There are 32,955 annotated protein-coding genes supported by EST sequences in the TAIR database. Of these, 18,285 include both the 5'UTR and the 3'UTR, 527 contain the 5'UTR but not the 3'UTR, 1,979 contain the 3'UTR but not the 5'UTR, and 12,171 contain neither. Considerably more genes were found to lack 5'UTR annotation than 3'UTR annotation, probably due to the directional construction of cDNA libraries from the polyA tail. A number of the annotated 5'UTR sequences are expected to be only partial sequences; however, this is not expected to greatly affect the conclusions of this paper (see below). Many protein-coding genes (72.0%) contain introns. While the non-coding regions are less likely to include introns, these are more commonly found in the 5'UTR: 19.9% of annotated 5'UTRs contain introns; by comparison only 5.6% of 3'UTRs are annotated to include introns. The high number of intron-containing 5'UTRs cannot be explained by the 5'UTR lengths, as the average 5'UTR that has been delimited by EST and/or cDNA sequences is considerably shorter than the average 3'UTR region (5'UTR: median = 99 nucleotides, LQ (lower quartile) = 56 nucleotides, UQ (upper quartile) = 175 nucleotides; 3'UTR: median = 208 nucleotides, LQ = 154 nucleotides, UQ = 283 nucleotides). Indeed, Table 1 shows that the average number of introns per nucleotide is, 1.6 × 10^-3, 2.7 × 10^-3 and 2.9 × 10^-4 in 5'UTRs, CDSs and 3'UTRs, respectively. Since these intron densities are normalized by the total length of sequence, incomplete UTR annotation should not greatly bias these results. Thus the intron density in 5'UTRs is ~60% of the intron density in CDSs and ~5.5 times the intron density in 3'UTRs. In mammals, a marked under-representation of introns in the 3'UTRs may be explained by the requirements for nonsense-mediated decay (NMD) 3839. The corresponding under-representation measured here in A. thaliana may indicate that plants utilize a similar NMD pathway.

Figure 1 shows the size distribution of introns within 5'UTRs, CDSs and 3'UTRs. The 3'UTR and CDS introns have very similar length average distributions (3'UTR: n = 1,468, mean = 208 nucleotides, median = 104 nucleotides, LQ = 88 nucleotides, UQ = 192 nucleotides, SD = 325 nucleotides; CDS: n = 127,141, mean = 158 nucleotides, median = 98 nucleotides, LQ = 85 nucleotides, UQ = 157 nucleotides, SD = 167 nucleotides). 5'UTR introns (n = 4,240, mean = 316 nucleotides, median = 253 nucleotides, LQ = 122 nucleotides, UQ = 416 nucleotides, SD = 292 nucleotides) are, on the other hand, significantly biased towards longer lengths (Kolmogorov-Smirnov test, 5'UTRs v. CDSs, D = 0.41, p-value < 10^-15). Notably, there is a significant under-representation of short introns between 50 nucleotides and 150 nucleotides and a notable increase of introns above 200 nucleotides. These results are not expected to be greatly biased by incomplete UTR annotation. The significant differences between the size of introns that reside within the 5'UTRs compared to introns that reside within CDSs and 3'UTRs may be a feature that influences their role in enhancing translation levels. It is also plausible that the large 5'UTR introns function as spacers within the genome, providing an AT-rich stretch of sequence between coding sequence and promoter.

It has been suggested that the splicing of 5'UTR introns would lead to deposition of EJCs, which interact with translation initiation, resulting in translational enhancement 2223242526. If the EJCs facilitate the recruitment of ribosomes, then there may be selection on the position of introns within 5'UTRs; specifically there may be an optimal length between the position of the intron within the 5'UTRs and the translation start codon. Introns closer to the ATG would recruit the EJC and facilitate an interaction between the RNA, trans-factors and the ribosome. We compared the observed 5'UTR intron position distributions with the distributions that would be expected if introns were distributed uniformly (i.e. constant insertion probability after any given nucleotide) throughout 5'UTRs – calculated using Monte-Carlo simulations (see Methods). Figure 2 presents the distribution of intron positions within the 5'UTRs, relative to the beginning and end of the corresponding 5'UTRs. Clearly, the actual position distribution of 5'UTR introns relative to either the beginning, or the end, of the 5'UTR deviates from the expected distribution. It appears that the introns are more frequently located distant from the beginning of the 5'UTR (80–300 nucleotides after the start of the 5'UTR), and more frequently located proximal to the end of the 5'UTR (1–40 nucleotides before the start codon). Although incomplete 5'UTR annotation may result in some bias in the absolute positional distributions, since the Monte Carlo model distribution is based on the same 5'UTR sequences, the deviations from the expected distribution are real. The proximity of 5'UTR introns to the translation start site is consistent with a role that involves a function in translation, given the simple model of translation we are using. This could be achieved through alteration in the secondary structure of the 5'UTR leader either directly through the act of recruiting the spliceosomal complex during splicing or through the deposition of EJC proteins following processing. Recently the secondary structure of 5'UTRs has been shown to influence post-transcriptional regulation 40; it is interesting to speculate whether the presence of an EJC could influence RNA folding and the consequences of this for gene regulation. Further experiments to alter the 5'UTR intron position will clarify if this has any biological role in post-transcriptional regulation.

An investigation of the nucleotide preference around the splice junctions allows a more detailed comparison between the UTR introns and the CDS introns. Sequence logos 41 were created to visualize the nucleotide conservation around the splicing donor (GT) and the splicing acceptor (AG) junctions for each of the intron categories 5'UTR, CDS and 3'UTR (see Methods). The resulting logos are presented in Figure 3 (donor) and Figure 4 (acceptor). Only six and eight nucleotides of exon sequence from the donor and acceptor sites, and 25 nucleotides of intron sequence were included in each sequence logo, as there was no noticeable nucleotide bias outside these regions. The sequence logos shown in Figures 3 and 4 present all the canonical fingerprints of introns, namely the G/GT and AG/G splice site consensus, and the AT-rich element, which is a key feature for efficient splicing in plant introns 4243. There is, however, a significant C-rich region near the donor site of 5'UTR introns (+8 to +25 bases after intron start) (Figure 3) that does not appear to be present in either CDS or 3'UTR introns. There is also an increase in the occurrence of G and C residues in the region near the acceptor sites of 5'UTR introns (-17 to -8 bases before intron end) (Figure 4). These sequence biases could be necessary for the spliceosomal recognition of introns within non-coding sequence; however, these sequence biases are particular to the 5'UTR and are not seen within the 3'UTR, so their role cannot be general to non-coding introns. Further experimental analysis of the post-transcriptional and pre-translational effect of altering these sequences is necessary before any biological significance can be attributed to these observations.

We also generated transgenic Arabidopsis to confirm the level of IME obtained in transient assay data for the pGEF1I and pGEF1Idel fusions. The plasmids used for our transient assay were modified to contain the plant kanamycin transformation selection gene and transformed into A. thaliana 'Columbia' by the floral dip method 45. Figure 7 shows the absolute LUC activity for two leaves from each of 10 independent T1 plants transformed with the intron-containing EF1α-A3 5'UTR, and from each of 7 independent T1 plants transformed with the intronless EF1α-A3 5'UTR. In all these transgenic lines, the Renilla luciferase gene was inactive, despite being under the transcriptional control of the 35S promoter. Therefore the relative expression of LUC was not determined. It is not unusual for some reporter genes to become inactivated during the transformation process 46, though zero out of 17 is unusual and may reflect a higher susceptibility of the Renilla luciferase gene to this form of gene-silencing. This transgene silencing is probably because of a post-transformation event, such as methylation-mediated transcriptional silencing 47. Nonetheless, although there is much variability in the absolute levels of LUC activity, expressed as relative light units per mg of leaf fresh weight, there is a striking elevation in the level of LUC expression from those transgenic lines transformed with an intron in the 5'UTR relative to those without a 5'UTR intron (t-test, p-value = 0.0037). This is consistent with the transient expression data. These T1 lines were not analyzed for T-DNA copy number, so some of the variability between transgenic lines containing the same construct could be due to the number of T-DNA insertions in each of the transgenic lines.

Post-transcriptional gene silencing (PTGS) or RNAi is a mechanism that is used to degrade RNA transcripts after they have been transcribed 48. PTGS is activated by dsRNA to produce siRNA that can act as signalling molecules, promoting a cascade of mRNA degradation 4950. As IME is a post-transcriptional regulation, it is possible that the inclusion of certain introns within the 5' region of a gene reduces the RNA's susceptibility to siRNA, through some unknown mechanism. In order to assess the involvement of PTGS in IME, a similar transient assay using plasmids pGEF1I and pGEF1Idel (Figure 5) was performed using a modified version of the pSoup helper plasmid that is able to suppress gene silencing 44. To suppress gene silencing the p19 expression cassette from pBIN 61-P19 was cloned into pSoup and therefore resident within the same Agrobacterium cell as the dual-luciferase reporter cassette. The P19 enzyme is a viral silencing suppressor from the tomato bushy top virus (TBTV), which prevents activation of PTGS 51. Although gene expression of both the LUC and REN reporter genes are enhanced when co-expressed with P19, both with and without the intron, under the presence of P19 (0.2185 ± 0.0183 with intron and 0.0692 ± 0.0073 without intron) the enhancement achieved by the presence of the 5'UTR intron was consistent with the relative enhancement that is obtained without P19 (0.1089 ± 0.0081 with intron and 0.0315 ± 0.0034 without intron). As the IME levels in the presence and absence of P19 are similar, we conclude that no component of this IME can be attributed to PTGS.

One of the most prominent observations from the bioinformatics analysis was the dramatic difference in the length distribution of 5'UTR introns compared with both CDS and 3'UTR introns (Figure 1). To test the influence of 5'UTR intron size on post-transcriptional enhancement, a series of truncated EF1α-A3 5'UTR introns were generated. Fifty nucleotides up and downstream of the intron acceptor and donor site were maintained for splicing efficiency so as not to interfere with the possible role of intron sequence proximal to these splicing junction sites. Figure 8 is a schematic representation of the nine intron deletions used in this experiment, grouped according to the position of the deletion point relative to the 3' region of the intron. Figure 9 presents the relative luciferase activity for each of these constructs, along with the complete intron, and the intronless reporter-LUC fusions. Comparing the differences in reporter gene expression for each series of constructs that retain the same 3' portion of the intron sequence (i.e. constructs 1A-C, 2A-C, and 3A-C), shows a clear positive correlation between gene expression and intron length: the longer the intron, the greater the relative level of LUC activity (t-tests: 1ABC p-value = 2.8 × 10^-5; 2ABC p-value = 6.7 × 10^-5; 3ABC p-value = 4.0 × 10^-4). On the other hand, if constructs are grouped by their 5' end, there is no obvious correlation between intron length and LUC activity.

These observations suggest that IME in EF1α-A3 is mediated by, not one, but three or more elements distributed over the 5' 350 nucleotides of the 5' UTR intron, with relatively little effect from the 3' 250 nucleotides. Indeed, multiple AT-rich stimulatory elements have been previously described in plants 1030313233. and, consistent with this, the EF1α-A3 5' UTR intron is AT-rich. It is interesting that the three inferred IME elements – distributed over 300 nucleotides – require a very significant 'increase' in intron length when compared with the median lengths (CDS introns 98 nucleotides; 5'UTR introns 253 nucleotides). Potentially, this offers an explaination as to why the median 5'UTR intron length is so much greater than the median CDS intron length, although experiments on more genes would be required to confirm this.