To check the possibility that Alu-derived exons were fixed into protein-coding genes through an RNA-editing-mediated process, we first used expressed sequence tags (ESTs) and cDNAs from GenBank version 136 aligned to the human genome (version hg16) to identify internal human exons that contain Alu elements, as described in 5. We looked for Alu-containing exons flanked by either AA at the 3' ss or AT at the 5'ss. These non-canonical splice sites will normally not be selected by the splicing machinery; however, each of these splice sites is theoretically capable of becoming a bona fide splice site through A-to-I RNA editing, because inosine is recognized by the splicing apparatus as a guanosine 19. We demanded that the Alu-exon will be supported by more than one EST/cDNA, and that it neither induced frame-shift nor contained a stop codon, as these parameters were shown to be indicative of functional alternatively spliced exons 5.

We were able to detect one such Alu-derived exon, having an AA 3' ss, that conformed to the above conditions (Figure 1). This exon is the eighth exon in the nuclear prelamin A recognition factor (NARF), a protein that interacts with the carboxyl-terminal tail of prelamin A and localizes to the nuclear lamina 20. As with all internal Alu-containing exons described to date, it is alternatively spliced, and exon-inclusion is supported by seven cDNAs, including the full-length cDNA (GenBank:BC000438; UCSC March 2006 version). The exon inserts 46 in-frame additional amino acids into the coding sequence of NARF.

A-to-I RNA editing takes place in the context of dsRNA, to which the ADAR proteins bind through a dsRNA-binding domain (reviewed in 2122). It was therefore shown that, for the vast majority of edited Alus in human exons, there is a nearby (up to 2,000 base-pairs apart) intronic Alu counterpart in the opposite orientation, which presumably serves as the template for dsRNA formation 1314. Indeed, we were able to find an oppositely oriented Alu sequence 25 bases upstream of the exonized Alu. The antisense Alu has 81% identity when aligned to the exon-producing Alu, suggesting that these two Alus might form a stable intramolecular dsRNA formation following transcription (Figure 1b). This, in addition to the non-canonical AA splice site, implies that RNA editing participates in the exonization of that Alu.

Editing within Alu elements frequently occurs in more than one position, due to the long RNA duplex usually formed by two oppositely oriented nearby Alus 111314151823. Indeed, by searching for A→G discrepancies in the alignments of cDNAs to the human genome we detected five additional potential editing sites in the Alu-exon (Figure 1b, E1-E5). The first of these, found in position 19 of the exon, is of particular interest, because it has the potential to change a TAG codon (termination of translation) to a TGG codon (coding for tryptophane). In the absence of RNA editing in the E1 position, the insertion of this Alu-exon would have caused a premature termination. It is important to note that the editing in the exonic E1-E5 sites is directly recognizable from the ESTs or cDNAs in comparison to human genomic DNA, whereas editing in the potential 3'ss is postulated based only on the genomic sequence.

To check whether this putative Alu-exon is indeed spliced into the mature mRNA of NARF, we tested the existence of the exon-inclusion and exon-skipping forms in endogenous mRNA from various normal human tissues and cell-lines. As shown in Figure 2, the inclusion form was detected in all cDNAs generated from normal human tissues as well as from different human cell lines. This indicates that the exonization of the NARF Alu is evolutionarily fixed in the human transcriptome. Moreover, exon inclusion levels in different tissues followed expected levels of RNA editing in those tissues. For example, brain, kidney and spleen showed the highest levels of exon inclusion while skeletal muscle showed the lowest levels of exon inclusion (Figure 2a; Additional data file 1). These results are in line with genome wide analysis of edited RNA in different tissues 5111315 and to the amount of inosine detected in RNA in various tissues 24. The above results further suggest that RNA editing is involved in the regulation of alternative splicing of the exonized Alu in the gene encoding NARF. Interestingly, we note high levels of exon inclusion in MCF7 and 293T cell-lines, but not in HeLa, SKOV3 and MDAH cell-lines (human cancer cell lines originated from breast, kidney, cervix and ovaries, respectively) (Figure 2a), although the global editing level in cell-lines is expected to be relatively low 11. This demonstrates that the amplitude of editing level in various cell line types is of a variable nature.

We sequenced all of the exon inclusion PCR products and analyzed the editing frequencies at the five editing sites (named E1-E5, Figure 1b) using the Discovery Studio (DS) Gene 1.5 program (Accelrys Inc., San Diego, CA, USA). Importantly, the first exonic editing site, E1 (at position 19 of the exon), was edited at nearly 100% efficiency in all tested tissues and cell lines, whereas the editing levels of the other sites varied (E2 being edited in an average of 53.6% of RNAs, E3 in an average of 26.1%, E4 in an average of 7.9% and E5 in an average of 37.5%) (Figure 2b). Notably, editing sites E1, E3 and E5 are mistakenly annotated as single nucleotide polymorphisms (SNPs) in dbSNP 25 (rs17855348, rs17849311 and rs17855349, respectively) based on the variance in cDNA data. Many similar examples of RNA editing sites erroneously deposited in dbSNP have been recently reported 26.

Usually, editing efficiency is much lower than 100% per site, depending on the expression levels of the ADAR enzymes in the given tissue, the secondary structure of the substrate, or the surrounding sequence. As shown above, position E1 in the NARF Alu-exon is edited in nearly all RNA molecules containing this exon. Inactivation of the nonsense-mediated mRNA decay (NMD) by adding puromycine (see Materials and methods) to 293T cell line did not affect the >97% editing efficiency in site E1 (data not shown). This indicates that the high level of editing in site E1 is not due to elimination of unedited, stop-codon containing mRNAs, but rather is indicative of a high efficiency of editing in that site. Apart from the Q/R site of gluR-B 27, which is restricted to brain, this is the highest editing efficiency documented in human, though it has a much broader tissue expression spectrum. This result suggests that additional regulatory mechanisms have evolved to ensure that the stop codon is edited to a Trp codon in all mRNAs containing the Alu-exon. It further implies that the exonization of the NARF Alu-exon is functional in the human transcriptome.

To substantiate the possibility that exon 8 in NARF was exonized through an RNA-editing-mediated process, we constructed a minigene containing the human genomic sequence of the gene encoding NARF from exon 7 to 9, including the two introns in between and the alternative Alu-exon. Following transfection of this minigene into 293T cells, total RNA was collected, and the splicing pattern of the NARF minigene was examined by RT-PCR analysis using primers specific to the plasmid cDNA and not the endogenous one (see Materials and methods). We then tested the effect of serial intronic and exonic mutations on the splicing of the Alu-exon (Figure 3a).

When the wild-type minigene was transfected into 293T cells, 23% of the mature mRNAs derived from this minigene represented the exon-inclusion form (Figure 3b, lane 1). However, deletion of the antisense Alu element upstream of the exonized Alu resulted in total abrogation of exon inclusion (Figure 3b, lane 2), indicating that these two adjacent Alus probably pair to create the dsRNA that is required for RNA editing. Without this dsRNA, editing does not occur, and functional AG in the 3'ss cannot be created. The effect of the antisense Alu deletion was reversed when the AA splice site near the Alu-exon was mutated to AG, indicating that a single AA→AG change is sufficient for exonization of this Alu (Figure 3b, lane 3). Interestingly, the AA→AG mutation increased exon-inclusion two-fold over the wild type, suggesting that, in 293T cells, about one-third of the AA pairs in the 3'ss are edited into a functional AG 3' ss. Also, a single AA→AT mutation at the 3'ss, created on the wild-type plasmid, resulted in full exon skipping, indicating the importance of editing at that site for exonization. Whereas, a single AA→AG mutation resulted in approximately 30% exonization (Figure 3b, lanes 4 and 5). The higher level of exonization after a single AA→AG mutation at the 3'ss without and with the antisense Alu presumably suggests that although the antisense Alu is essential for exonization, it also reduces the level of maximum exonization by interfering with spliceosome accessibility to the Alu-exon due to dsRNA formation (compare lanes 3 and 5). Combined together, these results demonstrate that the exonization of the Alu-exon 8 in NARF is mediated by RNA editing, and that this mechanism also controls the level of inclusion of this exon in different tissues.

To test the possibility that specific sequences within the Alu-Alu duplex are involved in the regulation of high efficiency editing at the E1 site, we mutated the two nucleotides surrounding the edited site, as well as the nucleotide in the antisense Alu that is postulated to be opposite to the edited nucleotide within the dsRNA (see Figure 4a for the mutated nucleotides). All these mutations substantially reduced the editing in the E1 site (Figure 4b,c), indicating the importance of the surrounding sequence and the postulated opposite nucleotide in the antisense Alu for editing at that site. Moreover, mutations M2 and M3 also resulted in a significant reduction of RNA editing in the other exonic sites - the most significant effect was on site E2 (Figure 4b,c). This might suggest that the edited position is part of a sequence motif that directs high efficiency RNA editing at the other sites as well.

Our results indicate that RNA editing not only enables the exonization of the NARF Alu-exon, but also regulates its inclusion levels in different tissues (Figure 2). This regulation is probably attributed to the efficiency by which the AA splice site is edited to AG. However, another possible mechanism by which RNA editing can control exon inclusion levels is by altering exonic splicing enhancers and silencers (ESEs and ESSs, also denoted exonic splicing regulatory sequences (ESRs)) within the Alu-exon (Table 1). Indeed, editing of the first exonic site (E1) is predicted to eliminate a putative ESR (28; see also ESRsearch 29. It also exchanges a putative ESS (GGT

A

G

To test the possibility that RNA editing regulates the inclusion levels of the NARF Alu-exon by altering ESRs within the exon, we serially mutated each of the exonic edited sites from A-to-G, simulating 100% editing efficiency. To examine the effect of the exonic sites only we used a minigene in which an A-to-G mutation mimics 100% editing in the 3'ss, and we also deleted the antisense Alu that affects editing of the exonic sites. As shown in Figure 5, an A-to-G mutation in E1 and E3, but not in E2 and E5, resulted in a significant increase in exon inclusion levels (Figure 5, compare lanes 2 and 4 with lanes 3 and 6). However, editing in position E4 significantly reduced the inclusion level, suggesting the creation of a putative SC35 site that functions as an ESS (Figure 5, lane 5). These results indicate that editing of three out of the five exonic edited sites affects alternative splicing levels. However, it is unlikely that alternative splicing is regulated through editing in the E1 site, because it is uniformly edited at high levels in all tissues tested (Figure 2b).

ESTs and cDNAs from GenBank version 136 were aligned to the human genome (version hg16) to identify internal human exons that contain Alu elements, as described in 5Alu-containing exons were identified using blastn analysis against the Alu consensus with a threshold of 1E-10. Alus having AA/GT or AG/AT 3' ss/5' ss, which flank exons in the protein-coding region of the genes, were taken for further analysis. Only Alu-exons supported by multiple cDNAs and not containing stop codons (in the ESTs and cDNA) were further considered. Exons were manually screened to remove false computational predictions.

Oligonucleotide primers were designed to amplify (from human genomic DNA) a minigene that contains exons 7, 8, and 9 (and the introns in between) of NARF (GenBank: NM_012336). Each primer contained an additional extension encoding a restriction enzyme sequence. The PCR product of NARF (2.8 kb) was restriction digested and inserted between the KpnI/XhoI sites in the pEGFP-C3 vector, which contains green fluorescent protein (GFP; Clontech, Palo Alto, CA, USA).

Overlapping oligonucleotide primers containing the desired mutations were used to amplify a mutation-containing replica of the wild-type minigene plasmid, using PfuTurbo DNA polymerase (Stratagene, La Jolla, CA, USA) (Additional data file 3). After PCR amplification the reaction was digested with DpnI restriction enzyme (New England Biolabs, Ipswich, MA, USA) for 1 h at 37°C, 1-3 μl of the reaction was transformed into the Escherichia coli XL-1 strain, and colonies were picked for Mini-prep extraction (Qiagen, Valencia, CA, USA). All plasmids were confirmed by sequencing.

293T cells were cultured in Dulbecco's modified Eagle's medium, supplemented with 4.5 g/ml glucose (Biological Industries, Bait-Haemek, Israel), 10% fetal calf serum, 100 U/ml penicillin, 0.1 mg/ml streptomycin, and 1 U/ml nystatin (Biological Industries, Bait-Haemek, Israel). Cells were grown on 6-well plates under standard conditions at 37°C with 5% CO₂. Cells were grown to 60% confluence, and transfection was performed using FuGENE6 (Roche Diagnostics, Basel, Switzerland) with 1 μg of plasmid DNA. Cells were harvested after 48 h. Total RNA was extracted using TRIzol Reagent (Sigma-Aldrich, St. Louis, MO, USA), followed by treatment with 1 U of RNase-free DNase (Ambion, Austin, TX, USA). Reverse transcription (RT) was preformed on 2 μg total RNA for 1 h at 42°C using Oligo-dT reverse primer and 2 U of reverse transcriptase avian myeloblastosis virus (A-AMV; Roche Diagnostics, Basel, Switzerland).

The spliced cDNA products derived from the expressed minigenes were detected by PCR using the pEGFP-C3-specific reverse primer and an exon 7 forward primer (Additional data file 3). Amplification was performed for 30 cycles, consisting of 1 minute at 94°C, 45 s at 61°C, and 1.5 minutes at 72°C. The products were resolved on 1.5% agarose gel and confirmed by sequencing. Band quantification was performed by densitometry scanning of ethidium bromide stained gels, using ImageJ software 47.

For the NMD treatment, cells 48 h post-transfection were subjected to 100 μg/ml puromycin (Sigma-Aldrich, St. Louis, MO, USA) for 4 h before RNA extraction.

Products from RT-PCR or from PCR obtained from commercial cDNAs (BioChain, Hayward, CA, USA) were separated by electrophoresis on 1.5% agarose gels. The appropriate PCR product was excised and the DNA was extracted and purified (Promega, Madison, WI, USA). Direct sequencing from both ends was done using the ABI PRISM (Applied Biosystems, Foster-City, CA, USA). The editing percentage from direct sequencing was calculated as GA+G×100 MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfeBSjuyZL2yd9gzLbvyNv2Caerbhv2BYDwAHbqedmvETj2BSbqee0evGueE0jxyaibaiKI8=vI8tuQ8FMI8Gi=hEeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciGacaGaaeqabaqadeqadaaakeaadaWcaaqaaiaadEeaaeaacaWGbbGaey4kaSIaam4raaaacqGHxdaTcaaIXaGaaGimaiaaicdaaaa@3AC3@ for the forward primer and CT+C×100 MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfeBSjuyZL2yd9gzLbvyNv2Caerbhv2BYDwAHbqedmvETj2BSbqee0evGueE0jxyaibaiKI8=vI8tuQ8FMI8Gi=hEeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciGacaGaaeqabaqadeqadaaakeaadaWcaaqaaiaadoeaaeaacaWGubGaey4kaSIaam4qaaaacqGHxdaTcaaIXaGaaGimaiaaicdaaaa@3ACE@ for the reverse primer; the presented percentages represent an average of three separated experiments or three independent amplifications. The nucleotides were quantified by the Discovery Studio (DS) Gene 1.5 program (Accelrys Inc., San Diego, CA, USA).