INTRODUCTION

Top Abstract Introduction Materials and Methods Results Discussion References

The removal of introns during pre-mRNA splicing represents a fundamental step in the transfer of information from DNA to protein. Introns are defined by at least three discrete elements: the 5' splice site, the branch point, and the 3' splice site. The spliceosome assembles around these three elements and excises introns with high fidelity in an ordered, stepwise process. In the yeast genes that contain introns, the almost perfect agreement of these elements with three consensus sequences appears sufficient to account for a high fidelity of splicing. Higher eukaryotes, on the other hand, have larger introns defined by more-degenerate splicing signals. In general, these large introns contain many close matches to the 5' and 3' splice site consensus sequences (54). These pseudosites often compare favorably to real splice sites yet are not used in the splicing reaction. Discovering what causes real sites to be used in favor of these sometimes stronger pseudosites is fundamental to our understanding of pre-mRNA processing.

Since consensus sequences for splice sites were first compiled (43), there has been much effort devoted to finding additional requirements that predict whether a site will be used. It has been proposed that a branch point, a 3' splice site, and a 5' splice site are recognized as a tripartite signal defining an exon rather than an intron (52). Several lines of experimental evidence support this exon definition model. (i) Downstream 5' splice sites have a stimulatory effect on splicing at an upstream 3' splice sites in single intron constructs (20, 52). (ii) The predominant phenotype of a mutation in a splice site of an internal exon is exon skipping rather than intron retention or the activation of a cryptic site (12, 36, 48). (iii) Mutations at a 5' splice site that eliminate splicing can be suppressed by additional mutations that strengthen the upstream 3' splice site across the exon (12).

The exon definition model implies a bridging activity across an exon, possibly leading to steric constraints on the size of an exon. Consistent with this idea, most human exons are between 50 and 200 nucleotides (nt) long (30, 67). Reducing exon size to <50 nt can impose a requirement for a splicing enhancer element (25, 32) or require the transient definition of a larger exon intermediate (59). However, a constraint based on maximum length is less clear (15, 29, 59, 61). A substantial minority of true exons fall outside this size range, and the close grouping of false splice sites within introns is such that many of them define pseudoexons that fall within the permissive size range (54). Thus, it seems unlikely that exon size could be a general factor in the distinction between true and false splice sites.

It has been known for some time that many alternatively spliced exons, small exons, or exons with weak splice sites rely upon the activity of enhancers for their inclusion in mRNA (65). These enhancers are often purine rich and in some cases have been shown to function by binding SR proteins (24, 39, 60). The SR proteins, in turn, are thought to act by recruiting additional splicing factors. For instance, enhancers of splicing at 3' sites have been shown to promote binding of U2AF65 to the polypyrimidine tract (64). However, a recent analysis of alternative splicing of immunoglobulin transcripts showed that enhancers can also act by countering the effect of an inhibitory element (33), rather than by a positive recruitment mechanism. Enhancers have also been proposed to act in the definition of constitutively spliced exons. The constitutively spliced -globin exon 2 has been shown to contain enhancer elements, with several discrete sequences interacting with specific SR proteins (40, 53). The presence of enhancers in -globin, however, do not seem to reflect a general requirement for enhancers in all exons since only a small minority of hundreds of known mammalian splicing mutants lie outside the splice sites themselves (12, 36, 44, 48, 49, 62).

In addition to sequences that promote exon inclusion, there are sequences that inhibit splicing, so-called exonic or intronic splicing silencers. The silencers are less well characterized; they can be purine or pyrimidine rich and bind a diverse array of proteins. Perhaps the best understood example of negative regulation is the role of the Drosophila sxl protein in the sex-specific processing of tra pre-mRNA, where sxl is thought to block access of U2AF65 to the 3' splice site by binding to a cis element (63). Polypyrimidine tract binding protein (PTB), also known as hnRNP I, can function by antagonizing U2AF65 action, as has been shown in the processing of alpha-tropomyosin and GABA(A) receptor gamma2 pre-mRNAs (4, 37). Other silencers have been shown to function through interactions with hnRNP A1 (7, 11, 21), hrp48 (a Drosophila protein in the hnRNP A family) (55), and hnRNP H (13).

Our incomplete understanding of splicing regulatory elements prevents us from predicting the presence of exons in a typically long vertebrate transcript, at least not without searching for open reading frames. Splicing at some true sites may be stimulated by enhancers, splicing at some pseudosites may be inhibited by silencers, and some sites may be defined by the interplay between these two types of elements. Due to the multitude of pseudo-splice sites within introns, we decided to search for negatively acting sequences that may be repressing intronic pseudosites. Early work identified several exonic and intronic sequences that inhibited splicing when ligated downstream of a 3' splice site (25, 51).

To look for intron sequences that inhibit splicing, we constructed several libraries and genetically selected for sequences that could disrupt splicing when inserted into the central exon of a three-exon minigene. We used human genomic DNA as a source rich in intron sequences. A library of human DNA inserts readily yielded sequences that inhibited exon splicing. A comparative screen of human and bacterial DNA sequences revealed that over a third of 19 randomly chosen human inserts caused aberrant processing of the pre-mRNA, whereas none of 27 bacterial inserts did so. We concluded that the human genome is rich in specific sequences that inhibit splicing, and we suggest that such sequences may play a role in silencing inappropriate splice sites.

	MATERIALS AND METHODS

Top Abstract Introduction Materials and Methods Results Discussion References

Constructs. All libraries were constructed by cloning restriction fragments into a pD2B-based minigene. pD2B is a three-exon dhfr minigene (for dihydrofolate reductase [DHFR]) driven from its own promoter in the vector pSP72 (Promega) and containing an extra copy of dhfr exon 2 as the internal exon (see Fig. 1, line 2). The upstream portion of the minigene, from the full promoter to 91 bp into the second intron, is genomic in sequence. The remaining 181 bp of the second intron is a repeat of genomic dhfr intron 1. The third exon is made up of exons 2 to 6 from cDNA, followed by genomic dhfr polyadenylation signals. The construction of pD2B has been described previously (15). pD2C3SfuI is a derivative of pD2B that contains XhoI and ApaI sites on either side of the SfuI site in the central exon.

Human libraries. Human placental DNA was fragmented with a cocktail of HinPI, TaqI, and MspI restriction enzymes; the 50- to 250-bp fraction was purified by gel electrophoresis. These fragments were ligated into the SfuI site of pD2B, whose 6-bp recognition site begins at position +16 in the 50-bp internal exon of pD2B. The products of ligation were used to transform the chemically competent SURE strain of Escherichia coli (Stratagene). After selection on ampicillin plates, the resistant colonies were pooled and the DNA was extracted. Redigestion with SfuI was used to eliminate residual uninserted vector.

Synthetic library. The two single stranded 5' phosphorylated oligomers CGCGGGCCCGGGCTGTGN₂₀TTTATGCTCTCGAGTA and CGTACTCGAGAGCATAAAN₂₀CACAGCCCGGGCCCG were allowed to anneal. The resulting double-stranded DNA contained a randomized 20-nt core flanked by clamps containing the unique restriction sites XhoI and ApaI and capped by two SfuI-compatible sticky ends. The 51-bp fragments were ligated into the SfuI site of exon 2 of pD2B as described above and transformed into the XL-1 Blue strain of E. coli. About 10,000 transformed colonies were pooled for DNA extraction.

Bacterial library. E. coli genomic DNA was fragmented with a cocktail of HinPI, TaqI, and MspI restriction enzymes, and the 80- to 120-bp fraction was gel purified. The fragments were ligated into pD2C3SfuI and used to transform the XL-1 strain of E. coli. The transformed bacteria were plated directly after heat shock to reduce the chance of sister colonies contributing duplicate sequences in the subsequent screen.

Recloning inserts. Two isolates (pD2C2 and pD2C3) of the synthetic library were used for recloning various inserts. The restriction sites XhoI and ApaI were used for the recloning. In pD2C3 the XhoI site is upstream of the ApaI sites, whereas in pD2C2 the ApaI site is upstream of the XhoI site. pD2C3 was used as the vector to remake the splicing constructs from the inserts found in dhfr⁺ recipients of the pD2B human library. The inhibitory inserts were amplified from genomic DNA and cloned via XhoI and ApaI sites introduced within the primers. The 5' primer was GAACGAACTCGAGTACTTC, and the 3' primer was TGAGGAGGTGGTGGGCCCTCTTT. pD2C2 was used as a host to invert inserts by recloning XhoI-ApaI fragments that had been originally cloned in pD2C3. To insert sequences B11 and B36 into the aprt gene, we cleaved plasmids carrying these inserts (pB11 and pB36) at Acc113I and HaeIII restriction sites in the flanking dhfr sequence and cloned the small fragment into the EcoRV site in exon 2 of pAPRTWT (34).

Constructs with known binding sites. The following double-stranded oligomers were cloned between the XhoI and the ApaI sites of pD2C2: hnRNP A1, TATGATAGGGACTTAGGGT; hnRNP H, TAAATGTGGGACCTAGA; PTB, CTGCAGCCTGGAGCTCCTCTCGTGGCC; and U2AF65, TTTTTTTTTCCTTTTTTTTTCCTTTTTTTTT. The hnRNPA 1 and PTB sequences were derived from SELEX experiments (9, 58), the hnRNP H sequence was identified as the hnRNP H binding site in the rat beta-tropomyosin gene (13), and the U2AF65 sequence contains the consensus U2AF binding site derived from a panel of SELEX winners (58).

Transfection. All transfections were performed with Lipofectamine (Life Technologies) using the conditions recommended by the manufacturer for CHO K1 cells, except for the DNA concentrations. Near-confluent 100-mm plates of dhfr null CHO DG44 cells (3 × 10⁶ cells) were transfected with various amounts of plasmid DNA and enough human genomic DNA to bring the total to 7 µg. In experiments in which transfectants were selected for a DHFR⁺ phenotype, DNA from the pD2B human library or synthetic library was used in a series of 1:10 dilutions. In the case of the pD2B human library, the starting plasmid concentration was 1.38 µg/dish. After a 24-h recovery period, the transfected cells were trypsinized and transferred to a larger dish for selection in F-12 medium lacking glycine, hypoxanthine, and thymidine and supplemented with 7% dialyzed fetal calf serum. In experiments in which no selection for DHFR was applied, transfectants were selected for G418 resistance conferred by a cotransfecting neo plasmid, pEGFP-N1 (Clontech). DG44 cells (5 × 10⁵) in a 35-mm dish were cotransfected with 0.5 µg of pEGFP-N1 and approximately 0.5 µg of a plasmid bearing an insert in the dhfr minigene. After a 24-h recovery period, cells were transferred to a 100-mm dish and challenged in F-12 medium containing G418 (400 µg of active compound/ml). Surviving colonies were pooled for further analysis.

RNA and DNA extraction. RNA was extracted by the method of Huang and High (31) for the experiments of Fig. 2. All other analyses were performed on total RNA isolated with s.n.a.p columns (Invitrogen) using the manufacturer's recommended procedure. DNA was extracted with DNAzol (MRC) using the protocol supplied by the manufacturer.

RT-PCR. Reverse transcriptase (RT) reactions were performed using approximately 1 µg of total RNA by the random primer protocol supplied by the manufacturer (Life Technologies). One-fifth of the 20-µl reaction mixture was used in the subsequent PCR. PCR was performed with HotWax Mg++ beads (Invitrogen) using the conditions recommended by the manufacturer, except for the inclusion of radioactive substrate (14). Unless otherwise specified, all reactions were performed with an annealing temperature of 55°C for 27 cycles. In the case of genomic PCR shown in Fig. 3, 1 µg of DNA was amplified for 35 cycles. PCR products were separated by electrophoresis in a 2% modified agarose gel (Trevigel 500; Trevigen). ImageQuant software was used for the quantitative comparison of spliced products after phosphorimaging.

	RESULTS

Top Abstract Introduction Materials and Methods Results Discussion References

Genetic selection of genomic sequences that inhibit splicing. Most splicing mutations result in exon skipping (12, 35, 47). We took advantage of that fact to select for sequences that inhibit splicing. Our strategy was to insert fragments of human genomic DNA into the central exon of a three-exon dhfr minigene. The central exon in this minigene is an extra exon; its inclusion into dhfr mRNA results in a 50-nt insertion that disrupts DHFR enzymatic activity. If an insert inhibits the splicing of this central exon, the two terminal exons are joined to form dhfr mRNA that codes for the functional enzyme. We have previously used this minigene, carried in pD2B, to select for a large number of base substitution mutants deficient in splicing of the central exon (14).

View larger version (33K): [in this window] [in a new window]   FIG. 1.   Selection scheme for sequences that inhibit splicing. The top diagram depicts a basic dhfr minigene (in pDCH1P) retaining the 300-nt dhfr intron 1 as the sole intron, with exons 2 through 6 originating from cDNA. This minigene confers a DHFR+ growth phenotype (ability to grow in the absence of purines and thymidine) when transfected into DHFR recipient cells. pDCH1P was modified to produce pD2B, carrying the minigene shown in the second diagram. This gene has an extra copy of exon 2 inserted into the intron. This extra 50-nt exon is efficiently included in the mRNA, resulting in a message that cannot code for a functional DHFR enzyme. Libraries were constructed by cloning DNA fragments into the unique SfuI restriction site engineered into the central exon. Inserts that reduce splicing result in partial or complete skipping of the central exon, thus restoring the production of functional mRNA.

View larger version (66K): [in this window] [in a new window]   FIG. 2.   Inhibition of splicing in transfectants selected for a DHFR+ growth phenotype. CHO dhfr-deficient cells were transfected with a plasmid library containing human DNA insertions in the central exon of a dhfr test minigene, as described in the text. Colonies that exhibited a DHFR+ growth phenotype (growth in GHT medium) were expanded and assayed for their splicing phenotype by RT-PCR, using primers located in exons 1 and 4. A phosphorimage of an electrophoretic gel is shown. The bracket indicates the positions of bands corresponding to inclusion of the exons with inserts of various sizes. Splicing in cells transfected with the uninserted parental plasmid (pD2B) is also shown.

View larger version (66K): [in this window] [in a new window]   FIG. 3.   Distribution of insert sizes in transfectants selected for dhfr splicing inhibition. Genomic DNA samples from clones selected for a DHFR+ growth phenotype were PCR amplified with primers flanking the insert site. The PCR products were stained with ethidium bromide after electrophoresis in 2% modified agarose gel. A 175-bp PCR product was sometimes seen. This size corresponds to a product containing exon 2 without an insert. Since the template contains a duplication of the intron 2 region (to which the 5' primer can anneal), such an artifactual product could have arisen by DNA recombination during PCR.

View larger version (48K): [in this window] [in a new window]   FIG. 4.   Inhibition of splicing in secondary transfectants carrying minigenes with recloned inserts. Inserts were PCR amplified from the DNA of eight selected primary transfectants and recloned into the central exon of the minigene vector. CHO dhfr null cells were retransfected with each of these plasmids, along with a neo vector. G418-resistant colonies were pooled and tested for their dhfr splicing phenotype by RT-PCR as described in the legend to Fig. 2. In some cases, as indicated, the transfected populations were selected for a DHFR+ phenotype in parallel with the G418 selection. The number above the lanes indicates the size of the RT-PCR product expected for exon inclusion. Below each lane is shown the proportion of exon skipping versus exon inclusion or cryptic splicing, as determined by PhosphorImager analysis.

The inhibitory sequences function in a heterologous context. It is possible that the insertion of a foreign sequence into dhfr exon 2 produced an inhibitory effect specific to dhfr exon 2. For instance, the insert might create an inhibitory secondary structure, either by sequestering positive signals or by disrupting the positive secondary structure. If this were the case, we would expect the splicing inhibitory sequences to be context specific and not to affect splicing if inserted into an unrelated pre-mRNA. To test this idea, we inserted the B36 and B11 sequences into position +4 of exon 2 of the cloned genomic aprt gene. Both B11 and B36 inhibited the correct splicing of aprt exon 2 (Fig. 5). B11 resulted in almost complete skipping of the exon. B36 also inhibited normal exon 2 splicing but spliced predominantly to a cryptic site (possibly the 3' splice site-like sequence, CTTCTCTCCCAACTCCCCGCAG/C) instead of the aprt 3' splice site. In contrast, insertion of an arbitrarily chosen 77-bp fragment (starting from position +8 in dhfr intron 1) at the same position had no effect on the splicing of this exon (Fig. 5). These results suggest that these inhibitory sequences can act autonomously.

View larger version (31K): [in this window] [in a new window]   FIG. 5.   Sequences selected for inhibition of dhfr splicing also inhibit aprt splicing. The inserts B11 and B36 were cloned into position +4 in aprt exon 2 in the plasmid pAPRTWT, which carries the full genomic hamster aprt gene. The control construct, shown in lane C, is identical to B11 and B36 except that it carries a 77-bp insert derived from dhfr intron 1 (positions +8 to +84) instead of an inhibitory insert in position +4 of exon 2. The inserted aprt plasmids, along with a neo plasmid, were used to transfect CHO U1S, an aprt double-deletion mutant. Transfectants resistant to G418 were pooled, expanded, and analyzed for aprt splicing by RT-PCR using primers located in exons 1 and 4. The letter "i" in the B36 and B11 lanes indicate the expected positions for bands corresponding to the inclusion of aprt exon 2. The c in lane B36 indicates the product of cryptic splicing.

Screening for genomic sequences that inhibit splicing. The ease with which inhibitory inserts were isolated from this pool of human genomic DNA fragments raised the possibility that many sequences may be inhibitory. We therefore attempted to select for such inhibitory sequences from a library of random synthetic 20-mers. However, an experiment carried out on the same scale as that described above for human genomic fragments yielded no insert-dependent DHFR-positive colonies. Although some colonies were produced in transfections carried out with very high concentrations of DNA, all the dhfr-positive cell clones were carrying minigenes that had eliminated their central exon (data not shown). They presumably originated by homologous recombination between the duplicated regions encompassing dhfr exon 2 in this minigene (14). We concluded that either the 20-mer was not long enough to specify an inhibitory sequence or the human genome was a richer source of such sequences than a random library.

View larger version (50K): [in this window] [in a new window]   FIG. 6.   Inhibitory sequences occur more frequently in human than in E. coli genomic DNA. (A) Pooled permanent transfectants carrying dhfr minigenes with unique inserts of human DNA were assayed for splicing by RT-PCR and PAGE. The additional lanes marked pD2B, pD2CSfu, pD2C2, and pCMVD2C3 are vector controls for the sample lanes immediately preceding or following them. M is a size marker (416 bp) for skipped mRNA. The size of this band is also marked with an arrow. The letter "c" indicates an mRNA produced by splicing to a cryptic site. The size range of mRNAs that included the inserted exon is indicated by a bracket in the right margin. Each lane was quantified using a PhosphorImager and ImageQuant software. The bar chart shows the percent skipping in black and percent cryptic splicing in gray. Open bars represent control constructs. (B) Pooled permanent transfectants carrying dhfr minigenes with unique E. coli DNA inserts were assayed by RT-PCR and PAGE. All inserts were introduced into pD2C3SfuI, whose splicing phenotype can be seen in both panels A and B. B36 and pD2C3SfuI provide size controls for skipped mRNA and included mRNA, respectively. M indicates size markers. Other features are labeled as for panel A.

Effect of recognized inhibitory sequences on minigene transcript splicing. The high frequency in the human genome of sequences that inhibited splicing focused our attention on sequences that are the targets of abundant RNA-binding proteins. We wondered whether much shorter sequences of this type could also act as inhibitors of constitutive splicing in this system. We chose sequences known to be bound by hnRNP A1, hnRNP H, PTB (hnRNP I), and U2AF65 to test the ability of those factors to inhibit the inclusion of our exon. hnRNP A1 has been shown to inhibit splicing in hnRNP A1 (7), human immunodeficiency virus (HIV) tat (11), and human fibroblast growth factor receptor 2 (21) transcripts. The insertion of a 19-nt sequence selected for tight binding to hnRNP A1 (1, 9) showed some inhibitory activity in dhfr minigene transcripts but still allowed over 75% exon inclusion (Fig. 7, lane 1). The hnRNP H sequence was that of the alternatively spliced exon 7 of the rat beta-tropomyosin transcript; binding of hnRNP H to this sequence correlates with its exclusion in favor of exon 6 (13). As shown in Fig. 7, lane 2, the insertion of this 17-nt sequence did not inhibit splicing of the central dhfr exon in pooled permanent transfectants. PTB and U2AF65 both can bind to polypyrimidine tracts (PPTs), although their preferred sequences emerge as different from iterative binding selection procedures (58, 66). To increase the chance of distinguishing these two activities, we chose a selected version of the PTB binding sequence (p53.6 in reference 56) that appeared to have the least overlap with the U2AF65 binding site consensus (Fig. 7). This 27-nt PTB element was without effect on splicing in dhfr minigene transcripts (Fig. 7, lane 3). The 31-nt U2AF65 sequence was chosen as a combination of sequences found by iterative selection for binding (58, 66) and sequences shown to enhance splicing in a functional assay (19). This sequence inhibited splicing strongly (91% skipping) (Fig. 7, lane 4). It should be noted that this sequence is not directly followed by an AG dinucleotide and so does not constitute a 3' splice site. This last result indicates that a sequence as short as 31 nt can strongly inhibit inclusion of this constitutive dhfr exon and suggested polypyrimidine tracts as potentially active components of our inhibitory inserts.

View larger version (45K): [in this window] [in a new window]   FIG. 7.   Effect of recognized protein binding sequences on splicing. Four oligomers corresponding to sequences known to bind the proteins hnRNP A1, hnRNP H, PTB, and U2AF65 were inserted between the XhoI and ApaI sites near the center of the central exon of pD2C2. The resulting plasmids were transfected into CHO cells and pooled permanent transfectants were assayed for splicing by RT-PCR as indicated. The oligomers were hnRNP A1 TATGATAGGGACTTAGGGT), hnRNP H (TAAATGTGGGACCTAGA), PTB (CTGCAGCCTGGAGCTCCTCTCGTGGCC), and U2AF65 (TTTTTTTTTCCTTTTTTTTTCCTTTTTTTTT). The numbers beneath the lanes show the percent skipped.

Characteristics of sequences. We sequenced 10 inhibitory inserts, obtained either by selection or screening. In a search for possible commonalities, all possible pairs of the 10 sequences were aligned. FastA was used to search each insert against a local database of all 10 inserts. The frequency and quality of the matches found was not different from that expected from a set of random sequences, suggesting that the inhibitory regions were too small and/or too degenerate to be detected within such a small group. However, the motif discovery program MEME (5) found several 10-mer motifs, the top three being the G-rich sequence GGCAGGGUGG and two pyrimidine-rich sequences (CUUACUCUUC and UCUUUCACCG). As can be seen at the bottom of Fig. 8, 6 of the 10 sequences contained good matches to the G-rich consensus sequence. An examination of the sequences that encompass this motif in the 6 inserts revealed an abundance of G triplet repeats (Fig. 8). Multiple sequence alignments with low gap and extension penalties confirmed the presence of clusters of G triplets, principally within the 5' region of B2, B36, P16, and B5. G triplets have been proposed as a distinguishing characteristic of the 5' and 3' edges of many introns (41, 46).

View larger version (53K): [in this window] [in a new window]   FIG. 8.   Human genomic sequences that inhibit splicing. Each of the 10 sequences inhibits splicing of the test exon by at least 25% (Fig. 6). Polypyrimidine tracts (length of 11 or more with a pyrimidine content of 85% or more) are underlined. Runs of three or more G's are in boldface. A G-rich consensus sequence found in 6 of the 10 sequences is double underlined; the agreements to this consensus are shown at the bottom of the figure.

	DISCUSSION

Top Abstract Introduction Materials and Methods Results Discussion References

Isolation of inhibitory sequences. We report here the isolation of sequences capable of inhibiting splicing when placed within an exon. The target exon, exon 2 of the hamster dhfr transcript, is flanked by 3' and 5' splice sites that have somewhat above-average consensus scores (52) of 84 and 86, respectively. There is no detectable skipping of this exon in endogenous dhfr transcripts and <5% skipping in cells transfected with the minigene used here. We first used a genetic selection to isolate transfectants that had incorporated minigenes containing inhibitory human DNA inserts. Inhibition of splicing to the extra exon 2 in this construct leads to exon skipping and to the production of functional DHFR and growth in selective medium. Given the low frequency of DHFR⁺ transfectants, we had to consider the prospect that the skipping phenotype was caused by heritable variation of a trans-acting factor rather than by cis inhibition caused by the insert. We ruled out the former possibility by rescuing the insert from exon-skipping clones, reinserting it into a fresh minigene and retesting the new construct. In all of the eight cases tested, the skipping phenotype followed the insert, indicating that a cis-acting element was responsible for the exon-skipping phenotype. In all of these selected cases, the inserts caused >50% exon 2 skipping. It is interesting to note that although selected for their ability to skip, two of the inserts also exhibited low levels of cryptic splicing to an unmapped position within the insert (A1 and A2 in Fig. 4).

Inhibitory sequences are frequent in the human genome and rare in a bacterial genome. Perhaps the most striking result of this work was the prevalence of inhibitory sequences found in human genomic DNA. More than one-third (7 of 19) of randomly cloned human genomic restriction fragments proved to be inhibitory in this in vivo splicing assay. In order to analyze as many distinct sequences as possible, we excluded inserts that were highly related; these amounted to about half of the original set (14 of 33). If we assume that these discarded sequences produce the same splicing phenotype as their tested homologues, the statistical results would be much the same, with 15 of 33 sequences being inhibitory. In contrast, DNA from two other sources did not generate inhibitory inserts. The first source was random 20-mers. These synthetic sequences yielded no splicing mutants in genetic selections that screened an estimated 9,000 different sequences. Since the short size of these inserts may have limited their effectiveness, we turned to E. coli as a second source of control DNA. Of 27 constructs containing E. coli restriction fragments with an average size of 120 nt, none skipped the test exon more than 15% of the time (Fig. 6B).

Sequence analysis of the inhibitory inserts. Compared to splicing enhancer sequences, there have been fewer splicing silencer sequences described. Nevertheless, it is interesting to note that the inhibitory inserts isolated here contain sequence elements that have been implicated previously in splicing inhibition.

A negative role for splice-like sites. Sequences that show good agreement to the consensus splice site sequences (pseudosites) far outnumber real splice sites within large introns (47, 54). A negative role for pseudo-splice sites would help to explain why these sites in vertebrate introns do not function as splice sites. It is an intriguing possibility that they are initially recognized as splice sites, nucleate partial spliceosome formation, and only fail at the later catalytic steps where they are discarded in favor of the sites that border real exons. Evidence from several well-studied systems of negative regulation of splicing supports the idea that splice-like sites can act as negative elements, recruiting components of the spliceosome to inappropriate places where they presumably compete with their legitimate counterpart for protein-protein interactions that are necessary for either recognition of an exon or a catalytic step in the splicing reaction. The exonic splicing silencer in the Drosophila P-element transcript functions by recruiting U1 snRNP into a nonproductive complex at pseudo-5' splice sites adjacent to the real 5' splice site (56). In the HIV tat or rev transcript, catalytically inactive complexes containing U1 snRNP and U2 snRNP repress exon 2 splicing (22). Rous sarcoma virus inhibits the splicing of most genomic copies of RNA via an element termed a negative regulator of splicing (NRS). The intronic NRS has two regions, a 5' purine-rich region that recruits ASF/SF2 and a 3' region that is capable of interacting with both U1 snRNP and U11 snRNP. The binding of U1 snRNP is predominantly responsible for the inhibition (18). The splicing inhibitor in influenza virus NS1 RNA forms a large complex containing U1, U2, U4, and U6 snRNPs that never proceeds to a functional spliceosome (45). U2 snRNP is also present in a complex associated with the immunoglobulin exon M2 inhibitor. U2AF65 binds to the real 3' splice site, but U2 snRNP is part of the complex that binds to the inhibitor (33). Mutations in endogenous genes provide additional support for the idea that pseudo-splice sites interact with the splicing machinery. In the hamster dhfr gene, mutations in a pseudo-5' splice site located 90 nt downstream of the real 5' splice site allow splicing of an otherwise-inactive mutant form of the site (12).