Alternative splicing is an important post-transcriptional process and mainly contributes to the complexity of a transcriptome and proteome 123. Alternative splicing often produces two or more proteins with functional differences from one gene 4 but can also downregulate the overall protein level by producing targets for nonsense-mediated mRNA decay 5, which is used, for example, in the autoregulation of splicing factors 6. Furthermore, defects in splicing are the basis for a number of diseases 7.

One major mechanism of alternative splicing to alter protein function is the insertion/deletion of functional units such as protein domains, transmembrane (TM) helices, signal peptides, or coiled-coil regions. Alternative splicing tends to insert/delete complete functional units instead of affecting parts of a unit 8. Moreover, several protein domains have a tendency to be spliced out in some transcripts 910. Many proteins occur in a soluble as well as a membrane-bound form. When encoded by a single gene, the soluble form can be produced by post-translational ectodomain shedding 11 or alternative splicing of exons that encode the TM helices. Indeed, 40-50% of the alternatively spliced, single-pass TM proteins have a splice form that specifically removes the TM domain 1213. Furthermore, protein forms can differ in their affinity to bind ligands 1415 or in their subcellular location 16.

In this paper, we present a novel mechanism to modulate function and/or subcellular localization of a protein by alternative splicing. Assuming a protein feature is encoded in two parts by two non-consecutive exons, for example, exon 2 and 4, inclusion of exon 3 results in a protein lacking this feature since it is disconnected at the sequence level. In contrast, the skipping of exon 3 leads to a protein with this feature. We provide evidence for this mechanism by considering four protein features: TM helices, phosphorylation and glycosylation sites, and Pfam domains. In general, this mechanism is conceivable for many other protein features and provides a novel strategy for ab initio prediction of alternative splice events.

In order to find genes that encode a protein feature by two non-consecutive exons, we searched all human RefSeq transcripts for annotated features that span an exon boundary. For these exon pairs, we searched dbEST to find alternative splice events that insert a sequence between them. Thus, we only selected pairs of exons if they had expressed sequence tag (EST)-confirmed, alternative exons between them that are skipped in the given RefSeq. Apart from alternative exons, intron retention or an alternative donor/acceptor site located in the intron can lead to such an insert. We only selected inserts that preserved the open reading frame. Then we evaluated whether the longer transcript (with the insert) still encodes the feature or not. We only considered two exons for small features like TM helices and post-translational modification contexts since it is unlikely that more than two exons encode the feature. For more complex features like Pfam domains, we allowed for the domain to be encoded by more than two exons.

The first protein feature we considered was TM domains. We annotated TM helices in all RefSeq transcripts with the TMHMM program 17. We found 1,807 TM domains (14% of all TM domains) that are encoded by two exons (Additional data file 1). For ten cases, we found EST evidence for an insert due to alternative splicing. As TM domains are short stretches of hydrophobic amino acids, an insert with polar residues will result in the destruction of the TM helix. Indeed, the evaluation of these ten longer transcripts with TMHMM showed that six clearly lacked the TM domain which, in three cases, leads to a soluble protein (Table 1). An example of the disruption of the single TM domain is depicted for DIABLO in Figure 1a. A more complex example is at the Rhesus blood group antigen gene (RHCE) where the inclusion of two exons resulted in a loss of one TM domain as well as the gain of three others (Figure 1b). The massive reconstruction of TM domains in the respective protein isoforms can have considerable consequences for the orientation of the proteins within the cellular membrane and for their interaction with other membrane components.

To find further cases of feature disruption by sequence insertion, we applied the procedure to experimentally verified post-translational modification sites. Post-translational modification of proteins plays a role in various important processes. For example, phosphorylation of splicing factors can influence splicing decisions 18 and glycosylation is associated with a modulation of proteolytic resistance and ligand binding 19. The residue to be modified must be located in a favorable sequence context to be recognized by the enzyme. If this residue is close to an exon boundary, an alternative splice event can change the context to an unfavorable one with the consequence that the modification cannot take place anymore. We inspected the O-GlycBase 19 and Phospho.ELM 20, and found 435 modified residues that are close to 213 different exon-exon junctions. Among them, four exon junctions showed an insert due to alternative splicing. CCL14 has a glycosylated serine at position 26, which is the last residue encoded by exon 1. We found two ESTs (AA612866, Z70293) with an included 48-nucleotide exon between exon 1 and 2. The NetOGlyc 21 score for the serine in the new sequence context dropped from 0.97 to 0.35 (threshold 0.5). Thus, the new context might prevent glycosylation of this residue. For CDK5, an alternative acceptor (BU529114) that inserts nine amino acids upstream of exon 8 alters the context of the phosphorylated serine at position 159 of the protein. The NetPhos 22 scores of both contexts differ (0.93 vs 0.43, threshold 0.5), which indicates that only one context allows recognition by the kinase and, thus, the phosphorylation of the serine. Additionally, we found two examples (MGP and CDK2) where an included exon alters the context of a phosphorylated residue, however, the scores for the new contexts dropped only marginally.

For the fourth feature, we considered functional protein domains using the Pfam database 23. We found 473 inserts into a Pfam domain and nine of those resulted in a disruption of the Pfam (Table 2). Additionally, using the algorithm described in 24, we found three cases where the skipping of a RefSeq exon creates a new Pfam (Table 2). For example, skipping exon 4 of NM_024565 created the cyclin N-terminal domain (Figure 2a). Since exons 5 to 7 of this transcript encode the cyclin C-terminal domain (PF02984), only the exon skipping variant might perform the function of a cyclin. Moreover, skipping exon 2 of NM_139174 resulted in a new double-stranded RNA binding domain (Figure 2b). Downstream of this domain, the transcript encodes an adenosine-deaminase (editase) domain. Thus, the loss of the RNA binding property might act as a negative regulation of the editase activity. Most Pfam domains fold into three-dimensional structures and we cannot rule out that these 12 domains also adopt the correct folding with the insert. However, using standard cut-off scores, these Pfam domains cannot be found in the longer transcripts since the scores for both individual parts are always below the threshold.

In general, any EST-based approach is hampered by the bias of publicly available EST databases towards cancer-related tissues or cell lines that may exhibit aberrant splicing 2526. Furthermore, a splice form that is only represented by a single EST may be a rare error by the spliceosome. Therefore, we determined the number and tissue source of the ESTs that match both splice variants for the described examples (Additional data file 2). For seven of the 20 examples, only one splice form is represented by a single EST or by cancer-related ESTs. However, the remaining examples are supported by several ESTs as well as ESTs from normal tissue, and in four cases both splice variants are contained in the RefSeq database. Thus, we conclude that the majority of the described examples are real splice variants and not artifacts or aberrant splice events.

Besides the four features investigated here, there are many others that can only function if they are connected on the sequence level. Such functional sites or motifs often have a linear structure and comprise, for example, signal peptides, post-translational cleavage sites and subcellular localization signals as well as sites for protein-protein interaction. Many of these motifs are collected in the Eukaryotic Linear Motif (ELM) database 27. Such features can lose their function if an insert separates them on the sequence level. For example, splicing at an alternative donor site of the protein kinase C delta leads to an insert of 26 amino acids into a caspase-3 cleavage site and to an isoform that is caspase-insensitive 28. We have not investigated such features here since only a fraction of them have been experimentally verified and a prediction results in a high number of false positives. With further efforts in verifying and characterizing these features, we expect an increasing number of examples for the proposed mechanism of modulating protein function by alternative splicing. Interestingly, the same principle was recently used to experimentally characterize exon splicing silencers (ESS) 29. In this study, ESS candidates were inserted in the middle exon of a three-exon minigene. If a candidate ESS acts as a silencer, the middle exon is skipped and only in this case a functional green fluorescent protein is encoded. Furthermore, this mechanism is not restricted to protein features but it is also conceivable for sequence and structural features at the mRNA level. For example, some of the variable first exons of NOS1 together with exon 2, form a hairpin structure that is involved in translational regulation, whereas other alternative first exons do not allow hairpin formation 30.

From an evolutionary viewpoint, this mechanism can be explained in two ways depending on whether the protein feature is ancestral or not. If the feature is ancestral, it means it is initially encoded by two neighboring exons and the inserted part must have appeared in the intronic sequences 313233. In this case, the insert simply has the function of a spacer. If the feature is not ancestral, it means the longer splice form is evolutionarily older and, therefore, the alternative exon or splice site must have been converted from a constitutive to an alternative one. This can happen, for example, by the weakening of splice sites or the creation of ESS 34. Complex features with a high sequence specificity such as Pfam domains are likely to be ancestral. In contrast, small features with a loose sequence motif such as the context of a post-translational modification site can arise just by chance and can therefore be evolutionarily younger.

Not all alternative splice events are represented in EST databases and, thus, the development of non-EST-based methods for ab initio prediction of splice events is a necessary but challenging task. Currently, there is only one method that mainly uses genomic conservation of exons and flanking introns to discriminate between alternative and constitutive exons 35. Although alternative splicing often deletes functional units, it is very hard to predict such events on the protein level without ESTs. However, a search for protein features that are put together by exon skipping would provide a new way to predict alternative splice events. For that purpose, it has to be assumed that the split feature is unlikely to be encoded by two non-consecutive exons just by chance. Since Pfam domains usually have a high sequence specificity, we tested this assumption for Pfams by skipping 10,962 constitutive exons. We found only four cases (0.036%) where skipping of a constitutive exon results in an additional Pfam domain (Additional data file 3). In contrast, nine of the 473 (1.9%) alternatively spliced inserts into Pfam domains resulted in a loss of the Pfam. The odds ratio of 53 indicates that Pfam domains are unlikely to be encoded by non-consecutive exons just by chance.

Alternative splicing frequently modulates protein function by insertion or deletion of functional units. In this case, the functional difference is directly associated with the sequence of the inserted or deleted part. Here, we provide evidence for an additional mechanism that acts by putting together a feature from two parts encoded by non-consecutive exons. Thus, the functional difference is not related to a specific insert and the two parts of the feature are present on both the long and the short splice form. The general idea is shown in Figure 3.

Recent alternative splicing databases include the annotation of the functional differences between two protein forms 36. For this purpose, the novel mechanism described here has to be taken into account since it is obviously not sufficient to inspect the alternative exons in the context of the splice form that includes these exons. The functional difference of the examples shown here can only be found if the complete shorter splice form is investigated simultaneously.

The following additional data are available with the online version of this paper. Additional data file 1 is a table listing the TM domains that are encoded by two exons. Additional data file 2 contains the number of ESTs/RefSeqs and information about the tissues or libraries for both splice variants of the examples. Additional data file 3 contains the four cases where skipping of a constitutive exon results in a new Pfam domain.