Volume

Volume 13 Number 5 1985

13 Number
5 1985

Nucleic

Nucleic Acids Research

Acids

A new principle of RNA folding basd on pseudoknotting

Cornelis W.A.Pleij, Krijn Rietveld and Leendert Bosch

Department of Biochemistry, State University of Leiden, Wassenaarseweg 64, 2333 AL Leiden, The
Netherlands
Received 3 December 1984; Revised and Accepted 5 February 1985

ABSTRACT Tertiary interactions involving hairpin or interior loops of RNA can lead to extended quasi-continuous double helical stem regions, consisting of coaxially stacked segments of duplex RNA, bridged by single-stranded connections. This type of compact folding plays a role in various strategic regions of RNA molecules. Their role in ribosome functioning, RNA splicing and recognition of tRNA-like structures is discussed.
INTRODUCTION Insight in the three-dimensional structure of single stranded RNA is essential for a full understanding of its functions. It is striking therefore how little we know about the specific folding of these biopolymers in space. Except for a few tRNAs (1-5) no high resolution structural information is available. Even of 5S RNA, a molecule whose structure has been studied in great detail (6) no definitive model for its secondary structure exists, let alone its tertiary structure. In many cases, the available information has led to proposals for the secondary structure in which single-stranded regions alternate with double-stranded regions. The latter contain the classical Watson-Crick G C and AMU base pairs and sometimes G U base pairs. It is increasingly becoming clear that also other combinations have to be considered (7). In such secondary structure models, characteristic features like hairpin, bulge, interior and bifurcation loops can be discerned beside the normal double-stranded stem regions. Frequently with little experimental data at hand these proposals are based on computer-aided nredictions (8,9) which unfortunately have a limited value only (10). In general, the folding of RNA molecules will reflect a delicate balance between base-base interactions and electrostatic repulsion of the negatively charged phosphate residues. In these foldings an imoortant contribution to the stability of the RNA structure is furnished by the stacking interactions of the base residues (11). Under high salt conditions with sufficient screeno I RL Press Umited, Oxford, England. 1717

Nucleic Acids Research

ing of the charged phosphates, tertiary interactions like triple base pairs of non-Watson-Crick type will lead to the formation of compact domains with hydrophobic cores as found in L-shaped tRNA (1-5). Although the studies of tRNA structure have contributed considerably to our understanding of the three-dimensional folding of RNA it remains to be seen whether the same principles can be extrapolated to other RNA structures. Recently we have found evidence for a new building principle that combines a compact folding of the RNA chain with an increase in base stacking. The principle was brought to light by structure mapping of the 3' termini of a number of plant viral RNAs with tRNA-like properties (12-14). Here we describe this principle in some detail and present evidence for its occurrence in the structure of other natural RNAs. It may play a role in ribosome functioning and RNA splicing.

FOLDING OF tRNA-LIKE STRUCTURES AT THE 3' TERMINI OF VIRAL RNAs The 3' termini of various plant viral RNAs like turnip yellow mosaic virus (TYMV) RNA, brome mosaic virus (BMV) RNA, and tobacco mosaic virus (TMV) RNA are recognized by a variety of tRNA-specific enzymes, e.g. tRNA nucleotidyltransferase and a specific aminoacyl tRNA synthetase (15-16). We recently described structure mapping studies on labelled 3' terminal RNA fragments of all three individual viral RNAs. Models of the secondary structure were proposed based upon enzymatic digestions with RNase Ti,Sl nuclease and the double-strand sDecific RNase from the venom of the cobra Naja naja oxiana and chemical modification of adenosine residues with diethyl pyrocarbonate and cytidine residues with dimethyl sulfate. The cloverleaf structure as found in canonical tRNA appeared to be absent in the tRNA-like structures of the viral RNAs. Similar nroposals were made by Florentz et al. for TYMV RNA (17) and by Ahlquist et al. for BMV RNA (18). We have shown that three-dimensional foldings appeared to have evolved, however, which show a great resemblance with the L-shane of the elongator tRNAs. These models provide a basis for our understanding how these viral RNAs have met the requirements for faithful recognition by tRNA-specific enzymes. For a more detailed discussion of these models, see ref. 12-14. Of great interest here is the construction of the RNA domain which comprises the aminoacyl acceptor arm of the tRNA-like structures. To illustrate this noint the secondary structure of the last 42 nucleotides at the 3' terminus of TYMV RNA is shown in fig. la. Experimental data and sequence comparisons with other plant viral RNAs pointed to a base pairing interaction involving the
1718

Nucleic Acids Research

a
S', AOH
C

b
E
UCA

U I!J3CdACUCCCAt d UCGACC

CGGGUG
UCCGN

6G

CA CC

mAUCCCGU

AAUCGCC

nt. ) Su 42.

Figur 1.i Theedie.Nsonalth foldinge of the aminoacyl aeptformarm of thsetRThe aminoacyl acceptor arm is formed by theia like starutranemofntYMN. ltastin the inict trple Geliand triplent Cro sequeince Ianvovd IInd base pairing arsin illustrthedbys thein dashedaclionesh restwe of3G1 the tRA2i5 strCture is) griven byea soli lie.Noe thre-iensoabseldnce ofane aminoacyl stempformdb barmseoin pairicdng betwee n"cdb the 3'aandd5'psid of the structure.ai stadar tRNA ()Larrangement ftN Tny rmyat. he aminoacyl acceptor fre arm is ycailsakn fromthere boraseoarng intrctmaison beween thatG TYM5 and -C27. (c) Artistr
detai ls about experimental detai ls and a di scussion of these models, see ref. 12-14.
1719

Nucleic Acids Research

triple G sequence G13-G15 with the triple C sequence C25-C27. The existence of the 3 G-C base pairs appears to be crucial for the construction of an aminoacyl acceptor arm which has striking similarities to that of canonical tRNA. Coaxial stacking of stem I and II with that formed bv the 3 G-C base pairs coming from the tertiary interaction (fig. lb) yields an augmented acceptor arm with 12 base pairs, identical to that found in the augmented helix of Tp and amino acid acceptor stems found in yeast tRNA Phe (fig. ld) and almost all other elongator tRNAs. Another essential feature of this Dart of the TYMV RNA structure is the presence of two short stretches of single-stranded RNA comnrisinq 3 (A1Q-U12) and 4 (U21-C24) nucleotides and snanning distances of a quarter and almost half a turn of an RNA duplex resDectively, thereby crossina the helix from one strand to the other. As can be seen in fia. lb these stretches connect two helical segments of the aminoacyl acceotor arm in such a way that a quasi-continuous double helix with 8 base pairs can be formed by the viral RNA chain segment running from A1-C27. Note that these two connecting stretches of RNA are located at roughly the same side of the arm (fig. lc). This may be important for the recognition of the viral 3' terminus by tRNAspecific enzymes (12). We have found similar tertiary foldinas in the 3' termini of the RNAs from the plant viruses BMV and TMV (13,14). We therefore wondered how widespread this peculiar foldina would be in other RNAs.
THE NEW BUILDING PRINCIPLE The principle underlying the construction of the aminoacyl acceptor arms of the viral RNAs can be generalized as deDicted in fig. 2a: a stretch of nucleotides in the loop of a hairoin and bordering the stem region, forms normal Watson-Crick base Dairs with nucleotides of a complementarv sequence located elsewhere in the RNA chain. This type of interaction favours the formation of an elongated double-stranded helix by coaxial stackinn of the two separate double helical segments: stem Si and the stem generated by the tertiary interaction (stem S2). Formally, it is also Dossible to consider stem S1 as the result of a tertiary interaction. This is clarified by other more schematic representations of the same interaction oattern (fig. 2b and 2c). It may already be noted here that intersection of the chords in fig. 2b, representing the two overlappina double helices, is not admissable in current methods for computer-aided predictions of RNA secondary structure, because they give rise to so-called "knotted structures" (8,9). Based on
1720

Nucleic Acids Research

bL

,.-.

s~~~

1> c 5

5
5

Figure 2. The formation of extended double hel ices in RNA chains, based on tertiary interactions of a special kind, shown in four presentations. Si and S2 represent stem regions formed by normal Watson-Crick base pairing and Li and L2 the single stranded RNA regions connecting the double helical segments Si and S2. (a) Conventional representation of the secondary structure in which nucleotides from the hairpin loop adjacent to stem region Si base pair with a complementary region at the 3' side of the hairpin. (b) Schematic illustration of the building principle in a graphical format. (c) Schematic folding. (d) Three-dimensi onal folding, showing the quasi-continuous doubl e stranded helix of 8 base pairs and the two crossing loops connecting the two double helical segments. Our experience with the plant viral RNA structures we here make the imDortant assumption that for all interactions of this kind stem Si and stem S2 are stacked on top of each other in such a way that a quasi-continuous, righthanded double helix is formed, comparable to A-RNA (fig. 2d). This assumption of course is only valid if the single-stranded connecting loops Li and L2 pose no sterical contraints upon this structure. Due to the handedness of the double helix and the polarity of the chain, loop Li and L2 will not be equivalent: loop Li crosses the deep groove and loop L2 the shallow groove of the double helix. This must have consequences for the length and the conformation of each of the two l OOpS. A first insight in the minimal number of nucleotides needed in loop Li or L2 is provided by an analysis of the data which emerge from our three-dimensional models of the tRNA-like 3' termini of the plant viral RNAs. In table 1 we have summnarized these data in terms of the number of base pairs in stem Si and S2 and the number of nucleotides in loop Li and L2. Beside the RNAs studied experimentally we also included related viral RNAs whose 3' terminal sequences are known and which can be folded in similar tertiary structures (12-14, i8, 19). The lowest number of base pairs found in the stem regions is 3, which presumably is also the
1721

Nucleic Acids Research

Table 1. Folding characteristics of the aminoacyl acceptor arms in the tRNAlike structure of some plant viral RNAs Number of base pairs Number of nucleotides in loops in stem regions Si S2 L2 Li
3 5 YMV RNAb 4 3 6 4 3 EMV RNA 3 6 6 2 BMV RNAc 94 7 5 2 TAV RNAd 109 3 MV RNA 4 3 2 3 5 4 CcTMV RNA 3 3 4 3 GTAMV RNA 2 a For definitions and symbols see fig. 2. b Abbreviations used are: TYMV, turnip yellow mosaic virus; EMV, eggplant mosaic virus; BMV, brome mosaic virus; TAV, tomato aspermy virus; TMV, tobacco mosaic virus; CcTMV, cowpea strain of TMV; GTAMV, green tomato atypical mosaic virus. c The BMV related viruses cowpea chlorotic mottle virus (CCMV), broad bean mottle virus (BBMV) and cucumber mosaic virus (CMV) have identical structures
d Cucumber green mottle mosaic virus (CGMMV) and the L and OM strain of TMV have identical structures (14).

(13919 ).

minimal number. The maximum found so tdr is 7. The lower limit for either loop seems to be 2 nucleotides, though bridgina a different number of base pairs. The upper limit is less well defined and can be hundred or even more than one thousand nucleotides (see below). Such large loons of course will possess secondary structure themselves as was obvious in the case of BMV RNA (13). An intriguing outcome is the finding that 2 nucleotides are sufficient to span a distance of 6 base pairs over the deep groove of the double helix. Analysis of the known geometry of the RNA-A helix shows that this is feasible (fig. 3a). Based on the known coordinates for synthetic RNA double helices (20) we have calculated the distance between one phosnhate atom (P6) on one strandand others on the opposite strand (fig. 3b). Depending on the direction chosen, either the deep groove (PO - P_6) or the shallow groove (P - P6) is crossed. (See also legend to fig. 3). Fiqure 3c shows that the distance between two phosphates is minimal when 7 base pairs over the deep groove are bridged (10.1 R). This distance is almost half of that between PO and PO (17.4 i) and similar to that between P0 and P2 on one strand (11.0 R). Bridging the deen groove with 2 nucleotides therefore seems possible. Connecting the two strands over the shallow groove takes at least 16.9 R, while the distances crossing 2, 3 and 4 base pairs do not differ very much and are actually of the same order of magnitude as for PO - PO. Because the
1722

Nucleic Acids Research

b
a ,
,-, I

DEEP GROOVE SHALLOW GROOVE

3'
P
0~~~D
0

P
0-

PO i
0-

|
0

5'

':6 ':4 ri 2 '_0

0<

l,

'6
D_

DEEP
GROOVE

;N P1

.fi

SHALLOW

r*
0

GROOVE

a-

I30
. .
6 8
10

.
-10
-8 -6

.
-4

6 -2 2 4 phosphate number

Figure 3. Distances between Dhosphate atoms in a regular RNA double helix of the A-type. The shortest distance between a fixed phosphate residue on one strand (P6) and the phosphates on the other strand is given thouoh it should be realized that in some cases elements of the double helix like base Dairs are penetrated. (a) Three-dimensional representation showing the deeo and shallow groove of the double helix. Going upwards from phosphate PO which is located opposite to P6 on the other strand, the phosphates are i ndi cated with a positive subscrint, whereas negative ones are used going downwards. (b) Two-dimensional scheme. Vertical bars represent base Dairs, the arrows the shortest distances between the phosphate residues involved. (c) Graphical representation. Distances were calculated based on Dublished coordinates of a regular RNA-A double helix (20).
minimal number of nucleotides required for closing a hairDin loop is accepted to be three (21), we assume that this number is also three for 100D L2. This is actually observed in the case of TYMV RNA (see table 1). Bridging the shallow groove with 2 nucleotides (see TMV RNA) may therefore require either a distortion of the double helix and/or a change in the regular nucleotide conformation(s) in the crossing loop. Stretching of the ribose ohosphate backbone, however, by changes in the ribose ouckering (C3 endo to C2 endo) and torsion angle y (C4 C5) will Drobably be not sufficient (22). If two nucleotides can cross the shallow groove, one has to envision that even one nucleotide might suffice for the deep groove. In the latter case it is conceivable that the two phosDhates concerned (PO and P or P7, fig. 3) miaht come into closer proximity by a bending of the double helix over this groove as found to be possible for DNA (23). Duolex RNA, however, is renorted to be less flexible than the DNA double helix (24,25). On the other hand, an
-

1723

Nucleic Acids Research

appreciable contraction of the wide groove upon bindinq of spermine to tRNA has been reported (26), resultina in a distance of only 8.6 R between phosphates on the oDposite sides. It remains to be seen whether such a one nucleotide connection is realized in RNA molecules. Whatever the number of nucleotides needed for crossing the double helix, the phosohate ribose backbone is generally forced to make a sharD turn at the junction of loop todouble helix (fig. 2d). Model building studies of the aminoacyl acceptor arm of TYMV RNA (Nicholson molecular models, Labquin, Readinn, England) showed that the connecting loops of 3 and 4 nucleotides resoectively (fig. lb) can be readily attached to the double helix provided that the P-0 torsion angles a and C of the phosphodiester bond concerned are adjusted. Such an abrupt change in the direction of the phosphodiester chain has been observed in the crystal structure of tRNAPhe from yeast (1,2). Interestinaly, this so-called "U-turn" or "ir3turn" (27,28) involvinq a channe in torsion angle a does occur in the anticodon loop immediately upstream of a helical stack of 5 bases out of 7 at the 3' side of the loon (including the anticodon). The resulting structure is reminiscent of that described in this paper because the two nucleotides at the 5' side of the anticodon "bridge" 5 stacked nucleotides over what can be considered to be the deep' qroove. It appears that the invariant U residue, near the sharD turn is involved in a peculiar tertiary hydrogen bonding (27). Whether similar tertiary interactions are present or even required in the foldinas described here is unknown. However, as suggested earlier (12) it cannot be excluded that tertiary hydrogen bonding or triole base pair interactions between the stem regions and the crossing loops take Dlace and will help to stabilize this special type of folding. So far, our enzymatic diaestion and chemical modification studies point to a good accessibility of most of the bases in the loops (12-14, our unpublished observations). Moreover sequence comparisons between various aminoacylatable plant viral RNAs indicate that these loops are not rich in U residues and are prone to base substitutions (12-14) suggesting a connecting function only.
ENERGETIC IMPLICATIONS OF THE NEW BUILDING PRINCIPLE An important aspect of the tertiary foldings described so far is the question of their stability. For the secondary structure of RNA a reasonable estimate of the stability of various elements like hairpins can be made on the basis of the well known Tinoco rules (29). Usually a considerable decrease in free energy is obtained upon increasing the number of base pairs
1724

Nucleic Acids Research

and the length of the stem regions. It probably is the main contribution to the stability of the foldings shown here (fig. 2). This is illustrated in the case of the aminoacyl acceptor arm of TYMV RNA which consists of three double helical segments (fig. lb). Two of them: hairpin I and II have a A G of -5.6 and -9.1 kcal/mol, respectively, according to the Tinoco rules. The entire arm consisting of the loop of hairpin II and a regular helix of 12 base pairs, which includes the triple G-triple C interaction, gives a A G of -34.2 kcal/mol. Such estimates obviously suffer from the uncertainties concerning the influence of the connecting loops. Presumably the latter will give a positive contribution to the A G depending mainly on the length and base sequence of the loops but up till now one can only speculate about the magnitude of these effects. In fact this situation is comparable to that of the bifurcation loops in classical RNA secondary structure (9). Insight in the thermodynamic properties of these structures will be gained only upon a detailed study of either natural RNA fragments like the tRNA-like 3' termini of viral RNAs or of synthetic oligonucleotides able to form these knotted structures. Studies to this aim are under way in our laboratory. The experimental results obtained so far indicate that the "tertiary" interaction is disrupted first on lowering the Mg2+ concentration in the case of the three plant viral RNAs studied. Interestingly a new hairpin is formed at the 3' end of BMV RNA upon partial denaturation of the aminoacyl accentor arm (13) pointing to an interesting conformational transition.
THE NEW BUILDING PRINCIPLE PLAYS A ROLE IN RIBOSOME FUNCTIONING AND RNA SPLICING During the last few years our insight in the secondary structure of ribosomal RNAs has increased. In particular phylogenetic comparisons showing preservation of various structural elements in the rRNAs of various organisms have lent strong support to the models proposed ( 30,31). In this context it is of interest that two helices at the 5' end of E. coli 16S rRNA, designated helix 1 and 2 (fig. 4a), can be recognized that are conserved in various species but whose coexistence is puzzling. According to Maly and Brimacombe (32), model building will be required to establish to what extent they can exist simultaneously. In the light of the foregoing it will be clear, however, that these helices can form an extended double helix of 9 base pairs (figure 4b) and an even longer quasi-continuous double helix when helix 3 is stacked on top of the 9 base pairs of helix 1 and 2 with
1725

Nucleic Acids Research

a
U .0u

~
Au
c a
A G f0 ^ G U

c
'

EXON

Pi1;

AC

'5....u 0

'A

0 0

AA

-A 0

U4

-UA

AU-

UA C Um3
^^ uAA~

5,

EXON a

.NTRON_

Figure 4. Models of RNA secondary structures showinq the possible role of " pseudoknots " i n RNA spl icing and ri bosome functi oni ng. (a) Part of the secondary structure of the 165 ribosomal RNA from E. coli. The formation of stem reaion 1 by base pairing between the two comnlemeintary boxed regions was confirmed by phylogenetic comparisons. For further details see ref. 31 and 32. (b) Coaxial stacking of stem regions 1 and 2 in 165 rRNA based on the new building principle described in the text. Helical segment 1 might be stacked on helix 3 by including the G A base pair at the end of helix 3. (c) Schematic representation of the generalized secondary structure model of fungal mitochondrial introns according to Davies et al. (34). Only key regions of the intron structure are given. The thiRE Te reDresents the exoni reoions. The thin line the intron, the arrows show the splice sites and the zigzag line the internal guide sequence. For a full description of this model and for the symbols used see ref. 34. (d) Coaxial stacking of the double helix segments P1, P1O and P2 which brings the two splice sites into close proximity.
either an A residue bulged out or with the incorporation of a GA pair. If this model is correct, it does not only offer a rational explanation for the enigmatic coexistence of helices 1 and 2 but it also implies that the molecule has to undergo a considerable conformational change for adopti ng another structure. Such a conformati onal transition may play an important role as a switch in ribosome functioning. So far this novel feature of RNA folding has only been recognized at the extremities of the RNA chains: at the 3 end of viral RNA and at the 5 end of rRNA. That it may also be involved in the folding of internal parts of the RNA chain is suggested by the recently proposed mechanism for autocatalytic RNA self-splicing. The latter phenomenon has been reported by Cech for the nuclear rRNA precursor of Tetrahymena and may also be
1726

Nucleic Acids Research

involved in the splicing of fungal mitochondrial RNA transcripts (33). Phylogenetic comparisons of the intron and exon sequences involved, led to a proposal for a generalized secondary structure model as shown in fig. 4c (34). Crucial in this model is the interaction between a few nucleotides in the loop closing the P1 stem and a complementary region of the 3' exon. This interaction, fully obeing the building principle outlined above (compare fig. 2a) will bring the splice junctions in close proximity (fig. 4d). It may create the stereochemical condition for qenerating an active site (35) and for engaginq in a transient state of the chain nucleotides involved in the cleavage and ligation reactions. Clearly the detailed mechanism remains to be elucidated but it may be annreciated here that no major conformational changes are needed durina the entire solicing event and that cleavage and ligation may take place autocatalytically in one concerted reaction mechanism. It is also noteworthy that stacking of stem P2 on ton of stem P1 and stem P10 as we have proposed in fin. 4d results in the formation of a quasicontinuous double helix consisting of three segments similar to the folding of the aminoacyl acceptor arm of TYMV RNA (fig. lb). The so-formed "orecise alignment structure" and "internal guide sequence" (34) are strongly reminiscent of the structure proposed for the sn RNP-mediated splicing of hn RNA (36).
VARIANTS OF THE BUILDING PRINCIPLE Having noted that the buildina nrinciple of fin. 2 most likely is used in the folding of natural RNAs, the question may be raised whether variants of this principle are theoretically possible. Figure 5 illustrates an examole showing that this is the case. Here the nucleotides of an interior loon base pair with the nucleotides of a comnlementary sequence distantly located on the chain. The arrangement of fig. 5 gives rise to the formation of a quasi-continuous helix consisting of three segments and as such is reminiscent of the aminoacyl acceptor arm of TYMV RNA (cf. fig. lb). Like in TYMV RNA, the two connecting looos are also found at the same side of the double helix. One of these looDS bridges the deen groove like loom Li of fig. 2c. It will be interesting to see whether nature has used this variant to construct aminoacyl acceptor arms of viral tRNA-like structures so far not identified. It is gratifying to note, however, that this variant of tertiary interaction fits in the intron model of fig. 4c near the strongly conserved
1727

Nucleic Acids Research

5'

Figure 5. Hypothetical structure of an aminoacyl acceptor arm in a tRNA-like structure. Coaxial stacking of 12 base pairs can be obtained by base pairing of nucleotides in the interior loop and a complementary region at the 5' side. The structure gives rise to two connecting loops of which one (most proximal to the 3' terminus) crosses the double helix. The hairpin loop containing 7 nucleotides corresponds to the Tip-loop of classical tRNA (see also fig. 1). The exact number of base pairs in the various double helical segments is here chosen arbitrarily as 4 in each segment, but may vary in principle from 3 to 6.
and presumably functionally important core of the structure. We suggest that the helix formed by base pairinq of E' and F stacks coaxially with helix P8. In principle other variants may be envisaged involving hairpin, interior, bulge, and bifurcation loops. They all notentially lead to the formation of extended double helices.

KNOTTED STRUCTURES IN RNA AND COMPUTER-AIDED PREDICTIONS OF RNA SECONDARY STRUCTURE All intramolecular tertiary interactions in RNA described above are examples of what have been called "knotted structures" (8) thounh of a special kind. If the tertiary interactions would comnrise one turn of adouble helix or more, the formation of a so-called "real knot" in the RNA chain is conceivable by the threading of one of both free ends through a loop (loop Li for the 3' end and loop L2 for the 5' end, fig. 2c). No true knots in RNA molecules however have been reported. We therefore orefer the term "pseudoknot" coined by Studnicka et al. (8), for the tertiary interaction we found in the plant viral RNAs. So far, not more than 6-7 base nairs are involved in this pseudoknot formation, which is well below one full turn of an RNA duplex (11 base pairs). Helices of this length might prevent that
1728

Nucleic Acids Research

the tertiary structure (locally) becomes too riaid, thereby enabling these interactions to play their outative role as functional conformational swi tches. As pointed out earlier (37) the right-handed double helix of the pseudoknot requires a compensating net left-handed twist in the connecting RNA regions (loops Li and L2 of fig. 2). If these regions are relatively short, they might have peculiar conformational pronerties and might serve as unique recognition sites for proteins. Knotted structures or pseudoknots were considered in computer-aided predictions of RNA secondary structure, but were not taken up in the algorithms for reasons of simplicity or because no examples of this tyoe of interaction were known (8,9,38,39). Nevertheless Salser tried to indicate possible tertiary interactions starting from a predicted secondary structure of globin mRNA (40). Interestingly some of the enumerated possible tertiary interactions are of the type discussed in this paper. As a first step to introduce pseudoknots (especially of the type of fig. 2) into a modified algorithm for RNA structure Drediction, we searched for possible pseudoknots in sequences like that of TYMV RNA, MS2 RNA and 16S rRNA from E. coli. Numerous examples were found though this number decreased upon restrictions in the number of base nairs in the stem and nucleotides in the loop regions. Whether they really occur depends on the stability of the other elements in the RNA structure and on factors like kinetic parameters which might trap an RNA region in a localized enerqy minimum from which it cannot readily escape at the expense of a thermodynamically more favourable long range interaction. This all together poses a formidable oroblem in solving the secondary and tertiary structure of RNA molecules by computeraided predictions alone, also because no energy rules are at hand for these pseudoknotted structures.

CONCLUDING REMARKS In this paper we have shown that pseudoknots are present in RNA molecules. In itself it is not surprising to see that under physiological conditions long range base pairing interactions in the three-dimensional folding of RNA do occur. It may be stressed here, however, that these tertiary interactions involving hairpin or interior loops can give rise to long extended double helical stem regions. More work is needed to get insight in their precise folding, their stability and their spread in natural RNAs.

1729

Nucleic Acids Research

ACKNOWLEDGEMENTS We thank Dr. P.H. van Knippenberg for stimulating discussions and critical reading of the manuscript and Mr. J.H. van Dierendonck for drawing the three-dimensional RNA foldings. We are indebted to Drs. C. Haasnoot and C.W. Hilbers for helpful discussions on the data oresented in fig. 3.
REFERENCES 1. Robertus, J.D., Ladner, J.E., Finch, J.T., Rhodes, D., Brown, R.S., Clark, B.F.C. and Klug, A. (1974) Nature 250, 546-551. 2. Kim, S.-H., Suddath, F.L., Quigley, G.J.,Th!cPherson, A., Sussmann, J.L., Wang, A.H.J., Seeman, N.C. and Rich.AI1974) Science 185, 435-440. 3. Moras, D., Comarmond, M.B., Fischer, J., Weiss, R.T,liierry, J.C., Ebel, J.P. and Giege, R. (1980) Nature 288, 699-674. 4. Woo, N.H., Roe, B.A. and Rich, A.J(980) Nature 286, 346-351. 5. Schevitzs. R.W., Podjarny, A.D., Krishnamachari, IT Hughes, J.J. and Sigler, P.B. (1979) Nature 278, 188-190. 6. Singhal, R.P. and Shaw, J.K71980) in "Progress in Nucleic Acids Research and Molecular Biology", Cohn, W.E. (1983) ed. Vol. 28, pn. 177-202, Academic Press, New York. 7. Traub, W. and Sussmann, J.L. (1982) Nucleic Acids. Res. 10, 2701-2708. 8. Studnicka, G.M., Rahn, G.M., Cummings, I.W. and Salser, W.-A. (1978) Nucleic Acids Res. 5, 3365-3387. 9. Zuker, M. and StiegTer, T. (1981) Nucleic Acids Res. 9, 133-148. 10. Auron, P.E., Rindone, W.P., Vary, C.P.H., Celentano, J.J. and Vournakis, J.N. (1982) Nucleic Acids Res. 10, 403-419. 11. Cantor, C.R. and Schimmel, P.R.71980) "Biophysical Chemistry", Part 1 Chapter 6, W.H. Freeman and Company, San Francisco. 12. Rietveld, K., van Poelgeest, R., Pleij, C.W.A., van Boom, J.H. and Bosch, L. (1982) Nucleic Acids. Res. 10, 1929-1946. 13. Rietveld, K., Pleij, C.W.A. and Bosch, L. (1983) EMBO J. 2, 1079-1085. 14. Rietveld, K., Linschooten, K., Pleij, C.W.A. and Bosch, L. (1984) EMBO J. 3, 2613-2619. 15. Tfaenni, A.-L., Joshi, S. and Chapeville, F. (1982) in "Progress in Nucleic Acids Research and Molecular Biology", Cohn, W.E., ed., Academic Press,New York. 16. Hall, T.C. (1979) Int. Rev. Cytol. 60, 1-26. 17. Florentz, C., Briand, J.P., Romby, TP7, Hirth, L., Ebel, J.P. and Gieqe, R. (1982) EMBO J. 1, 269-276. 18. Ahlquist, P., Dasgupta, R. and Kaesberg, P. (1981) Cell 23, 183-189. 19. Joshi, R.L., Joshi, S., Chapeville, F. and Haenni, A.-L.71983) EMBO J. 2, 1123-1127. 20. Arnott, S., Hukins, D.W.L. and Dover, S.D. (1972) Biochem. Biophys. Res. Commun. 48, 1392-1399. 21. Gralla, rF and Crothers, D.M. (1973) J. Mol. Biol. 73, 497-511. 22. The description of the conformation of oolynucleotie chains is according to the new IUPAC-IUB notation (Eur. J. Biochem. (1983) 131,

23. Levitt, M. (1983) Cold Spring Harbor Symp. Quant. Biol. 47, 251-262. 24. Bolton, P.H. and Kearns, D.R. (1979) J. Am. Chem. Soc. 1U1, 479-484. 25. Arnott, S., Chandrasekaran, R. and Selsing, E. (1975) iTFwStructure and Conformation of Nucleic Acids and Protein-Nucleic Acid Interactions" Sundaralingam, M. and Rao, S.T., eds, pp. 577-596, University Park Press, Baltimore.
1730

9-15).

Nucleic Acids Research

26. Quigley, G.J., Teeter, M.M. and Rich, A. (1978) Proc. Natl. Acad. Sci. USA 75, 64-68. 27. QuigT-y, G.J. and Rich, A. (1976) Science 194, 796-806. 28. Saenger, W. (1984) "Principles of Nucleic Acid Structure", Chapter 15, Springer Verlag, New York. 29. Tinoco, I., Borer, P.N., Dengler, B., Levine, M.D., Uhlenbeck, O.C., Crothers, D.M. and Gralla, J. (1973) Nature New Biol. 246, 40-41. 30. Noller, H.F. and van Knippenberg, P.H. (1983) in "Genes: SIfucture and Expression", Kroon, A.M, ed., pp. 71-99, Wiley and Sons, New York. 31. Brimacombe, R., Maly, P. and Zwieb, C. (1983) Progress Nucleic Acid Research and Molecular Biology 28, 1-48. 32. Maly, P. and Brimacombe, R. (191) Nucleic Acids. Res. 11, 7263-7286. 33. Cech, T.R. (1983) Cell 34, 713-716. 34. Davies, R.W., Waring, RB., Ray, J.A., Brown, T.A. and Scazzocchio, C. (1982) Nature 300, 719-724. 35. Altman, S. (198WJ Cell 36, 237-239. 36. Lerner, M.R., Boyle, J.AT, Mount, S.M., Wolin, S.L. and Steitz, J.A. (1980) Nature 283, 220-224. 37. Cantor, C.R. iiFTRibosomes", Chambliss, G., Craven, G., Davies, J., Davis, K., Kahan, L. and Nomura, M., (1980) eds., nn. 23-49, University Park Press, Baltimore. 38. Nussinov, R. and Jacobson, A.B. (1980) Proc. Natl. Acad. Sci. USA 77, 6309-6313. 39. Quigley, G.J., Gehrke, L., Roth, D.A. and Auron, P.E. (1984) Nucleic Acids Res. 12, 347-366. 40. Salser, W. (1T77) Cold Sprinq Harbor Symp. Quant. Biol. 42, 985-1002.

1731