In eukaryotes, alternative splicing creates a diversity of proteins with a limited number of genes. Producing variants of the same protein may be beneficial for tissue specialization at different developmental stages, or when subject to changing physiological conditions. Regulation of alternative splicing also provides an additional layer of control over gene expression. The importance of alternative splicing has been shown in multiple studies of development and cancer [1–3]. Identification of new alternative splice variants may provide additional knowledge about gene regulation and function. Such information is essential for developing treatments for diseases associated with splicing abnormalities, for instance, by using inhibitors of the aberrant transcript expression [4].

Although non-coding sequences (introns) are present in the genomes of all eukaryotes, alternative splicing is more common in complex, multicellular organisms. This bias may be caused by the difficulties in developing such a mechanism by fast-growing unicellular organisms, as production of splice variants, although helpful in achieving protein diversity, also poses a risk of generating aberrant protein products [5].

Splicing studies identified different varieties of transcript rearrangements together with several key proteins involved in this process [6]. The most prevalent form of alternative splicing is exon skipping (cassette exons), comprising about 40% of all splicing events conserved between humans and mice [7]. Comparative studies of exon skipping in mice and humans also indicate the presence of selective pressure for retaining 'functional splice variants', in which exon skipping does not shift the open reading frame (ORF) for the encoded protein [8]. In C. elegans, 77% of cassette exon splice variants retain the original ORFs according to the data available in release WS130 of public Wormbase database [9].

C. elegans is a well-studied model organism with a fully sequenced genome, and its alternative splicing has been thoroughly investigated using computational analysis of EST (Expressed Sequence Tags) sequences [10]. The results of this analysis are available through Wormbase. Comparison of ESTs with genomic sequence revealed 1782 genes with alternatively spliced transcripts (Wormbase release WS130), which accounts for about 9% of all C. elegans genes. By comparison, it is estimated that 40–80% of all human genes may be alternatively spliced [11, 12].

We used data from serial analysis of gene expression (SAGE) for computational prediction of novel alternative exon skipping and intron retention events to discover previously unidentified splice variants in C. elegans. Unlike microarrays, SAGE provides information for previously unknown polyadenylated mRNA. We analyzed the data from six C. elegans SAGE libraries using a set of custom Perl scripts. For computational predictions we used C. elegans DNA sequence information from Wormbase release WS130. Applying strict selection criteria, we chose the eighteen most probable predictions of novel alternative splicing events for validation experiments with RT-PCR. Three of the eight predicted exon skipping and two of ten intron retention cases were confirmed in these experiments, demonstrating that computational predictions based on genomic and SAGE data are useful for discovery of novel alternative splice variants. This study is aimed at testing the possibility of predicting alternative splice using computational analysis of SAGE data and genome sequence. To our knowledge, this is the first such study in C. elegans.

By simulating virtual splicing in silico we were able to predict novel alternatively spliced transcripts for previously annotated genes. Hence, this method can produce new information even in the case of a well-studied organism. We analyzed two types of alternative splicing (exon skipping and intron retention) using strict filtering criteria. A similar approach to other types of alternative splicing in C. elegans would likely reveal additional splice variants. This approach is also applicable to other organisms, such as the mouse [16], for which adequate gene annotations and SAGE data are available. Although we cannot readily estimate how many additional alternatively spliced variants C. elegans may have, we have shown that SAGE data can be efficiently used for discovery of novel transcript isoforms.

SAGE allowed us to examine the transcript levels for a few thousand genes (typically, 4000–7000 per library) in one experiment. This approach may significantly expand our ability to study alternative splicing and improve our understanding of its mechanism. Alternatively spliced transcripts in C. elegans have been characterized using EST data [10], indicating that about 10% of C. elegans genes are alternatively spliced. As the authors comment, this number may be an underestimate.

It is interesting that both of our confirmed candidates with intron retention also had ESTs aligned with retained introns, supporting the presence of these transcripts in the mRNA population. However, if taken alone the EST data may not provide sufficient evidence of intron retention. In fact, two other candidates, D1054.10 and R09E10.3, although both having ESTs aligned with predicted retained introns, were not confirmed by our RT-PCR tests.

Microarray analyses of exon skipping events have used oligonucleotide probes designed to overlap the annotated or predicted splice junctions [17, 18], but their use as a discovery tool is limited because the design of the oligonucleotide probes requires sequence data for the splice variants being tested. The task of analyzing all possible rearrangements for every annotated mRNA is beyond the capacity of all modern arrays, and prediction of all splicing events resulting in a new RNA sequence is nearly impossible.

Recently, Kuo et al. [16] used SAGE data to analyze the mouse genome for novel splicing sites in annotated genes. These authors hypothesized that tags neither mapped to known transcripts nor to the genome might span novel splice junctions. They developed an algorithm (SAGE2Splice) for mapping SAGE tags to potential splice junctions. These authors focused on predicting novel splicing sites in the genome rather than discovery of alternative splicing events, which was the goal of our study.

In contrast to microarrays, the SAGE protocol does not require pre-existing information about analyzed transcripts. In principle, experimental data are obtained for every polyadenylated mRNA in the cell. Both the SAGE protocol and the RT-PCR validation experiments sampled the polyA(+) mRNA population. Although the results of our validation experiments showed that confirmed candidates belong to the pool of poly A(+) mRNA, we do not know if those transcripts are functional. Demonstration of their ability to produce active proteins would require additional work, e.g. analysis of polysome-bound fraction of cytoplasmic mRNA. Nevertheless, the computational predictions based on SAGE data can provide the initial guidelines for identification of novel alternative splice variants. Numerous C. elegans SAGE data sets are currently available via online public databases such as GEO [19, 20]. Mining these data using our approach should improve our understanding of alternative splicing mechanisms.

Our results demonstrate a practical application of SAGE data analysis for discovery of alternative mRNA isoforms. SAGE allows sampling of the whole mRNA population including uncharacterized transcripts, which would be missed in analysis with alternative large-scale methods such as microarray. Our results also imply that C. elegans has a larger number of alternative mRNA isoforms than initially predicted.