RNA 1:425-436

RNA (1995), 1:425-436. Cambridge University Press. Printed in the USA.Copyright 1995 RNA Society.

RNA structural patterns and splicing: Molecular basisfor an RNA-based enhancer

D. LIBRI,1'2 F. STUTZ,1 T. McCARTHY,1 and MICHAEL ROSBASH11 Department of Biology, Howard Hughes Medical Institute, Brandeis University, Waltham, Massachusetts 02254, USA2 Institut Pasteur, Unite de Gdndtique Moldculaire du Ddveloppement (C.N.R.S. URAl 947), 25 rue du Dr. Roux,75724 Paris Cedex 15, France

ABSTRACTEfficient splicing of the 325-nt yeast (Saccharomyces cerevisiae) rp5l b intron requires the presence of two shortinteracting sequences located 200 nt apart. We used the powerful technique of randomization-selection to probethe overall structure of the intron and to investigate its role in pre-mRNA splicing. We identified a number ofalternative RNA-RNA interactions in the intron that promote efficient splicing, and we showed that similar basepairings can also improve splicing efficiency in artificially designed introns. Only a very limited amount ofstructural information is necessary to create or maintain such a mechanism. Our results suggest that the basepairing contributes transiently to the spliceosome assembly process, most likely by complementing interactionsbetween splicing factors. We propose that splicing enhancement by structure represents a general mecha-nism operating in large yeast introns that evolutionarily preceded the protein-based splicing enhancers ofhigher eukaryotes.

Keywords: pre-mRNA; splicing; structure; yeast

INTRODUCTION

In nuclear pre-mRNA splicing, RNA-RNA interactionsamong snRNPs as well as between the snRNPs and thepre-mRNA substrate play a major role in recognitionof proper splicing signals and in maturation of thespliceosome (Moore et al., 1993). A still unresolved is-sue, however, is the extent to which intramolecularRNA-RNA interactions in the pre-mRNA affect thesplicing process. In higher eukaryotes, it has beenshown that artificial hairpins can hinder splice site ac-

cess both in vivo and in vitro (Eperon et al., 1986, 1988;Fu & Manley, 1987), but only one example is knownof a naturally occurring secondary structure that affectssplice site selection in vivo (Clouet-d'Orval et al., 1991;Libri et al., 1991). It has also been shown that second-ary structure formation can improve the in vitro splic-ing efficiency of an adenovirus intron, possibly byreducing the spacing between the branch point and the3' splice site (Chebli et al., 1989).

In yeast (Saccharomyces cerevisiae), short artificial hair-pins that sequester the 5' splice site or the branch point

are efficient inhibitors of splicing both in vivo and invitro (Goguel et al., 1993). There is also a natural case

in which a 3' splice site is skipped by virtue of beingsequestered in a stem (Deshler & Rossi, 1991). Theexistence of natural pre-mRNA intramolecular inter-actions that favor the splicing of yeast introns was ini-tially suggested based on sequence analysis of yeastintrons (Parker & Patterson, 1987). These authorspointed out that the yeast introns could be divided intotwo main size classes: short introns (on average 90 ntlong) and large introns (400 nt on average). They sug-gested that large introns contain short complementarysegments whose interaction would have the effect ofshortening the distance between the donor site andthe branch point to an "operational" length similar tothe one found in small introns. Newman (1987) sub-sequently showed that the interaction of two short re-

gions in the CYH2 intron was essential for optimalsplicing efficiency and suggested that the role of the in-teraction was to prevent one of the interacting segmentsfrom somehow inhibiting splicing. More recently, theribosomal protein rp5lb gene intron in yeast has beenshown to contain two sequence segments whose inter-action contributes to enhance splicing efficiency of thisintron and influences splice site choice in competitionexperiments (Goguel & Rosbash, 1993).

425

Reprint requests to: Michael Rosbash, Department of Biology,Howard Hughes Medical Institute, Brandeis University, 415 SouthStreet, Waltham, Massachusetts 02254, USA; e-mail: [email protected].

D. Libri et al.

Important questions remained unanswered by all ofthe previous studies. (1) The position of the invertedrepeats (usually located some 200 nt apart) raises thequestion of the thermodynamic stability of such an in-teraction and suggests that the stem might be part ofa more complex secondary structure. (2) The exact roleplayed by this secondary structure in improving splic-ing efficiency is unclear. (3) The general applicabilityof splicing enhancement by secondary structure toother yeast introns is also uncertain.Madhani and Guthrie (1994) combined in vitro ran-

domization and in vivo selection techniques to inves-tigate RNA-RNA intermolecular interactions in thespliceosome. We have extensively applied this meth-odology to investigate the role played by the rp5lb in-tron intramolecular base pairing as well as its overallstructural context. Our results point to the existence ofa general mechanism by which intramolecular RNA-RNA interaction between the two ends of large yeastintrons enhances splicing efficiency.

RESULTS

To investigate the role of pre-mRNA secondary struc-ture in splicing of large yeast introns, we employed areporter gene whose expression is dependent on splic-ing. The rp5lb intron was inserted in the coding regionof the CUP1 gene whose product confers copper resis-tance to yeast (Fig. 1A). Under these conditions, splic-ing is the limiting factor in the expression of the CUP1product, and copper resistance of the strain correlateswell with the excision efficiency of the intron (Stutz &Rosbash, 1994; our unpubl. data). The stem involvedin splicing of the rp5l intron is shown in Figure 1B. Wedesignated UB1 (upstream box 1) and DB1 (down-stream box 1) the upstream and downstream constit-uents of the stem.To investigate the sensitivity of splicing efficiency to

minor alterations in the stem, changes of 3 nt (3mUB1),5 nt (5mUB1), 4 nt (4mUB1), 6 nt (6mUBI), and 8 nt(8mUB1) were introduced in UB1; the sequence in DB1was mutated in 3mDB1 and 5mDB1, which carry thecompensatory mutations of 3mUB1 and 5mUB1, re-spectively (Fig. 2A). All the single mutants listed aboveare expected to weaken or disrupt the secondary struc-ture, whereas the double mutants (3mUB1/3mDB1 and5mUB1/5mDB1) should restore the stem. This set ofmutants was tested both for copper sensitivity and forthe pre-mRNA/mRNA ratio in primer extension exper-iments (Fig. 2C and data not shown).Mutation of three or five base pairing nucleotides on

both sides of the stem had an inhibitory effect on splic-ing of the intron. The inhibition was partially relievedin the double mutant 5mUB1/5mDB1, although not tolevels sufficient to support copper-resistant growth aswell as the wild-type construction (5mUB1/5mDB1grows only as 5mUB1 although better than 5mDB1;

A50 at

5@ss UB1ISu L -

S.,I

RplbintronDB1 BP 3'55

-325t

:cup-:Mr _ |..1.T.

_..._s _

B5'ss

UBI

GGAUAAUGAUGGACUG GCACCUAUU CUAUC%ACACG G4

(BP 'DB1'

FIGURE 1. A: Schematic diagram of the CUP1/rp51b reporter con-struct. Positions of UB1, DB1, branch point (BP), and splice sites areindicated. Only the spliced product gives a functional CUP1 protein.B: Schematic drawing of the UB1/DB1 interaction, as reported byGoguel and Rosbash (1993). Nucleotide numbering starts with thefirst nucleotide in the intron. Note that nucleotides 33 and 246 areA and C instead of G and C, respectively (contrary to a previous re-port [Abovich & Rosbash, 1984]).

Fig. 2C). However, no compensation was observed forthe double mutant 3mUB1/3mDB1. Even more surpris-ing was mutant 8mUB1, which caused no significantdrop in splicing efficiency despite containing an exten-sion of the same mutations introduced in 3mUB1 and5mUB1.One possible explanation for the unexpected splic-

ing phenotypes is that these mutations cause signifi-cant changes in pre-mRNA structure, which affectssplicing efficiency either positively or negatively. Forexample, the mutations 3mDB1 and 5mDB1 might in-duce the formation of an alternative inhibitory stemnear the 3' splice site (Fig. 2B), which would be favoredover the reestablishment of the original base pairing.The presence of an appropriately located reverse tran-scriptase strong stop with RNA templates containingthe DB1 mutations (Fig. 2B and data not shown) sup-ports this hypothesis. On the other hand, the mutant8mUB1 might exhibit an unexpected efficient splicingphenotype because the mutated bases can find a dif-ferent and permissive base pairing partner as con-firmed below.

Multiple and alternative intramolecular RNAinteractions affect splicing efficiency

If multiple foldings of the intron are compatible withefficient splicing, their analysis might be informativeabout the mechanism that relates them to splicing ef-

426

RNA secondary structures and splicing enhancement

A

B

8mUbl5mUb1 cuacuacu3mUbl\*I 6mUbl

cacu cu~~cacdmb

i7CGGAUM GAUGGACJ GG C

AACCUAUU CUAUC%ACACGGUgalu 'I3mDbl 241

3 L.. Jgaugg 5mDbl

5' SmDbl BP

UAUUCUAUC gguagCAAUGGUPUGAAM)AUGGGALUA UUAUCGULAUUAG UUU

3'

-4 -

1.0mm

1.2mM 1.4mM

FIGURE 2. A: Positions of the mutations introduced in UB1 and DB1.Wild-type sequence is indicated in capital letters. Mutant 3mUB1 in-troduces four changes instead of three due to a mistake in the origi-nal published sequence (Abovich & Rosbash, 1984). However, we

decided to maintain the name to avoid confusions with 4mUB1. Notethat an A to G change in position 33 does not affect splicing efficiency(see Fig. 3). B: Proposed inhibitory secondary structure for mutant5mDB1. Mutations introduced are shown in lowercase letters; thebranch point (BP) is indicated and the AG acceptor is underlined.Arrow indicates the position of a strong reverse transcriptase stopin primer extension experiments (data not shown). C: Copper growthassay for the mutants of the UB1/DB1 stem. Left part of the figureis a schematic drawing showing the positions of the different mu-tants. CuS04 concentration is indicated under each panel. Thehigher copper-resistant phenotype displayed by 5mUB1 (and5mUB1/5mDB1) as opposed to the other supposedly less severe mu-

tants of the stem is most likely due to imperfect base pairing of the5mUB1 with DB3 (data not shown). The latter mutants can be fullydistinguished from the Wt and 8mUB1 at 1.6mM Cu2+ (see Fig. 6B).

ficiency. To this end, the same 8 nt mutated in 8mUB1were randomized, and the pool obtained was intro-duced into yeast. A similar experiment was done byrandomizing the 5 nt mutated in 5mUB1 (Fig. 3). Thepools were selected by yeast growth in copper-contain-

ing liquid media or by growth on copper plates, andtheir evolution was monitored by sequence analysis atintermediate time points of the liquid assay (data notshown).The pool containing five randomized bases gave an

essentially homogeneous population containing mainlythe wild-type sequence (Fig. 3). Analysis of the clonesselected on copper plates confirmed this result. Thepool containing eight randomized bases, however,gave rise to "winning" sequences that were assignedto four families (named selected box 1-4, SB1-SB4),present to different extents in the total pool.One of the families, SB1, contained sequences that

are variants of the wild-type UB1. A second family(SB3) contained sequences identical to that of 8mUB1.Analysis of the two remaining families and inspectionof the sequence of the rp5lb intron revealed the exis-tence of potential base pairing regions for every fam-ily. The proposed stems are shown in Figure 3 and arecompatible with the nucleotide variations observedwithin the families. (Only SB3, with one family mem-ber, has no variation.) Note that family SB4 is predictedto interact with a sequence immediately downstreamof the UACUAAC box.

In Figure 4A, SB1-SB4 are represented by differentshaped boxes, whereas the putative complementary re-gions are named DB1-DB4 and drawn as matchingboxes, according to their interacting counterparts.These interactions have been tested by mutating everydownstream box, either in the presence of the match-ing upstream selected box or in the presence of thewild-type sequence UB1. The plasmids were tested fortheir ability to confer a copper-resistant growth pheno-type to yeast as well as by primer extension (Fig. 4Band data not shown).Confirming the importance of the pairing inter-

actions, mutations in DB2, Dj3, and DB4 impairedsplicing in the presence of SB2, SB3, or SB4, respec-tively; the same modifications in a wild-type (UB1) con-text did not reduce splicing efficiency (as expectedbecause the original base pairing [UB1/DB1] should notbe affected by the modifications). In some experiments(Fig. 4B), mutation of DB4 seemed to affect splicing ef-ficiency also in the presence of the wild-type sequencein the upstream box (UB1), although less markedlythan in an SB4 context. Given the proximity of DB4 tothe UACUAAC box, it was not unexpected that somesequences in this position might have a general nega-tive effect on splicing efficiency. To ensure that the ob-served effect of mutating DB4 in the presence of SB4was indeed a result of disrupting of their base pairing,we constructed two new pools of sequences in whichrandomization of eight bases in the DB4 region (db4N)was combined either with the wild-type sequence inUB1 (Wt/db4N) or with SB4 (SB4/db4N).The comparison of the growth curves, the plating ef-

ficiency of the two pools in the presence of copper, and

427

D. Libri et al.

5'ss UblN.

C4"rCGgaugaoUGAUGGA UG GGACCUAUU CL4JACACG G4

3'ss

f 5N Randomization/selection

5'5 gg(4) Sbl (10)Cgfl3 C GG ',;'G,aauaagaUGGA UG GC

AA CCUAUUCUAUCLhACA C G Gq _1P 'ss bi

SbZ5's

U gC(5)C%CUGGaaaaaUUGG41L9A GCUUL%GUUAACCAA

246 Db 24

BP ss

FIGURE 3. Scheme of the randomization-selec-tion experiment performed on UBL. Position ofthe randomized nucleotides (5N or 8N) is indi-cated in the central part. Arrows indicate thewinners of the selection, one essentially homo-geneous sequence for the 5N and four families forthe 8N randomization. The base pairing inter-actions of all the winners are indicated. For eachfamily, the most represented clone is shown andthe variations found in the second and third mostrepresented sequences are also indicated. Thenumber of times that independent clones withthe same sequence were isolated is indicated inparentheses.

the copper sensitivity of randomly chosen individualclones confirmed the SB4/DB4 proposed interaction(data not shown). Analysis of individual sequencesfrom the two pools after copper selection demonstratedthe absence of a significant evolution of the Wt/db4Npool but revealed the existence of strong selection pres-sure on the DB4 region in the presence of SB4. Of thethree "winning" families in the SB4/db4N pool, one re-

sembles closely the starting sequence (wild-type DB4),a second contains sequences that can base pair with an

alternative site in the intron, and the third containshighly U-rich sequences (see Discussion).

How complex is the pre-mRNA structure?

The middle of the intron (the roughly 200 nt betweenUB1 and DB1) might also be structured and contributesignificantly to the free energy of intron folding. Forexample, the sequence in UB1, when submitted torandomization-selection, might only "find" base pair-

ing partners in the proximity of DB1 because of addi-tional structural features that juxtapose the two endsof the intron. The putative presence of a more complexintron structure predicts that other regions of the in-tron will be sensitive to mutations. Because only threemutations in the stem defined by UB1 and DB1 in-duced a significant drop in splicing efficiency (mutants3mUB1 and 3mDB1) and even a single nucleotidechange could modify the copper sensitivity in some ex-periments (data not shown), we reasoned that lightmutagenesis of the whole intron should be sufficientto perturb optimal folding.

Therefore, we constructed a "doped" pool in whichthe sequence of the wild-type intron was mutagenizedat the level of about 3-6% per position in two indepen-dent experiments. Under these conditions, about 75%,40%, and 15% of the sequences have at least one, two,or three changes, respectively, in any stem of lengthcomparable to the UB1/DB1 interaction. The splice sitesand the UACUAAC box were excluded from the mu-

G?U'CG JGA UG GCACCUAUU CUAUCUIJACACGG4

3'ss Dbl

| 8N Randomization/selection

S'Ss

NC Sb3 (2)

uAGAUGAUGAAG U ADb3 191

BP 3Ss

51ss Sb4 (8)

Nb4s iAUGAAGCCGauuuacgaUGGe4WAC UCUGUUGAAU UCA BP

C ~DW4

428


A SB1 (wt-like)E SB2

E2J SB3

E SB4ATG 5'ss AAw

DB3 DB2 DB1 DB4nr,L rb- rnn r'-'-i AG clones Growth

A UACUAAC

r--

Wt/db2

SB2/Wt

SB2/db2

Wt/db3

SB3/Wt

SB3/db3

Wt/db4

SB4/Wt

SB4/db4

+

+

+

21l hn2! :00 .'0%'I

pre-mRNA

FIGURE 4. A: Schematic drawing showing the sequencesselected in UB1 (SB1-SB4) arid their interacting partners(DBI-DB4) in the downstream portion of the intron. Ev-ery family of selected clones is indicated by a differentlyshaped box and the corresponding downstream elementsare indicated as matcing boxes. The effect of every down-stream box mutation, in the presence either of the match-ing upstream box or the wild-type UBI, is shown in thesummary of the copper sensitivity of the different dones.Mutated sequences are indicated by a rectangle and low-ercase characters (e.g., Wt/db2 contains a mutation in theDB2 box, and is otherwise wild type). B: Primer extensionanalysis of RNAs from the selected dones and their mu-tant derivatives. Last lane (labeled Wt/db4) comes froma different experiment and shows a somewhat differentpre-mRNA/mRNA ratio (see text).

mRNA

tagenesis, and the pool complexity after transfer toyeast (which is the limiting factor) was estimated to beon the order of 4* 104. Subsets of the pools wereplated and selected on copper, and a total of 50 cloneswere fully sequenced. "Cold" mutagenic spots are ex-

pected in regions sensitive to mutations for functionalreasons (as in RNA double-stranded segments) and inregions modified to a lesser extent in the starting pool,either as a result of a systematic positional bias (i.e., theTaq polymerase might systematically mutate different

B

429

D. Libri et al.

A

CS cs Cs

r 0 _ 0 ' 0 _ 0 r CD w 0 _ e r 0_ 0 * 0 N e 0 N 0X0 C 0) rT- N

r ivr_r_- N_ _ cN

"G" clonesN da'C"GA" clones

R . I me .1I

DB1 CS

L. D _ Lw 0 r C '- CO* 10 N 0 0 rN N N N X 0

BDB 1 (246)

II' It k

II I

CD _ eD <- D _ CD CD _-'t N 0 10 0 0 N 0

N N N N

'o aadlCCDO r-CNQ O

C

5'ss UB1 Is,

"G

UGAGCCGCLtASU GAUGGAUSGGCGUUCUGGCgkCUA4UAcCLACACG GU

UA "x u

V ~~~~~DBIJ.

regions at different rates) or as a consequence of ran-

dom variation (as obtained in a computer simulation ofthe experiment, data not shown).The plot in Figure 5A shows the mutation frequency

of the doped pool after selection on copper (24 se-

quences). As expected, two "cold" spots were presentat UB1 and DB1. Analysis of the few mutations found

FIGURE 5. A: Plot showing the mutation frequency in the "winners" from theselection of the "doped" pool derived from the wild-type intron. Positions onthe x-axis indicate the first nucleotide of the 6-nt window used to compute thenumber of mutations (y-axis) along the intron. UB1 and DB1 are indicated inbold letters. The UACUAAC box and the splice sites are also shown on thegraph. CS indicates four of the five cold spots that have been directly tested bysite-directed mutagenesis. Dotted line indicates the average number of muta-tions expected in every 6-nt window assuming a random distribution with a 3.4%mutational rate per position. Solid line idicates a 96.4% level of confidence (i.e.,there is a 3.6% chance of finding only 0 or 1 mutation per 6-nt window in a ran-dom distribution). B: Same as in A, with the difference that the doped pool wasobtained from an intron containing also the 5mUB1 mutation. Mutational rateper position is 4% in this case. Three arrows indicate the hot mutational spots(two in UB1 and one in DB1). Two peaks in UB1 belong to two independentfamilies of sequences (G dlones and GA clones). When replotted separately, bothfamilies display a peak of mutations in the DB1 region (data not shown). C: Pro-posed pairing for one representative of the GA family. Original sequence of mu-tant 5mUB1 before "doping" is shown in uppercase letters and changes thatallow base pairing and growth of the winner clone are in lowercase.

in these regions strengthened the wild-type base pair-ing model (Fig. 2A; data not shown). A few additionalcold spots were identified whose functional relevancewas tested by site-directed mutagenesis in four cases

(Fig. 5A). Random sequences were introduced at ev-

ery site, and a few arbitrarily chosen clones were testedfor survival on copper plates. All the clones displayed

25 T

UBI

20

1 5

10

5

0

25 T

20 +

15 +

10 -

5,

0

430


normal copper-resistant growth (data not shown).Moreover, the percentage survival of the pools on cop-per plates was indistinguishable from that of wild type.Because similar cold spots were also observed in otherexperiments (Fig. 5b and data not shown), we con-clude that they were biases generated in the initial poolby an uneven distribution of introduced mutations. To-gether with experiments presented below, the resultsargue against the involvement of a more complex pre-mRNA structure in the splicing of the rp5lb intron.

What is the function of the UB1IDB1 interaction?

The interaction UB1/DB1 might function to neutralizea negative element, such as the inhibitory binding ofa protein or an alternate folding that might limit splicesite accessibility. When this interaction is disrupted, asin mutant 5mUB1 for example, the "poison" sequence(perhaps in collaboration with a protein) would ac-tively block splicing. Restoration of efficient process-ing should thus be possible by mutation of the poisonsequence as well as by reconstitution of the original (oran equivalent) pairing. To examine this possibility, weperformed a selection experiment on a doped pool con-taining the mutation 5mUB1, which disrupts the inter-action UB1/DB1. Winners of the selection are expectedto display high mutation frequencies in the regions ofUB1 and DB1 (due to the reestablishment of the pair-ing), and additionally, "hot" spots might be expectedwhere poison sequences are located. About 50 winnerclones were fully sequenced, and the distribution ofmutations is plotted in Figure 5B.Two major peaks fall in the UB1 region, and a third

is in the DB1 region. Closer inspection of the sequencesrevealed that all selected clones could be divided intotwo mutually exclusive families: in the first (around70% of the total), all the sequences contain an A to Gmutation at position 25 of the intron; in the second(30% of the total) a CT doublet (positions 30-31) be-longing to the original mutation in 5mUB1 reverted tothe wild type GA. Both the "G" and the "GA" classesstill displayed a highly mutagenic peak in the regionof DB1 when the distribution of mutations was plottedseparately (data not shown). Moreover, the G or GAmutations alone are not sufficient to restore splicing in5mUB1; the clustered mutations in the DB1 region arealso required (data not shown; note also that the poorgrowing mutant 3mUB1 has a GA in the same positionas the GA dones). The analysis of the covariations fullysupports the reformation of the base pairing inter-actions between the regions of UB1 and DB1 (shownin Fig. 5C for the GA clones).The almost complete coincidence of the hot spots

with the two regions involved in base pairing arguesagainst the existence of a negative cis-acting elementwhose action is somehow counteracted by the inter-

In vivo splicing of artificial substratesand their evolution

The absence of an ordered inhibitory state for the struc-ture mutants (i.e., the apparent absence of sequenceinformation which would actively inhibit splicing) sug-gests that splicing of a randomly chosen long intron se-quence should be inefficient (i.e., the default state isinefficient splicing). To test this hypothesis, two arbi-trarily chosen sequences were inserted between theUACUAAC box and the 5' splice site of the wild-typeintron. The constructions b/bI/b and b/Tm/b are twochimeric introns containing an inversion of the wholecentral portion of the rp5lb intron and a fragment ofthe chicken ,3-tropomyosin gene in inverted orienta-tion, respectively. Consistent with expectation, neitherconstruct conferred substantial copper-resistant growthto yeast, and this phenotype was due to poor splicingefficiency as verified by primer extension experiments.A third construct containing another tropomyosin frag-ment gave essentially similar results (data not shown).To test whether poor splicing was due to the lack of

some limited structural information, we performed anin vivo selection experiment similar to the one alreadyperformed on the wild-type and the 5mUB1 mutant in-tron. Two doped pools were constructed from thestarting chimeric intron sequences. Two clones wereisolated from the evolution of b/Tm/b (b/Tm/b.S1, sixtimes, and b/Tm/b.S6, once), and five different cloneswere chosen among the "winners" of the b/bIl/b evo-lution (survival rate 10-3_10-4). The improvement insplicing efficiency after evolution varies from 2.5-foldfor the b/bIl/b based clones to almost 15-fold forb/Tm/b.S6 (data not shown; Fig. 6A,B).

Clones b/Tm/b.S1 and b/Tm/b.S6 share the same Gto A change at position +7 of the intron (i.e., the firstposition outside the conserved 6-nt sequence for the 5'splice site), which likely improves base pairing with Ul(Rosbash & Seraphin, 1991). When introduced into anotherwise unmodified b/Tm/b intron, this change im-proved splicing efficiency, but not to levels sufficientto achieve the same copper-resistant growth pheno-type as b/Tm/b.S1 and b/Tm/b.S6. For all the selectedclones, pairings similar to the ones previously de-scribed for the rp5lb intron can be drawn; the one forclone b/Tm/b.S1 is shown in Figure 6C. The results in-dicate that even very limited in vivo evolution of an in-efficiently spliced artificial intron can improve splicingefficiency up to 15-fold. Taken together, the data indi-cate that structural modifications of the pre-mRNA ac-count for this improvement.

Intramolecular interactions can be designed toimprove splicing in chimeric introns

To investigate the possibility of designing de novo astructural element required to achieve efficient splicing

431

action of UB1 with DBl.

D. Libri et al.

A

n _ n st

0).0

B

SntJB1 3niJBl Wt0 0 0

b/bVb.st b/bVb.S14 b6b

b/bl/bSl9 b/bi/b.S13

b/Tm/bS6 b/li/b0 0 0 0

b/Tm/b.st b/Tm/bS1

1.2mM

lAmM

1.6iM

C

51 A GA ;A

:UWAUCGUUAcUgGUWACU GUAu I t~~~~G

AUAG U A

3'

in the two artificial introns, we introduced in the 5' endof both b/bI/b and b/Tm/b introns a 15-nt mutation thatwould be complementary to a region upstream of theUACUAAC box; the position of these inverted repeatsin the artificial introns is identical to the position of UB1and DB1 in the wild-type intron with respect to the 5'splice site and the UACUAAC box. The constructionderived from b/bI/b (b/bI/b.St) displayed a copper-resistance growth phenotype and a splicing efficiencycomparable to the clones derived by in vivo selectionof the same intron (Fig. 6B). In contrast, the modifiedintron derived from b/Tm/b (b/Tm/b. St) was unable toconfer copper-resistant growth to yeast (Fig. 6B), sug-gesting perhaps that this stem is not allowed to formin vivo or is not fully compatible with the splicing en-

hancement mechanism.

FIGURE 6. A: Splicing improvement plot for the randomnization-selection experiment performed on artificial introns. The splicingefficency (ratio mRNA/pre-mRNA) of the best splicers after random-ization-selection of b/bi/b and b/Tm/b (b/bi/b.S14 and b/Tm/b.S1,respectively) was set equal to 100 and the splicing efficiency of thestarting clone was scored accordingly; all the other selected cloneshave intermediate phenotypes (data not shown). Splicing efficien-cies of the wild type and its mutant derivative 5mUB1 are also shownfor comparison. Note that, because the pre-mRNAs from the threekind of clones (i.e., the rp5lb, the b/Tm/b and the b/bi/b based in-trons) are considerably different (and most likely have different turn-over rates), it is not possible to compare their splicing efficiencies(pre-mRNA/mRNA ratio) directly. Rough quantification of themRNA amounts is however consistent with the copper-sensitivityresults. B: Copper growth assay performed on selected artificial in-trons and on artificial introns containing the designed inverted re-peats. Left part of the figure is a schematic drawing showing thepositions of the different mutants. b/bi/b.S13, 14 and 19 derive fromb/bi/b, and b/Tm/b.S1 and 6 derive from b/Tm/b. The suffix "st"indicates that the clone contains two designed inverted repeats.C: Base pairing proposed for selected clone b/Tm/b. S1. Original se-quence and mutations selected are shown in uppercase and lower-case, respectively. Bottom part shows diagrammatically the positionsof the interacting sequences.

DISCUSSION

The wild-type pairing

The presence of an intramolecular interaction withinthe rp5lb intron was first proposed by Parker and Pat-terson (1987) and more recently demonstrated byGoguel and Rosbash (1993) in this laboratory. Theopen questions of the overall structural context of thisinteraction and of its mode of action have been ad-dressed in this report using the sensitive and highly in-formative technique of in vivo selection.The results of two different experiments allowed us

to confirm the previously reported interaction of UB1and DB1 (Goguel & Rosbash, 1993). First, the molecu-lar evolution of 8 nt contained in UB1 yielded four fam-

432


ilies of sequences, one of which included the wild-typesequence. All the nucleotide variations contained inthis family either maintained or improved the strengthof the original interaction. For example, a bulged Upresent in the original stem disappeared in most of theselected sequences (Fig. 3 and data not shown). Sec-ond, evolution of the doped wild-type intron gave avery low frequency of mutations in UB1 or DB1 (abouthalf the expected value; Fig. 5A and data not shown),almost all of which are compatible with the proposedstem. Only two positions were modified in UB1 (a to-tal of 5 clones out of the 25 sequenced): U35 to A (threeclones), which most likely allows stronger stacking in-teractions with the neighboring purines, and G39 to A(two clones), which gives rise to a more canonical A-Uinteraction. This same base pair is affected by a muta-tion in DB1 (U251 to C, two clones), which allows for-mation of a more stable G-C. Only one nonconservativechange is observed in these two regions (U247 to A,one clone), and it is located in a region that most likelydoes not provide a major contribution to the overallstem stability.

A stronger interaction might not be necessarilyfavored for splicing enhancement

The molecular evolution of UB1 and some site-directedmutagenesis experiments revealed the existence of atleast three additional base pairing possibilities for se-quences in UB1. One intriguing observation stemsfrom the analysis of these interactions and their vari-ation. Because most of the winners of the UB1 poolevolution have been competitively selected in copper-containing medium before separation on nonselectiveplates, their relative abundance should reflect relativegrowth advantage, due in principle to stronger inter-actions between the two regions of the intron. Indeed,sequences with lower base pairing potentials are ingeneral underrepresented within a given family. Thereare, however, some exceptions.

In the first family of sequences (interaction SB1/DB1),the best base pairing sequence is only half as abundantas the most frequent family member (Fig. 3). In thesecond family (interaction SB2/DB2), the most abun-dant clone contains a C-A mismatch in the center ofan otherwise perfect 13-nt stem. Clones that contain aWatson-Crick interaction at this position exist, but theyare four or five times less abundant and contain at leastone additional noncanonical interaction. Because themost abundant second family clone accounts for 40%of the family (and 25% of the whole selection), it is veryunlikely that its high frequency derives from an initialbias in the starting pool. Indeed, a comparison of thecopper growth curves of the first and third most abun-dant clones of this family (which have almost identi-cal sequences with the exception of a G-C pair instead

pair; Fig. 3) directly proved the existence of a selectiveadvantage for the clone containing the less stable stem(data not shown). This shows that a stronger inter-action does not necessarily favor higher splicing effi-ciency. The extent of splicing enhancement is alsolikely to depend on the sequence of the stem as wellas its stability (data not shown; Goguel & Rosbash,1993).Stem disruption as well as stem formation might also

be limiting processes for splicing efficiency enhance-ment. This disruption is most likely accomplished be-fore the first splicing step, at least in the case of thesequences belonging to the fourth family (interactionSB4/DB4; Fig. 3) and possibly those of the evolved ar-tificial intron b/Tm/b.S1 (Fig. 6C). In both cases, one

of the two interacting sequences is located downstreamof the branch point and maintaining this interactionmight be incompatible with function in a branchedstructure.

How complex is the intronic secondary structure?

The experiments presented in this report indicate thata wide range of structural patterns that improve splic-ing efficiency share the feature of having the 5' and 3'end of the intron in close proximity. It may seem sur-prising that the UB1 or the 5mU`B1 doped pool evolu-tions do not identify shorter range interactions, whichare expected to be favored by thermodynamic criteria.Particularly striking is the case of the UB1 evolution,because essentially all possible sequences are expectedto be contained in the starting pool and, taking into ac-count the contribution of the nucleotides flanking therandomized bases, a number of alternative interactionsites scattered throughout the intron can be identifiedby inspection of the sequence.These observations can be explained by two alterna-

tive hypotheses: (1) UB1 pairing is restricted to the DB1region (the region between DB4 and DB3) becausethere is a more complex intronic structure that restrictspairing to these two regions. Undetermined featuresof this complex folding account for the effect on splic-ing efficiency. (2) There is functional selective pressurethat favors base pairings between the two ends of theintrons, because their close proximity is the feature thatenhances splicing efficiency. A distinction between thetwo models is the extent to which the region betweenUB1 and DB1 is structured.The experiment in which a high number of func-

tional intron variants were selected from a doped poolderived from the wild-type sequence (Fig. 5) arguesagainst the presence of a complex structure; the fewcold spots identified (besides UB1 and DB1) were in-sensitive to site-directed mutagenesis (data not shown),indicating that their occurrence is likely due to an un-

even distribution of mutations in the starting pool.of the A-C mismatch, and a G-U instead of an A-U

433

The finding that a small stem in an artificial intron

D. Libri et al.

(b/bI/b.St, which is very unlikely to share the samestructural organization as the wild-type rp5lb intron)is sufficient as well as necessary for efficient splicingalso argues against the existence of a more complexstructure.The data indicate that the interaction of UB1 and DB1

(or any of the similar possible interactions) is the pri-mary contributor to splicing efficiency enhancement.Its AG is mostly independent from other free energycontributions and likely near an optimum value.

The role of the interaction

The evolution of the doped pool obtained by light mu-tagenesis of a non-base pairing mutant (5mUB1) failedto reveal any event other than the formation of a basepairing interaction between the same regions of the in-tron. Given the frequency of the G and GA clones andthe number of sequences analyzed, we would likelyhave detected events from as frequent as the reestab-lishment of a pairing (about 1/1,000) to probably 10times less frequent, i.e., any actively inhibited splicingstate that would be sensitive to the simultaneous mu-tation of 3-4 nt. The results argue against the existenceof an "ordered" nonfunctional state characterized bythe existence of negative cis-acting elements in therp5lb intron; rather, they suggest a direct and positiverole for intramolecular RNA-RNA interactions in largeyeast introns. Using these same in vivo evolution tech-niques, we found very similar structural elements inthe otherwise poorly conserved rp5la intron (data notshown; see also Pikielny & Rosbash, 1985); togetherwith other data presented in this report and in the lit-erature (Newman, 1987; Parker & Patterson, 1987), theresults suggest that this mechanism is of general sig-nificance for large yeast introns.We favor the hypothesis that the base pairing con-

tributes to the overall strength of 5' splice site-branchpoint interactions and acts cooperatively with splicingfactor interactions that bridge these two intronic re-gions. The increased local concentration of bindingsites and factors would contribute to lowering the en-tropic cost of one or more interactions that take placeduring spliceosome assembly. For example, inter-actions between yeast Ul snRNP and the yeast Mud2protein might be aided by the pre-mRNA intramolec-ular base pairing. Both factors are known to bridge in-teractions between the branch site and the 5' splice site.Mud2p is known to associate functionally with UlsnRNP, to have homology with mammalian U2AF, andto crosslink to pre-mRNA (Abovich et al., 1994). Al-though the binding site of Mud2p has not yet beenidentified, like U2AF it might be a U-rich region nearthe branch site. It is intriguing to note that one of thefamilies emerging from the evolution of the DB4 region(immediately downstream of the branch point) con-

overcome the usual requirement for a base-paired stem(see also Patterson & Guthrie, 1991). Members of thefamily defined by the SB4/DB4 interaction also exhibita very poor growth phenotype in the absence of theMud2 protein (data not shown), suggesting a tight linkbetween this particular structural pattern and the bind-ing of Mud2p. It is interesting to note that none ofthese elements (the U-rich tract, Mud2 or the base pair-ing interaction) is absolutely required for splicing, butthey all contribute to its efficiency.

Splicing enhancement by structure might be consid-ered an evolutionarily "primitive" enhancer mechanism,whose presence may be related to a less prominent roleplayed in yeast by SR or HnRNP proteins (or their as

yet unidentified equivalents). In higher eukaryotes,these proteins are believed to be involved in splice siterecognition and pairing, by contributing to spliceoso-mal protein-protein interactions (Wu & Maniatis, 1993;Amrein et al., 1994; Staknis & Reed, 1994) and by help-ing to neutralize nonproductive pre-mRNA foldingpatterns (Burd & Dreyfuss, 1994). In this context, it isinteresting to note that negative secondary structuresof the pre-mRNA seem to play a more prominent rolein yeast than in higher eukaryotes (Eperon et al., 1986,1988; Fu & Manley, 1987), which might also be relatedto the striking intron-length dependence of splicing inyeast (Klinz & Gallwitz, 1985).Although more detailed considerations must await

in vitro experiments, this view suggests that the basepairing interaction may catalytically or thermodynam-ically enhance yeast pre-mRNA splicing in a manner

that resembles the proposed roles of HnRNP or SRproteins in mammalian systems. It also emphasizesthe related roles of RNA-RNA and RNA-protein inter-actions (e.g., Mohr et al., 1994) and underscores thedual origins of nuclear pre-mRNA splicing: the ancientRNA world and the more recent world of proteincatalysis.

MATERIALS AND METHODS

Cloning, mutagenesis, and pool constructions

The rp5lb intron was cloned between a BamH I and a Sma Isite in a reporter construct containing the yeast CUP1 gene(Stutz & Rosbash, 1994) so that the starting ATG is locatedimmediately upstream of the donor site.

Site-directed mutagenesis has been performed either as de-scribed in Kunkel et al. (1987) or as follows: a mutagenic oli-gonucleotide, an oligonucleotide containing one of thecloning sites (e.g., Sma I), and 100 ng of the wild-type tem-plate were used to obtain a primary PCR product. One of thestrands of this product was subsequently used as primer ina secondary PCR reaction that contained, besides the primaryPCR product, a wild-type template and an oligonucleotidecorresponding to the second cloning site (e.g., BamH I). The

tains an extremely U-rich sequence that might partially

434

resulting PCR product was subsequently cloned in the CUP


reporter plasmid. In most cases, Vent polymerase was usedto minimize unwanted errors.The pools of sequences containing random nucleotides in

the UB1 region were constructed by cloning a PCR productobtained with a degenerate 5' oligonucleotide (containing anNco I site) and a 3' primer (containing a Sma I site) betweenthe Sma I site and an Nco I site created by mutagenesis inthe reporter gene. All the other pools containing completelyrandomized regions were obtained using the mutagenesisprocedure described above. In all cases, the PCR product con-taining the degenerate sequences was cloned in a vector pre-pared from a plasmid containing a nonfunctional intron. Thiseffectively ruled out the presence in the cloned pools of func-tional introns that do not derive from the mutagenesis pro-cedure. The percentage of nonfunctional introns derivingfrom the vector (uncut or cut only once) was generally lowerthan 1%.The doped pools were constructed by mutagenic PCR on

the whole intron essentially as described by Cadwell andJoyce (1992) and modified by Bartel and Szostak (1993). Theconditions described by Leung et al. (1989), which allow ahigher mutagenic rate but with a strong AT to GC bias, werealso employed in duplicate experiments. The mutation ratewas estimated a priori by monitoring the disappearance ofa restriction site in the intermediate PCR products. The mu-tation rate of the different pools was as follows: 3-6% (wild-type intron); 4-10% (intron containing the mutation 5mUB1);and 5% and 8%, respectively, for the evolution of b/Tm/b andb/bIl/b. The splice sites were excluded from mutagenesis andthe UACUAAC box was "repaired" by subsequent site-directed mutagenesis on the final doped pool.A reduced mutational bias was observed with the condi-

tions described by Cadwell and Joyce (1992) as opposed tothe conditions of Leung et al. (1989). However, A/T to X/Ymutations occurred with a frequency about four times higherthan G/C to X/Y. Notably, no G/C to C/G changes were ob-served. A positional bias was also observed in the distribu-tion of mutations, maybe due to some sequence-dependentvariations in the polymerase speed (see text).The artificial intron b/bI/b was constructed by inverting

the intron sequence between positions +12 and +281 withappropriately designed oligonucleotides. The intron b/Tm/bwas obtained by cloning a fragment from the chicken tro-pomyosin gene (sequence positions 7401-7593; Libri et al.,1989) in the antisense orientation between the same intronpositions in the reporter gene (UB1 and DB1 are not includedin the chimeric introns). The chimeric introns containing theartificially designed inverted repeats were obtained by mu-tagenizing 15 nt starting at intron position 31 and by intro-ducing changes complementary to a sequence located atposition -41 with respect to the UACUAAC box.

Transformations

The pools of sequences were first transferred to bacteria byelectroporation and the DNA prepared was then transferredto yeast by lithium acetate/PEG transformation as described(Gietz et al., 1992). The complexity of the pools varied be-tween 104 and 5 . 105. The transformants were first plated onsynthetic Leu- plates and harvested as soon as colonieswere visible. After overnight growth of representative ali-quots of the pools, they were submitted to copper selections.

Copper-selection assays

All selections and growth assays on plates were performed atCuS04 concentrations of 1.2, 1.4, and 1.6 mM. The selectiveconcentrations employed in liquid media were 0.4-0.8 mM.This last competitive assay proved to be extremely sensitive,being able to detect growth differences almost undetectableby other methods.

Sequence analysis

Most sequencing was performed directly on PCR productsderived from nucleic acid preparations or cell suspensions ofindividual clones or pools, with the femtomol sequencing kit(Promega).The "winners" from the doped pool selections were ana-

lyzed as follows: 25 clones were sequenced from two inde-pendent experiments (a total of 50); all the sequences obtainedwere introduced in Microsoft Excel worksheets and the to-tal number of mutations contained in each window of 6 ntstarting from every position in the intron has been plottedto obtain the graphics shown in Figure 5A and B.The secondary structure analysis was performed either by

eye or by searching the whole intron sequence for homolo-gies to the degenerate complement of the sequence ofinterest.

Yeast strain

All the experiments were performed using the copper-sensitive yeast strain Y69ACUPAB::URA3 (oa, leu2-3, leu2-112,ura3-52, trpl-289, his3-A1, cyhR, ACUP1, Arp5lB::URA3).This strain was obtained by deleting the tandemly repeatedX-CUP1 genes from strain MGD 353-46D (Seraphin et al.,1988) as described earlier (Stutz & Rosbash, 1994). In a sec-ond step, the nonessential rp51B gene was completely de-leted by homologous recombination using a rp51B genomicconstruct in which the rp51B sequences between the EcoR Vsite (position -210) and the Hinc II site in the 3' flanking re-gion (Abovich & Rosbash, 1984) had been replaced by theURA3 gene.

Various

Primer extension of total RNAs and ligations was performedessentially as suggested by the enzyme manufacturer (Gibco-BRL and USB, respectively). For some experiments, a ther-mostable reverse transcriptase from Epicentre was employed.Taq and Vent polymerase were purchased from BoehringerMannheim and New England Biolabs, respectively.

ACKNOWLEDGMENTS

We thank D. Bartel, M. Buvoli, M. Moore, C. Pikielny, andH.V. Colot for critical reading of the manuscript and all themembers of the Rosbash lab for fruitful discussions. The tech-nical assistance of D. Johns-Gordon and the expert secretarialhelp of L.A. Monaghan are kindly acknowledged. D.L. is

435

436 D. Libri et al.

supported by the C.N.R.S., France. This work was supportedin part by a grant from the National Institutes of Health toM.R. (GM 23549).

Received April 24, 1995; returned for revision May 8, 1995;revised manuscript received May 15, 1995

REFERENCES

Abovich N, Liao XC, Rosbash M. 1994. The yeast Mud2 protein: Aninteraction with prpll defines a bridge between commitmentcomplexes and U2 snRNP addition. Genes & Dev 8:843-854.

Abovich N, Rosbash M. 1984. Two genes for ribosomal protein 51of Saccharomyces cerevisiae complement and contribute to the ri-bosomes. Mol Cell Biol 4:1871-1879.

Amrein H, Hedley ML, Maniatis T. 1994. The role of specific protein-RNA and protein-protein interactions in positive and negativecontrol of pre-mRNA splicing by transformer 2. Cell 76:735-746.

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

Burd CG, Dreyfuss G. 1994. Conserved structures and diversity offunctions of RNA-binding proteins. Science 265:615-621.

Cadwell RC, Joyce GF. 1992. Randomization of genes by PCR mu-tagenesis. PCR Methods Applic 2:28-33.

Chebli K, Gattoni R, Schmitt P, Hildwein G, Stevenin J. 1989. The216-nucleotide intron of the ElA pre-mRNA contains a hairpinstructure that permits utilization of unusually distant branch ac-ceptors. Mol Cell Biol 9:4852-4861.

Clouet-d'Orval B, d'Aubenton-Carafa Y, Sirand-Pugnet P, Brody E,Marie J. 1991. RNA structure represses utilization of a muscle spe-cific exon in HeLa cell nuclear extracts. Science 252:1823-1828.

Deshler JO, Rossi JJ. 1991. Unexpected point mutations activate cryp-tic 3' splice sites by perturbing a natural secondary structurewithin a yeast intron. Genes & Dev 5:1252-1263.

Eperon LP, Estibeiro JP, Eperon IC. 1986. The role of nucleotide se-quences in splice site selection in eukaryotic pre-messenger RNA.Nature 324:280-282.

Eperon LP, Graham IR, Griffiths AD, Eperon IC. 1988. Effects of RNAsecondary structure on alternative splicing of pre-mRNA: Is fold-ing limited to a region behind the transcribing RNA polymerase?Cell 54:393-401.

Fu XY, Manley JL. 1987. Factors influencing alternative splice site uti-lization in vivo. Mol Cell Biol 7:738-748.

Gietz D, Jean AS, Woods RA, Schiest RH. 1992. Improved methodfor high efficiency transformation of intact yeast cells. Nucleic AcidsRes 20:1425.

Goguel V, Rosbash M. 1993. Splice site choice and splicing efficiencyare positively influenced by pre-mRNA intramolecular base pair-ing in yeast. Cell 72:893-901.

Goguel V, Wang Y, Rosbash M. 1993. Short artificial hairpins seques-

ter splicing signals and inhibit yeast pre-mRNA splicing. Mol CellBiol 13:6841-6848.

Klinz F, Gallwitz D. 1985. Size and position of intervening sequencesare critical for the splicing efficiency of pre-mRNA in the yeastSaccharomyces cerevisiae. Nucleic Acids Res 13:3791-3804.

Kunkel TA, Roberts JD, Zakour RA. 1987. Rapid and efficient site-specific mutagenesis without phenotypic selection. Methods En-zymol 154:367-382.

Leung DW, Chen E, Goeddel DV. 1989. A method for random mu-tagenesis of a defined DNA segment using a modified polymer-ase chain reaction. Technique 1:11-15.

Libri D, Lemonnier M, Meinnel T, Fiszman MY. 1989. A single genecodes for the ,B subunit of smooth and skeletal muscle tropomy-osin in the chicken. J Biol Chem 264:2935-2944.

Libri D, Piseri A, Fiszman MY. 1991. Tissue specific splicing in vivoof the 3 tropomyosin gene: Dependence on an RNA secondarystructure. Science 252:1842-1845.

Madhani HD, Guthrie C. 1994. Randomization-selection analysis ofsnRNAs in vivo: Evidence for a tertiary interaction in the splice-osome. Genes & Dev 8:1071-1086.

Mohr G, Caprara MG, Guo Q, Lambowitz AM. 1994. A tyrosyl-tRNAsynthetase can function similarly to an RNA structure in the Tetra-hymena ribozyme. Nature 370:147-150.

Moore MJ, Query CC, Sharp PA. 1993. Splicing of precursors tomRNAs by the spliceosome. In: Gesteland RF, Atkins JF, eds. TheRNA world. Cold Spring Harbor, New York: Cold Spring HarborLaboratory Press. pp 303-357.

Newman A. 1987. Specific accessory sequences in Saccharomyces ce-revisiae introns control assembly of pre-mRNAs into spliceosomes.EMBO J 6:3833-3839.

Parker R, Patterson B. 1987. Architecture of fungal introns: Implica-tions for spliceosome assembly. In: Dudock B, Inouye M, eds. Mo-lecular biology ofRNA: New perspectives. New York: Academic Press.pp 133-149.

Patterson B, Guthrie C. 1991. A U-rich tract enhances usage of analternative 3' splice site in yeast. Cell 64:181-187.

Pikielny CW, Rosbash M. 1985. mRNA splicing efficiency in yeast andthe contribution of nonconserved sequences. Cell 41:119-126.

Rosbash M, Seraphin B. 1991. Who's on first? The Ul snRNP-5' splicesite interaction and splicing. Trends Biochem Sci 15:187-190.

Seraphin B, Kretzner L, Rosbash M. 1988. A Ul snRNA:pre-mRNAbase pairing interaction is required early in yeast spliceosomeassembly but does not uniquely define the 5'-cleavage site.EMBO J 7:2533-2538.

Staknis D, Reed R. 1994. SR proteins promote the first specific rec-ognition of pre-mRNA and are present together with the Ul smallnuclear ribonucleoprotein particle in a general splicing enhancercomplex. Mol Cell Biol 14:7670-7682.

Stutz F, Rosbash M. 1994. A functional interaction between Rev andyeast pre-mRNA is related to splicing complex formation. EMBOJ 13:4096-4104.

Wu JY, Maniatis T. 1993. Specific interactions between proteins im-plicated in splice site selection and regulated alternative splicing.Cell 75:1061-1070.

RNA 1:425-436

Documents

Transcript of RNA 1:425-436