Sequence features responsible for intron retention in human
© Sakabe and de Souza; licensee BioMed Central Ltd. 2007
Received: 11 October 2006
Accepted: 26 February 2007
Published: 26 February 2007
One of the least common types of alternative splicing is the complete retention of an intron in a mature transcript. Intron retention (IR) is believed to be the result of intron, rather than exon, definition associated with failure of the recognition of weak splice sites flanking short introns. Although studies on individual retained introns have been published, few systematic surveys of large amounts of data have been conducted on the mechanisms that lead to IR.
TTo understand how sequence features are associated with or control IR, and to produce a generalized model that could reveal previously unknown signals that regulate this type of alternative splicing, we partitioned intron retention events observed in human cDNAs into two groups based on the relative abundance of both isoforms and compared relevant features. We found that a higher frequency of IR in human is associated with individual introns that have weaker splice sites, genes with shorter intron lengths, higher expression levels and lower density of both a set of exon splicing silencers (ESSs) and the intronic splicing enhancer GGG. Both groups of retained introns presented events conserved in mouse, in which the retained introns were also short and presented weaker splice sites.
Although our results confirmed that weaker splice sites are associated with IR, they showed that this feature alone cannot explain a non-negligible fraction of events. Our analysis suggests that cis-regulatory elements are likely to play a crucial role in regulating IR and also reveals previously unknown features that seem to influence its occurrence. These results highlight the importance of considering the interplay among these features in the regulation of the relative frequency of IR.
Most eukaryotic genes are composed by exons and introns, requiring pre-mRNA splicing. Although splicing occurs with incredible fidelity, a high rate of alternative joining of exons has been observed. Alternative splicing can be mainly of three types: exon skipping, whereby an exon may be included or not in the mature mRNA, alternative use of splice sites, resulting in longer or shorter exons and intron retention (IR), whereby an intron sequence is maintained in the mature transcript or spliced out of it (reviewed in ).
Intron retention is unexpected since it affects mRNA transport to the cytoplasm  and can insert premature stop codons in the mature transcript that would then be degraded by non-sense mediated decay . In fact, decreasing the levels of mRNAs of a given gene can be a function of alternative splicing . This, however, is not the fate of all mRNAs with a retained intron. For example, an isoform of the mouse tgif2 gene that retains an intron in the coding region was shown to have biological activity . Furthermore, we have previously shown that introns retained in the human transcriptome bear signals of being biologically functional as GC content and codon usage similar to those of coding exons and the capacity of encoding protein domains .
In respect to the mechanistic nature of IR, the most intuitive explanation is that these introns are flanked by weak splice sites that occasionally are not properly recognized. Stamm and coworkers  observed weaker conservation of splice sites flanking retained introns. Studies with human genes have shown that strengthening sub-optimal splice sites flanking retained introns caused an increase in their removal levels  or completely abolished retention . On the other hand, other studies showed that weakening optimal splice sites leads to intron retention (chicken troponin I gene , human α-globin 2 , Drosophila melanogaster's zeste , S. pombe's cdc2 gene ).
Interestingly, all the introns involved in these studies were short (< 274 nt). This may have a mechanistic reason according to current models of splice site recognition. In the exon definition model , splice sites are recognized throughout the exons, which in vertebrates are usually shorter than introns. In this case, the non-recognition of a pair of weak splice sites would lead to the skipping of an exon, rather than to IR. However, in organisms whose exons are normally longer than introns (S. pombe,C. elegans and Drosophila, for instance), splice site pairs seem to be recognized throughout introns, the intron definition model . The prevalence of one mechanism over the other seems to be based on the recognition of the shortest unit, either an intron or an exon over which a protein "bridge" connects two exon borders . Therefore, even in vertebrates, short introns could be the recognition unit for splicing and in this case IR is expected if splice sites are not properly recognized .
In addition to splice site variation, some studies showed that the expression of the intron retaining and intron spliced isoforms varies along tissues [5, 14, 15] indicating that cellular factors are also involved in the control of intron retention. In fact, splice site recognition depends on several factors as for example regulatory elements located in the mRNA sequence and on the concentration of protein factors (for review, see ). It has been shown that the presence of specific cis-regulatory elements can indeed counterbalance retention of individual introns [8, 16].
Although the studies mentioned above provide an experimental basis for IR on specific genes, a generalized model based on a large quantity of data is still missing. This model could reveal how different factors contribute to the frequency of IR. We sought to identify potential sequence features that may be associated with this type of alternative splicing by examining the contrasting features of two sets of retained introns found in human cDNAs. One set corresponded to those cases in which the intron retaining isoform was the minor splice form and the other in which it was the major form. We found that a higher relative frequency of IR is associated with weaker splice sites, genes with overall short intron lengths, higher expression levels and particular densities of regulatory elements.
IR events were identified on human cDNA sequences aligned to the genome. An event was considered to be any occurrence of at least one cDNA defining an intron through the existence of its flanking exons plus at least one other spliced cDNA containing at least parts (> 1 nt) of the flanking exons and the intron (see Methods).
In order to understand how splice site strength controls IR, we analyzed retained introns partitioned into two categories, according to their relative isoform frequencies (RIF = number of cDNAs with retained intron/number of cDNAs defining the intron). The low-RIF group contained IR events where the intron retaining cDNA was the minor form (RIF < 1, which corresponds to < 50 % of the cDNAs with IR) and the high-RIF group contained those events where it was the major form (RIF > 1, > 50% of IR). As we wanted to contrast features of the groups of IR events with different relative frequencies of IR, events with an equal number of intron retaining and intron defining evidences (3% of all filtered cases) were not considered.
The initial set of events (6370 low and 1838 high-RIF) was filtered to produce a higher quality data set. Only those events where both the intron retaining and intron defining forms were confirmed by at least two cDNAs each, presenting no alternative borders (see Methods) and retaining introns longer than 50 nt were accepted. The filtered data set was composed by 1516 low and 296 high-RIF IR events.
Partitioning of intron retention events in the human transcriptome according to their relative isoform frequencies (RIF).
number of cDNA clusters*
number of clusters w/> 1 event of the same RIF**
number of events/cluster
low (< 50% retaining forms)
high (> 50% retaining forms)
Splice site strength as a cause of intron retention
Frequency of retained introns with varying Shapiro and Senapathy splice site scores.
% of splice sites
S&S score (5' ss + 3' ss)
around constitutive exons
The influence of flanking exon and intron lengths in intron retention
In organisms like Drosophila  and S. pombe , introns shorter than exons are recognized instead of the exons and weakening of splice sites frequently leads to intron retention. In vertebrates, IR frequently involves short introns (mean length ~100–200 nt [5, 6, 19, 20] and individual examples in Table S2 of Additional file 2: Lengths of retained introns and flanking exons). Therefore, the length of introns is probably an important feature in the mechanism of IR, in agreement with the intron definition model. Indeed, Figure 3C shows a bias for short retained introns in both RIF groups. However, it is difficult to assess whether this trend is real using cDNA data, since ESTs bias the data for short introns. Although in  the authors found a difference in the length of introns retained in different frequencies, the lack of such difference in our data may be due to the fact that by requiring each event to be represented by at least 2 cDNAs, the events identified by us were those with short introns in both RIF groups that had a higher chance of being sequenced multiple times. One study that did not use ESTs to identify intron retention events, employed instead only the extent of genomic sequence conserved in human and mouse as a predictor . The mean length of the 21 retained introns identified was also small (187 nt ± 156).
Although this bias for retention of short introns may also be a result of selection for inclusion of short sequences, another indirect evidence that supports a possible mechanistic relation of short introns and retention is the higher frequency of shorter non-retained introns in intron retaining genes. The average length of introns in our data set is 5580 nt (s.d. 17836, 120,949 introns), close to previous findings (5025 nt ). Genes presenting a low RIF have shorter introns (2005 ± 11797), close to the length of introns of housekeeping genes (2573), according to . Non-retained introns of the high-RIF group are also shorter (3207 ± 12160, the distribution of lengths is presented in Table S3 of Additional file 2: Lengths of retained introns and flanking exons). In agreement with this observation, the mean length of non-retaining introns of the genes with intron retention identified by  without EST data is 3513 nt ± 8337 (166 introns, removing one gene that has an abnormally high mean intron length (LACE1, ~19,000 nt) of the 21). It is possible that this overall shorter length of introns increases the chance of a gene to undergo intron retention.
The last feature related to short introns analyzed was the size of "exon + retained intron + exon" units. Whereas the mean length of retained introns + flanking exons units is around 500 nt (Figure 3D), the mean lengths of "exon + non-retained intron + exon" units are considerably greater (all genes: 4690 nt ± 12572, n = 87824; low-RIF: 2022 nt ± 10997, n = 13486 and high-RIF: 2987 nt ± 10675, n = 2275), following the lengths of non-retained introns alone. The means are not greater than those of the corresponding non-retained introns alone, as it would be expected, because the lengths of "exon + intron + exon" units were calculated only for internal exons, excluding first introns in the 5' UTR, which are typically longer ( and data not shown). Figure 3D shows the distribution of lengths of "exon + retained intron + exon" units (see also Figure S1 in Additional File 2: Lengths of retained introns and flanking exons for lengths of "exon + non-retained intron + exon" units). Independently of a length bias for retention, the fact that many of the IR units are shorter than 400 nt, only slightly above the length limit for exon definition of 300 nt , suggests that this observation may have a mechanistic reason for at least part of the events as we discuss later.
Genes presenting intron retention are more highly and broadly expressed than other genes
We also observed that there is a correlation between RIF and expression levels of genes (Figure 2, panels C and D, low-RIF: ρ = -0.31, P < 10-4, high-RIF: ρ = 0.16, P = 0.01). Note that the correlation is negative for the low-RIF and positive for the high-RIF group. One explanation is that higher expression levels increase the amount of all transcripts, both of the minor and the major forms, but under these circumstances, very rare forms also appear. Rare splice forms in the low-RIF group will be those with strong splice sites that very rarely retain an intron, resulting in the appearance of events with low retention/non-retention ratios only with higher expression (negative correlation). The reverse would be true for the high-RIF group, in which rare forms will be those with very weak splice sites that only rarely remove the intron, and only with higher expression (positive correlation).
Such association of higher expression and RIF may be either because higher expression increases the error rate of the splicing machinery, specially because it seems that splicing occurs co-transcriptionally , or simply allows IR (or intron removal, in the case of the high-RIF group) to be detected since more transcripts are produced. This latter explanation is attractive because we verified that clusters that presented intron retention have more cDNAs than all other genes (for example, ~17% of the intron retaining and 6% of all other clusters, have > 400 cDNAs, χ2-test, 1 d.f., P < 10-10, see distribution of cluster sizes in Figure S2 of the Additional file 3: Analysis of gene expression).
Distribution and density of splicing cis-regulators
Cis-regulatory elements (exonic splicing enhancers and silencers (ESE and ESS), intronic splicing enhancers and silencers (ISE and ISS)), play an important role in constitutive and alternative splicing by acting as binding sites for trans-factors that promote (enhancers) or inhibit (silencers) splicing (for review, see ). For example, a GAA ESE was shown to be crucial to counterbalance retention of intron D of the bovine growth hormone gene  and of intron 3 of the 9G8 splicing factor gene .
As not all motifs corresponding to cis-regulatory elements found in exons or introns are functional, one hypothesis is that they are present in higher densities (motifs/nt) in the sequences they act . Recently, Wang et al  showed that a particular set of ESSs that enhance splicing of retained introns is enriched in these sequences.
Densities of putative cis-regulatory motifs in exons, retained and non-retained introns of genes of the low and high-RIF groups.
FAS-ESS hex-3 class 1
FAS-ESS hex-3 class 2
FAS-ESS hex-3 class 1
FAS-ESS hex-3 class 2
Retained introns were also different from exons, but with slightly higher densities of SELEX-ESEs, intermediary to exons and non-retained introns (Table 3). This trend was not observed for RESCUE-ESEs and GAA, probably because these motifs are purine-rich (40% of RESCUE motifs have at least one GAA; notice that RESCUE-ESE and GAA data follow each other's tendencies closely). It seems that the peculiar nucleotide composition of retained introns can harbor less purine-rich motifs, but SELEX-ESEs are not affected in the same way, since they have a different nucleotide composition (see Table S5 in Additional File 5: Density of cis-regulatory elements in long exons).
Major differences between the low and high-RIF groups are those related to the densities of GGG and ESSs. Low-RIF retained introns presented a slightly higher density of GGG than non-retained introns, whereas high-RIF retained introns presented the opposite trend (Table 3 and Figure 5A). In regard to FAS-hex3 ESSs, Wang et al  showed that class 1 ESSs are actually under-represented in retained introns in relation to non-retained introns, which was also observed in our data for both RIF groups. Class 2 FAS-ESSs, however, followed the same trend of GGG, showing higher densities in the low-RIF than in non-retained introns and lower densities in the high-RIF group (Table 3 and Figure 5B). This difference between class 1 and class 2 ESSs is probably related to the fact that the latter are G-rich (5/26 class 1 ESSs and 28/46 class 2 ESS have at least one GGG triplet).
As IR will result in one long exon, being the commonest isoform of the high-RIF group, we evaluated the possibility that such differential motif distribution between flanking exons and retained introns could be actually typical of an architecture of long exons (divided in flanking segments and a central segment corresponding to the intron, i.e. a pseudo-retained intron), and not to retained introns themselves. Analysis of exons of length between 300–600 nt and > 600 nt showed that these sequences do not present the pattern observed for "exon + retained intron + exon" units (Additional file 5: Density of cis-regulatory elements in long exons). Rather, the flanking segments and pseudo-retained introns presented motif densities similar to exons in general and therefore higher than those of retained and non-retained introns, indicating that the observed differences in Table 3 are specific to introns that are retained.
The ESS/GGG densities observed could suggest that cis-regulatory elements play a role in differentially regulating IR in the two RIF groups. A higher ESS/GGG density in low-RIF retained introns could increase the efficiency of intron splicing (thus, lower RIFs) whereas for high-RIF introns, the inverse would be true. The basis for this argument is that knocking out different GGG triplets in the second intron of α-globin 2 leads to different levels of intron retention  and that inserting ESSs in the frequently retained intron 3 of the NKIRAS2 gene increased the levels of the spliced form . This would explain why introns with S&S score in the range of 140–160 can be retained in both low and high RIFs (see the distribution of S&S scores in Table 2 and GGG/ESS density as a function of S&S scores in Figures 5C and 5D).
The interpretation of the slightly higher SF2/ASF and SC35 densities observed in retained introns is not straightforward. Although it might be only due to a biased nucleotide composition, another possibility is that these ESEs could help defining short "exon + retained intron + exon" units as a single exon (see Discussion), increasing retention. The other possibility is an action in the recognition of splice sites. Although ESEs are exonic by definition, SR proteins binding to introns was shown to both increase and decrease splicing in different studies (see discussion in [25, 27]). If these ESEs increase intron splicing in the events analyzed here, they would counterbalance intron retention. Both cases could be happening in different intron retention events, but the fact that low-RIF densities for SF2/ASF and SC35 are higher than in high-RIF could indicate that their role is in counterbalancing intron retention. However, the signals are too subtle to allow definite conclusions to be made.
One could argue that both the higher density of SELEX-ESEs and lower ESS/GGG density in the high-RIF set of retained introns could be due to selection on sequences that occasionally are exonic, but this possibility is challenged by RESCUE-ESE and class 1 FAS-ESS, whose densities are similar to bona fide non-retained introns.
Conservation of the frequency of intron retention in mouse
Evolutionary conservation of given features is often taken as a signal of biological function, under the assumption that negative selection is acting on them. Studying conserved IR events could support the observations made with human data and therefore help understanding the mechanism of intron retention. As the criteria we used to select high-confidence human IR events may have filtered out true events, in this analysis we searched the mouse genome and transcriptome for all the IR events initially detected. With the exception of the number of conserved events, all other data (identities and percentages) were the same when searching only filtered human events.
Conservation of human intron retention events in the mouse.
1. mean %ID (human retained introns versus mouse genome)1
59% ± 10 (1582 introns)
81% ± 9 (408 introns)
2. mean %ID (exons)
91% ± 3
91% ± 3
3. number of events presenting only the non-retaining form
4. number of events presenting only the retaining form
5. number of conserved events2
6. number of conserved events with conserved RIF
7. mean %ID of conserved retained introns with conserved RIF
86% ± 10 (22 retained introns)
91% ± 9 (7 retained introns)
8. mean %ID of non-conserved retained introns2
59% ± 9 (no retention in mouse; 1252 introns)
82% ± 9 (no non-retained form in mouse; 177 introns)
9. mean length of conserved retained intron
152 ± 83 (22 introns)
142 ± 87 (7 introns)
10. information content of splice sites4
7.2 bits ± 0.2/9.7 bits ± 0.4 (22 introns, all GT..AG)
5.4 bits ± 1.3/7.4 bits ± 1.3(7 introns, all GT..AG)
Most IR events presented only the major form in mouse (Table 4, rows 3 and 4), being considered non-conserved. Those events that presented both minor and major forms in both human and mouse were deemed conserved (29 low-RIF and 8 high-RIF). Among these events, 22/29 in the low-RIF and 7/8 in the high-RIF group presented conserved RIFs. According to our supposition, 8/22 and 2/7 of the human low and high-RIF events, respectively, had been filtered out. This indicates that our filter was stringent, removing possible true events, and considering all the events in the mouse analysis was adequate. The small percentage of events found to be conserved is probably due to the lack of cDNA data covering the homologous region, rather than lack of conservation.
Interestingly, conserved retained introns presented a high percent identity, irrespective of the RIF group (Table 4, row 7), close to that of exons in general, whereas those human IR events of the low-RIF group which presented only the major form in mouse presented the same percent identity of non-retained introns (Table 4, row 8).
As observed for human there was a difference in the information content of splice sites, those of the low-RIF group showing higher conservation (Table 4, last row, see sequence logos as Additional file 6: Sequence logos of splice sites of retained introns conserved in mouse). Irrespective of the frequency of retention, retained introns in the mouse were considerably shorter than the average non-retained introns. Data for the distribution and density of cis-regulatory elements are not provided for mouse conserved events due to the extremely reduced size of the data set. It is worth noting, however, that the trend of GGG/ESS density in conserved events was consistent with that of human, i.e. notably higher in low-RIF retained introns.
The main problem in analyzing IR is to guarantee that the events are authentic alternative splicing occurrences and not part of partially processed messages. However, the fact that we found a mechanistically coherent association of splice site strength and differential densities of cis-regulatory motifs with the two RIF groups (corroborated by conserved events) indicates that our data set contains a large fraction of authentic IR events.
In agreement with previous experimental observations on individual vertebrate retained introns [7–9, 16] and with bioinformatic analyses [6, 20], we found that IR is associated with weaker splice sites. We extended this general observation showing that a decrease in splice site strength leads to higher relative frequencies of retention. However, a non-negligible fraction of events bear strong splice sites. Extending a previous observation that a specific set of ESSs was able to inhibit intron retention , we showed that there is an inverse association of the density of these regulatory elements (and also of the ISE GGG) with the frequency of IR. The fact that a considerable fraction of low-RIF retained introns presented strong splice sites and high GGG/ESS density indicates that other factors more complex than those studied here are also involved in the regulation of this type of alternative splicing as for example the relative concentration of trans-factors, the formation of secondary structures in the RNA molecule or the dependence of splicing of upstream introns for retention of a specific intron, as observed for the thrombopoietin gene .
Although cDNA data sets like ours are possibly biased for short retained introns, other observations, not directly based on cDNA data, support the notion that such introns are short ([6, 21] and see Additional file 2: Lengths of retained introns and flanking exons). Together with our observation that genes with IR present overall shorter introns these data suggest that this may have mechanistic reasons for at least part of the events. One attractive possibility is that these short retained introns are part of longer units (up to 300 nt, the limit for exon definition ) that are at times recognized as single long exons and at times as split "exon + intron + exon" structures. In support to this view, flanking exons seem to have the length of exons in general. In addition, the retained introns (or intra-exonic segments) contain intronic features necessary to be occasionally spliced out, such as (weak) cryptic splice sites and splicing enhancers as GGG triplets that enhance intron removal through intron definition  and ESSs. According to our data, in this model the relative frequency of retention would be regulated by the strength of splice sites and GGG/ESS density.
The mechanism in model A could be related to the fact that handling long exons is complicated for vertebrate spliceosomes  and it is tempting to speculate that the model could be extended to "exon + retained intron + exon" units longer than 300–400 nt. It was observed that an expanded exon of 566 nt could not be correctly spliced, but suffered internal splicing, creating three new small exons of 63, 90 and 73 nt and two introns shorter than 90 nt each. When expanding to 894 nt, two exons of 63 and 73 nucleotides and one intron were created . These observations support the idea proposed in model A that retained introns would use weak internal cryptic splice sites that are recognized during exon definition of a long exon.
However, we also verified that there are cases where retained introns are much longer than those known to be defined and would not fit in the relatively short "exon + intron + exon" units above. For these cases, we propose an alternative, and speculative, model (mechanism B in Figure 6) that does not involve intron definition. This mechanism implies that the introns flanking the retained one should be removed first, i.e., intron removal order may dictate the outcome of the splicing pattern, which has already been observed to influence exon skipping . Occasionally, the transition from E to A or even from A to B complex (during exon-definition to intron-bridging complex conversion), which are possible points of alternative splicing regulation , could suffer some kind of interruption due to the weakness of splice sites, halting later steps and retaining the intron.
Gene expression level would not be an intrinsic part of the model proposed above, but may contribute by even abolishing intron retention in tissues where expression is lower. The occurrence of more than one event per gene specially in the low-RIF group may be related to the fact that these genes present overall shorter introns, increasing the probability of an intron being retained through model A. With higher gene expression, retention of introns with moderately strong splice sites would be facilitated, though in low frequency. For introns to be retained in high frequency, splice sites would have to be considerably weaker, which due to selection are found in a minority of the introns, and therefore the chance of multiple events happening in the same gene is probably smaller.
Although splice site strength is associated with a large fraction of IR events and in these cases is likely to determine whether an intron will be retained, and also in which frequency, other factors seem to be crucial in driving the regulation of this type of alternative splicing. The presence of specific regulatory motifs, by counterbalancing splice site strength, mainly for those introns that have moderately high splice sites scores seem to be very important in driving the frequency of IR. Other features as intron length and even gene expression, together with still unknown factors, also seem to play important roles in the regulation of this phenomenon.
Sequence data, cDNA alignment and clustering
The human genome sequence (build 35), mRNA and EST sequences (May, 2004) were downloaded from the UCSC Genome Browser  and cDNA clustering was performed as explained in [5, 31]. Briefly, mRNAs and ESTs were aligned to the genome using SIM4  to obtain alignments with splice sites precisely defined. Only the best alignment for each sequence was kept. Clustering of cDNAs was performed by comparing genomic coordinates of aligned sequences. In order to be included in the same cluster, two sequences had to present at least one similar exon/intron boundary (± 3 nt) or an overlap of at least 30 nt if the sequences only overlapped in the extremities.
Annotation of intron retention events and data filtering
Only spliced sequences were used for alternative splicing annotation. Intron retention, exon skipping and constitutive exons were identified in clustered cDNAs by comparison of genomic coordinates with a tool developed by us . An IR event was annotated for any occurrence of at least one cDNA (ESTs and mRNAs) defining an intron through the existence of its flanking exons plus at least one other spliced cDNA containing at least parts (> 1 nt) of the flanking exons and the intron (see Additional file 7: Examples of intron retention events).
In order to obtain a high confidence set of IR events, the initial sets of 6370 low and 1838 high-RIF events were filtered to 1515 and 250 events, respectively (see Additional file 8: List of low-RIF events and Additional file 9: List of high-RIF events). Only those events where both the intron retaining and intron defining forms were confirmed by at least two cDNAs each (in the case of ESTs, from different libraries) were accepted. Also, retained introns with alternative splice borders were discarded. The only exception was when the most frequent border was confirmed by > 10 cDNAs and there was only one alternative border evidenced by a single cDNA. Only the most frequent border was used for the analyses. Retained introns shorter than 50 nt were discarded.
Splice site identification
For each IR event of the filtered data set, one representative sequence was selected to guide extraction of splice sites from the genome. To avoid misaligned splice borders, only exon/intron borders identified by SIM4 that did not present gaps or mismatches in a window of 20 nt in the exon from both borders were accepted. Only alignments with splice site directionality coherent with the alignment strand of mRNAs in the cDNA cluster were accepted. Information content of splice borders was calculated using DELILA's rseq  for positions -3 to +6 for 5'ss and for -20 to +3 for 3'ss. Sequence logos were generated using WebLogo .
Gene expression measurements
The number of tissues in which a given gene is expressed (breadth) and the expression levels were determined through SAGE tag counts . Short SAGE (10 nt) libraries were downloaded from SAGE Genie .
Expression levels: for one mRNA with a poly(A) tail (at least 5 consecutive As in the 3' end of the sequence) from each cDNA cluster, the 3' most NlaIII site was searched and 10 nt downstream were extracted as the bona-fide SAGE tag produced by the given gene . The tag was then searched in all SAGE libraries and its frequency counted. Redundant tags (same tag for at least 2 different genes) were excluded. Tag frequencies were normalized for library size by dividing the tag frequency by the total number of tags in the library multiplied by an arbitrary factor of 200,000. Normalized counts greater than 500 were considered outliers, representing about 0.3 % of the counts.
Comparison of gene expression levels among sets of genes: the global expression level of a group of genes for a given tissue was taken as all normalized expression levels of all libraries available for the tissue. To compare expression levels between a given experimental group (housekeeping, low or high-RIF) and all other genes, a bootstrap-like process of data randomization was performed. Ten thousand random subsets of the size of the experimental group were generated from the set of expression levels of all genes and had their mean expression level compared to that of the experimental set. If the random expression was different than (higher or lower, depending on the tissue) that of the experimental set, the counter was increased by 1. Expression levels were considered different when at most 5% of the random subsets compared had an expression level higher (or lower, depending on the tissue) than or equal to that of the experimental set.
Density of splicing cis-regulatory motifs
Putative and experimentally tested cis-regulatory elements from the literature were searched in exon, non-retained intron and retained intron sequences. Four scoring matrices for SF2/ASF, SC35, SRp40 and SRp55 (SELEX-ESEs) and their default cut-off scores  were used as sample ESEs. In addition, 238 putative ESEs  were scanned together as one category (RESCUE-ESEs). The ESE motif GAA , ESSs from  (FAS-hex3) and GGG, a known ISE  were also scanned. To normalize for length, only the initial and final 200 nt were scanned, sliding 1 nt at a time. For sequences shorter than 400 nt, the entire sequence was scanned. All occurrences were then counted and summed in their respective categories and divided by the total number of nucleotides scanned to yield motif densities (motifs/nt). The evaluation of the statistical significance of different densities in each sequence set was carried out with a bootstrap-like procedure similar to that used for gene expression measurements.
Conservation of intron retention in mouse
Identity of human/mouse introns and exons: human cDNAs without the retained intron were aligned with Blast  (E-value = 10-10) to the mouse genome sequence (build 33) obtained from UCSC (May, 2004)  and homologous intron positions were determined. Only introns that presented similar borders (± 10 nt) and lengths (± 15%) in both organisms were accepted. Also, the hit had to present continuous cDNA coordinates, i.e., the two exons defining the mouse intron could have a maximum gap of 20 nt in the cDNA, to ensure the intron was not simply a long alignment gap. When multiple hits existed, the one with the lowest E-value was preferred. Introns found in mouse were further globally aligned to human retained intron sequences with needle  using default parameters and had their percent identities averaged. The same procedure was used to determine the identity of non-retained introns. The identity of exons was obtained from the alignment of flanking exons to mouse cDNAs.
Conservation of IR events: human cDNA blocks corresponding to "exon + retained intron + exon" segments were aligned to mouse ESTs and mRNAs. Matches corresponding to the coordinates of the retained intron plus parts of the flanking exons (at least -20 nt upstream and +20 nt downstream) were accepted as evidence of intron retention in the mouse (only spliced alignments were accepted). Matches corresponding to exon + exon where the exon/intron borders were similar by ± 10 nt were taken as evidence of the intron being excised in a given mouse cDNA. The cDNA also had to be continuous (gap < 20 nt). Splice sites were identified by aligning mouse cDNAs without the retained intron to the genome with SIM4, using the same procedure applied for human IR events.
List of abbreviations
5' splice site
3' splice site
exonic splicing enhancer
exonic splicing silencer
fluorescence-activated screen for exonic splicing silencers
intronic splicing enhancer
relative isoform frequency
We thank Maria Vibranovski, Pedro AF Galante and Robson de Souza for discussions along the work and comments on the manuscript. We also thank PAFG for providing the cDNA mapping and clustering, the SAGE database and technical help and CB Burge for kindly providing the sequences of the FAS-hex3 ESSs. NJS was supported by a PhD fellowship from FAPESP.
- Matlin AJ, Clark F, Smith CW: Understanding alternative splicing: towards a cellular code. Nat Rev Mol Cell Biol. 2005, 6: 386-398. 10.1038/nrm1645.PubMedView ArticleGoogle Scholar
- Nott A, Meislin SH, Moore MJ: A quantitative analysis of intron effects on mammalian gene expression. RNA. 2003, 9: 607-617. 10.1261/rna.5250403.PubMed CentralPubMedView ArticleGoogle Scholar
- Lareau LF, Green RE, Bhatnagar RS, Brenner SE: The evolving roles of alternative splicing. Curr Opin Struct Biol. 2004, 14: 273-282. 10.1016/j.sbi.2004.05.002.PubMedView ArticleGoogle Scholar
- Melhuish TA, Wotton D: The Tgif2 gene contains a retained intron within the coding sequence. BMC Mol Biol. 2006, 7: 2-10.1186/1471-2199-7-2.PubMed CentralPubMedView ArticleGoogle Scholar
- Galante PA, Sakabe NJ, Kirschbaum-Slager N, de Souza SJ: Detection and evaluation of intron retention events in the human transcriptome. RNA. 2004, 10: 757-765. 10.1261/rna.5123504.PubMed CentralPubMedView ArticleGoogle Scholar
- Stamm S, Zhu J, Nakai K, Stoilov P, Stoss O, Zhang MQ: An alternative-exon database and its statistical analysis. DNA Cell Biol. 2000, 19: 739-756. 10.1089/104454900750058107.PubMedView ArticleGoogle Scholar
- Romano M, Marcucci R, Baralle FE: Splicing of constitutive upstream introns is essential for the recognition of intra-exonic suboptimal splice sites in the thrombopoietin gene. Nucleic Acids Res. 2001, 29: 886-894. 10.1093/nar/29.4.886.PubMed CentralPubMedView ArticleGoogle Scholar
- Dirksen WP, Sun Q, Rottman FM: Multiple splicing signals control alternative intron retention of bovine growth hormone pre-mRNA. J Biol Chem. 1995, 270: 5346-5352. 10.1074/jbc.270.10.5346.PubMedView ArticleGoogle Scholar
- Sterner DA, Berget SM: In vivo recognition of a vertebrate mini-exon as an exon-intron-exon unit. Mol Cell Biol. 1993, 13: 2677-2687.PubMed CentralPubMedView ArticleGoogle Scholar
- McCullough AJ, Berget SM: G triplets located throughout a class of small vertebrate introns enforce intron borders and regulate splice site selection. Mol Cell Biol. 1997, 17: 4562-4571.PubMed CentralPubMedView ArticleGoogle Scholar
- Talerico M, Berget SM: Intron definition in splicing of small Drosophila introns. Mol Cell Biol. 1994, 14: 3434-3445.PubMed CentralPubMedView ArticleGoogle Scholar
- Romfo CM, Alvarez CJ, van Heeckeren WJ, Webb CJ, Wise JA: Evidence for splice site pairing via intron definition in Schizosaccharomyces pombe. Mol Cell Biol. 2000, 20: 7955-7970. 10.1128/MCB.20.21.7955-7970.2000.PubMed CentralPubMedView ArticleGoogle Scholar
- Berget SM: Exon recognition in vertebrate splicing. J Biol Chem. 1995, 270: 2411-2414.PubMedView ArticleGoogle Scholar
- Popielarz M, Cavaloc Y, Mattei MG, Gattoni R, Stevenin J: The gene encoding human splicing factor 9G8. Structure, chromosomal localization, and expression of alternatively processed transcripts. J Biol Chem. 1995, 270: 17830-17835. 10.1074/jbc.270.30.17830.PubMedView ArticleGoogle Scholar
- Ledee DR, Chen J, Tonelli LH, Takase H, Gery I, Zelenka PS: Differential expression of splice variants of chemokine CCL27 mRNA in lens, cornea, and retina of the normal mouse eye. Mol Vis. 2004, 10: 663-667.PubMedGoogle Scholar
- Lejeune F, Cavaloc Y, Stevenin J: Alternative splicing of intron 3 of the serine/arginine-rich protein 9G8 gene. Identification of flanking exonic splicing enhancers and involvement of 9G8 as a trans-acting factor. J Biol Chem. 2001, 276: 7850-7858. 10.1074/jbc.M009510200.PubMedView ArticleGoogle Scholar
- Mount SM, Burks C, Hertz G, Stormo GD, White O, Fields C: Splicing signals in Drosophila: intron size, information content, and consensus sequences. Nucleic Acids Res. 1992, 20: 4255-4262. 10.1093/nar/20.16.4255.PubMed CentralPubMedView ArticleGoogle Scholar
- Shapiro MB, Senapathy P: RNA splice junctions of different classes of eukaryotes: sequence statistics and functional implications in gene expression. Nucleic Acids Res. 1987, 15: 7155-7174. 10.1093/nar/15.17.7155.PubMed CentralPubMedView ArticleGoogle Scholar
- Sugnet CW, Kent WJ, Ares MJ, Haussler D: Transcriptome and genome conservation of alternative splicing events in humans and mice. Pac Symp Biocomput. 2004, 66-77.Google Scholar
- Zheng CL, Fu XD, Gribskov M: Characteristics and regulatory elements defining constitutive splicing and different modes of alternative splicing in human and mouse. RNA. 2005, 11: 1777-1787. 10.1261/rna.2660805.PubMed CentralPubMedView ArticleGoogle Scholar
- Ohler U, Shorom N, Burge CB: Recognition of unknown conserved alternatively spliced exons. PLoS Comput Biol. 2005, 1: 113-122. 10.1371/journal.pcbi.0010015.PubMedView ArticleGoogle Scholar
- Eisenberg E, Levanon EY: Human housekeeping genes are compact. Trends Genet. 2003, 19: 362-365. 10.1016/S0168-9525(03)00140-9.PubMedView ArticleGoogle Scholar
- Hong X, Scofield DG, Lynch M: Intron Size, Abundance, and Distribution within Untranslated Regions of Genes. Mol Biol Evol. 2006, 23: 2392-2404. 10.1093/molbev/msl111.PubMedView ArticleGoogle Scholar
- Kornblihtt AR, de la MM, Fededa JP, Munoz MJ, Nogues G: Multiple links between transcription and splicing. RNA. 2004, 10: 1489-1498. 10.1261/rna.7100104.PubMed CentralPubMedView ArticleGoogle Scholar
- Wu Y, Zhang Y, Zhang J: Distribution of exonic splicing enhancer elements in human genes. Genomics. 2005, 86: 329-336. 10.1016/j.ygeno.2005.05.011.PubMedView ArticleGoogle Scholar
- Wang Z, Xiao X, Nostrand EV, Burge CB: General and specific functions of exonic splicing silencers in splicing control. Mol Cell. 2006, 23: 61-70. 10.1016/j.molcel.2006.05.018.PubMed CentralPubMedView ArticleGoogle Scholar
- Ibrahim C, Schaal TD, Hertel KJ, Reed R, Maniatis T: Serine/arginine-rich protein-dependent suppression of exon skipping by exonic splicing enhancers. Proc Natl Acad Sci U S A. 2005, 102: 5002-5007. 10.1073/pnas.0500543102.PubMed CentralPubMedView ArticleGoogle Scholar
- Sterner DA, Carlo T, Berget SM: Architectural limits on split genes. Proc Natl Acad Sci U S A. 1996, 93: 15081-15085. 10.1073/pnas.93.26.15081.PubMed CentralPubMedView ArticleGoogle Scholar
- Takahara K, Schwarze U, Imamura Y, Hoffman GG, Toriello H, Smith LT, Byers PH, Greenspan DS: Order of intron removal influences multiple splice outcomes, including a two-exon skip, in a COL5A1 acceptor-site mutation that results in abnormal pro-alpha1(V) N-propeptides and Ehlers-Danlos syndrome type I. Am J Hum Genet. 2002, 71: 451-465. 10.1086/342099.PubMed CentralPubMedView ArticleGoogle Scholar
- Karolchik D, Baertsch R, Diekhans R, Furey TS, Hinrichs A, Lu YT, Roskin KM, Schwartz M, Sugnet CW, Thomas DJ: The UCSC Genome Browser Database. Nucleic Acids Res. 2003, 31: 51-54. 10.1093/nar/gkg129.PubMed CentralPubMedView ArticleGoogle Scholar
- Automatic annotation of alternative splicing on cDNA Clusters. 2006, [http://www.compbio.ludwig.org.br/asc]
- Florea L, Hartzell G, Zhang Z, Rubin GM, Miller W: A computer program for aligning a cDNA sequence with a genomic DNA sequence. Genome Res. 1998, 8: 967-974.PubMed CentralPubMedGoogle Scholar
- Schneider TD, Stormo GD, Yarus MA, Gold L: Delila system tools. Nucleic Acids Res. 1984, 12: 129-140. 10.1093/nar/12.1Part1.129.PubMed CentralPubMedView ArticleGoogle Scholar
- Crooks GE, Hon G, Chandonia JM, Brenner SE: WebLogo: a sequence logo generator. Genome Res. 2004, 14: 1188-1190. 10.1101/gr.849004.PubMed CentralPubMedView ArticleGoogle Scholar
- Velculescu VE, Zhang L, Vogelstein B, Kinzler KW: Serial analysis of gene expression. Science. 1995, 270: 484-487. 10.1126/science.270.5235.484.PubMedView ArticleGoogle Scholar
- Boon K, Osorio EC, Greenhut SF, Schaefer CF, Shoemaker J, Polyak K, Morin PJ, Buetow KH, Strausberg RL, de Souza SJ, Riggins GJ: An anatomy of normal and malignant gene expression. Proc Natl Acad Sci U S A. 2002, 99: 11287-11292. 10.1073/pnas.152324199.PubMed CentralPubMedView ArticleGoogle Scholar
- Cartegni L, Wang J, Zhu Z, Zhang MQ, Krainer AR: ESEfinder: A web resource to identify exonic splicing enhancers. Nucleic Acids Res. 2003, 31: 3568-3571. 10.1093/nar/gkg616.PubMed CentralPubMedView ArticleGoogle Scholar
- Fairbrother WG, Yeh RF, Sharp PA, Burge CB: Predictive identification of exonic splicing enhancers in human genes. Science. 2002, 297: 1007-1013. 10.1126/science.1073774.PubMedView ArticleGoogle Scholar
- Louie E, Ott J, Majewski J: Nucleotide frequency variation across human genes. Genome Res. 2003, 13: 2594-2601. 10.1101/gr.1317703.PubMed CentralPubMedView ArticleGoogle Scholar
- Altschul SF, Madden TL, Schaffer AA, Zhang J, Zhang Z, Miller W, Lipman DJ: Gapped BLAST and PSI-BLAST: a new generation of protein database search programs. Nucleic Acids Res. 1997, 25: 3389-3402. 10.1093/nar/25.17.3389.PubMed CentralPubMedView ArticleGoogle Scholar
- Rice P, Longden I, Bleasby A: EMBOSS: the European Molecular Biology Open Software Suite. Trends Genet. 2000, 16: 276-277. 10.1016/S0168-9525(00)02024-2.PubMedView ArticleGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.