin. Mol. Gen. Genet. 158:47–54.
41. Bückel, P., W. Lubitz, and A. Böck. 1971. Substrate specificity of a mutant alanyl-transfer ribonucleic acid synthetase of Escherichia coli. J. Bacteriol. 108:1008–1016.
42. Bückel, P., D. Ruffler, W. Piepersberg, and A. Böck. 1972. RNA overproducing revertants of an alanyl-tRNA synthetase mutant of Escherichia coli. Mol. Gen. Genet. 119:323–335.
43. Bukari, A. I., and D. Zipser. 1973. Mutants of Escherichia coli with a defect in the degradation of nonsense fragments. Nature (London) 243:238–241.
44. Cannistraro, V. J., M. N. Subbarao, and D. Kennell. 1986. Specific endonucleolytic cleavage sites for decay of Escherichia coli mRNA. J. Mol. Biol. 192:257–274.
45. Cannon, M., R. Krug, and W. Gilbert. 1963. The binding of S-RNA by Escherichia coli ribosomes. J. Mol. Biol. 7:360–378.
46. Caplan, A. B., and J. R. Menninger. 1979. Tests of the ribosomal editing hypothesis: amino acid starvation differentially enhances the dissociation of peptidyl-tRNA from the ribosome. J. Mol. Biol. 134:621–637.
47. Caplan, A. B., and J. R. Menninger. 1984. Dissociation of peptidyl-tRNA from ribosomes is perturbed by streptomycin and by strA mutations. Mol. Gen. Genet. 194:534–538.
48. Chernoff, Y. O., A. Vincent, and S. W. Liebman. 1994. Mutations in eukaryotic 18S ribosomal RNA affect translational fidelity and resistance to aminogylcoside antibiotics. EMBO J. 13:906–913.
49. Churchward, G., H. Bremer, and R. Young. 1982. Macro-molecular composition of bacteria. J. Theor. Biol. 94:651–670.
50. Churchward, G., E. Estiva, and H. Bremer. 1981. Growth rate dependent control of chromosome replication initiation in Escherichia coli. J. Bacteriol. 145:1232–1238.
51. Couturier, M., L. Desmet, and R. Thomas. 1964. High pleiotropy of streptomycin mutations in Escherichia coli. Biochem. Biophys. Res. Commun. 16:244–248.
52. Cox, E. C., J. R. White, and J. G. Flaks. 1964. Streptomycin action and the ribosome. Proc. Natl. Acad. Sci. USA 51:703–709.
53. Crick, F. H. C., L. Barnett, S. Brenner, and R. J. Watts-Tobin. 1961. General nature of the genetic code for proteins. Nature (London) 192:1227–1232.
54. Cundliffe, E. 1971. The mode of action of thiostrepton in vivo. Biochem. Biophys. Res. Commun. 44:912–917.
55. Czworkowski, J., J. Wang, T. A. Steitz, and P. B. Moore. 1994. The crystal structure of elongation factor G complexed with GDP, at 2.7 Å resolution. EMBO J. 13:3661–3668.
56. Davies, J. 1964. The ribosomes of streptomycin-sensitive and resistant strains of Escherichia coli. Proc. Natl. Acad. Sci. USA 51:659–664.
57. Davies, J., W. Gilbert, and L. Gorini. 1964. Streptomycin, suppression and the code. Proc. Natl. Acad. Sci. USA 51:883–890.
58. Davies, J., L. Gorini, and B. D. Davis. 1965. Misreading of ribonucleic acid codewords induced by aminoglycoside antibiotics. Mol. Pharmacol. 1:93–106.
59. Deuser, E., G. Stöffler, H. G. Wittmann, and G. Apirion. 1970. Altered S4 proteins in Escherichia coli mutants from streptomycin dependence to independence. Mol. Gen. Genet. 109:298–302.
60. De Wilde, M., and B. Wittmann-Liebold. 1973. Localization of the amino-acid exchange in protein S5 from an Escherichia coli mutant resistant to spectinomycin. Mol. Gen. Genet. 127:273–276.
61. Diaz, I., and M. Ehrenberg. 1991. ms2i6A deficiency enhances proofreading in translation. J. Mol. Biol. 222:1161–1171.
62. Dong, H., and C. G. Kurland. 1995. Ribosome mutants with altered accuracy translate with reduced processivity. J. Mol. Biol. 248:551–561.
63. Donner, D., and C. G. Kurland. 1972. Changes in the primary structure of a mutationally altered ribosomal protein S4 of Escherichia coli. Mol. Gen. Genet. 115:49–53.
64. Eccleston, J. F., D. B. Dix, and R. C. Thompson. 1985. The rate of cleavage of GTP on the binding of Phe-tRNA elongation factor Tu·GTP to poly(U)-programmed ribosomes of Escherichia coli. J. Biol. Chem. 260:16237–16241.
65. Edelmann, P., and J. Gallant. 1977. Mistranslation in E. coli. Cell 10:131–137.
66. Ehrenberg, M., and C. Blomberg. 1980. Thermodynamic constraints on kinetic proofreading in biosynthetic pathways. Biophys. J. 31:333–358.
67. Ehrenberg, M., and C. G. Kurland. 1984. Costs of accuracy determined by a maximal growth rate constraint. Q. Rev. Biophys. 17:45–82.
68. Ehrenberg, M., A.-M. Rojas, J. Weiser, and C. G. Kurland. 1990. How many EF-Tu molecules participate in aminoacyl-tRNA binding and peptide bond formation in Escherichia coli translation? J. Mol. Biol. 211:739–749.
69. Eisenberg, D., and D. Crothers. 1979. Physical Chemistry with Applications for the Life Sciences. Benjamin, Cummings Publishing Company, London.
70. Ellis, N., and J. Gallant. 1982. An estimate of the global error frequency in translation. Mol. Gen. Genet. 188:169–172.
71. Fayet, O., J.-M. Louarn, and C. Georgopoulos. 1986. Suppression of the Escherichia coli dnaA46 mutation by amplification of the groES and groEL genes. Mol. Gen. Genet. 202:435–445.
72. Fayet, O., T. Ziegelhoffer, and C. Georgopoulos. 1989. The groES and groEL heat shock genes of Escherichia coli are essential for bacterial growth at all temperatures. J. Bacteriol. 171:1379–1385.
73. Fersht, A. R., and L. Serrano. 1993. Principles of protein stability derived from protein engineering experiments. Curr. Opin. Struct. Biol. 3:75–83.
74. Fitzgerald, R. J., F. Bernheim, and D. B. Fitzgerald. 1948. The inhibition by streptomycin of adaptive enzyme formation in Mycobacteriae. J. Biol. Chem. 175:195–200.
75. Flaks, J. G., E. C. Cox, M. L. Witting, and J.R. White. 1962. Polypeptide synthesis with ribosomes from streptomycin-resistant and -dependent Escherichia coli. Biochem. Biophys. Res. Commun. 7:390–393.
76. Förster, Y. O., S. Limmer, S. Ribeiro, R. Hilgenfeld, and M. Sprinzl. 1993. Ternary complex between elongation factor Tu·GTP and Phe-tRNAPhe. Biochimie 75:1159–1166.
77. Friedman, D. I., M. J. Imperiale, and S. L. Adhya. 1987. RNA 3' end formation in the control of gene expression. Annu. Rev. Genet. 21:453–488.
78. Fromm, H., E. Galun, and M. Edelman. 1989. A novel site for streptomycin resistance in the 530 loop of chloroplast 16S ribosomal RNA. Plant Mol. Biol. 12:499–505.
79. Funatsu, G., W. Puls, E. Schiltz, J. Reinbot, and H. G. Wittmann. 1972. Ribosomal proteins. XXXI. Comparative studies on altered protein S4 of six Escherichia coli revertants from streptomycin dependence. Mol. Gen. Genet. 115:131–139.
80. Funatsu, G., E. Schiltz, and H. G. Wittmann. 1971. Ribosomal proteins. XXVII. Localization of amino acid exchanges in protein S5 from two Escherichia coli mutants resistant to spectinomycin. Mol. Gen. Genet. 114:106–111.
81. Funatsu, G., and H. G. Wittmann. 1972. Localisation of amino acid replacements in protein S12 isolated from E. coli mutants resistant to streptomycin. J. Mol. Biol. 68:547–550.
82. Gallant, J., and D. Foley. 1980. On the causes and prevention of mistranslation, p. 615–638. In G. Chambliss, G. R. Craven, J. Davies, K. Davis, L. Kahan, and M. Nomura (ed.), Ribosomes, Structure, Function, and Genetics. University Park Press, Baltimore.
83. Gallant, J. A., and D. Lindsley. 1992. Leftward frameshifting at a hungry codon. J. Mol. Biol. 223:31–40.
84. Garen, A., and O. Siddiqi. 1962. Suppression of mutations in the alkaline phosphotase structural cistron of Escherichia coli. Proc. Natl. Acad. Sci. USA 48:1121–1127.
85. Gauthier, A., M. Turmel, and C. Lemieux. 1988. Mapping of chloroplast mutations conferring resistance to antibiotics in Chlamydomonas: evidence for a novel site of streptomycin resistance in the small subunit rRNA. Mol. Gen. Genet. 214:192–197.
86. Gilbert, W. 1963. Polypeptide synthesis in Escherichia coli. II. The polypeptide chain and S-RNA. J. Mol. Biol. 6:389–403.
87. Girshovich, A. S., E. S. Bochkavera, and V. D. Vasiliev. 1986. Localization of elongation factor Tu on the ribosome. FEBS Lett. 197:192–198.
88. Girshovich, A. S., T. V. Kurtskhalia, A. Ovchinnikov Yu, and V. D. Vasiliev. 1981. Localization of the elongation factor G on Escherichia coli ribosomes. FEBS Lett. 130:54–59.
89. Gnirke, A., U. Geigenmüller, H.J. Rheinberger, and K. H. Nierhaus. 1989. The allosteric three-site model for the ribosomal elongation cycle. J. Biol. Chem. 264:7291–7301.
90. Goff, S. A., L. P. Casson, and A. L. Goldberg. 1984. Heat-shock regulatory gene htpR influences rates of protein degradation and expression of the lon gene in Escherichia coli. Proc. Natl. Acad. Sci. USA 81:6647–6651.
91. Goff, S. A., and A. L. Goldberg. 1985. Production of abnormal proteins in Escherichia coli stimulates transcription of lon and other heat-shock genes. Cell 41:587–595.
92. Goloubinoff, P., J. T. Christeller, A. A. Gatenby, and G. H. Lorimer. 1989. Reconstitution of active dimeric ribulose biphosphate carboxylase from an unfolded state depends on two chaperonin proteins and Mg-ATP. Nature (London) 342:884–889.
93. Gordon, J. 1967. Interaction of guanosine 5'-triphosphate with a supernatant fraction from E. coli and aminoacyl-SRNA. Proc. Natl. Acad. Sci. USA 58:1574–1578.
94. Gordon, J. 1969. Hydrolysis of guanosine 5'-triphosphate associated with binding of aminoacyl transfer ribonucleic acid to ribosomes. J. Biol. Chem. 244:5680–5686.
95. Göringer, H. U., K. A. Hijazi, E. J. Murgola, and A. E. Dahlberg. 1991. Mutations in 16S rRNA that affect UGA (stop codon)-directed translation termination. Proc. Natl. Acad. Sci. USA 88:6603–6607.
96. Gorini, L., W. B. Gundersen, and M. Burger. 1961. Genetics of regulation of enzyme synthesis in the arginine biosynthetic pathway of Escherichia coli. Cold Spring Harbor Symp. Quant. Biol. 26:173–182.
97. Gorini, L., G. A. Jacoby, and L. Breckenridge. 1966. Ribosomal ambiguity. Cold Spring Harbor Symp. Quant. Biol. 31:657–664.
98. Gorini, L., and E. Kataja. 1965. Suppression activated by streptomycin and related antibiotics in drug sensitive strains. Biochem. Biophys. Res. Commun. 18:656–663.
99. Gottesman, S., and D. Zipser. 1978. Deg phenotype of Escherichia coli lon mutants. J. Bacteriol. 133:844–851.
100. Gregory, S. T., K. R. Lieberman, and A. E. Dahlberg. 1994. Mutations in the peptidyl transferase region of E. coli 23S rRNA affecting translational accuracy. Nucleic Acids Res. 22:279–284.
101. Grosjean, J., S. D. Henaue, and D. M. Crothers. 1978. The physical basis for ambiguity in genetic coding interactions. Proc. Natl. Acad. Sci. USA 75:610–614.
102. Harris, E. H., B. D. Burkhart, N. W. Gillham, and J. E. Boynton. 1989. Antibiotic resistance mutations in the chloroplast 16S and 23S rRNA genes of Chlamydomonas reinhardtii: correlation of genetic and physical maps of the chloroplast genome. Genetics 123:281–292.
103. Hasenbank, R., C. Guthrie, G. Stöffler, H. G. Wittmann, L. Rosen, and D. Apirion. 1973. Electrophoretic and immunological studies on ribosomal proteins of 100 Escherichia coli revertants from streptomycin dependence. Mol. Gen. Genet. 127:1–18.
104. Hashimoto, K. 1960. Streptomycin resistance in Escherichia coli analysed by transduction. Genetics 45:49–62.
105. Hausner, T.-P., J. Atmadja, and K. N. Nierhaus. 1987. Evidence that the G2661 region of 23S rRNA is located at the ribosomal binding sites of both elongation factors. Biochimie 69:911–923.
106. Hecht, M. H., J. M. Sturtevant, and R. T. Sauer. 1984. Effect of single amino acid replacements on the thermal stability of the NH2-terminal domain of phage lambda repressor. Proc. Natl. Acad. Sci. USA 81:5685–5689.
107. Hellinga, H. W., R. Wynn, and F. M. Richards. 1992. The hydrophobic core of Escherichia coli thioredoxin shows a high tolerance to nonconservative single amino acid substitutions. Biochemistry 31:11203–11209.
108. Highland, J. H., L. Lin, and J. W. Bodley. 1971. Protection of ribosomes from thiostrepton inactivation by the binding of G factor and guanosine diphosphate. Biochemistry 10:4404–4409.
109. Hopfield, J. J. 1974. Kinetic proofreading: a new mechanism for reducing errors in biosynthetic processes requiring high specificity. Proc. Natl. Acad. Sci. USA 71:4135–4139.
110. Hou, Y., Y.P. Lin, J. D. Sharer, and P. E. March. 1994. In vivo selection of conditional-lethal mutations in the gene encoding elongation factor G of Escherichia coli. J. Bacteriol. 176:123–129.
111. Hughes, D. 1986. The isolation and mapping of EF-Tu mutations in Salmonella typhimurium. Mol. Gen. Genet. 202:108–111.
112. Hughes, D. 1987. Mutant forms of tufA and tufB independently suppress nonsense mutations. J. Mol. Biol. 197:611–615.
113. Hughes, D. 1990. Both genes for EF-Tu in Salmonella typhimurium are individually dispensable for growth. J. Mol. Biol. 215:41–51.
114. Hughes, D. 1991. Error-prone EF-Tu reduces in vivo enzyme activity and cellular growth rate. Mol. Microbiol. 5:623–630.
115. Hughes, D., J. F. Atkins, and S. Thompson. 1987. Mutants of elongation factor Tu promote ribosomal frameshifting and nonsense readthrough. EMBO J. 6:4235–4239.
116. Hughes, D., and I. Tubulekas. 1993. Ternary complex-ribosome interaction: its influence on protein synthesis and on growth rate. Biochem. Soc. Trans. 21:851–857.
117. Itoh, T., and H. G. Wittmann. 1973. Amino acid replacements in proteins S5 and S12 of two Escherichia coli revertants from streptomycin dependence to independence. Mol. Gen. Genet. 127:19–32.
118. Jelenc, P. C., and C. G. Kurland. 1979. Nucleoside triphosphate regeneration decreases the frequency of translation errors. Proc. Natl. Acad. Sci. USA 76:3174–3178.
119. Jenkins, A. J., J. B. Marsh, I. R. Oliver, and M. Master. 1986. A DNA fragment containing the groE genes can suppress mutations in the Escherichia coli dnaA gene. Mol. Gen. Genet. 202:446–454.
120. Johanson, U., and D. Hughes. 1994. Fusidic acid-resistant mutants define three regions in elongation factor G of Salmonella typhimurium. Gene 143:55–59.
121. Jonák, J., P. H. Anborgh, and A. Parmeggiani. 1994. Histidine-118 of elongation factor Tu: its role in aminoacyl-tRNA binding and regulation of the GTPase activity. FEBS Lett. 343:94–98.
122. Jørgensen, F., F. M. Adamski, W. P. Tate, and C. G. Kurland. 1993. Release factor-dependent false stops are infrequent in Escherichia coli. J. Mol. Biol. 230:41–50.
123. Jørgensen, F., and C. G. Kurland. 1990. Processivity errors of gene expression in Escherichia coli. J. Mol. Biol. 215:511–521.
124. Kaji, H., and A. Kaji. 1965. Specific binding of sRNA to ribosomes: effects of streptomycin. Proc. Natl. Acad. Sci. USA 54:213–219.
125. Kakniashvili, D. G., S. K. Smailov, and L. P. Gavrilova. 1986. The excess GTP hydrolysed during mistranslation is expended at the stage of EF-Tu-promoted binding of non-cognate aminoacyl-tRNA. FEBS Lett. 196:103–107.
126. Karimi, R., and M. Ehrenberg. 1994. Dissociation rate of cognate peptidyl-tRNA from the A-site of hyper-accurate and error-prone ribosomes. Eur. J. Biochem. 226:355–360.
127. Kaziro, Y. 1978. The role of guanosine 5'-triphosphate in polypeptide chain elongation. Biochim. Biophys. Acta 505:95–127.
128. King, T. C., and D. Schlessinger. 1987. Processing of RNA transcripts, p. 703–718. In F. C. Neidhardt, J. C. Ingraham, K. B. Low, B. Magasanik, M. Schaechter, and H. E. Umbarger (ed.), Escherichia coli and Salmonella typhimurium: Cellular and Molecular Biology. American Society for Microbiology, Washington, D.C.
129. Kinzy, T. G., J. P. Freeman, A. E. Johnson, and W. C. Merrick. 1992. A model for the aminoacyl-tRNA binding site of eukaryotic elongation factor 1α. J. Biol. Chem. 267:1623–1632.
130. Kirsebom, L. A., R. Amons, and L. A. Isaksson. 1986. Primary structure of mutationally altered ribosomal protein L7/L12 and their effects on cellular growth and translational accuracy. Eur. J. Biochem. 156:669–675.
131. Kirsebom, L. A., and L. A. Isaksson. 1985. Involvement of ribosomal protein L7/L12 in control of translation accuracy. Proc. Natl. Acad. Sci. USA 82:717–721.
132. Kjeldgaard, M., P. Nissen, S. Thirup, and J. Nyborg. 1993. The crystal structure of elongation factor EF-Tu from Thermus aquaticus in the GTP conformation. Structure 1:35–50.
133. Kjeldgaard, M., and J. Nyborg. 1992. Refined structure of elongation factor EF-Tu from Escherichia coli. J. Mol. Biol. 223:721–742.
134. Kreider, G., and B. L. Brownstein. 1972. Ribosomal proteins involved in the suppression of streptomycin dependence in Escherichia coli. J. Bacteriol. 109:780–785.
135. Kühberger, R., W. Piepersberg, A. Petzel, P. Buckel, and A. Böck. 1979. Alteration of ribosomal protein L6 in gentamicin-resistant strains of Escherichia coli. Effects on fidelity of protein synthesis. Biochemistry 18:187–193.
136. Kurland, C. G. 1978. The role of guanine nucleotides in protein biosynthesis. Biophys. J. 22:373–392.
137. Kurland, C. G. 1979. Reading frame errors on ribosomes, p. 98–108. In J. E. Celis and J.D. Smith (ed.), Nonsense Mutations and tRNA Suppressors. Academic Press, London.
138. Kurland, C. G. 1992. Translational accuracy and the fitness of bacteria. Annu. Rev. Genet. 26:29–50.
139. Kurland, C. G., D. I. Andersson, S. G. E. Andersson, K. Bohman, F. Bouadloun, M. Ehrenberg, P. C. Jelenc, and T. Ruusala. 1984. Translational accuracy and bacterial growth, p. 193–207. In B. F. C. Clark and H. U. Petersen (ed.), Gene Expression: the Translational Step and Its Control. Alfred Benzon Symposium. Munksgaard, Copenhagen.
140. Kurland, C. G., and M. Ehrenberg. 1984. Optimization of translation accuracy. Prog. Nucleic Acid Res. Mol. Biol. 31:191–219.
141. Kurland, C. G., and M. Ehrenberg. 1985. Constraints on the accuracy of messenger RNA movement. Q. Rev. Biophys. 18:423–450.
142. Kurland, C. G., and M. Ehrenberg. 1987. Growth-optimizing accuracy of gene expression. Annu. Rev. Biophys. Biophys. Chem. 16:291–317.
143. Kurland, C. G., R. Rigler, M. Ehrenberg, and C. Blomberg. 1975. Allosteric mechanism for codon-dependent tRNA selection on ribosomes. Proc. Natl. Acad. Sci. USA 72:4248–4251.
144. Langer, J. A.. and J. A. Lake. 1986. Elongation factor Tu localized on the exterior surface of the small ribosomal subunit. J. Mol. Biol. 187:617–621.
145. Langridge, J., and J. H. Campbell. 1969. Classification and intragenic position of mutations in the β-galactosidase gene of Escherichia coli. Mol. Gen. Genet. 103:339–347.
146. Lederberg, E. M., L. Cavalli-Sforza, and J. Lederberg. 1964. Interaction of streptomycin and a suppressor for galactose fermentation in E. coli K-12. Proc. Natl. Acad. Sci. USA 51:678–682.
147. Leijonmarck, M., S. Eriksson, and A. Liljas. 1980. Crystal structure of a ribosomal component at 2.6Å resolution. Nature (London) 286:824–826.
148. Likover, T. E., and C. G. Kurland. 1967. Ribosomes from a streptomycin-dependent strain of Escherichia coli. J. Mol. Biol. 25:497–504.
149. Liljas, A. 1990. Some structural aspects of elongation, p. 309–317. In W. E. Hill, A. Dahlberg, R. A. Garrett, P. B. Moore, D. Schlessinger, and J. R. Warner (ed.), The Ribosome: Structure, Function, and Evolution. American Society for Microbiology, Washington, D.C.
150. Lim, V., C. Venclovas, A. Spirin, R. Brimacombe, P. Mitchell, and F. Müller. 1992. How are tRNA’s and mRNA arranged in the ribosome? An attempt to correlate the stereochemistry of the tRNA-mRNA interaction with constraints imposed by the ribosomal topography. Nucleic Acids Res. 20:2627–2637.
151. Lindsley, D., and J.A. Gallant. 1993. On the directional specificity of ribosome frameshifting at a "hungry" codon. Proc. Natl. Acad. Sci. USA 90:5469–5473.
152. Loftfield, R., and D. Vanderjagt. 1972. The frequency of errors in protein synthesis. Biochem. J. 128:1353–1356.
153. Loftfield, R. B. 1963. The frequency of errors in protein biosynthesis. Biochem. J. 89:82–87.
154. Maaløe, O. 1979. Regulation of the protein-synthesizing machinery: ribosomes, tRNA, factors and so on, p. 487–542. In R. F. Goldberger (ed.), Biological Regulation and Development. Plenum Publishing Corp., New York.
155. Maaløe, O., and N. O. Kjeldgaard. 1966. Control of Macromolecular Synthesis: a Study of DNA, RNA and Protein Synthesis in Bacteria. W. A. Benjamin, Inc., New York.
156. Mager, J., M. Benedict, and M. Artman. 1962. Common site of action for polyamines and streptomycin. Biochim. Biophys. Acta 62:202–204.
157. Manley, J. L. 1978. Synthesis and degradation of termination and premature-termination fragments of β-galactosidase in vitro and in vivo. J. Mol. Biol. 125:407–432.
158. Marshall, D. R., and A. H. D. Brown. 1975. The charge-state model of protein polymorphisms in natural populations. J. Mol. Evol. 6:149–163.
159. Matthews, B. W. 1991. Mutational analysis of protein stability. Curr. Opin. Struct. Biol. 1:17–21.
160. Matthews, B. W. 1993. Structural and genetic analysis of protein folding and stability. Curr. Opin. Struct. Biol. 3:589–593.
161. McClain, W. H. 1993. Rules that govern tRNA identity in protein synthesis. J. Mol. Biol. 234:257–280.
162. Melançon, P., C. Lemieux, and L. Brakier-Gingras. 1988. A mutation in the 530 loop of Escherichia coli 16S ribosomal RNA causes resistance to streptomycin. Nucleic Acids Res. 16:9631–9639.
163. Melançon, P., W. E. Tapprich, and L. Brakier-Gingras. 1992. Single-base mutations at position 2661 of Escherichia coli 23S rRNA increase the efficiency of translational proofreading. J. Bacteriol. 174:7896–7901.
164. Menninger, J. R. 1976. Peptidyl transfer RNA dissociates during protein synthesis from ribosomes of Escherichia coli. J. Biol. Chem. 251:3392–3398.
165. Menninger, J. R. 1977. Ribosome editing and the error catastrophe hypothesis of cellular ageing. Mech. Ageing Dev. 6:131–142.
166. Menninger, J. R. 1978. The accumulation as peptidyl-transfer RNA of isoaccepting transfer RNA families in Escherichia coli with temperature-sensitive peptidyl-transfer RNA hydrolase. J. Biol. Chem. 253:6808–6813.
167. Menninger, J. R., A. B. Caplan, P. K. E. Gingrich, and A. G. Atherly. 1983. Tests of the ribosome editor hypothesis. II. Relaxed (relA) and stringent (relA +) E. coli differ in rates of dissociation of peptidyl-tRNA from ribosomes. Mol. Gen. Genet. 190:215–221.
168. Metz-Boutigue, M.-H., J. Reibolt, J.-P. Ebel, C. Ehresmann, and B. Ehresmann. 1989. Crosslinking of elongation factor Tu to tRNAPhe by trans-diamminedichloroplatinum(II). FEBS Lett. 245:194–200.
169. Mikkola, R., and C. G. Kurland. 1991. Is there a unique ribosome phenotype for naturally occurring Escherichia coli? Biochimie 73:1061–1066.
170. Mikkola, R., and C. G. Kurland. 1992. Selection of laboratory wild type phenotype from natural isolates of E. coli in chemostats. Mol. Biol. Evol. 9:394–402.
171. Mikkola, R., and C. G. Kurland. Slowly growing bacteria are selected in nature. J. Mol. Ecol., in press.
172. Miller, C. P., and M. Bohnhoff. 1947. Two streptomycin-resistant variants of Meningococcus. J. Bacteriol. 54:467–481.
173. Miller, D. L., and H. Weissbach. 1977. Factors involved in the transfer of aminoacyl-tRNA to the ribosome, p. 323–374. In H. Weissbach and S. Pestka (ed.), Molecular Mechanisms of Protein Biosynthesis. Academic Press, New York.
174. Miller, J. H., C. Coulondre, M. Hofer, U. Schmeissner, H. Sommer, A. Schmitz, and P. Lu. 1979. Genetic studies of the lac repressor. J. Mol. Biol. 131:191–222.
175. Moazed, D., and H. F. Noller. 1989. Interaction of tRNA with 23S rRNA in the ribosomal A, P, and E sites. Cell 57:585–597.
176. Moazed, D., and H. F. Noller. 1989. Intermediate states in the movement of transfer RNA in the ribosome. Nature (London) 342:142–148.
177. Moazed, D., J. M. Robertson, and H. F. Noller. 1988. Interaction of elongation factors EF-G and EF-Tu with a conserved loop in 23S RNA. Nature (London) 334:362–364.
178. Modolell, J., B. Cabrer, A. Parmeggiani, and D. Vazquez. 1971. Inhibition of siomycin and thiostrepton of both aminoacyl-tRNA and factor G binding to ribosomes. Proc. Natl. Acad. Sci. USA 68:1796–1800.
179. Moine, H., and A. E. Dahlberg. 1994. Mutations in helix 34 of Escherichia coli 16S ribosomal RNA have multiple effects on ribosome function and synthesis. J. Mol. Biol. 243:402–412.
180. Montandon, P. E., R. Wagner, and E. Stutz. 1986. E. coli ribosomes with a C912 to U base change in the 16S rRNA are streptomycin resistent. EMBO J. 5:3705–3708.
181. Murgola, E. J., K. A. Hijazi, H. U. Göringer, and A. E. Dahlberg. 1988. Mutant 16S ribosomal RNA: a codon-specific translational suppressor. Proc. Natl. Acad. Sci. USA 85:4162–4165.
182. Nei, M. 1975. Molecular Population Genetics and Evolution. Elsevier North-Holland, Amsterdam,
183. Nei, M., and R. Chakraborty. 1973. Genetic distance and electrophoretic identity of proteins between taxa. J. Mol. Evol. 2:323–328.
184. Newcombe, H. B., and M. H. Nyholm. 1950. The inheritance of streptomycin resistance and dependence in crosses of Escherichia coli. Genetics 35:603–611.
185. Nierhaus, K. H. 1990. The allosteric three-site model for the ribosomal elongation cycle: features and future. Biochemistry 29:4997–5008.
186. Nierhaus, K. H. 1993. Solution of the ribosome riddle: how the ribosome selects the correct aminoacyl-tRNA out of 41 similar contestants. Mol. Microbiol. 9:661–669.
187. Ninio, J. 1974. A semiquantitative treatment of missense and nonsense suppression in the strA and ram ribosomal mutants of Escherichia coli. Evaluation of some molecular parameters of translation in vitro. J. Mol. Biol. 84:297–313.
188. Ninio, J. 1975. Kinetic amplification of enzyme discrimination. Biochimie 57:587–595.
189. Noller, H. F. 1991. Ribosomal RNA and translation. Annu. Rev. Biochem. 60:191–227.
190. Noller, H. F., D. Moazed, S. Stern, T. Powers, P. N. Allen, J. M. Robertson, B. Weiser, and K. Triman. 1990. Structure of rRNA and its functional interactions in translation, p. 73–92. In W. E. Hill, A. Dahlberg, R. A. Garrett, P. B. Moore, D. Schlessinger, and J. R. Warner (ed.), The Ribosome: Structure, Function, and Evolution. American Society for Microbiology, Washington, D.C.
191. Ochman, H., and R. K. Selander. 1984. Standard reference strains of Escherichia coli from natural populations. J. Bacteriol. 157:690–693.
192. O’Connor, M., and A. Dahlberg. 1993. Mutations at U2555, a tRNA-protected base in 23S rRNA, affect translational fidelity. Proc. Natl. Acad. Sci. USA 90:9214–9218.
193. O’Connor, M., H. U. Göringer, and A. E. Dahlberg. 1992. A ribosomal ambiguity mutation in the 530 loop of E. coli 16S rRNA. Nucleic Acids Res. 20:4221–4227.
194. Orias, E., and T. K. Gartner. 1966. Suppression of amber and ochre rII mutants of bacteriophage T4 by streptomycin. J. Bacteriol. 91:2210–2215.
195. Ozaki, M., S. Mizushima, and M. Nomura. 1969. Identification and functional characterization of the protein controlled by the streptomycin resistant locus in E. coli. Nature (London) 222:333–339.
196. Pakula, A. A., V. B. Young, and R. T. Sauer. 1986. Bacteriophage lambda cro mutations: effects on activity and intracellular degradation. Proc. Natl. Acad. Sci. USA 83:8829–8833.
197. Parker, J. 1989. Errors and alternatives in reading the universal genetic code. Microbiol. Rev. 53:273–298.
198. Parker, J. 1992. Variations in reading the genetic code, p. 191–267. In D. L. Hatfield, B. J. Lee, and R. M. Pirtle (ed.), Transfer RNA in Protein Synthesis. CRC Press, London.
199. Parker, J., and G. Holtz. 1984. Control of basal-level codon misreading in Escherichia coli. Biochem. Biophys. Res. Commun. 121:487–492.
200. Parker, J., T. C. Johnston, P. T. Borgia, G. Holtz, E. Remaut, and W. Fiers. 1983. Codon usage and mistranslation: in vivo basal level misreading of the MS2 coat protein message. J. Biol. Chem. 258:10007–10012.
201. Pestka, S., R. Marshall, and M. Nirenberg. 1965. Ribonucleic acid codewords and protein synthesis. V. Effects of streptomycin on the formation of ribosome-sRNA complexes. Proc. Natl. Acad. Sci. USA 53:639–646.
202. Peter, K., D. Lindsley, L. Peng, and J. A. Gallant. 1992. Context rules of rightward overlapping reading. New Biol. 4:520–526.
203. Phillips, T. A., R. A. VanBogelen, and F. C. Neidhardt. 1984. lon gene product of Escherichia coli is a heat-shock protein. J. Bacteriol. 159:283–287.
204. Piepersberg, W., A. Böck, and H. G. Wittmann. 1975. Effect of different mutations in ribosomal protein S5 of Escherichia coli on translational fidelity. Mol. Gen. Genet. 140:91–100.
205. Pinard, R., M. Côté, C. Payard, and L. Brakier-Gingras. 1994. Positions 13 and 914 in Escherichia coli 16S ribosomal RNA are involved in the control of translational accuracy. Nucleic Acids Res. 22:619–624.
206. Poteete, A. R., D. Rennell, and S. E. Bouvier. 1992. Functional significance of conserved amino acid residues. Proteins 13:38–40.
207. Powers, T., and H. Noller. 1991. A functional pseudoknot in 16S ribosomal RNA. EMBO J. 10:2203–2214.
208. Ramakrishnan, V., and S. W. White. 1992. The structure of ribosomal protein S5 reveals sites of interaction with 16S rRNA. Nature (London) 358:768–771.
209. Rennell, D., S. E. Bouvier, L. W. Hardy, and A. R. Poteete. 1991. Systematic mutation of bacteriophage T4 lysozyme. J. Mol. Biol. 222:67–88.
210. Rheinberger, H. J. 1986. Allosteric interactions between the transfer-RNA binding sites A and E. J. Biol. Chem. 261:9133–9139.
211. Rheinberger, H. J. 1991. The function of the translating ribosome. Biochimie 73:1067–1088.
212. Rheinberger, H. J., and K.H. Nierhaus. 1983. Testing an alternative model of the ribosomal peptide elongation cycle. Proc. Natl. Acad. Sci. USA 80:4213–4217.
213. Rheinberger, H. J., H. Sternbach, and K. H. Nierhaus. 1981. Three tRNA binding sites on Escherichia coli ribosomes. Proc. Natl. Acad. Sci. USA 78:5310–5314.
214. Richman, N., and J. W. Bodley. 1972. Ribosomes cannot interact simultaneously with elongation factors EF Tu and EF G. Proc. Natl. Acad. Sci. USA 69:686–689.
215. Richter Dahlfors, A. A., and C. G. Kurland. 1990. Novel mutants of elongation factor G. J. Mol. Biol. 215:549–557.
216. Richter Dahlfors, A. A., and C. G. Kurland. 1990. Stoichiometry of elongation factor G function in translation. J. Mol. Biol. 216:311–314.
217. Riddle, D. L., and J. Carbon. 1973. Frameshift suppression: a nucleotide addition in the anticodon of a glycine transfer RNA. Nature (London) New Biol. 242:230–234.
218. Riddle, D. L., and J. R. Roth. 1970. Suppressors of frameshift mutations in Salmonella typhimurium. J. Mol. Biol. 54:131–134.
219. Rossett, R., and L. Gorini. 1969. Ribosomal ambiguity mutation. J. Mol. Biol. 39:95–112.
220. Roth, J. R. 1974. Frameshift suppression. Annu. Rev. Genet. 8:319–346.
221. Roth, J. R. 1981. Frameshift suppression. Cell 24:601–602.
222. Ruusala, T., D. I. Andersson, M. Ehrenberg, and C. G. Kurland. 1984. Hyper-accurate ribosomes inhibit growth. EMBO J. 3:2575–2580.
223. Ruusala, T., M. Ehrenberg, and C. G. Kurland. 1982. Is there proofreading during polypeptide synthesis? EMBO J. 1:741–745.
224. Ruusala, T., and C. G. Kurland. 1984. Streptomycin preferentially perturbs ribosomal proofreading. Mol. Gen. Genet. 198:100–104.
225. Rychlik, W., O. W. Odom, and B. Hardesty. 1983. Localization of elongation factor Tu binding site on Escherichia coli ribosomes. Biochemistry 22:85–93.
226. Rydén Aulin, M., and D. Hughes. 1990. Overproduction of release factor reduces spontaneous frameshifting and frameshift suppression by mutant elongation factor Tu. J. Bacteriol. 172:6721–6726.
227. Saarma, U., and J. Remme. 1992. Novel mutants of 23S RNA: characterization and functional properties. Nucleic Acids Res. 20:3147–3152.
228. Sauer, R. T., and W. A. Lim. 1992. Mutational analysis of protein stability. Curr. Opin. Struct. Biol. 2:46–51.
229. Schilling-Bartezko, S., F. Franceschi, H. Sternbach, and K. H. Nierhaus. 1992. Apparent association constants of tRNA’s for the ribosomal A, P, and E-sites. J. Biol. Chem. 267:4693–4702.
230. Schnier, J., S. Isono, A. G. Cumberlidge, and K. Isono. 1985. Unstable mutations caused by regional tandem multiplications in the gene for ribosomal protein S4 show thermosensitivity in Escherichia coli. Mol. Gen. Genet. 199:265–270.
231. Scoble, J., N. Bilgin, and M. Ehrenberg. 1994. Two GTP’s are hydrolyzed on two molecules of EF-Tu for each elongation cycle during code translation. Biochimie 76:59–62.
232. Shen, Z., and T. D. Fox. 1989. Substitution of an invariant nucleotide at the base of the highly conserved ‘530-loop’ of 15S rRNA causes suppression of yeast mitochondrial ochre mutations. Nucleic Acids Res. 17:4535–4539.
233. Shineberg, J. B., and D. Zipser. 1973. The lon gene and degradation of β-galactosidase nonsense fragments. J. Bacteriol. 116:1469–1471.
234. Shorey, R. L., J. M. Ravel, C. W. Garner, and W. Shive. 1969. Formation and properties of the aminoacyl transfer ribonucleic acid-guanosine triphosphate-protein complex. J. Biol. Chem. 244:4555–4564.
235. Shortle, D., and B. Lin. 1985. Genetic analysis of staphylococcal nuclease: identification of three intragenic "global" suppressors of nuclease-minus mutations. Genetics 110:539–555.
236. Sköld, S.-E. 1983. Chemical crosslinking of elongation factor G to the 23S RNA in 70S ribosomes from Escherichia coli. Nucleic Acids Res. 11:4923–4932.
237. Sonneborn, T. M. 1965. Degeneracy of the genetic code: extent, nature and genetic implications, p. 377–397. In V. Bryson and H. J. Vogel (ed.), Evolving Genes and Proteins. Academic Press, New York.
238. Speyer, J. F., P. Lengyel, and C. Basilio. 1962. Ribosomal localization of streptomycin sensitivity. Proc. Natl. Acad. Sci. USA 48:684–686.
239. Spirin, A. S. 1969. A model of the functioning ribosome: locking and unlocking of the ribosome subparticles. Cold Spring Harbor Symp. Quant. Biol. 34:197–207.
240. Spirin, A.S. 1985. Ribosomal translocation: facts and models. Prog. Nucleic Acids Res. Mol. Biol. 32:75–114.
241. Spirin, A. S., V. I. Lim, and R. Brimacombe. 1992. The arrangement of tRNA in the ribosome, p. 445–454. In K. H. Nierhaus, F. Franceschi, A. R. Subramanian, V. A. Erdmann, and B. Wittmann-Liebold (ed.), The Translational Apparatus. Plenum Publishing Corp., New York.
242. Spirin, A. S., and V. D. Vasiliev. 1989. Localization of functional centers on the prokaryotic ribosome: immuno-electron microscopy approach. Biol. Cell 66:215–223.
243. Spotts, C. R., and R. Y. Stanier. 1961. Mechanism of streptomycin action on bacteria: a unitary hypothesis. Nature (London) 192:633–637.
244. Straus, D. B., W. A. Walter, and C. A. Gross. 1988. Escherichia coli heat shock gene mutants are defective in proteolysis. Genes Dev. 2:1851–1858.
245. Strigini, P., and L. Gorini. 1970. Ribosomal mutations affecting the efficiency of amber suppression. J. Mol. Biol. 47:517–530.
246. Swan, D., G. Sander, E. Bermek, W. Krämer, T. Kreuzer, C. Arglebe, R. Zöllner, K. Eckert, and H. Matthaei. 1969. On the mechanism of coded binding of aminoacyl-tRNA to ribosomes: number and properties of sites. Cold Spring Harbor Symp. Quant. Biol. 34:179–195.
247. Swart, G. W. M., A. Parmeggiani, B. Kraal, and L. Bosch. 1987. Effects of the mutation Gly222Asp on the functions of elongation factor Tu. Biochemistry 26:2047–2054.
248. Tapio, S., N. Bilgin, and M. Ehrenberg. 1990. Impaired in vitro kinetics of EF-Tu mutant Aa. Eur. J. Biochem. 188:347–354.
249. Tapio, S., and L. Isaksson. 1988. Antagonistic effects of mutant elongation factor Tu and ribosomal protein S12 on control of translational accuracy, suppression and cellular growth. Biochimie 70:273–281.
250. Tapio, S., and L. Isaksson. 1991. Base 2661 in Escherichia coli 23S rRNA influences the binding of elongation factor Tu during protein synthesis in vivo. Eur. J. Biochem. 202:981–984.
251. Tapio, S., and C. G. Kurland. 1986. Mutant EF-Tu increases missense error in vitro. Mol. Gen. Genet. 205:186–188.
252. Tapprich, W. E., and A. E. Dahlberg. 1990. A single base mutation at position 2661 in E. coli 23S ribosomal RNA affects the binding of ternary complex to the ribosome. EMBO J. 9:2649–2655.
253. Thompson, J., E. Cundliffe, and A.E. Dahlberg. 1988. Site-directed mutagenesis of Escherichia coli 23S ribosomal RNA at position 1067 within the GTP hydrolysis centre. J. Mol. Biol. 203:457–465.
254. Thompson, J., F. Schmidt, and E. Cundliffe. 1982. Site of action of a ribosomal RNA methylase conferring resistance to thiostrepton. J. Biol. Chem. 257:7915–7917.
255. Thompson, R. C., and D. B. Dix. 1982. Accuracy of protein biosynthesis. A kinetic study of the reaction of poly(U)-programmed ribosomes with a leucyl-tRNA2-elongation factor Tu-GTP complex. J. Biol. Chem. 257:6677–6682.
256. Thompson, R. C., D. B. Dix, and J. F. Eccleston. 1980. Single turnover kinetic studies of guanosine triphosphate hydrolysis and peptide formation in the elongation factor Tu-dependent binding of aminoacyl-tRNA to Escherichia coli ribosomes. J. Biol. Chem. 255:11088–11090.
257. Thompson, R. C., D. B. Dix, R. B. Gerson, and A. M. Karim. 1981. A GTPase reaction accompanying the rejection of Leu-tRNA2 by UUU-programmed ribosomes. J. Biol. Chem. 256:81–86.
258. Thompson, R. C., and A. M. Karim. 1982. The accuracy of protein biosynthesis is limited by its speed: high fidelity selection by ribosomes of aminoacyl-tRNA ternary complexes containing GTP[γ-S]. Proc. Natl. Acad. Sci. USA 79:4922–4926.
259. Thompson, R. C., and P. J. Stone. 1977. Proofreading of the codon-anticodon interaction on ribosomes. Proc. Natl. Acad. Sci. USA 74:198–202.
260. Timms, A. R., and B. A. Bridges. 1993. Double, independent mutational events in the rpsL gene of Escherichia coli: an example of hypermutability? Mol. Microbiol. 9:335–342.
261. Timms, A. R., H. Syeingrimsdottir, A. R. Lehman, and B. A. Bridges. 1992. Mutant sequences in the rpsL gene of Escherichia coli B/r: mechanistic implications for spontaneous and ultraviolet light mutagenesis. Mol. Gen. Genet. 232:89–96.
262. Topisirovic, L., R. Villarroel, M. De Wilde, A. Herzog, T. Cabezon, and A. Bollen. 1977. Translational fidelity in Escherichia coli: contrasting role of neaA and ramA gene products in the ribosome. Mol. Gen. Genet. 151:89–94.
263. Travers, A. A., and H. A. F. Mace. 1982. The heat-shock phenomenon in bacteria—a protection against DNA relaxation?, p. 127–130. In M. J. Schlesinger, M. Ashburner, and A. Tissieres (ed.), Heat-Shock from Bacteria to Man. Cold Spring Harbor Laboratory, Cold Spring Harbor, N.Y.
264. Tsung, K., S. Inouye, and M. Inouye. 1989. Factors affecting the efficiency of protein synthesis in Escherichia coli. J. Biol. Chem. 264:4428–4433.
265. Tubulekas, I., R. H. Buckingham, and D. Hughes. 1991. Mutant ribosomes can generate dominant kirromycin resistance. J. Bacteriol. 173:3635–3643.
266. Tubulekas, I., and D. Hughes. 1993. Growth and translation elongation rate are sensitive to the concentration of EF-Tu. Mol. Microbiol. 8:761–770.
267. Tubulekas, I., and D. Hughes. 1993. A single amino acid substitution in elongation factor Tu disrupts interaction between the ternary complex and the ribosome. J. Bacteriol. 175:240–250.
268. Tubulekas, I., and D. Hughes. 1993. Suppression of rpsL phenotypes by tuf mutations reveals a unique relationship between translation elongation and growth rate. Mol. Microbiol. 7:275–284.
269. van Acken, U. 1975. Protein chemical studies on ribosomal proteins S4 and S12 from ram (ribosomal ambiguity) mutants of Escherichia coli. Mol. Gen. Genet. 140:61–68.
270. VanBogelen, R. A., and F. C. Neidhardt. 1990. Ribosomes as sensors of heat and cold shock in Escherichia coli. Proc. Natl. Acad. Sci. USA 87:5589–5593.
271. van der Meide, P. H., R. A. Kastelein, E. Vijgenboom, and L. Bosch. 1983. tuf gene dosage effects on the intracellular concentration of EF-TuB. Eur. J. Biochem. 130:409–417.
272. Van Dyk, T. K., A. A. Gatenby, and R. A. LaRossa. 1989. Demonstration by genetic suppression of interaction of GroE products with many proteins. Nature (London) 342:451–453.
273. Vijgenboom, E., and L. Bosch. 1989. Translational frameshifts induced by mutant species of the polypeptide chain elongation factor Tu of Escherichia coli. J. Biol. Chem. 264:13012–13017.
274. Vijgenboom, E., T. Vink, B. Kraal, and L. Bosch. 1985. Mutants of the elongation factor EF-Tu, a new class of nonsense suppressors. EMBO J. 4:1049–1052.
275. Wagner, E. G. H., P. C. Jelenc, M. Ehrenberg, and C. G. Kurland. 1982. Rate of elongation of polyphenylalanine in vitro. Eur. J. Biochem. 122:193–197.
276. Watson, J. D. 1964. The synthesis of proteins upon ribosomes. Bull. Soc. Chem. Biol. 46:1399–1425.
277. Weijland, A., G. Parlato, and A. Parmeggiani. 1994. Elongation factor Tu D138N, a mutant with modified substrate specificity, as a tool to study energy consumption in protein biosynthesis. Biochemistry 33:10711–10717.
278. Weijland, A., and A. Parmeggiani. 1993. Toward a model for the interaction between elongation factor-Tu and the ribosome. Science 259:1311–1314.
279. Weisblum, B., and V. Demohn. 1970. Inhibition by thiostrepton of the formation of a ribosome-bound guanine nucleotide complex. FEBS Lett. 11:149–152.
280. Weiss, R., and J. Gallant. 1986. Frameshift suppression in aminoacyl-tRNA limited cells. Genetics 112:727–739.
281. Weiss, R., and J. A. Gallant. 1983. Mechanism of ribosome frameshifting during translation of the genetic code. Nature (London) 302:389–393.
282. Weiss, R., D. Lindsley, B. Falahee, and J. Gallant. 1988. On the mechanism of ribosomal frameshifting at hungry codons. J. Mol. Biol. 203:403–410.
283. Weiss, R. B., D. M. Dunn, J. F. Atkins, and R. F. Gesteland. 1990. Ribosomal frameshifting from –2 to +50 nucleotides. Prog. Nucleic Acid Res. Mol. Biol. 39:159–183.
284. Weiss, R. B., W. M. Huang, and D. M. Dunn. 1990. A nascent peptide is required for ribosomal bypass of the coding gap in bacteriophage T4 gene 60. Cell 62:117–126.
285. Weissbach, H., B. Redfield, and N. Brod. 1971. Aminoacyl-tRNA-Tu-GTP interaction with ribosomes. Arch. Biochem. Biophys. 145:676–684.
286. Weiss-Brummer, B., and A. Hüttenhofer. 1989. The paromomycin resistance mutation (parR-454) in the 15S rRNA gene of the yeast Saccharomyces cerevisiae is involved in ribosomal frameshifting. Mol. Gen. Genet. 217:362–369.
287. Wettstein, F. O., and H. Noll. 1965. Binding of transfer ribonucleic acid to ribosomes engaged in protein synthesis: number and properties of ribosomal binding sites. J. Mol. Biol. 11:35–53.
288. Wittmann, H. G., G. Stöffler, W. Piepersberg, P. Bückel, D. Ruffler, and A. Böck. 1974. Altered S5 and S20 ribosomal proteins in revertants of an alanyl-tRNA synthetase mutant of Escherichia coli. Mol. Gen. Genet. 134:225–236.
289. Woese, C. R. 1965. On the evolution of the genetic code. Proc. Natl. Acad. Sci. USA 54:1546–1552.
290. Wolf, H., G. Chinali, and A. Parmeggiani. 1977. Mechanism of the inhibition of protein synthesis by kirromycin. Eur. J. Biochem. 75:67–75.
291. Yaguchi, M., and H.G. Wittmann. 1976. Alteration of ribosomal protein S17 by mutation linked to neamine resistance in Escherichia coli. II. Localization of the amino acid replacement in protein S17 from a neaA mutant. J. Mol. Biol. 104:617–620.
292. Yourno, J., D. Barr, and S. Tannemura. 1969. Externally suppressible frameshift mutant of Salmonella typhimurium. J. Bacteriol. 100:453–459.
293. Zengel, J. M., R. Young, D. P. Dennis, and M. Nomura. 1977. Role of ribosomal protein S12 in peptide chain elongation: analysis of pleiotropic, streptomycin resistance mutants of Escherichia coli. J. Bacteriol. 129:1320–1329.
294. Zimmermann, R. A., T. Garvin, and L. Gorini. 1971. Alteration of a 30S ribosomal protein accompanying the ram mutation in Escherichia coli. Proc. Natl. Acad. Sci. USA 68:2263–2267.
Chapter65
