Theoretically, five different types of molecular events can inactivate introns or
cause the deletion of an intron from a gene, thereby contributing to a decrease in
intron abundance (Additional file 1). The first type of event is precise intron loss (PIL); in this case, intron losses
do not affect the integrity of flanking exons. The second type of event is imprecise
intron loss (IIL), which is accompanied by the insertion and/or deletion (indel) of
nucleotides into/from flanking exons. The third type of event is termed de-intronization;
in this case, sequences are not deleted from the genome, but rather an intronic sequence
is converted into an exonic sequence by mutations that deactivate splicing signals.
The fourth type of event is termed de-exonization, which is the conversion of an internal
exon into an internal portion of an intron by mutations that deactivate splicing signals.
This process leads to the fusion of an exon and its flanking two introns, which creates
a larger intron and therefore decreases the intron number. Finally, the deletion of
an internal exon also results in the fusion of two neighboring introns and consequently
decreases the intron number by one. In this paper, we used the term “intron lossâ€
in a broad sense to include all five of the above types of intron variations. Almost
all previously observed intron losses are PILs; IILs and other types of intron losses
appear only rarely 1]-15]. There are three possibilities for the observed patterns. The first possibility is
that they occur at quite different frequencies. For example, if intron losses are
mediated by mRNA molecules, then all intron losses should be PILs 16]. The second possibility is that intron losses that change coding sequences have essentially
been eliminated by purifying selection. The third possibility is that there is a methodological
bias toward the identification of PILs. It is possible for intron losses to introduce
indels into coding sequences and therefore significantly reduce the similarities between
flanking coding sequences and their orthologous regions. To be confident in identifying
cases of intron loss, researchers generally discard poorly aligned regions 9]-11].
IILs are less frequent than PILs but are not rare in Solanum or Arabidopsis thaliana
As the genomes of tomatoes (Solanum lycopersicum) and potatoes (Solanum tuberosum) diverge by less than 10% 17], we obtained reliable alignments for most of their orthologs. By surveying intron-exon
structural changes in tomato and potato, we found 11 cases of PIL and six cases of
IIL (Figure 1 and Additional file 2).
Figure 1. Two examples of imprecise intron losses. (A) The S. tuberosum gene PGSC0003DMG400000276 lost both an intron and a 21-bp-long downstream exon. The downstream intron was intact
and did not lose any nucleotides; its splicing was supported by??10 RNA-Seq reads.
The splicing of the target intron in S. lycopersicum was also supported by??10 RNA-Seq reads and by EST asmbl_392.tomatov23pasa_pasa4.
The assembly of the variation site in S. tuberosum was supported by nine different Whole Genome Shotgun (WGS) reads. (B) The S. lycopersicum gene Solyc04g007270.2 lost a 204-bp-long intron, a 22-bp-long segment from an upstream exon and a 9-bp-long
segment from a downstream exon. This deletion occurred in the 3?-UTR and did not inactivate
the gene, as supported by 10 RNA-Seq reads. The successful splicing of the target
intron in S. tuberosum was supported by??10 RNA-Seq reads. The assembly of the variation site in S. lycopersicum was supported by??10 WGS reads. For details of the RNA-Seq and WGS reads and other
intron losses, see Additional file 2 and Additional file 3.
The species A. thaliana diverged from its relative Arabidopsis lyrata less than ten million years ago 18]. In comparing the genomes of these species, we found 17 IILs in A. thaliana (Additional file 3). A close examination of 114 cases of intron loss from A. thaliana that were reported in a previous study 19] revealed that 104 of these cases, which occurred in 98 genes, were PILs, two were
IILs, and eight lacked support due to insufficient numbers of RNA-Seq reads. The two
cases of IILs were included in our 17-case dataset of IILs.
Nearly all ILs have been fixed during evolution
Transcriptome data showed that all of the variant genes in our study were still actively
expressed (Additional files 2 and 3). A close examination of these genes did not reveal any premature stop codons that
were introduced by intron loss mutations. The IIL genes are unlikely to be pseudogenized.
Indels caused by IILs in coding sequences are more likely to be selected against.
It is possible that the IILs that we observed were recent events that would soon be
eliminated. This possibility could be excluded if the variations that we observed
in one species were also found to exist in another. For this reason, we investigated
whether tomato’s wild relative, Solanum pimpinellifolium, shares intron variations with tomato. In S. lycopersicum, there are five IILs and five PILs. We surveyed the orthologous genes in the genome
of S. pimpinellifolium and found all ten of the variations. Therefore, all of the intron losses that we
observed in S. lycopersicum have been fixed during evolution.
Taking advantage of the availability of genome sequences corresponding to multiple
A. thaliana lines, we also tested whether the intron variations that were observed in A. thaliana have been fixed during evolution. By surveying 17 IIL genes in 180 lines of A. thaliana from Sweden 20], we found that all 17 cases of IILs had been fixed. Similarly, by surveying 104 PILs
in 180 lines of A. thaliana, we found that 101 cases of PIL have been fixed during evolution and only three PILs
remained polymorphic, in genes AT1G48420, AT3G23080, and AT4G00350. However, the allele frequencies of these three PILs were high: 97.7% for AT1G48420, 36.1% for AT3G23080 and 97.2% for AT4G00350.
IILs are expected to be under negative selection because of the indels that they cause
in coding sequences; therefore, a majority of IILs might be eliminated. Nevertheless,
the fixed IILs were still found to comprise an appreciable proportion of intron losses
in Solanum and Arabidopsis.
The relative frequencies of IILs were underestimated when comparing distantly related
genomes
The practice of filtering unreliable alignments may have led to underestimates of
the frequencies of IILs in previous studies. If this were the case, a higher proportion
of IILs than PILs would have been undetectable when we compared each of the two Solanum genomes with a distantly related species versus when we compared the two Solanum genomes to each other. To test this possibility, we surveyed for the presence of
Solanum IL genes in the genome of a rice, Oryza sativa, that diverged from Solanum 163 million years ago 18]. Among the 11 PIL genes and six IIL genes that we surveyed, we found O. sativa orthologs for all 11 PIL genes and only two IIL genes. In principle, when a very
low identity is observed between two aligned sequences, it indicates that either a
large number of mutations accumulated after divergence or that a low-quality alignment
was produced. To maintain accuracy, these alignments are generally discarded. We calculated
the identities of the coding sequence alignments of the sites flanking the intron
losses. For each intron loss, 45-bp-long regions of coding sequence (not including
gaps in the alignments) corresponding to positions on each side of the loss were included
in the calculation. Using the first quartile of the identities of all aligned coding
sequences that were generated between Solanum and O. sativa (0.53) as the threshold to filter unreliable alignments, one IIL and two PILs in
Solanum were discarded. In summary, only one of the six IILs that occurred in Solanum could be detected when comparing its genome to rice. In contrast, nine of the 11
PILs could be detected using the same method. This difference is statistically significant
(?2 test, P?=?0.03). Similar analyses have been carried out on the introns that have been lost
from the genome of A. thaliana. Rice and A. thaliana also diverged 163 million years ago 18]. Among 17 IIL genes and 98 PIL genes that were found in A. thaliana, we found four IILs and 47 PILs when comparing its genome against rice. Thus, a lower
frequency of IILs than PILs was observed in A. thaliana when the reference genome was O. sativa (23.5% vs. 45.1%, respectively).
Similar to intron losses, the majority of previous studies on intron gains have been
restricted to highly conserved orthologous genes. Among these genes, very few or no
intron gains were found in humans, mice, and Arabidopsis thaliana19],21]-23]. This is in stark contrast to a study that specifically explored intron gains by
evaluating segmental duplications, in which tens of intron gains had been revealed
in each of these three species 24]. In that study, intron gains that were accompanied by insertions and/or deletions
of coding sequences were not excluded.
Identifying the true relative frequencies of PILs and IILs presents a dilemma: accurate
IIL to PIL ratios can only be obtained when recently diverged genomes are compared.
However, recent divergence means that only a limited amount of time has elapsed to
enable the accumulation of intron loss variations. For these reasons, it would be
helpful in the future to extend the current study to additional eukaryotic lineages.
De-exonization and exon deletion in Solanum
We identified one case of de-exonization, one case of exon deletion, two cases of
intronization, and one case of exonization in tomato and potato (Additional file 2). In the tomato gene Solyc09g016940.2, an internal exon and the 5? splicing signal of its downstream intron were both lost.
In the potato gene PGSC0003DMG400004043, an internal exon has been converted into an internal region of a larger intron.
In addition to GT-AG boundaries, there are many cis-acting sequence elements and trans-acting
factors that facilitate intron recognition and splicing 25]; therefore, it is reasonable to hypothesize that some intron variations do not involve
changes to these boundaries.