REVIEW

Re-creating the RNA worldIchiro Hirao and Andrew D. Ellington

Department of Chemistry, Indiana University, Bloomington, Indiana 47405, USA.

Results from in vitro selection experiments can be used to construct andtest models for the evolution of the RNA world. Surprisingly, the success ofselected RNAs at binding ligands and catalyzing reactions may make itdifficult to determine precisely the lineage of molecular fossils, molecules

that are believed to have survived from the RNA world to the present.

Introduction

The discovery of catalytic RNA has dramatically shapedspeculations about the origin of life and the evolution ofmetabolism. The primordial 'chicken and egg' conun-drum - which came first, informational polynucleotidesor functional polypeptides? - was obviated by thesimple but elegant compaction of both genetic informa-tion and catalytic function into the same molecule.Moreover, the chemical complexity of catalytic RNAs(ribozymes) led to speculations that the biochemistry ofthe earliest organisms may have been a consequence ofthe shapes and functions that nucleic acids could assume.In the most intricate scenarios, the earliest RNA replica-tors may have directly emerged from the 'prebiotic soup'and subsequently evolved to become an 'RNA world'[1], in which primordial biochemistry and metabolismmirrored that found in modern life [2].

The conceptual revolution in our understanding ofnucleic acid evolution has been paralleled by an equallyprofound technical revolution. Techniques such as DNAsequencing, in vitro transcription and the polymerasechain reaction have made it possible to manipulate andcharacterize nucleic acid sequences almost completely exvivo. In turn, the principles that govern the natural selec-tion of organisms can now be applied to molecules in atest tube: a pool of heritably variable nucleic acid mol-ecules can be prepared and sieved for sequences thatconfer an improved phenotype, these phenotypes can beallowed to compete among themselves, and the survivorscan be preferentially amplified. Although the notion ofin vitro selection is sometimes claimed to have itself arisende novo [3], it is a logical and obvious consequence of abody of research into molecular evolution [4,5]: it wasthe available techniques that advanced to the point whereold ideas became more experimentally accessible.

Models for molecular evolution

The intersection of the discovery of ribozymes with thedevelopment of techniques for nucleic acid amplification

allowed models of molecular evolution to be recapitu-lated in a test tube. As extant catalytic RNAs are eitheranalogous to, or descended from, early ribozymes, manyof the ideas advanced under the general aegis of the'RNA world' hypothesis can be subjected to experimen-tal tests. Both the data gathered and their interpretationcan be scrutinized to determine whether a particularaspect of a particular model for the RNA world mighthave occurred, or whether an alternative scenario wasmore likely. But although the data garnered from in vitroselection experiments can be used to guide model-build-ing, care must still be taken to ensure that the modelsthemselves are compatible with other facts and logicallyself-consistent.

As a guide to how primordial molecular biology canpotentially be modelled by experiment and inference, wewill examine three claims that have been advancedregarding different aspects of the 'RNA world' hypo-thesis: that RNA catalysts can be functionally diverse,that the origin of the genetic code may lie in aminoacid-RNA interactions, and that the evolution of thetranslation apparatus and other functional RNAs couldhave been guided by interactions with low molecularweight effectors. By comparing and contrasting how evi-dence is used to support or contradict these claims, itshould be possible to gain an understanding of the valueand limitations of using artificial evolutionary experi-ments today to describe natural evolutionary processesthat may have occurred in the past.

The RNA world and the diversity of RNAcatalysts

In order to establish a common ground for discussion, ageneralized view of the RNA world is shown in Figure 1.In this model, the earliest self-replicating nucleic acids -whether RNA, DNA or some analogous polymer -were elaborated into all-RNA genomes and catalysts. Thesurvival of virtually any self-replicating system is depen-dent on avoiding dissipation by diffusion, and it isexpected that cellularization was one of the earliest

© Current Biology 1995, Vol 5 No 9 1017

1018 Current Biology 1995, Vol 5 No 9

Fig. 1. A generalized view of the RNAworld. Events hypothesized to haveoccurred between the origin of life andthe progenote (the last common ances-tor of modern life) are roughly ordered.The chronology flows from 'early' at thetop of the page (- 3.5 billion years ago)to 'late' at the bottom. Cellularizationrepresents the immobilization of self-replicators, and could as easily haveoccurred on a two-dimensional surfaceas within a lipid vesicle. As diffusionwould have severely reduced the fitnessof replicators, cellularization may havebeen concomitant with the origin of life.Similarly, as the resupply of metabolitesfrom the soup would probably havesuperseded the need for an error-resis-tant genome, metabolism is shown tooccur before the invention of DNA.'Molecular fossils' that may have per-sisted from earlier versions of life arerepresented on the right.

phenotypic properties of a ribo-organism. DNA wasfound to be a superior macromolecule for storing geneticinformation, and proteins were a fortuitous side-productof a complex ribozyme that could catalyze template-directed peptide-bond formation. The invention of thetranslation apparatus was likely to have been a cataclysmicevent that allowed superior protein catalysts to displacebiochemically complex, but less efficient, ribozymes.Most of these evolutionary events occurred long beforethe last common ancestor of modern life, the progenote,gave rise to the three modern domains (eubacteria,archaebacteria, and eukaryotes). Some vestiges of RNA'sformer greatness can, however, still be found in (catalyti-cally competent?) ribosomal RNA and its attendant trans-fer RNAs, the nucleotide 'signatures' of ubiquitous cofac-tors such as NAD and ATP, and perhaps in idiosyncraticmolecular fossils such as the group I self-splicing intron.

This scenario makes a strong prediction regarding the cat-alytic functions that can be assumed by ribozymes. Forexample, if NAD is a 'ribo-cofactor' descended from theRNA world, then it should be possible for at least someRNA sequences to act as dehydrogenases [6]. Similarly,the fact that the cellular currency for energy, ATP, is anucleotide suggests that there may have been enzymessuch as ribo-kinases. In fact, the RNA world hypothesisnecessarily assumes the existence of a wide variety of ribo-zymes, with functions that would have spanned the bio-chemical reactions of intermediary metabolism. If these

catalytic functions were assumed by ribozymes in the past,then it should be possible to recreate such ribozymes inthe present. This prediction has been made implicitly orexplicitly by a number of authors, both prior to, andfollowing, the actual discovery of catalytic RNA [2,7-12].

In accord with these predictions, in vitro selection hasbeen used to generate a wide variety of new ribozymes.Two types of selection procedure have yielded catalysts.Indirect selection is similar to the protocols originallyused to isolate catalytic antibodies. Prudent et al. [13]selected RNA molecules that can bind to a bridgedbiphenyl analogue of the transition state of an isomeriza-tion reaction. The binding species were found to beribozymes able to catalyze the isomerization of substrateschemically similar to the analogue. Direct selectionrequires that active RNA molecules in a pool alter them-selves during the selection process in such a way as toallow their isolation. For example, Bartel and Szostak [14]were able to select RNA molecules able to ligate a primersequence onto themselves from a random sequence pool.Conversely, Pan and Uhlenbeck [15] selected ribozymesthat can use a lead cofactor to cleave themselves.

Lorsch and Szostak [16] forged a link between cofactor-binding and catalysis by selecting ribo-kinases that canphosphorylate themselves from a pool of RNA mol-ecules in which a randomly varied sequence was placednext to a previously selected binding site for ATP. The

REVIEW 1019

selection of binding sites for redox cofactors similarlyopens the way to the selection of ribo-dehydrogenases orribo-oxidases [17,18]. Finally, direct selection for aribozyme that can catalyze a reaction that did not involvea phosphodiester bond rearrangement has recently beencarried out by Wilson and Szostak [19]. These authorsidentified ribozymes that can biotinylate themselves, analkylation reaction akin to those carried out by methyltransferases throughout metabolism.

If at least some ribozymes were cobbled together fromrandom sequences during early evolution, then newfunctions might have been quickly acquired by muta-tional walks. Again, in vitro selection experiments givecredence to this idea. In particular, Joyce and his co-workers have evolved the group I self-splicing intron toperform reactions other than self-splicing, includingDNA cleavage [20-22] and RNA cleavage with calcium[23]. Indeed, ribozymes that have evolved to catalyze onereaction might intrinsically harbor the catalytic machin-ery to carry out heterologous reactions. The group I self-splicing intron has been shown to catalyze ester (asopposed to phosphodiester) hydrolysis [24], and the vari-ant evolved to catalyze DNA cleavage has been shownalso to hydrolyze amide bonds [25].

Amino acid-RNA interactions and the origin ofthe genetic code

In the genetic code, the assignment of triplet codons toparticular amino acids is clearly non-random, yet thelogic or formula that could account for the standardarrangement found in most organisms remains unknown.Although it is possible that the genetic code is a 'frozenaccident' [7], its basic importance has driven variousauthors to devise schemes that posit rational linkagesbetween genetic information and amino-acid functional-ity. Perhaps the most beguiling of these hypotheses is theidea that genetic information does not just code for, butsomehow begets, amino-acid functionality [26,27]. Forexample, it has been proposed that amino acids may havedirectly associated with their cognate triplet codons oranticodons [28,29].

Once again, support for this hypothesis may be found inthe minutiae of biochemistry. The group I self-splicingintron can interact specifically with a guanosine cofactor.Yarus has shown that arginine can competitively inhibitguanosine-binding to group I self-splicing introns [30],and notes that the primary sequence of the guanosine-binding site contains several trinucleotide tracts that couldbe read as arginine codons [31]. This was interpreted tobe evidence in favor of the 'direct association' hypothesis,as opposed to being just a fortuitous cross-reaction basedon structural similarities between the aptly named guano-sine and the guanidino group of arginine (Fig. 2).

The generality of the amino-acid association observedwith the group I self-splicing intron can be readily

Fig. 2. Structural similarities between arginine and guanosine. Thetwo structures are drawn so that the guanidino head group of argi-nine is in a similar orientation to related chemical moieties foundon guanosine. The similarity can be extended through the baseand into the sugar, where either the primary amine or carboxylateof the amino acid might overlap with hydroxyls on ribose.

assessed using in vitro selection. RNA molecules that canspecifically bind to arginine - 'aptamers' - were itera-tively sieved from a completely random sequence pool[32]. Aptamer sequences were determined and arginine-binding sites mapped. The results obtained appeared tobe in accord with those observed for the group I intron.One of the predominant aptamers did contain argininecodons that overlapped with the arginine-binding site.

The ability of arginine to associate with RNA moleculesmotivated Yarus [33,34] to propose a model for the originof the genetic code (Fig. 3). The model proposes thatguanosine-binding sites, such as those found in the group Iintron, were originally selected for during the evolution ofself-repli&tion. The structural similarities between guano-sine and arginine allowed this amino acid to bind near theactive site of a ribozyme. This ribozyme may have beenable to use arginine as a substrate, catalyzing its activationand then the aminoacylation of the ribozyme by the acti-vated arginine. The arginine codon or anticodon (or both)would have been a key feature of the binding site of thenewly evolved tRNA synthetase-like ribozyme. OthertRNA synthetase-like ribozymes would have evolved withtheir own coded interactions with amino acids. As pro-teins displaced the RNA world, the ribozymes might havefaded, but the code would have remained.

Aspects of this model are supported by results from invitro selection experiments. The aptamers selected to bindarginine were found also to bind guanosine. Aptamersselected to bind both arginine and guanosine were foundto contain some arginine codons [35]. An RNA

RNA world

1020 Current Biology 1995, Vol 5 No 9

arginine-binding motifs gave rise to a tRNA synthetase-like ribozyme, there are also multiple other ways to bindarginine that do not necessarily involve arginine codons(or anticodons). For example, Famulok [37] has selectedarginine-binding motifs that are not obviously dependenton arginine codons or anticodons. More importantly, theamino-acid specificity of these aptamers can be altered (tocitrulline) by substitutions at three residues, yet theseresidues are not part of a single arginine codon or anti-codon or codon-anticodon interaction. Moreover, thereis no indication from selection experiments that targetedarginine-rich motifs on proteins [38], or other aminoacids such as tryptophan [39] or valine [40], that there isany particular correspondence between selected motifsand cognate codons or anticodons. If anything, selectionexperiments seem to indicate that there are a huge num-ber of possible binding motifs for a given amino acid.Given the large number of possibilities, it may be impossi-ble to prove or disprove any particular model for the evo-lution of the genetic code. If the evolutionary 'film' wereto be run again, an entirely different, but equally deter-mined, genetic code might be equally likely to emerge.

Antibiotics and the evolution of translation

Fig. 3. Model for the evolution of the translation apparatus and theorigin of the genetic code. This figure summarizes many of thespeculations put forth by Yarus and his co-workers [30,31,33,34].Primordial replicases may have contained binding sites for nucleo-tides such as guanosine (G) or adenosine (A). These binding sitesmay have fortuitously also been able to bind some amino acids,and could have served as 'proto-codons'. The ability of the guano-sine-binding site of the Tetrahymena intron also to bind arginine(R) is typically used as an example. Adjacent binding of amino acidand nucleotide cofactors (A) could have resulted in the formationof activated amino acids, such as amino acid adenylates. Activatedamino acids could in turn have reacted with terminal (or internal)hydroxyls of the RNAs that held them. If these compounds orevents imparted function or phenotype to a ribo-organism, thestage would have been set for the invention of translation. ThetRNA molecules seen in modern organisms can thus be viewed asmolecular fossils of ancient amino-acid-binding sites. The amazingconsistency of modern tRNA structures stands in stark contrast tothe apparent sequence and structural plasticity of selected amino-acid-binding sites, and is not addressed by this model.

molecule that can catalyze an aminoacyl-RNA-synthe-tase-like reaction using activated amino acids was selectedfrom a random sequence RNA pool. This demonstratesthe plausibility of the suggestion that ribozymes withsuch catalytic activity may have existed in the RNAworld [36]. Taken together, these results affirm the plausi-bility of the 'direct association' hypothesis and models forthe subsequent evolution of the genetic code.

Results from a similar number of selection experiments,however, contradict these hypotheses and models. Whileit is conceivable that one or more of the guanosine/

Antibiotics are known to bind to, and disrupt the func-tion of, ribosomal RNA. Conventionally, it was believedthat such interactions are due more to the structure of theantibiotic than to the structure of the RNA. In this view,antibiotics are compounds that were carefully crafted formicrobiological warfare and enhanced the survival of thespecies that produced them. Julian Davies [41,42], how-ever, has postulated that ribosomal RNA may be predis-posed to interact with antibiotic inhibitors because ofhistorical constraints. For example, if ribosomes wereonce regulated by small organic molecules (so-called low-molecular-weight effectors or LMEs) then the bindingsites for these molecules may have been conservedthrough the course of evolution, waiting to be exploitedby the combinatorial chemistry programs of bacteria.

More recently, it has been shown that antibiotics can bindto, and inhibit, other functional RNAs, such as the groupI self-splicing intron and the hammerhead ribozyme[43,44]. These results have led Renee Schroeder andco-workers [45] to speculate that the active centers offunctional RNAs may form binding pockets that havesimilar shapes or functions. In turn, if the ribosome andthe group I intron have similar LME-binding sites, thenthey may also share a common ancestry [45,46]. Thisnovel hypothesis suggests that antibiotics can be used asstructural and functional probes to indicate relatedness,much as antibodies were used to examine homologousprotein epitopes before the advent of phylogeneticsequence analysis. Measuring the relatedness betweenmolecules such as ribosomal RNA and the group I intronwith a structural probe might be the only way by whichtheir ancestry could be confirmed, as these functionalRNAs are unrelated in primary and secondary structure.

RNA world REVIEW 1021

Several groups have now carried out in vitro selectionexperiments that target aminoglycoside antibiotics. Wangand Rando [48] and Lato et al. [49] both find that evenlimited sieving of random sequence pools can produce amultitude of aptamers with moderate affinity(Kd < 10 tIM) and specificity for a given aminoglycoside.As has been shown for other targets, more extensiveselection results in the isolation of a limited number ofhigh-affinity (Kd < 1 FM) binding species. Even themoderate-affinity aptamers can bind aminoglycosides aswell as ribosomal RNA and the group I self-splicingintron can, and they can similarly discriminate betweenrelated aminoglycosides.

Fig. 4. Models for the evolution of aminoglycoside-binding sites.Two contrasting models for the presence of aminoglycoside-binding sites in modern RNAs are shown. (a) The 'LME hypothe-sis' suggests that modern aminoglycoside-binding sites aremolecular fossils of a relatively limited number of binding sitesthat occurred in ancient RNAs. The sites have been conservedthrough time because of their association with critical RNA struc-tures or functions. (b) The 'stochastic hypothesis' suggests that, asaminoglycosides have probably evolved to bind RNA, it is notsurprising that aminoglycoside-binding sites can be found bychance alone on functional RNAs. Data from in vitro selectionexperiments confirm that there are a wide variety of unnaturalaminoglycoside-binding sites 48,49].

The conjectures of Davies and Schroeder are particularlyintriguing because they tie together a variety of disparatefacts about functional RNAs and could validate the linealdescent of modern organisms from ancient sequences.The hypotheses must, however, be weighed against thenull hypothesis - that antibiotic-binding sites on differ-ent RNAs are the result of chance rather than history.This alternative is consistent with the fact that high-affinity antibiotic-binding sites can be found on RNAsthat are probably not descended from primordialsequences, such as the Rev-binding element of humanimmunodeficiency virus 1 [47].

In vitro selection can be used to assess the likelihood ofthe historical/LME hypothesis by assessing the likelihoodof its converse: the null (or 'stochastic') hypothesis (Fig.4). If it is relatively difficult to find antibiotic-bindingsequences in a random sequence pool, then those anti-biotic-binding sites that are identified in nature may bedue to common descent. If it is relatively easy to isolateantibiotic-binding sequences, then there is no reason toassume that binding sites that are otherwise unrelatedshare a common ancestor.

These results suggest that natural aminoglycoside-bindingsites are far from unique, but are likely to be a subset of adiverse universe of functionally related RNAs. However,this result does not in itself necessarily invalidate the LMEhypothesis. For example, if there were distinguishingcharacteristics that allowed aminoglycoside-binding apta-mers to be parsed into classes, such as common primaryor secondary structural signatures, then it might still bepossible to show that natural binding sites are somehowdistinct from the plethora of unnatural binding sites.Recently, Schroeder and her co-workers [50] have identi-fied RNA motifs that can bind to neomycin, and suggestthat these motifs may contain a common (but not unique)secondary structural feature - a widened major groove.

Comparisons and conclusions

The three examples examined above are purposefullyordered in terms of chronology, specificity and implausi-bility. This correspondence is not accidental: overall, themain conclusion that can be drawn from in vitro selectionexperiments is that the total number of potential bindingmotifs and catalysts is gargantuan. Therefore, while it issafe to speculate on the acquisition of a particular func-tion or event (such as ribo-kinases or cellularization), thevery plurality of sequence and function may make itimpossible to prove the origins of a particular sequence(the genetic code) or structure (the antibiotic-bindingsite of the ribosome).

If this is the case, then it is not merely the informationwithin the genetic code that is a frozen accident ofbiology, but perhaps all remnants of the inferred RNAworld. Cofactors, transfer RNAs and the ribosomemight all have turned out significantly different thanthe versions we are familiar with, and deterministicattempts to reconcile their sequences or structures withone another or with other molecular traits, such asself-splicing introns, become all the more difficult.Paradoxically, examination of the precise nature of ourorigins may be thwarted by the same factors that origi-nally led to the success of the RNA world.

Acknowledgements: This work was supported by a National ScienceFoundation National Young Investigator Award (A.E.), a Scholar

RNA world REVIEW 1021

1022 Current Biology 1995, Vol 5 No 9

Award from the American Foundation for AIDS Research (A.E.),the Pew Scholar Award in the Biomedical Sciences (A.E.), theCottrell Scholar Award (A.E.), the Office of Naval Research grant#N00014-93-1-0430 (A.E.), and the National Institutes of Healthgrant #PHS R01GM48175 (A.E.).

References1. Gilbert W: The RNA world. Nature 1986, 319:618.2. Benner SA, Ellington AD, Tauer A: Modern metabolism as a

palimpsest of the RNA world. Proc Natl Acad Sci USA 1989, 86:7054-7058.

3. Gold L, Tuerk C: Methods for Identifying Nucleic Acid Ligands.United States Patent #5270163, 1993.

4. Kramer FR, Mills DR, Cole PE, Nishihara T, Spiegelman S: Evolutionin vitro: sequence and phenotype of a mutant RNA resistant toethidium bromide. J Mol Biol 1974, 89:719-736.

5. Orgel LE: Selection in vitro. Proc R Soc Lnd [Biol] 1979, 205:435-442.

6. Visser CM: Evolution of bio catalysis 2. Nicotinamide and/or flavin-containing RNA molecules as possible pre-genetic-code replicatingoxido-reductases. J Mol Evol 1984, 14:301-306.

7. Crick FHC: The origin of the genetic code. J Mol Biol 1968, 38:367-379.

8. Orgel LE: Evolution of the genetic apparatus. Mol Biol 1968, 38:381-393.

9. White HB: Coenzymes as fossils of an earlier metabolic state. J MolEvol 1976, 7:101-104.

10. Visser CM, Kellogg RM: Biotin. Its place in evolution. Mol Evol1978, 11:171-178.

11. Joyce GF: RNA evolution and the origins of life. Nature 1989,338:21 7-224.

12. Benner SA, Cohen MA, Gonnet GH, Berkowitz DB, Johnsson KP:Reading the palimpsest: contemporary biochemical data andthe RNA world. In The RNA World. Edited by Gesteland RF, AtkinsIF. New York: Cold Spring Harbor Laboratory Press; 1993:27-70.

13. Prudent JR, Uno T, Schultz PG: Expanding the scope of RNA cata-lysis. Science 1994, 264:1924-1927.

14. Bartel DP, Szostak JW: Isolation of new ribozymes from a large poolof random sequences. Science 1993, 261:1411-1418.

15. Pan T, Uhlenbeck OC: In vitro selection of RNAs that undergoautolytic cleavage with Pb2+. Biochemistry 1992, 31:3887-3995.

16. Lorsch JW, Szostak JW: In vitro evolution of new ribozymes withpolynucleotide kinase activity. Nature 1994, 371:31-36.

17. Burgstaller P, Famulok M: Isolation of RNA aptamers for biologicalcofactors by in vitro selection. Angew Chem Int Ed Engl 1994,33:1084-1087.

18. Lauhon CT, Szostak JW: RNA aptamers that bind flavin and nicotin-amide redox cofactors. J Am Chem Soc 1995, 117:1246-1257.

19. Wilson C, Szostak JW: In vitro evolution of a self-alkylatingribozyme. Nature 1995, 374:777-782.

20. Robertson DL, Joyce, GF: Selection in vitro of an RNA enzyme thatspecifically cleaves single-stranded DNA. Nature 1990, 344:467-468.

21. Beaudry A, Joyce GF: Directed evolution of an RNA enzyme.Science 1992, 257:635-641.

22. Tsang J, Joyce GF: Evolutionary optimization of the catalyticproperties of a DNA-cleaving ribozyme. Biochemistry 1994, 33:5966-5973.

23. Lehman N, Joyce GF: Evolution in vitro of an RNA enzyme withaltered metal dependence. Nature 1993, 361:182-185.

24. Piccirilli JA, McConnell TS, Zaug AJ, Noller HF, Cech TR: Amino-acyl esterase activity of the Tetrahymena ribozyme. Science 1992,256:1420-1424.

25. Dai X, Mesmaekder AD, Joyce GF: Cleavage of an amide bond by aribozyme. Science 1995, 267:237-240.

26. Szathmary E: Coding coenzyme handles: a hypothesis for the originof the genetic code. Proc Natl Acad Sci USA 1993, 90:9916-9920.

27. Giulio MD, Capobianco MR, Medugno M: On the optimization ofthe physiochemical distances between amino acids in the evolutionof the genetic code. Theor Biol 1994, 168:43-51.

28. Woese CR, Dugre DH, Dugre SA, Kondo M, Saxinger WC: On thefundamental nature and evolution of the genetic code. Cold SpringHarbor Symp Quant Biol 1966, 31:723-736.

29. Shimizu M: Molecular basis for the genetic code. J Mol Evol 1982,18:297-303.

30. Yarus M: A specific amino-acid binding site composed of RNA.Science 1988, 240:1751-1758.

31. Yarus M, Christian E: Genetic code origins. Nature 1989, 342:349-350.

32. Connell GJ, Illangasekare M, Yarus M: Three small ribooligonucleo-tides with specific arginine sites. Biochemistry 1993, 32:5497-5502.

33. Yarus M: An RNA-amino acid complex and the origin of thegenetic code. New Biol 1991, 3:183-189.

34. Yarus M: An RNA-amino acid affinity. In The RNA World. Edited byGesteland RF, Atkins JF. New York: Cold Spring Harbor LaboratoryPress; 1993:205-217.

35. Connell GJ, Yarus M: RNAs with dual specificity and dual RNAswith similar specificity. Science 1994, 264:1137-1141.

36. Illangasekare M, Sanchez G, Nickles T, Yarus M: Aminoacyl-RNAsynthesis catalyzed by an RNA. Science 1994, 267:643-647.

37. Famulok M: Molecular recognition of amino acids by RNA-aptamers: an L-citrulline binding RNA motif and its evolution intoan L-arginine binder. J Am Chem Soc 1994, 116:1698-1706.

38. Giver L, Bartel D, Zapp M, Pawul A, Green M, Ellington AD: Selec-tive optimization of the Rev-binding element of HIV-1. NucleicAcids Res 1993, 21:5509-5516.

39. Famulok M, Szostak JW: Stereospecific recognition of tryptophanagarose by in vitro selected RNA. Am Chem Soc 1992 114:3990-3991.

40. Majerfeld I, Yarus M: An RNA pocket for an aliphatic hydrophobe.Nature Struct Biol 1994, 1:287-292.

41. Davies J: What are antibiotics? Archaic functions for modern activi-ties. Mol Microbiol 1990, 4:1227-1232.

42. Davies J, von Ahsen U, Schroeder R: Antibiotics and the RNAworld: a role for low-molecular-weight effectors in biochemicalevolution? In The RNA World. Edited by Gesteland RF, Atkins F.New York: Cold Spring Harbor Laboratory Press; 1993:185-204.

43. von Ahsen U, Davies J, Schroeder R: Antibiotic inhibition of group Iribozyme function. Nature 1991, 353:368-370.

44. Stage TK, Hertel KJ, Uhlenbeck OC: Inhibition of the hammerheadribozyme by neomycin. RNA 1995, 1:95-101.

45. Schroeder R, Streicher B, Wank H: Splice-site selection and de-coding: are they related? Science 1993, 260:1443-1444.

46. Noller HF: Drugs and the RNA world. Nature 1991, 353:302-303.47. Zapp ML, Stern S, Green MR: Small molecules that selectively block

RNA binding of HIV-1 Rev protein inhibit Rev function and viralproduction. Cell 1993, 74:969-978.

48. Wang Y, Rando RR: Specific binding of aminoglycoside antibioticsto RNA. Chem & Biol 1995, 2:281-290.

49. Lato SM, Boles AR, Ellington AD: In vitro selection of RNA lectins:using combinatorial chemistry to interpret ribozyme evolution.Chem & Biol 1995, 2:291-303.

50. Wallis MG, von Ahsen U, Schroeder R, Famulok: A novel RNAmotif for neomycin recognition. Chem & Biol 1995, 2:543-552.

Received: 14 July 1995.