Spliceosomal intron size expansion in domesticated grapevine (Vitis vinifera)

EST collection, processing and assembly

In total, 353,748 publicly available EST sequences of V. vinifera were clustered into 26,111 clusters, 8,560 of which represented by only one EST sequence. The largest cluster is composed of 1,998 EST sequences. The distribution of cluster size suggested that one-third of V. vinifera genes are not expressed at very high levels (32.8% singletons) and a few (3 clusters contain more than 1000 ESTs) genes are highly expressed across the representative EST libraries. Among the 26,111 clusters, 5,678 clusters contain no intron, while 20,433 clusters were predicted to be spliced. Among all clusters, 17,059 clusters were predicted to encode proteins, in which 15,369 predicted proteins show homology to Arabidopsis proteins and 15,683 proteins show homology to Populus proteins (BLASTP, E-value < 0.001). BLAST searches against an EST database of other Vitis species showed that 15,089 clusters (60.5%) had high similarity hits in at least one other species of grapevine.

Identification and annotation of large genes and introns

The average intron size, predicted by the annotation of the first draft sequences of Vitis vinifera Pinot Noir, is slightly larger than other sequenced plants [17]. After mapping the assembled consensus sequences to V. vinifera genomic regions, 2,697 introns having size between 3 kb and 100 kb were identified in 2,563 genes (10% of all assembled clusters, 13% of spliced clusters). Among these genes, 1,179 (46.0%) only have one EST sequence (singletons), a higher proportion than among all predicted genes from EST clusters (32.8%). This indicates that if the number of EST sequences in cluster is a good indictor of expression level, genes with large introns had lower expression level compared to all genes (46.0% vs 32.8% singletons).

Comparisons between the 2563 clusters (predicted genes from EST clusters) and existing gene annotations of the V. vinifera genome revealed that 1,812 clusters matched to known annotations with 100% identity, with 1,137 matches that showing identical transcription start/end sites and exon/intron structure between our predictions and known annotations. However, in 217 matches, multiple EST-predicted genes (487 clusters) matched to single annotated genes, while 188 clusters matched to multiple annotated genes. The later case represented one type of annotation error in V. vinifera genome, in which a single gene supported by expression data was incorrectly split into multiple annotated genes due to a large intron between coding regions. Among the 188 clusters, 99 clusters correspond to multiple, consecutively annotated genes, suggesting that additional introns are needed between annotated genes to correct the annotations (Figure 1).

The function or gene ontology of the 2,563 genes, suggested by their Arabidopsis and Populus homologs and Pfam protein domain families, does not show bias to any particular gene family, indicating that intron size is not correlated with gene function. In the 2,563 genes, 2,697 introns between 3 kb to 100 kb (cf. 20 in Arabidopsis and 54 in Populus, respectively) were identified (Figure 2). Among Arabidopsis, Populus and Vitis, genes with conserved exon/intron structure, which means identical exons/intron numbers, intron positions and phase, were identified for detailed intron size comparisons. Because Vitis genome is still in an intensive annotation process, causing constant modifications of annotation, UTR regions are not considered in this study. Only 39 genes having conserved exon/intron structure in Arabidopsis and Populus were identified from the 2,563 genes and subjected to one-to-one comparison, which will be an underestimate because of the uncertainties in Vitis annotation. This is also supported by the fact that a lot of remaining genes showing an extra or missing exon at 3' or 5' end in Vitis. Among the 39 genes, 74 introns larger than 3 kb were identified. The one-to-one comparison of intron sizes revealed that the V. vinifera genome, four times larger than Arabidopsis overall, has experienced a 12-fold expansion of intron size in these 39 genes. The genome size of V. vinifera is nearly identical to Populus but experienced a 5-fold expansion of intron size in these 39 genes (Figure 3b). The increase in intron size is generally the same in each individual gene (Figure 3a). In addition, large introns in the 39 genes are predominantly 5'-introns (over 40% intron 1 and intron 2).

Contents of large introns and TEs

RepeatMasker screening and BLAST searches of the 2,697 large introns suggested that over 80% of the intron contents were made up of repetitive elements. The dominant types of repetitive elements are low complexity sequences, simple repeats, LTR retrotransposons and LINEs. The number of Copia-type LTR retrotransposons is about 6.5 times that of the Gypsy-type, in contrast to less Copia than Gypsy in the V. vinifera genome overall [17]. The distribution of similarities between flanking LTR regions of the 591 complete LTR retrotransposons is shown in Figure 4a. When compared to the similarities between LTR regions identified in V. vinifera genome overall (Figure 4b), LTR regions in introns contain more highly similar pairs (about 30% of the pairs with identity higher than 99%, |², P < 0.01), suggesting younger LTR retrotransposon insertions and recent LTR retrotransposition activity in these introns. Based on an estimate of substitution rate in LTR regions, age estimates of LTR retrotransposons in introns identified a burst of LTR retrotransposon activities: <10,000 ya (LTR similarity 100%) (Figure 4a). For these young LTR-retrotransposons, Copia/Gypsy ratio is 5.8, close to the ratio for all LTR-retrotransposons in large introns, indicating that recent LTR-retrotransposon proliferation is not biased towards either retrotransposon family.

In the 39 genes with identical exon/intron structure/number in Arabidopsis and Populus, repetitive elements comprised roughly 90% of the 74 intronic regions, in which 10 complete LTR retrotransposons were identified. Among the 74 introns, 1 intron contains only a LINE; 1 intron contains only LTR elements; 15 introns contain LINEs and other repetitive elements; 15 introns contain LTR elements and other repetitive elements; 5 introns contain LTR, LINEs and other repetitive elements; 34 introns contain only other repetitive elements; and 3 introns do not contain repetitive elements. According to a BLAST search against the V. vinifera genome, 'other repetitive elements' were highly repetitive within V. vinifera and likely represented un-classified or unnamed grapevine-specific TEs. The three introns without any identified repetitive elements all contain LINE-like ORFs otherwise highly repetitive in the V. vinifera genome.

Characterizations of four selected introns (all containing LINE insertions) in Vitis species and domesticated varieties revealed intron length variation across these species and varieties (Table 1). None of the native Vitis species have LINE insertions in introns. For the first two introns, the size expansion was not found in any Vitis species or variety, including Pinot Noir, possibly suggesting that the LINE insertion is only present in the specific individual or lineage that was used for genome sequencing. The third intron contained identical LINE insertions in both alleles of Pinot Noir (homozygous), but was heterozygous in Dolcetto (Figure 5a). The fourth intron had LINE insertions in all varieties of domesticated grapevines, but was absent from introns of wild species. Pinot Noir, as well as Cabernet Sauvignon, Chardonnay and Riesling, are homozygous for this LINE insertion, but Dolcetto is heterozygous with one insertion-negative and one insertion-positive allele. In addition, Sangiovese and Zinfandel contain an additional LINE insertion in one allele. Together with a 'Pinot Noir' allele, both varieties are heterozygous for the intron length and carrying additional TE insertions not apparent in Pinot Noir genomic sequences (Figure 5b).

Gene name		GSVIVT00033984001	GSVIVT00022278001	GSVIVT00027725001	GSVIVT0003715001
Gene functions		nucleotide binding	signal transduction	ATP binding	hydrolase activity
Intron		4	12	2	1
	V. rotundifolia	-/-	-/-	-/-	-/-
	V. californica	-/-	-/-	-/-	-/-
Wild Species	V. girdiana	-/-	-/-	-/-	-/-
	V. aestivalis	-/-	-/-	-/-	-/-
	V. labrusca	-/-	-/-	-/-	-/-
	V. jacquemontii	-/-	-/-	-/-	-/-
	Cabernet Sauvignon	-/-	-/-	-/-	P2/P2
	Chardonnay	-/-	-/-	-/-	P2/P2
	Dolcetto	-/-	-/-	P1/-	P2/-
V. vinifera	Pinot Noir	-/- a	-/- a	P1/P1b	P2/P2c
	Riesling	-/-	-/-	-/-	P2/P2
	Sangiovese	-/-	-/-	-/-	P2/SZd
	Zinfandel	-/-	-/-	-/-	P2/SZ

Gene names and functions follow Genoscope terminology and annotations [17]. Dashes (-) represent wild-type alleles. Letters (P1, P2, SZ) represent alleles with TE insertions, discovered by genomic analyses. a. Wild-type alleles according to PCR, TE inserted alleles according to genomic sequences, may be unique to the sequenced individual of Genoscope. b. Pinot Noir allele according to both PCR and genomic sequences (P1: 6421 bp). c. Pinot noir allele according to both PCR and genomic sequences (P2: 8175 bp). d. Sangiovese and Zinfandel alleles according to PCR, not found in genomic sequences (SZ: 3246 bp)

Transcriptome and intron identification

Genome-wide intron analysis has only been done in very few plants [18]. In this study, we identified introns on a whole genome scale by combining a large EST data set and genomic sequences in an automated workflow. This approach has been rarely used before because sufficient EST and genomic data are needed but available for only a few plant species [19]. Domesticated grapevine ranks in the top 25 taxa with the largest collections of ESTs and is the fourth whole plant genome sequence available, making it ideal for the EST mapping approach described here. EST cluster number in this study is higher than the predicted gene number of V. vinifera [17], suggesting that gene prediction based on EST consensus sequences covered a gene range comparable to genomic annotation. By manually checking selected matches between EST consensus sequences and genomic regions, the rate of correct mapping is 100%. Therefore, our workflow accurately identified the corresponding genomic regions of EST consensus sequences, thus correct intron positions and sizes.

Annotation errors such as splitting single gene into multiple ones as a result of unsually long introns, were revealed by comparing ESTs to current annotations (Figure 1). This was expected in two ways: correct prediction of genes in genomic sequences is very sensitive to gene size because large size genes are more likely to be predicted incorrectly due to fragmentation [20]; the large introns present in these genes tend to cause prediction algorithms to split exons at both ends of large introns into two separate genes. Incorporating ESTs into gene prediction algorithms can clearly refine genome annotation as suggested by Coyne et al. [21].

Properties of genes with large introns

Large size introns are expected to require more energy to be transcribed and spliced, so large introns may decrease gene expression (energy cost hypothesis) [9]. Such hypothesis was supported by this study if we consider that the number of ESTs is a good indicator of gene expression level. However, there is evidence against this hypothesis in genes expressed in specific mammal organs/tissues [22] and plant genomes [23].

In contrast, gene size and/or intron size seems to be not related to gene function in this study. Among identified large genes, there is no particularly overrepresented gene family. There are still about 20% of expanded intronic regions not accounted for by similarity searches or repeat masking, suggesting unique sequences in introns are also evolving. Most of these unique sequences may be evolving neutrally since there may not be any selective constraint on them. However, a few of them may contain very important regulatory elements controlling gene expression patterns thus under selective constraints.

Intron size expansion, TEs, grapevine evolution and domestication

Comparisons of genome sizes and intron sizes between Populus and Vitis revealed that intron sizes has increased much more than genome size. Either grapevine specific intron expansion or Populus specific intron contraction can explain the intron size difference between the two plants. The young LTR retrotransposons and unique presence of these TEs in domesticated grapevine both suggested that an intron size expansion is more likely to be the case. Wendel et al. [18, 24] proposed that intron size dynamics may be decoupled with genome size evolution by showing that intron size stayed static in multiple rounds of genome expansion and contraction in cotton (Gossypium). In contrast, we showed that Vitis intron size has experienced a 5-fold expansion compared to Populus, with either unknown fluctuation in genome size, or none at all. Intron size dynamics in Vitis seem to be decoupled from genome size evolution but our data suggested expansion instead of conservation of intron size.

A major force behind the intron size expansion in grapevine is the proliferation of TEs, consistent with previous observations [17]. Among various TEs, LTR retrotransposons provides a proxy to estimate temporal aspects of their activities due to their unique structure. Several studies have estimated the age of LTR retrotransposon insertions in plant genomes such as rice, Medicago and maize [25–27]. Peaks of LTR retrotransposon activity were identified including some that were very recent and possibly related to plant domestication [28, 29]. Vitte and Panaud [30] suggested that plant LTR retrotransposons evolved following a 'burst and contraction' model, which could explain the observed 'peaks' of LTR retrotransposon activities in various plant genomes. Different age estimates of Vitaceae family all placed its origin at around 110 million years ago [31–33]. As a result, LTR retrotransposon bursts observed in the introns are much younger than the divergence of Vitales (with a single family Vitaceae) from the common ancestor with rosids, suggesting that increased retrotransposon activity, as well as intron size expansion, is also associated with unique evolutionary changes along the grapevine lineage such as recent domestication.

V. vinifera expanded introns contain significantly more identical LTR pairs, suggesting that more young LTR retrotransposons have been preserved in introns of domesticated grapevine (Figure 4a). Here we suggest that the most recent burst of LTR retrotransposons in grapevine may be associated with domestication, either as a result of artificial selection on the nearby genes or genomic regions (local effect), or a non-adaptive by-product of the domestication bottleneck and specific reproduction modes in domesticated grapevines (genome-wide effect). There is no enrichment of any particular gene families in genes with TE invaded introns. Threrfore it is very unlikely that all of the TEs and introns, or nearby genes with diverse functions, which are randomly distributed in genome, were selected by humans in the domestication process. Instead, LTR retrotransposons in the most recent burst were more likely to be preserved in introns than previous bursts due to vegetative clonal propagation of domesticated grapevine plants. Recent and variety-specific TE insertions in introns were discovered in our characterization of large introns from several commonly cultivated grapevine varieties, suggesting that initial TE invasion of introns spread with clonal propagations (see results, Table 1) [34]. This may be explained by the following reasons: first, somaclonal propagation of plants (or tissue culture) was suggested to promote or induce LTR retrotransposon activity [35, 36]. Domesticated plants with vegetative propagation may generate more LTR retrotransposons than wild species and non-clonal cultivated plants. Second, plant genome expansions caused by TE proliferation are counteracted by rapid DNA removal as a result of recombination [37]. However, due to a domestication bottleneck and vegetative propagation, the linkage disequilibrium (LD) in cultivated grapevines can be found at extremely long distance compared to wild species [38], suggesting that recombination is repressed in large chunks of chromosomes. TEs, in particular LTR retrotransposons, in domesticated grapevines, may not be removed as effectively as in wild species. Third, LTR retrotransposons, as mutagenesis factors, may cause deleterious mutations if inserted in genes. The mutants were selected against and removed from populations very quickly in natural environments [39]. In domesticated plants, intensive human care produces an environment with relaxed selective constraints, in which LTR retrotransposons invaded genes were more likely to persist. Particularly, vegetatively propagated cultivars such as domesticated grapevine have a uniform genetic background as a result of clonal reproduction to preserve desired traits. This reproduction mode facilitated the spread and fixation of introns with LTR retrotransposons (and other TEs) to the entire clonal population or variety. This may be the major reason why excessive TEs were not observed within genes of other plant genomes, including domesticated plants without clonal propagation such as rice [40, 41].

EST data collection, processing and assembly

All currently available V. vinifera Expressed Sequence Tags (ESTs) were obtained from NCBI GenBank using the PartiGene package [42]. Only V. vinifera ESTs were included in this study, including over 140 varieties and cultivars. Vector and poly-A sequences were trimmed by the built-in functions of PartiGene. Cleaned EST sequences were clustered using both CLOBB (the default clustering method of PartiGene [43]) and TGICL [44]. Both methods employ megaBLAST [45] searches to produce clusters based on sequence similarity. For both methods, a similarity cut-off of 99% and greater than 100 bp overlap were employed. Both approaches yielded similar numbers of EST clusters and the clustering results of TGICL were adopted for further analyses. The clusters were assembled and the consensus sequences for each cluster were predicted by PHRAP [46]. The protein coding potential of the clusters were predicted by ESTScan [47].

Identification of large genes and introns

EST consensus sequences were mapped to genomic sequences of V. vinifera Pinot Noir [17] to identify genomic locations and exon/intron structures of the predicted genes. The mapping process was performed with BLAT, by which ESTs or cDNAs were aligned to genomic regions with near identity [48]. Most alignment gaps in BLAT results represent intronic regions. Genes with putative large introns were selected by two criteria: first, the similarity score from BLAT search is higher than 99%; second, the mapped genomic region (excluding introns) cover 95% of the EST consensus sequence. These criteria ensure that predicted gene sequences were mapped to the correct genomic positions. The process was automated with a Perl script for parsing BLAT results, in which the gap sizes were calculated and gap sequences (size range 3 kb~100 kb) and coding regions associated with them were extracted. Extracted gap regions were manually inspected for splicing signals. Gaps with canonical donor and acceptor splice sites (5'-end GT and 3'-end AG) were considered introns. The entire process was also conducted on Arabidopsis and Populus ESTs and genomes to characterize large introns in those plant species.

Genes with putative large introns were submitted to the following analysis: 1, the coding regions were predicted and translated into protein sequences by ESTScan [47]; 2, predicted coding regions were compared to Genoscope annotations by similarity searches and manual corrections; 3, BLAST searches were conducted using the predicted coding regions as queries against an EST database of Vitis species other than V. vinifera. Presence of high similarity hits in non-V. vinifera EST database suggested that the genes have EST data in at least two closely related species, thus they are very likely truly expressed; 4. BLAST searches using predicted protein sequences were conducted against the Arabidopsis and Populus genomic databases to identify the homologous genes in the two plant genomes as an indicator of possible function. In addition, the predicted V. vinifera proteins were subjected to HMM search of protein functional domains against Pfam database [49].

Analysis of large intron contents and selected individual introns

Identified large introns (3 kb~100 kb-long) were subjected to the following analysis: 1, the intronic sequences were screened by RepeatMasker [50] to identify repetitive elements using an Arabidopsis repetitive element library; 2, the introns without any repetitive elements detectable by RepeatMasker were subjected to BLAST searches against the V. vinifera genome. Presence of large number of high scoring hits (cut-off E-value = 0.001) suggests grapevine-specific repetitive elements. 3. Full length LTR retrotransposons in intronic regions were identified by LTR_FINDER [51]. The sequence similarity between two LTR regions were used to estimate the age of LTR retrotransposon insertions using the methods described by SanMiguel et al. [52], with the JC substitution model used to correct for multiple substitutions in LTR regions [53]. All statistical analyses were conducted with the statistical package R [54].

To characterize homologous introns in other Vitis species and varieties, intronic regions that met the following criteria were chosen: 1, there is only one repetitive element in the intron, suggesting a relatively recent expansion; 2, the total intron size without repetitive elements does not exceed 3 kb (to facilitate PCR amplifications). Primers were designed for the flanking exon sequence as well as conserved regions within the repetitive elements to allow scoring either the presence or absence of the elements in a given intron (primer sequences available upon request). PCR amplifications of selected intronic regions were performed for one individual each of six wild grapevine species: V. rotundifolia, V. californica, V. girdiana, V. aestivalis and V. labrusca, V. jacquemontii, and seven V. vinifera cultivars: Cabernert Sauvignon, Chardonnay, Dolcetto, Pinot Noir, Riesling, Sangiovese and Zinfandel. Alleles with or without TE insertions were determined by the size of the PCR product.