Complete genome annotation data of animals and fungi were downloaded from Ensembl database (release 67) (http://www.ensembl.org/). Genomic data of plants were downloaded from JGI (http://www.jgi.doe.gov/). For convenience, we classified the 72 species into four groups: fungus, plant, invertebrate and vertebrate. Statistical analyses were performed using the Perl package. Gene structure information including the numbers and lengths of exon/intron and their sequences were extracted from the corresponding genome data. To obtain only reliable data, we applied the following relatively stringent criteria for the quality of the alignment. 1) The intron must be longer than 5 nt, as intron splicing requires a “minimum” of five nucleotides (GU-AG plus an A for the branch point) [14] . 2) For genes with many alternative splicing isoforms, we retained the isoform which produces the longest mRNA for statistical analysis.

In addition to the overall exon/intron numbers and lengths data created from the available sequences, we also obtained exon/intron boundary data for 6 organisms; Homo sapiens, Danio rerio, Drosophila melanogaster, Caenorhabditis elegans, Saccharomyces cerevisiae and Arabidopsis thaliana. We constructed the motif profiles in these 6 representative species, using the extracted intron sequences. Sequence motifs for 5’ splice site (5’ss) and 3’ splice site (3’ss) are depicted as sequence logos by the WebLogo http://weblogo.berkeley.edu/. We also extracted the adjacent 10 nucleotides (nt) of the upstream and downstream of each splice site, and analyzed the conservation of 5’ss and 3’ss splice-site signals.

A comprehensive survey of the 72 eukaryotic organisms shows that most eukaryotic genes contain less than 5 exons across different groups. Basically, the ratio of gene numbers decreases as the exon number increases (Table 1). In summary, the proportion of genes containing one exon varies from 28% to 9% in four groups. In fungi, the percentage of genes with 1 - 5 exons is 91.21%, which indicates that fungal genes are simpler than the other groups. The percentages of genes with 1 - 5 exons in plants and invertebrates account for approximately two-thirds. On the contrary, of those genes that contain more than five exons, their proportions are incremental from fungi to vertebrates. An extreme case is that almost all genes in S. cerevisiae contain 1 - 5 exons (99.97%), compared with only 33.85% in meleagris (meleagris gallopavo, vertebrate)

(Table S2). Taken together, these results indicate that the genes have more exons in vertebrates than in non-vertebrates.

Table 2 shows the varied distributions of exon length in the four groups. It is clear that short exons (<250 nt) are widespread across various eukaryotes. In fungi, the percentage of short exons is only 42.740% and the mean length of fungal exons is larger (589 nt) than in the other three groups (188 nt, 257 nt and 386 nt, respectively). In vertebrates, most of the exons (87.737%) are less than 250 nt in length (Table 2 and Table S1). The percentage of long exons (>500 nt) is 36.575% in fungi, while the corresponding proportions decrease from 21.685%, 9.977% to 5.582% in plants, invertebrates and vertebrates respectively. These results indicate that exon lengths vary across the eukaryotic kingdom with more short exons in vertebrates.

According to the data we used (Ensembl release 67), the human genome contains 20,687 protein coding genes with introns and 1713 (7%) intron-free protein coding genes. Altogether, there are 200,220 introns in human protein coding genes, so the average number of introns per gene is 8.94 in human genome. The number of introns per gene varies dramatically among diverse eukaryotes, including fungi (0.05 - 3.43 introns per gene), plants (0.33 - 7.30 introns per gene), invertebrates (2.92 - 7.42 introns per gene) and vertebrates (7.35 - 10.09 introns per gene) (Table S1). This statistical analysis showed that there is a wide variety of intron-densities in eukaryotic genomes; complex genomic organizations are much more common in the higher eukaryotes than lower eukaryotes.

Consistent with other studies [15] [16] , our results show that abundant long introns are present in vertebrates. Approximately 48.512% of the introns in vertebrates are >1000 nt in length (Table 3). In general, fungal introns are relatively short, 93.627% of the introns in fungi are shorter than 250 nt. In invertebrates and plants, the average percentages of short introns (<250 nt) are 48.320% and 59.847% respectively. Exceptionally, there is a specific distribution of short introns in teleosts. The average length of introns in teleost fishes was significantly smaller than that of other vertebrates. Furthermore, the percentage of short introns (<250 nt) is in the range of 32.17% - 67.06% (with an average of 52.89%) in the five teleost fishes, but only ~18% in all other vertebrates (Figure 1 and Table S1).

In all observed species, as an extreme example, the smallest percentage of short introns is only 5% in invertebrate (Strongylocentrotus purpuratus, sea urchin). However, the number of introns (157,214) in sea urchin is exceedingly large, which is about twice of other invertebrates (82,398). In the plant group the length of introns was small (183 nt) in three algae of Ostreococcus, with significantly smaller than the average value of other plants (329 nt), while exons were much larger (912 nt) than other plants (386 nt) (Table S1).

Although the total number of introns is similar among teleosts, the mean intron length differs significantly in the five teleost fishes (Table 4 and Table S1). Most introns in teleosts are small and similar in length, yet introns of zebrafish are much longer (2820 nt) than the other teleosts (480 - 1180 nt) and 49.911% of introns in zebrafish is more than 1000 nt. In addition, our results indicated that the peak of the intron length distribution is in the range of 50 - 110 nt in teleosts (Figure S1) and most eukaryotes. The peaks are consistent with previous reports, which show a typical bimodal distribution in many eukaryotes [17] [18] [19] .

¹No.: Number of introns; ²(%): The percentage of introns.

We analyzed the classical splicing signal motifs for each organism. The results of six representative species from four groups (H. sapiens, D. rerio, D. melanogaster, C. elegans, S. cerevisiae and A. thaliana) reveal well-known highly conserved motif profiles for introns within the range 51 - 70 nt (Figure 2) and longer. Although resembling one another, the motif profiles exhibit some differences and specificities among different species. The adjacent nucleotides around each splice site are far from random. They comprise two distinguished consensus sequences of the 5’ splice site (5’ss) and the 3’ splice site (3’ss) on the exon-intron boundaries [20] . The conservation of the 5’ss and 3’ss is lower in zebrafish and human than in the other species (Figure 2). For the introns with length in 6 - 50 nt, the splice sites are not conserved in yeast, zebrafish and human (Figure S2). Many eukaryotic genomic architectures are typified by small exons and flanking introns with variable length. Splice site recognition is more efficient when introns or exons are small, which appears to favor diverse splicing factors for alternative splicing [21] .

A comparison of exon-intron structures could elucidate the complexity of genetic diversity among eukaryotes. There is a trend showing a general increase in

intron densities and lengths in species from fungi, plants, invertebrates to vertebrates. The trend is inverse in relation to exon lengths (Figure 3).

Intron sizes vary widely within each group (fungi, plants, invertebrates and vertebrates). In contrast to intron length, the average lengths of exons are more similar in each group. An increasing body of evidence indicates that introns play a number of functional roles. Many introns contain functional non-coding RNAs, which play vital roles in fine-tuning gene expression [23] . Intron length appears to be positively correlated with expression in unicellular eukaryotes and negatively correlated with expression in multicellular eukaryotes [24] . Furthermore, it is a negative correlation between intron size and the level of expression of genes in nematodes and humans, which suggests that natural selection favors short introns in highly expressed genes to minimize the cost of transcription [25] . In contrast to intron size, the density of introns in a gene does not strongly depend on the level of gene expression [25] . Jeffares et al. found that intron density correlates with the logarithm of generation time. The organisms that reproduce rapidly tend to have fewer introns than organisms that have longer life

cycles [8] . This might be a result of selection for rapid cell division or gene expression.

The exon-intron architecture has also been shown to influence splice-site recognition. The splice-site recognition is more efficient when introns or exons are small [21] [26] . Lower eukaryotes have a genomic architecture that is typified by small introns and flanking exons with variable lengths, suggesting that splice-site recognition occurs across the intron [27] . Our analysis showed some small introns and large exons in most fungi and some algae, which is consistent with a previous report [21] . Jeffares et al. proposed that some genes are apparently under selective pressure to minimize introns [8] . As an example, the average intron size is only 124 bp in Ostreococcus tauri, which is the world’s smallest free-living eukaryote known to date [28] . It is a plausiblestrategy that green alga could select small introns to economize energetic cost from decreased transcript length, adapting changing marine environment to bypass the constraints imposed by light or nutrient limitation [29] .

The number and length of introns varies greatly between different organisms. Intron sequences constitute 24% of mammalian genomes and more than 95% of human gene sequences [30] [31] . Our study shows that teleosts have more and smaller introns (<250 nt) than the other vertebrates (Figure 1 and Table S1). This specific exon-intron structure may be related with the specific gene duplication event in teleosts since the genomic complexity of the teleosts was assumed to be caused by the fish-specific whole-genome duplication event (FSGD) [32] . Remarkably, introns of zebrafish are much bigger compared to other teleosts. Large introns can present several problems for organisms, including the expense of transcription and the difficulty of splicing large introns [33] . Comparative analysis of teleost genome sequences has revealed an ancient intron size expansion in the zebrafish lineage [14] . One possible explanation for the small intron size in other teleosts could be the pressure to maintain a constrained genome size in these fast-replicating organisms. It could also be associated with the FSGD event that triggered the stunning diversity observed in teleost fishes (~29,000 species, nearly half of all vertebrates) [32] .

Our analysis showed that introns are arranged non-randomly in diverse eukaryotes. The vertebrate genes are typically split into numerous small exons interrupted by much larger introns. In our statistical analysis, there are relatively long introns and short exons in 32 vertebrate species. It is a trend that intron length has gradually expanded in fish, amphibians, reptiles, aves and mammals (Table S1). Our analysis suggests that vertebrate introns increased in length during vertebrate evolution. Previous studies indicated that intron length has gradually expanded among mammals, whereas the length of exons has remained relatively constant [34] . Some findings have led to speculations that the spliceosome in mammals recognizes primarily the exons in a process termed exon definition, as opposed to that in fungi where introns are kept short and are thought to be the recognized unit in a process termed intron definition [34] [35] .

Intron and exon lengths can reflect the constraints imposed by splicing recognition, based on whether the exon is identified through the intron or exon definition mechanism. A large number of long introns could be a reservoir of genetic diversity in vertebrates, and they can facilitate the selection of different splicing factors for AS during evolution. Different intron lengths are associated with different types of AS [36] . Long introns could hinder the activity of the splicesome through interfering with the proper positioning of the splicesome upon exon-intron junctions [36] . Short introns tend to flank weak splice sites and long introns tend to flank exons with strong splice sites [16] [37] . AS is more abundant in higher eukaryotes than in lower eukaryotes, and the percentage of genes that undergo AS is higher in vertebrates than in invertebrates [7] . Recently, a genome-wide investigation of AS profiles across organs and species in vertebrate species, suggested that AS changes may be a driving force towards an increase in cellular complexity during vertebrate speciation [38] . However, a latest research corroborated that boundary shifts and complete intron sliding are only accidental in eukaryotic genome evolution [39] . The number of introns in vertebrates is more than in the other lineages, so it is reasonable to assume that the prevalence of AS in vertebrates is pivotal for their higher phenotypic complexity [40] .

Overall, our results show both general and genome-specific features of the exon-intron structures of eukaryotic genes. The evolution of exon-intron structures increases in complexity from lower eukaryotes to higher eukaryotes. Some species-specific characteristics of genomes were found in many teleosts and lower eukaryotes. This re-analysis of eukaryotic genomic organization revealed some lineage-specific characteristics of exons and introns, which paves way for further research on the conservation and diversification of eukaryotic evolution.