INTRODUCTION

Higher eukaryotes have evolved alternative splicing as a common means of producing several different mRNAs and polypeptides from a single primary transcript (for review, see Ref. 1). Transcripts that undergo alternative splicing can produce mRNAs and proteins with different structures, properties, and functions. The utilization of alternative splice sites can be regulated to prevent the inappropriate accumulation of one or more alternative products. To date, several examples of cis-acting regulatory elements, distinct from bona fide splice sites, have been identified which modulate splicing in a diverse assortment of genes from viruses (2-10), Drosophila (11-14), and higher eukaryotes (15-33).

The fibronectin gene is a classical example of a gene that undergoes alternative splicing (34, 35). The fibronectin gene is evolutionarily conserved and is predicted to yield up to 20 different mRNAs in human cells. The generation of this remarkable number of different mRNAs, and consequently polypeptides, is made possible by alternative splicing in three different coding regions of the fibronectin primary transcript (for review, see Refs. 36 and 37). One of the alternatively spliced regions encompasses the EDA (also referred to as EIIIA or EDI) exon. This exon is excluded selectively in fibronectin mRNAs produced by hepatocytes, but it is included to various extents by other cell types (Fig. 1A). EDA exon splicing is also subject to developmental regulation as EDA exon inclusion generally declines with age and differentiation (36, 37). In addition to tissue-specific and developmental regulation, inclusion of the EDA exon is modulated during the process of tissue repair and in certain diseases.

For these reasons, regulation of fibronectin EDA exon splicing has been the subject of several studies. In fact, the EDA exon was the first exon with which it was demonstrated that proper alternative splicing could occur in the context of transfected minigenes (38). Using the same experimental approach, it was determined subsequently that the central 81 nucleotides of the 270-nucleotide EDA exon were required for its inclusion into mRNA (31). Using a chimeric adenovirus/fibronectin EDA intron substrate in an in vitro splicing system, Lavigueur et al. (32) further mapped the element required for EDA exon inclusion to a 9-nucleotide purine-rich motif (5-GAAGAAGAC-3). Caputi et al. (33) identified the same purine-rich sequence as a positive cis-acting element by transfection experiments. Similar exonic purine-rich motifs have been demonstrated to stimulate splicing (39-41) and have been termed exon recognition sequences or exon splicing enhancers (ESEs).¹ A negative element (5-CAAGG-3) involved in down-regulation of EDA exon inclusion was also identified within the EDA exon just downstream of the ESE (33). Although the ESE within the EDA exon has been demonstrated (32) to interact with splicing factors belonging to a family of serine/arginine-rich RNA-binding proteins collectively termed SR proteins (42), it is not yet known if trans-acting factors interact with the negative element.

Using a chimeric intron composed of sequences flanking the 5-splice site (5ss) of the first intron of human -globin and sequences flanking the 3-splice site (3ss) of the human fibronectin EDA exon (2), we observed that deletion of most of the fibronectin EDA exon sequences resulted in a dramatic increase in splicing efficiency. We have carried out mapping experiments to characterize this novel exon splicing silencer (ESS) element. Closer examination of the ESS revealed an adjacent positive element that is able to stimulate splicing when placed downstream of a heterologous intron. This novel ESS and adjacent positive element represent the third and fourth splicing regulatory elements to be identified within the 270-nucleotide EDA exon. The presence of so many different regulatory elements is of particular interest because they may explain the complex, dynamic splicing pattern exhibited by the fibronectin EDA exon.

EXPERIMENTAL PROCEDURES

All plasmids are derivatives of pSVFN- (2). Truncation of fibronectin exon sequences (accession no. X07718) was carried out by PCR using the sense primer FN3ss, 5-CGT CGA CAA AGA AAA TGG TAT CTG C-3 (nucleotides 1146-1164) and the following antisense primers: , 5-CCC CCG GAT CCA ATG CCA GTC CTT TAG GG-3 (nucleotides 1255-1272); 1, 5-CGG GAT CCC ACC CTG TAC CTG GA-3 (nucleotides 1331-1345); 2, 5-CGG GAT CCA ACT TGC CCC TGT GG-3 (nucleotides 1316-1330); 3, 5-CGG GAT CCG CTT TCC CAA GCA AT-3 (nucleotides 1301-1315); and 4, 5-CGG GAT CCT TTG ATG GAA TCG AC-3 (nucleotides 1286-1300) to generate the respective constructs. The mod series of fibronectin exon mutations was carried out by PCR-mediated site-directed mutagenesis as described previously (43) using the FN3ss sense primer and the antisense primer CATR1 (described below) along with the following mutagenic primers: mod1, 5-CAC AGG GGC AAG atc TAC AGG GTG ACC-3 (nucleotides 1317-1348); mod2, 5-CAT CAA AAT TGC aga tct CAC AGG GGC AAG-3 (nucleotides 1294-1328); and mod3, 5-CAC TGA TGT GGA gat cTC AAA ATT GCT T-3 (nucleotides 1273-1307) to generate the respective constructs. Lowercase letters indicate nucleotides introduced to create a unique BglII restriction enzyme site (underlined). The single stem-loop 1 mutation (SL1B) was generated in a similar fashion using the mutagenic primer SL1B, 5-TCA CTG ATG TGG tac TCG ATT CCA T-3 (nucleotides 1272-1296), except that it was introduced in the context of the 1 truncation. The single stem-loop 2 mutation (SL2B) was generated in the context of the 1 truncation by PCR using FN3ss sense primer and the SL2B antisense primer 5-CCG GAT CCC ACC CTG TAC Cac cAA ACT TGC CCC TGT G-3 (nucleotides 1317-1345). The double stem-loop 2 mutation (SL2TB) was derived from pSVFNSL2B by PCR-mediated site-directed mutagenesis using the mutagenic primer SL2T, 5-GAT TCC ATC AAA ATT GCT acc GAA AGC CCA CAG-3 (nucleotides 1289-1321). The double stem-loop 1 mutation (SL1TB) was generated as follows. The PCR product generated using the FN3ss/SL1T primer pair was gel purified and mixed with an equimolar amount of gel purified PCR product generated with the SL1T, 5-CCA CAT CAG TGA tac CCA GTC CTT T-3 (nucleotides 1259-1283)/CATR1 (described below) primer pair. One end of each of these two DNA fragments contains a 12-bp overlap flanked by either the SL1B or SL1T mutation. By virtue of this 12-bp overlap, the two fragments were annealed and elongated by 10 cycles of low stringency PCR (25 °C annealing step) and then amplified by 25 cycles of normal stringency PCR (55 °C annealing step) using the FN3ss/CATR1 primer pair. Approximately 75% of clones sequenced contained the desired double mutation. The SalI-BamHI fragments of truncated or mutated PCR products were reinserted into the SalI and BamHI sites of pSVFN-.

The plasmid pSVCBSB has been described previously (43). Its derivatives, containing fibronectin EDA exon sequences in the sense (pSVCBSBFs) or antisense (pSVCBSBFas) orientation, were generated by inserting the 30-bp BglII-BamHI fragment of pSVFNmod1 into the unique BamHI site of pSVCBSB. The +1C, 5ss, and AC mutations were introduced in the context of pSVCBSBFs by PCR using the sense primer PCR1, 5-CTT AAG TTG GTG GTG AGG-3 (43), in conjunction with the following antisense primers: +1C, 5-CGG ATC CGG TCA GGG CTA GAG TAG GTC AgC CTG TAG ATC CG-3 (nucleotides 1337-1362); 5ss, 5-CCG GAT CCG GTC AGG GCT CGA GTA GGT TGT AGA TCC GGT CTG-3 (nucleotides 1337-1340, 1346-1362); and AC, 5-CCG GAT CCG GTC ACC CTG TAG ATC CG-3 (nucleotides 1337-1348). The SalI-BamHI fragments of the resulting PCR products were reinserted into the SalI and BamHI sites of pSVCBSB.

COS-7 cells (ATCC CRL1651) were transfected as described previously (44) with 5 µg of plasmid DNA/100-mm Petri dish. 48 h post-transfection, cells were washed once with ice-cold phosphate-buffered saline, lysed directly on the culture dish by the addition of 1 ml of 4 M guanidine thiocyanate, 25 mM sodium citrate (pH 7.0), 0.5% sarkosyl, and 0.1 M -mercaptoethanol, and total RNA was isolated as described previously (45). When actinomycin D was used, the drug was added to the culture medium 48 h post-transfection at a final concentration of 5 µg/ml, and total RNA was extracted as described above at the indicated times.

To generate probes for S1 nuclease analysis, the XhoI-SalI fragment of each construct was deleted, and the resulting plasmids were linearized with PvuI. Linearized plasmids were used as templates for primer extension using the 5-end labeled antisense primer CATR1, 5-CGG AAT TCC GGA TGA GCA TT-3, which overlaps the EcoRI restriction endonuclease site of the chloramphenicol acetyltransferase (CAT) open reading frame. 50 fmol of PvuI-linearized template DNA and 1 pmol of 5-end-labeled CATR1 primer were subjected to 15 thermal cycles (1 min at 94 °C, 30 s at 55 °C, and 1.5 min at 72 °C) with 2 units of Taq DNA polymerase in a 20-µl reaction. Free CATR1 primer was removed by agarose gel purification of the extended primer. S1 nuclease protection assays were carried out as described previously (43) using 10 µg of total RNA and 5-10 × 10⁴ cpm (approximately 50-100 fmol) of the appropriate probe. S1-resistant probe DNA fragments were subjected to electrophoresis on 4% polyacrylamide, 8 M urea gels and subsequently visualized and quantitated using a PhosphorImager (Molecular Dynamics). All cited ratios of spliced to unspliced RNAs are averages of results obtained in at least three independent experiments.

RESULTS

In previous studies, our laboratory generated a chimeric gene composed of sequences flanking the 5ss of the first intron of the human -globin gene and sequences flanking the 3ss upstream of the EDA exon of the human fibronectin gene (2). The fibronectin sequences within the chimeric construct include 99 nucleotides of the intron and the first 118 nucleotides of the EDA exon (Fig. 1B). The previously characterized purine-rich ESE and the adjacent negative element, which are located between 155 and 180 nucleotides downstream of the EDA 3ss, are not present in pSVFN- and thus do not contribute to the level of splicing observed. S1 analysis of RNA isolated from transfected COS-7 cells showed that splicing of pSVFN- RNA is relatively inefficient and results in the accumulation of roughly equal amounts of spliced and unspliced RNA (Fig. 1C, lane 1). In an attempt to study the positional effect of a heterologous ESS element, we generated a truncated form of the EDA exon (designated pSVFN; see Fig. 2) in which sequences from +28 to +118 were deleted. Deletion of these sequences resulted in a 20-fold increase in the ratio of spliced to unspliced (S:U) RNA (Fig. 1C, lane 2). This finding indicates that a splicing silencer element may exist within this region of the fibronectin EDA exon and that the EDA 3ss may be intrinsically efficient.

 5ss

FN 3ss

[View Larger Version of this Image (23K GIF file)]

[View Larger Version of this Image (18K GIF file)]

To map the fibronectin ESS element, we generated sequential 3-truncations of the EDA exon sequences (Fig. 2). S1 analysis of RNA from COS-7 cells transfected with pSVFN1, 2, 3, or 4 (Fig. 3) revealed that, with the exception of 1 (lane 2), splicing efficiency steadily increased as fibronectin EDA exon sequences were progressively shortened (lanes 3-5); the ratio of spliced to unspliced RNA approached that obtained with pSVFN. The 1 truncation defines the 3-boundary of the fibronectin ESS as nucleotide +101 relative to the EDA 3ss. The correlation between the size of the truncation and the splicing efficiency observed with 2, 3, and 4 suggests that the fibronectin ESS may be comprised of multiple smaller elements dispersed within the first 100 nucleotides of the EDA exon or that its function is dependent upon the stability of a secondary structure.

Mapping the ESS using 3-deletions of the EDA exon.

[View Larger Version of this Image (77K GIF file)]

However, based on the effect of these 3-deletions alone, we could not exclude the possibility that the progressive increase in splicing is caused by the positioning of CAT sequences closer to the EDA 3ss and not by disruption of an ESS within the EDA exon. To test this possibility, we used site-directed mutagenesis to introduce small internal deletions in the EDA exon within the context of pSVFN- (Fig. 2). The net effect of these internal modifications is a 5-nucleotide deletion in pSVFNmod1 and pSVFNmod2 and a 7-nucleotide deletion in pSVFNmod3. All three mutations resulted in a moderate (4-8-fold) increase in splicing efficiency (Fig. 4). These results suggest that the effects seen in Fig. 3 are caused by deletion of a splicing silencer element between nucleotides +28 and +101 rather than by movement of CAT sequences closer to the EDA 3ss. To verify that the alteration in the spliced:unspliced RNA ratio is not attributable to differences in RNA stability, the half-life of the spliced and unspliced RNAs of several constructs was examined. As indicated in Fig. 5, all three constructs tested had half-lives of approximately 6 h for spliced RNA (filled circles) and 2-4 h for unspliced RNA (empty circles). Thus, no apparent difference in the stability of spliced RNA was observed among pSVFN-, pSVFNmod3, and pSVFN4. Therefore, the marked increase in accumulation of spliced RNA cannot be attributed to a longer half-life. The small decrease in half-life observed for the unspliced RNA of pSVFN4 compared with that of pSVFN- or pSVFNmod3 is consistent with more efficient splicing kinetics.

[View Larger Version of this Image (63K GIF file)]

[View Larger Version of this Image (18K GIF file)]

The complete abrogation of splicing observed with the pSVFN1 construct suggests that the sequences deleted in pSVFN1 (nucleotides 102-118) harbor a splicing enhancer element. However, if that is the case, it differs from "classical" ESE elements (39-41) in that it is not purine-rich. To test if these EDA exon sequences could stimulate splicing when placed in a heterologous context, nucleotides 93-118 of the EDA exon were inserted into pSVCBSB (Fig. 6A) downstream of a chimeric -globin/HIV-1 intron. The pSVCBSB chimeric intron is not spliced (43) since critical HIV-1 exon splicing regulatory elements have been deleted (2). We have demonstrated previously that splicing of this intron can be stimulated significantly by several classical purine-rich ESEs, including the ESE within the fibronectin EDA exon. Consistent with our previously published results (43), only trace amounts of spliced RNA were detected with pSVCBSB (Fig. 6B, lane 1). Insertion of nucleotides 93-118 of the EDA exon in the sense orientation (pSVCBSBFs) stimulated splicing at least 25-fold (Fig. 6B, lane 2). Conversely, insertion of the same EDA sequences in the antisense orientation did not have a stimulatory effect on splicing (Fig. 6B, lane 3).

[View Larger Version of this Image (30K GIF file)]

The region associated with enhancer activity contains the sequence 5-AGGGTGACC-3, which bears homology (a 6 out of 9 match; underlined) to the 5ss consensus sequence, 5-(C/A)AGGT(A/G)AGT-3. One model to explain how this novel enhancer functions is that this pseudo-5ss element recruits U1 snRNP to stimulate recognition of the EDA 3ss much like authentic 5-splice sites have been demonstrated to stimulate exon recognition (46, 47). To examine this possibility, we introduced a point mutation that converted the critical GT dinucleotide of the pseudo-5ss to CT (designated +1C in Fig. 6C), deleted the majority of the pseudo-5ss (Fig. 6C, 5ss), or deleted the A/C-rich region immediately downstream of the pseudo-5ss (Fig. 6C, AC). Analysis of the effect of these mutations revealed that none of the mutations tested resulted in loss of enhancer function. Rather, in all cases the mutation resulted in a further enhancement in the ratio of spliced to unspliced RNA (Fig. 6D, compare lanes 3-5 with lane 2). Therefore, the splicing enhancer activity associated with this non-purine-rich sequence cannot be attributed solely to the 5ss-like element.

The relatively large size of the fibronectin ESS prompted us to examine its ability to form a secondary structure. The 118 nucleotides of the EDA exon present in pSVFN- are predicted to form three stem-loops (Fig. 7A). Each of the two stem-loops closest to the EDA 3ss contains a hairpin loop of 4 nucleotides and an internal loop that interrupts the 13-bp stem. The third and most distal stem-loop is predicted to consist of a stem of only 6 bp and a relatively large loop of 11 nucleotides. The fibronectin gene and its alternative splicing patterns are highly conserved among human, rat, and chicken (36, 37). Therefore, we examined whether the first 118 nucleotides of the EDA exon of other species retained the potential to form a secondary structure similar to that in Fig. 7A. Fibronectin sequences of different species were obtained from sequence data bases and aligned (Fig. 7B). Compared with the human fibronectin sequence, the EDA exons from fibronectin of other mammals such as dog, rat, and mouse differ by only a few nucleotides, whereas those of chicken, newt, and frog are slightly more divergent. As expected, most of these nucleotide substitutions are silent at the amino acid level. The interesting observation is the occurrence of covariance and conservative substitutions.

[View Larger Version of this Image (35K GIF file)]

The region comprising step-loop 1 is completely conserved among human, dog, and rat fibronectin sequences. Chicken, newt, and frog fibronectin sequences contain several substitutions that do not compromise the ability of this region to form step-loop 1. An example of covariance is seen in chicken fibronectin involving positions +23 and +44. As depicted in Fig. 7A, the nucleotides at these two positions are predicted to form a G-C bp in stem-loop 1. In chicken fibronectin, the nucleotide at +23 is replaced by an A, and a compensatory C  T substitution occurs at nucleotide +44, thus maintaining the ability to form a canonical Watson-Crick bp. Several examples of conservative substitutions involving putative base pairing between G and U residues are also observed. In both chicken and frog, a G-U bp in stem-loop 1 is converted to a canonical A-U bp by a G  A substitution at position +24. Examples in which canonical Watson-Crick bp are replaced by a wobble bp are also observed. In frog, a U-A bp is converted to a U-G bp by an A G substitution at +40, and a G-C bp is converted to a G-U bp by a C  T substitution at +44. Taken together, these examples of covariance and conservative nucleotide substitutions support the notion that the stem-loop 1 secondary structure depicted in Fig. 7A is absolutely conserved and is indicative of selective pressure for maintenance of this secondary structure in addition to the protein-coding potential.

Compared with stem-loop 1, there are only a few examples of covariance or conservative substitutions in the region comprising the putative stem-loop 2. For example, a U-G bp in stem-loop 2 is converted to a canonical C-G bp by a T  C substitution at +62 in rat fibronectin. In newt fibronectin, the C at +61 is replaced by a T, and a compensatory G  A substitution occurs at nucleotide +92, thus maintaining the ability to form a canonical Watson-Crick bp. However, the secondary structure of stem-loop 2 is not as evolutionarily conserved as that of stem-loop 1. For example, the bulge in stem-loop 2 is enlarged by substitutions at positions +86 and +87 in chicken fibronectin. These same two substitutions combined with others just downstream abolish the ability to form the base of stem-loop 2 in both newt and frog.

To test the validity of the predicted secondary structure, we introduced mutations in the predicted stem-loop 1 or 2 which would disrupt the secondary structure, and we then made second site mutations that were predicted to restore the secondary structure but alter the primary sequence (Fig. 7B). We targeted the most extensive base paired regions of the stem-loops: in the middle of stem-loop 1 and at the base of stem-loop 2 (boxed in Fig. 7A). To maximize the sensitivity of the assay, all mutations were introduced in the context of the 1 truncation so that the effect of the adjacent ESE would not confound interpretation of the results. Disruption of stem-loop 1 (SL1B) resulted in partial loss of ESS function (Fig. 7C, lane 2) which was restored (lane 3) upon introducing the second site mutation (SL1TB), which regenerated the secondary structure. In contrast, disruption of the base pairing at the base of stem-loop 2 (SL2B) was found to have no effect on ESS function (lane 4). A second site mutant (SL2TB), predicted to restore the base pairing of the lower stem of stem-loop 2, was found to have a very slight reduction of ESS function. Based on the SL2 mutations, it appears that formation of the lower stem of stem-loop 2 is not essential for ESS activity.

DISCUSSION

Transient transfection of COS-7 cells with a plasmid containing a chimeric gene containing the first 118 nucleotides of the fibronectin EDA exon and upstream intron sequences has led to the identification of an ESS and an adjacent positive element within that part of the EDA exon. The construct used in this study places the 3ss of the alternatively spliced EDA exon in the context of a terminal exon. It has been demonstrated that even in this simplified context, recognition of the EDA 3ss remains responsive to regulatory exon sequences both in vitro (32) and in vivo (2). Therefore, we believe that conclusions drawn from this chimeric gene should be applicable to authentic EDA exon splicing regulation. Deletion studies mapped the 3-boundary of the fibronectin ESS to nucleotide +101 of the EDA exon. Actinomycin D time course experiments failed to detect a significant effect of exonic deletions on RNA stability; therefore, it can be concluded that the fibronectin ESS and the adjacent enhancer act at the level of splicing. Secondary structure prediction of the human EDA sequence and the alignment of the first 118 nucleotides of the EDA exon from various species combined with the mutagenesis data presented here implicate conserved secondary structures as critical determinants of ESS activity.

Because the 1 truncation (which defines the 3-boundary of the ESS) eliminates the third stem-loop, stem-loop 3 cannot be essential for ESS activity. In fact, we demonstrate that the region comprising stem-loop 3 harbors an ESE activity. This is discussed in more detail below. The mod3 mutation, which affects base pairs in stem-loop 1, resulted in a moderately increased (8-fold) level of splicing. Therefore, it can be concluded that the region comprising stem-loop 1 is required for maximal ESS activity. However, the relatively high splicing levels observed with the 4 truncation indicate that sequences comprising stem-loop 1 have little ESS activity on their own in COS-7 cells. The SL1 mutations provide strong evidence that formation of the stem-loop 1 secondary structure rather than the primary sequence per se is essential for complete abrogation of splicing in the context of the 1 truncation. Therefore, stem-loop 1 is required but not sufficient for maximal ESS activity in COS-7 cells.

Results with truncation mutants 2, 3, and 4 revealed that splicing efficiency increased as increasing amounts of stem-loop 2 were removed. Both the mod1 and mod2 mutations, which affect base pairs in stem 2, resulted in a moderately increased (4-6-fold) splicing efficiency. These results support the conclusion that the region comprising stem-loop 2 is an important determinant of ESS activity; however, they do not provide direct evidence that the predicted stem-loop 2 secondary structure is functionally important. The SL2 mutations, which target the lower stem of stem-loop 2, were designed to address this question. Analysis of these mutations revealed that formation of the lower stem of stem-loop 2 is not essential for ESS activity. Consistent with this observation, the ability to form the lower stem of stem-loop 2 is not well conserved in fibronectin sequences of chicken, newt, and frog. In these species, several disruptive substitutions are present on one strand of the predicted helix (Fig. 7B). In contrast to the results obtained with the SL2B mutation, two other mutations that map to the base of stem-loop 2, mod1 and 2, decreased ESS activity significantly. Because the mod1 mutation is in the context of FN-, some of the observed increase in splicing efficiency may be partly attributable to the downstream ESE, making the direct comparison of mod1 to 2 and SL2B difficult. The fact that the mod1 mutation increases splicing and that the 2 and SL2B constructs, both of which do not contain the downstream ESE, have markedly different splicing efficiencies suggests that the affected nucleotide sequence at the base of putative stem-loop 2 plays an important role in ESS function. Computer prediction analysis of possible secondary structures carried out on the 2 and SL2B mutations (data not shown) reveals that both destroy the ability to form putative stem-loop 2. However, in the case of the SL2B mutation, an alternate stable stem-loop containing extensive base pairing can form in its place. In contrast, no equally stable structure is predicted to form with the 2 deletion. Therefore, it is possible that in addition to the primary sequence, certain base pairing interactions within the region of stem-loop 2 may also be important for ESS function. However, we do not yet understand the structural requirements of this region for ESS function, and in the absence of significant evolutionary covariation or conclusive experimental evidence we should emphasize the putative nature of stem-loop 2. Nonetheless, the data presented here clearly demonstrate that the region comprising putative stem-loop 2 displays very strong ESS activity in COS-7 cells.

The total abrogation of splicing brought about by the 1 deletion suggests the presence of an ESE that limits the effect of the adjacent ESS. This organization of splicing control elements is reminiscent of the adjacent enhancer and silencer elements identified in the terminal HIV-1 tat/rev exon (2). Unlike most exonic splicing enhancers identified to date (39-41), the sequence associated with this particular splicing enhancer is not purine-rich and thus not a classical ESE. Nonetheless, this novel enhancer is functionally equivalent to classical ESEs because it too can stimulate splicing of an ESE-dependent intron in an orientation-dependent fashion. Sequence analysis of this region raised the possibility that the enhancing activity could be attributed to the presence of a pseudo-5ss. In other systems, 5ss-like elements have been shown to contribute to modulation of splicing both negatively as in the case of the Drosophila P element (11) and positively as in the case of the human calcitonin/calcitonin gene-related peptide gene (17). However, analogous regions of homology to the 5ss consensus can also be identified within the antisense orientation, which was found to have no ESE activity. Disruption of the pseudo-5ss by either a point mutation or deletion did not abolish ESE activity; rather, the mutations resulted in a marked enhancement of splicing. These results indicate that the 5ss-like element is not required for enhancer activity of this sequence. In contrast, the result obtained with the AC derivative suggests that the region comprising the 5ss-like element can stimulate splicing in the absence of downstream fibronectin sequences. Therefore, it is possible that two distinct non-purine-rich stimulatory elements are present between nucleotides +93 and +118 of the fibronectin EDA exon. A similar non-purine-rich exon element that stimulates splicing of the upstream intron by stabilizing the interaction between U2AF and the polypyrimidine tract has been described in troponin T exon 16 (48). Interestingly, the two mutations designed to alter the 5ss-like element increase the overall A/C content of the test sequence. Therefore, this newly identified positive element may be an example of a class of A/C-rich enhancer elements which has been described to be functionally equivalent to purine-rich ESEs (49). It is noteworthy to point out that most of the substitutions that are observed between +86 and +118 in fibronectin sequences of chicken, newt, and frog also increase the overall A/C content. We speculate that in these organisms perhaps the strength of the enhancer is increased at the expense of the strength of the silencer.

At least two general mechanisms could explain the modulatory effect of the elements identified in the EDA exon: (i) they contribute to a secondary structure that simply impedes proper recognition of the 3ss; or (ii) specific cellular trans-acting factors interact with these elements to modulate the efficiency with which the 3ss is recognized. The sequences at the intron/exon junction surrounding the EDA 3ss have previously been proposed to form an alternate secondary structure involving base pairing between intronic and exonic sequences to form a single stem-loop interrupted by several bulges (50). Within this structure, the AG dinucleotide of the EDA 3ss is buried within the predicted secondary structure. Data presented in this study are inconsistent with the structure reported previously as being of functional importance. The SL1 mutations provide compelling evidence that stem-loop 1 as predicted in Fig. 7A is indeed a determinant of the ESS. Consequently, if RNA structure alone is involved in modulation of 3ss recognition, it must involve more complex tertiary interactions that cannot be predicted by the folding programs used.

We describe here the initial identification of an exonic element within the EDA exon of human fibronectin which has a dramatic down-regulatory effect on splicing. This novel ESS lies adjacent to a non-purine-rich region that is demonstrated to stimulate splicing in several contexts. Why should an alternatively spliced exon have so many cis-acting elements that modulate its inclusion into mRNA? In the case of the fibronectin EDA exon, it is possible that several distinct elements regulate its inclusion in both a tissue-specific and developmental stage-specific fashion as well as in response to certain external stimuli. Therefore, we can envision several parallel splicing regulatory circuits that interact with various regulatory trans-acting factors to bring about the spectrum of EDA exon inclusion observed in different tissues at different stages under different conditions. Now that several cis-acting modulatory elements have been identified, it will be of interest to identify the trans-acting factors and determine their individual as well as their collective mechanisms of action.