INTRODUCTION

TOP ABSTRACT INTRODUCTION RESULTS DISCUSSION METHODS REFERENCES

Gene expression products can have variable forms, characterized by alternate start sites of transcription and polyadenylation (Gautheret et al. 1998), exon skipping, and alternate donor and acceptor sites at exon boundaries (Mironov et al. 1999a; Brett et al. 2000; Croft et al. 2000). Exon skipping in transcript isoforms is the most frequent event altering the protein coding sequence of genes (Mironov et al. 1999b; International Human Genome Sequencing Consortium 2001) (http://industry.ebi.ac.uk/~thanaraj/gene.html). Surveys of the incidence of alternative splicing, including exon skipping, have been performed (Andreadis et al. 1987; Iida 1997; Valentine 1998; Thanaraj 1999), and a growing number of anecdotal observations confirm the utilization of exon-skipped transcripts in developmental (Dufour et al. 1998; Lambert de Rouvroit et al. 1999; Lim et al. 1999; Unsworth et al. 1999; Kawahara et al. 2000), tissue-specific (Zacharias et al. 1995), and disease-specific (Jiang and Wu 1999; Mercatante and Kole 2000; Strehler and Zacharias 2001) states.

Several approaches have successfully used hybridization experiments both in silico (Wolfsberg and Landsman 1997; Gautheret et al. 1998; Mironov et al. 1999b; Thanaraj 1999; Beaudoing et al. 2000; Brett et al. 2000; Croft et al. 2000; Schweighhoffer 2000; International Human Genome Sequencing Consortium 2001) and in vitro (Schweighhoffer 2000; Strehler and Zacharias 2001) to assess alternate transcript diversity. Nevertheless there are difficulties with interpretation of the results that include (1) the existence of gene families, paralogs, gene copies, and pseudogenes that have similar DNA sequences, providing false positive hybridization; (2) the existence of orphan genes that are located in the complementary strand of intronic or flanking regions (Mironov et al. 1999a); (3) insufficient representation of expressed sequence data in public expressed sequence tag (EST) databases to identify all transcript isoforms.

We have taken an exhaustive approach to the detection of exon skipping from carefully annotated, protein-confirmed genes to maximize the accurate assessment of the degree of isoform diversity.

	RESULTS

TOP ABSTRACT INTRODUCTION RESULTS DISCUSSION METHODS REFERENCES

To develop an unambiguous assessment of the degree to which exon skipping contributes to expressed transcript isoform diversity, and to assess the impact on protein coding of exon-skipping events within coding regions of transcripts from known genomic loci, we have compared ESTs to 545 annotated genes on chromosome 22. Although no standard measure of relative spliceform frequencies for human genes exists, coverage of exon boundaries by ESTs provides a measure of the diversity of isoforms for a particular gene. The incidence of captured ESTs spanning exon junctions may also provide a reasonable, though noncomprehensive, view of transcript diversity and expression. Detection of transcripts displaying exon skipping was performed using novel software, j_explorer, which reduced the complexity of the gene sequences to a set of possible splice junctions that were used to search public EST databases to identify ESTs spanning the annotated exon-exon junctions. The software employs standard data format (EMBL sequence format) and visualization tools (ARTEMIS, Rutherford et al. 2000) in the analysis (http://www.sanbi.ac.za/exon_skipping/). Removal of single and double exon genes reduced the set to 347 multi-exon genes (Table 1), of which 10 were annotated previously in literature or public databases as having experimentally confirmed exon skips (Table 2). Exon-skipping events were recorded when all original junctions involved in the skipping event, including flanking exons, were confirmed by EST sequences. All ESTs supporting exon-skipping events were subsequently confirmed to be unambiguous transcripts of the corresponding gene and not products of paralogous genes, pseudogenes, or related members of an extended gene family by BLAST searches against the nonredundant (nr) database at NCBI. Highly specific identification of exon-skipping and exon-repetition events has resulted.

Sensitivity was assessed using the 10 genes with experimentally confirmed exon skipping. j_explorer accurately identified the previously reported skipped exons in four of the genes (NF2, ADSL, CLTCL, and GGT1). Novel isoforms were detected in EWSR1, PLA2G6, and GGT1 (Table 2), whereas previously described exon-skipping events in four genes (CACNA1I, BZRP, MTMR3, SEP3) were not detected because ESTs mapping to these exon junctions were not available in the public EST databases. The approach yields zero false positives, as confirmed by available mRNA and genomic data, and provides a solid basis for the development of models of transcript diversity that can be generated from a single gene.

We have discovered 62 exon-skipping events in 52 genes (Table 2); 57 of the 62 (92%) exon-skipping events occur within the protein-coding region. The remainder occur in either the 3' (1/62) or 5' (4/62) untranslated region (UTR). In 31/62 (50%) of cases the reading frame is maintained but regions are deleted. In 18/62 cases (29%) the introduction of a skip destroys the reading frame resulting in a frame shift. Proteins for the remaining 8/62 (13%) could not be reconstructed. In four cases an alternative stop codon is used, whereas in five cases there is an alternative start codon introduced.

Gene transcripts were scanned for exon repetition using similarity searching of repeated exon constructs against public EST data. A single tandem repetition of exon 2 of the F-box protein (NM_012179) was detected with high identity to EST AA569698. Exon repetition has only previously been reported in rodents (Frantz et al. 1999).

Ratios of transcript isoforms are difficult to resolve using only EST data, however using the relative capture frequency of skipped exons as a measure provides an indication of the incidence of more commonly occurring isoforms (four or more ESTs confirm the isoform with exon skipping in: CLTCL1, ADSL, GGT1, GSTT1, HMG2L1, MFNG, dJ222E13.1) as compared to rarer isoforms. In 47/62 (76%) cases, the reference isoform, constructed from the genomic EMBL entry, is represented more frequently than a skipped exon isoform (Table 2).

The degree to which the level of gene expression, and hence database representation, affects the probability of finding a skipped transcript was assessed using the number of EST exon-exon junction captures per gene as a relative measure of transcript representation. Three categories comprising equal gene numbers were selected: low capture, which corresponds to <14 EST matches per gene; medium capture, those from 14 to 50 EST matches per gene; and high capture, those with 50 or more EST matches per gene (Table 3). Forty-four genes had no matches to ESTs. We found that 33 have >50 EST matches per gene and that >60% of genes that demonstrate exon skipping have large numbers of ESTs matching to them. Although no relationship between degree of gene expression and extent of skipping can be determined from this study, the degree to which exon junctions are represented in transcripts reveals that highly represented genes demonstrate skipping more frequently. Ten of the 17 (59%) most highly represented multi-exon genes show exon skipping and of these, three (18%) express more than one isoform (Table 2, http://www.sanbi.ac.za/exon_skipping).

To calculate the proportion of mRNAs that may contain a skipped exon we treat each EST spanning an exon junction as an independent sample of the exon junctions. The number of times an EST spans an exon junction is 23,922. The number of times an EST spans a nonconsecutive junction is 149. The number of times an EST spans a consecutive junction is 23,773. From this we estimate that the probability that a given exon junction in a given mRNA is nonconsecutive is ~f = 149/23,773. There are 2893 exon junctions in the 347 multi-exon genes, therefore the average number of exon junctions per multi-exon genes is m = 2893/347. The probability that a given multi-exon mRNA has at least one nonconsecutive exon junction is therefore 1  (1  f)^m = 0.051. As this estimate is derived from a large sample of exon junctions, it may be applicable as a genome-wide estimate.

	DISCUSSION

TOP ABSTRACT INTRODUCTION RESULTS DISCUSSION METHODS REFERENCES

Our approach precisely identifies exon skipping when EST transcript data that spans exon boundaries is available. The number of ESTs that cover an exon-exon boundary determines the likelihood of discovering an exon skip, but capture of exon-skipping events are dependent on the ratio of low-abundance to high-abundance isoforms of transcripts from the gene. The depth of transcript representation in EST databases, level of expression, and number and length of exons all contribute to the complexity of estimation of the number of genes that may have exon-skipped expressed transcripts. Estimation of the genome-wide extent of exon skipping is supported here by 52 of 347 multi-exon genes (~15%). This conservative estimate reflects the fact that only 68% of exon-exon junctions have EST coverage, and that this coverage is skewed towards over-representation of the 3' UTRs. In contrast, 59% of multi-exon, highly EST-represented genes present exon skipping. If exon skipping is independent of the level of expression of a gene, then 59% of all multi-exon genes could exhibit skipping. The fact that increasing EST coverage results in the detection of increasing numbers of exon-skipping events indicates that exon-skipped transcripts are relatively rare. We have estimated the probability of detecting an exon-skipped transcript from a pool of multi-exon transcripts to be ~5%. More sensitive transcript capture techniques may discover exon skipping to be far more widespread than the previous estimates of ~10%-~20% (Mironov et al. 1999a; Croft et al. 2000) (http://industry.ebi.ac.uk/~thanaraj/gene.html), which have been based on EST frequency-independent measures. Expression studies will clarify the relationship between level of expression and degree of exon skipping in transcripts. The diversity of skipped-exon transcript forms is likely to contribute significantly to the diversity of protein products encoded by the genome, especially because the ratio of skipped isoforms of transcripts appears to vary widely, which is likely to have significant functional impact on the proteins for which they code. At least 50% of exon skips that we have detected result in in-frame deletions in the predicted protein products. In 29% of cases, exon skipping results in a disruption of the reading frame which may change or disrupt the function of the protein product. Functional roles for these protein isoforms remain to be explored experimentally.

	METHODS

TOP ABSTRACT INTRODUCTION RESULTS DISCUSSION METHODS REFERENCES

j_explorer (available for download from http://www.sanbi.ac.za/exon_skipping) was used to assemble exon constructs from mRNA-annotated genomic sequences produced by the Human Chromosome 22 Sequencing Group at the Sanger Centre (Chr22.genes.dna file at http://www.sanger.ac.uk/HGP/Chr22/cwa_archive/Nature_02-12-1999/Chr22Genes.tar.gz). Using a 50-bp tag from the 3' terminus of the preceding exon and a 50-bp tag from the 5' terminus of all downstream exons, a set of all consecutive and nonconsecutive exon-exon junctions for each gene was created. Each junction was submitted for similarity searching against dbEST (human) using BLAST 2.0 (Altschul et al. 1990). By combining junctions in a consecutive (i.e., exon 1-exon 2 junction) and nonconsecutive (i.e., exon 1-exon 3 junction) manner the incidence of exon skipping was assessed. A skipping event is reported when an EST is detected that does not contain the exon(s) in question, but does contain an uninterrupted tag made up of 50 bp from each of the flanking exons. Exon repetition was investigated by creating splice junctions composed of the concatenation of the 3' and 5' 50-bp splice junctions of the same exon. ESTs showing significant (P<1 × 10⁴⁰) homology to an exon junction were extracted and aligned to the corresponding genomic sequence using sim4 (Florea et al. 1998). To exclude the possibility that ESTs confirming exon-skipping events were the products of paralogous genes or members of gene families, all ESTs identifying exon skipping were confirmed to be unique to a single target gene from Chromosome 22. Both interchromosomal and intrachromosomal specificity of the transcripts was confirmed using BLAST with a cut-off score of 1 × 10³⁰. sim4 was employed where ambiguous matches were encountered. The resulting `unambiguous transcripts' can therefore be assigned unambiguously to the correct gene of origin. The effect of these transcripts on the reading frame of the protein for which they code was assessed for frameshifts and in-frame deletions. We have confirmed that the results contain no false positives by manual analysis of all cases of possible misalignments. All exon skips reported by j_explorer therefore represent valid skipping events. The identity and genomic location of each the ESTs was converted into EMBL format and added as annotation to the relevant EMBL sequence file. Sequences were then analyzed using ARTEMIS (Rutherford et al. 2000) and are presented together with supplemental information, annotated EMBL entries, and links to ENSEMBL genes and transcripts at http://www.sanbi.ac.za/exon_skipping. All exon structure annotations for the genes used (both confirmed and predicted) were confirmed by manual inspection to be correct. To prevent the detection of skips as a result of incorrectly annotated exon boundaries we required that an EST spanning consecutive (or linear) exon boundaries was present in addition to the ESTs confirming the skip. All linear junctions that could not be confirmed by ESTs resulted in that junction being excluded from further analysis. To address data consistency, we confirmed that in EMBL release 64 (GenBank 119) and 65 (GenBank 121) ~68% of splice junctions are covered with an EST. This figure does not vary significantly between the two releases.

To determine the effect of exon skipping on protein production, the genomic sequences of the 52 genes with exon-skipping events affecting the coding sequence were compared to the cognate mRNA using sim4. Exon-skipping events in the first or last protein-coding exon were recorded as altering the start/stop codon. A UTR-skipping event was recorded when the exon skipped was located in either the 3' or 5' UTR. When the skipped exon was located between two coding exons, framefinder (Slater 2000) was used to predict open reading frames (ORFs) for the ESTs confirming the skip. If the ORF showed homology to the protein isoform then it was considered to be a valid representation of the protein isoform. BLASTX alignment of the EST and the predicted protein was used to determine whether or not the exon-skipping event introduced a frameshift.