Both constitutive and alternative splicing require the assembly of the basal splicing machinery in spliceosome complexes on consensus sequences present at all boundaries between introns and exons (the 5' and 3' splice sites). The spliceosome has two functions: to recognize and select splice sites, and to catalyze the two sequential transesterification reactions that remove the introns and join the exons together. The efficiency with which the spliceosome acts on an exon is determined by a balance of several features, including the strength of a splice site (that is, its conformity to consensus splice-site sequences), exon size, and the presence of auxiliary cis elements. Exons of ideal size (typically 50 to 300 nucleotides) with strong splice-site sequences are recognized efficiently by the splicing machinery and are constitutively included in the transcript, whereas suboptimal exons require auxiliary elements for recognition. But it is becoming increasingly clear that many constitutive exons also use auxiliary elements to ensure their recognition in all cells expressing the pre-mRNA. For example, a bioinformatic approach was used to identify intronic G-rich elements that facilitate the recognition of small constitutive exons [10]. Alternatively spliced exons are often small and have weak 5' and/or 3' splice sites, but for these exons auxiliary elements serve not only to improve splice site recognition, but also to modulate selection of splice sites used in specific cell contexts.

In general, the auxiliary elements that regulate the usage of alternative splice sites share several common features: they are small, variable in sequence, individually weak, and present in multiple copies. They are usually single-stranded, although secondary structure has been implicated in the function of a few elements (see, for example, [11]). Auxiliary elements are often conserved between species and perhaps between similarly regulated genes, but they contain degenerate sequence motifs, making it difficult to identify them. They can be exonic or intronic, and when they are intronic they can lie upstream, downstream, or flanking both sides of the regulated exon. Intronic elements can also be proximal (within 100 nucleotides) or distal (more than one kilobase away from the regulated exon), although they are often located close to the exon. And finally, auxiliary elements can enhance or repress splice-site selection. Depending on their location and their effect on the recognition of alternative splice sites, the elements are referred to as exonic splicing enhancers or silencers or intronic splicing enhancers or silencers (Figure 2). Table 2 lists the intronic splicing enhancers and silencers that have been identified to date; exonic splicing enhancers have recently been cataloged elsewhere [12].

Growing evidence suggests that many alternative splice sites are associated with both enhancers and silencers, and that regulation of alternative splicing is often the result of dynamic antagonism between trans-acting factors binding to these elements (reviewed in [13]). Indeed, results from the best-characterized vertebrate experimental systems (see below) argue that for most alternatively spliced transcripts there is no 'default' or unregulated state; instead, the ratio of alternative splice forms observed for a given pre-mRNA results from a balance between positive and negative regulation.

Numerous exonic splicing enhancers have been shown to bind serine/arginine-rich (SR) proteins, a family of essential splicing factors, to promote inclusion of alternative exons with weak splice sites in the pre-mRNA [14]. The effects of SR proteins on alternative splicing are antagonized by the constitutive splicing factor heterogeneous nuclear ribonucleoprotein A1 (hnRNP A1). Overexpression of SR proteins or hnRNP A1 has a general effect on splicing of alternatively spliced pre-mRNAs in vitro or in transient transfection assays, suggesting that alternative splicing of some RNAs can be globally regulated in cell populations by changing the relative ratios of constitutive splicing factors. In at least some cases, hnRNP A1 cooperatively binds exonic splicing silencer sequences within the same exon and blocks binding of the SR proteins to exonic splicing enhancers [15].

Other auxiliary elements mediate their effects by binding to auxiliary splicing factors. For example, muscle-specific splicing elements in the intronic regions flanking the alternative exon 5 of the cardiac troponin T protein regulate inclusion of exon 5 in transcripts in embryonic striated muscle. Positive muscle-specific splicing elements downstream of exon 5 bind members of the CUG-binding protein (CUG-BP) and embryonically lethal abnormal vision-type RNA binding protein 3 (ETR-3)-like factor (CELF) family; binding of CELF splicing factors promotes exon inclusion [16]. Negative elements antagonize this muscle-specific activity by binding the ubiquitously expressed pyrimidine tract-binding protein (PTB) upstream and downstream of exon 5 [17]. The neuron-specific N1 exon of the c-src proto-oncogene is similarly regulated. PTB binds to intronic sequences flanking the N1 exon to inhibit its inclusion in the c-src mRNA in non-neuronal cells [18]. In neurons, a downstream intronic enhancer region binds several factors to derepress the N1 exon, including KH-type splicing-regulatory protein (KSRP), far upstream element binding protein (FBP), and the heterogeneous nuclear ribonucleoproteins hnRNP H and hnRNP F [18]. The neuron-specific splicing of several other targets is similarly repressed by PTB, and is antagonized in at least some of these cases by binding of the neuron-specific activator neurooncological ventral antigen 1 (Nova-1) to enhancer elements [18].

Most cis elements known to regulate alternative splicing were identified using deletion or site-directed mutational analysis of minigenes that were tested in transient transfection assays. This approach is limited by the difficulty in making minigene constructs that preserve the ability to regulate the exon in a cell-culture system (reviewed in [19]). Another caveat of this approach has been that multiple small elements often display functional redundancy, making it hard to identify them by a loss-of-function approach. In contrast, this repetitiveness should be helpful for identification of auxiliary cis elements in large-scale genomic analyses. Sequencing of the human genome has provided large data sets that are invaluable for finding new elements involved in cell-context-specific alternative splicing.

Large genomic data sets have been used to identify elements by computational analyses. In one study, Fedorov and colleagues [20] found differences in the distribution of pentameric and hexameric nucleotide sequences in the exons of intron-containing and intron-lacking genes, some of which may represent exonic regulatory elements. The differences in nucleotide distributions that they reported were not created by a few strong signals, but rather by the accumulation of multiple weak signals, consistent with our current understanding of exon recognition. The putative exonic elements described by Fedorov et al. [20] did not generally match known exonic splicing enhancers and silencers, but experimentation to test whether these elements do indeed affect splicing was left to future studies.

Fairbrother and colleagues [12] used a computational approach to predict exonic splicing enhancers sequences by statistical analysis, followed by experimental verification of enhancer activity. They looked for hexameric sequences that are enriched in exons relative to introns and that are near weak splice sites relative to strong ones. They identified ten classes of putative exonic splicing enhancers, five of which matched known motifs for such elements and five of which were novel. Representatives of all ten motifs had enhancer activity when tested in minigene constructs in vivo. Fairbrother et al. [12] also showed that nine out of the ten elements tested had greater enhancer activity than corresponding mutant sequences that were statistically predicted to lack enhancer activity due to single base-pair substitutions. The predictive ability of their method was further demonstrated by comparing their predicted motifs to hexamers within the human hypoxanthine phosphoribosyl transferase (HPRT) gene. Numerous natural exonic mutations are known to cause exon skipping in the HPRT gene, mutation of which is associated with Lesch-Nyhan syndrome, a neurogenetic disorder caused by a defect in the purine biosynthesis pathway. Of 30 mutations, more than half disrupt predicted exonic splicing enhancer motifs. The results by Fairbrother et al. [12] suggest that it is possible to predict accurately the splicing phenotype of point mutations in human diseases by computational analysis and demonstrate the striking possibilities for predicting splicing elements from genomic sequence. Interestingly, each of the ten predicted exonic splicing enhancer motifs occurred as often or slightly less often in alternative exons than in constitutive exons, so it is likely that the motifs identified in this study play a role in exon recognition in both alternative and constitutive splicing.

On a small scale, a comparative genomic approach was used to identify intronic elements regulating splicing of a single alternative exon in the transcript for the splicing factor hnRNP A1. Human and mouse hnRNP A1 genomic sequences were aligned and conserved intronic stretches were tested to determine whether they play a role in splicing of the hnRNP A1 pre-mRNA [21]. Brudno and colleagues [22] used a computational approach to find candidate intronic regulatory elements involved in cell-type-specific alternative splicing. In this study, a kilobase of intronic sequence from each side of 25 brain-specific internal alternative exons was contrasted to a larger set of intronic sequences flanking constitutive exons. Brudno et al. [22] found a significant enrichment of the hexanucleotide UGCAUG and related pentamers in the region of the downstream intron proximal to the regulated exon. They also found this sequence at a high frequency downstream of a smaller set of known muscle-specific exons, suggesting that this element may play a broader role in cell-type-specific alternative splicing. Strikingly, UGCAUG was previously identified in intronic splicing enhancers defined by functional assays ([23] and references within), and, furthermore, the splicing factors KSRP and FBP bind to this sequence in the intronic downstream control element of c-src transcripts [18]. Brudno et al. [22] also report that computational analysis of PTB consensus binding sites revealed a statistical over-representation of these sites near brain-specific exons, supporting models in which dynamic antagonism between cell-type-specific activators and the more ubiquitous repressor PTB modulate splicing of alternative exons by binding to positive and negative auxiliary elements, respectively [17,18]. In future studies, this approach could be used to look at other subsets of cell-type-specific alternatively spliced genes, or expanded to encompass more distal intronic regions.

The most common experimental approach used to identify RNA motifs associated with regulatory factors is systematic evolution of ligands by exponential enrichment (SELEX), in which preferred functional sites or binding sites are selected and amplified from an initial pool of random RNA sequences. Several studies have identified potential exonic splicing enhancers by functional selection of exonic sequences that enhance splicing in cell-free splicing assays [24,25,26,27] or in cultured cells [28]. Of these, studies from the Krainer laboratory [26,27] are most striking. In this work, Liu and colleagues used a splicing substrate in which the known exonic splicing enhancer was replaced with a pool of random sequences. To select for sequences that mediate splicing via individual SR proteins, they used a splicing complementation assay in which recombinant SR proteins were added to splicing-deficient cytoplasmic extracts. Liu et al. [26,27] identified four sets of exonic splicing enhancer motifs that are specifically activated by SR proteins, namely ASF/SF2, SRp40, SRp55, and SC35. These motifs matched almost a dozen natural exonic splicing enhancers for which binding specificity of individual SR proteins has been determined. The score matrices generated for the four SR proteins were then used as tools to predict exonic splicing enhancers. Disruption of such a predicted enhancer explains aberrant splicing resulting from a single nonsense point mutation in the BRCA1 gene in breast and ovarian cancer patients [29]. The mutation perturbs an exonic splicing enhancer regulated by ASF/SF2, causing exon skipping, which in turn disrupts the carboxy-terminal end of the BRCA1 protein. This mutation had previously been thought to contribute to disease by introducing a premature termination codon, but this analysis [29] demonstrated that the substitution gives rise to a splicing mutation instead.

Similar analysis of a single base-pair substitution within exon 7 of the survival of motor neuron 2 (SMN2) gene showed that exon skipping and subsequent truncation of the SMN2 protein is due to disruption of an exonic splicing enhancer [30]. Truncation of the SMN2 protein prevents compensation for loss of a nearly identical SMN1 gene in patients with spinal muscular atrophy. These studies dramatically demonstrate the importance of exonic splicing enhancers for exon recognition, as well as the significant role disabled elements play in human disease. Thus far, the sequence motifs identified by SELEX for the four SR proteins studied by Liu et al. [26,27] have been compared to established exonic splicing enhancers and point mutations known to affect splicing. In the future, analyses can be extended to include additional SR proteins and the elements identified by SELEX can be compared to databases of genomic sequences, alternatively spliced exons, or other point mutations associated with human diseases.

One study used sequences found by SELEX to identify alternative splicing elements from genomic sequence. Buckanovich and Darnell [31] used SELEX to identify binding sites for the neuron-specific splicing factor Nova-1. The selected sequence of three intact UCAU repeats was shown experimentally to be both necessary and sufficient for high-affinity Nova-1 binding. Although the sequence was selected on the basis of optimal binding and not function, Buckanovich and Darnell [31] were able to identify natural targets of Nova-1-mediated cell-specific splicing, by searching GenBank, a neuron-specific alternatively spliced exon database, and the Nova-1 genomic sequence with the consensus RNA selection sequence. Only two targets (exon 3A of the glycine receptor α2 and exon H of Nova-1) were identified, but at the time of this study, few genomic sequences were available and alternative splicing databases contained few entries. It is interesting to note, however, that Brudno and colleagues [22] found no increase in the frequency of UCAY sequences (where Y is a pyrimidine) in brain-specific alternative introns, suggesting that Nova-1 activity mediated through this element may be important only for a minority of brain-specific exons. Today the approach used by Buckanovich and Darnell [31] could presumably be used to find many natural targets of other cell-context-specific regulators by searching the large genomic and alternative splicing data sets for elements identified by SELEX.