Genome-wide searches in the genomes of several species have identified a surprisingly high proportion of overlapping gene pairs. Depending on the sample sizes analyzed and search criteria, the frequencies for overlapping gene pairs vary between 4% and 9% for the human genome, 1.7%-14% in the murine genome, and up to 22% in the fly genome 1. The predominant composition of overlapping gene pairs is an antiparallel convergent arrangement 23, where sense and antisense genes overlap within their 3' regions. Joint expression of both these genes in the same cell would allow the partly overlapping transcripts to associate as dsRNA molecules, which may interfere with RNA processing, transport, stability or other molecular mechanisms. Convergently overlapping gene pairs (COPs) can therefore provide the source for natural antisense transcripts (NATs) that may act as regulators of the sense gene. In addition to NATs being transcribed from the same locus as the sense transcript (cis-NATs), NATs can be transcribed from a different locus (trans-NATs), as illustrated by a search for overlapping transcripts with coding capacity in the human genome, which identified 87 cis-NATs and 80 trans-NATs 3.

In bacteria, more than 100 NATs are involved in the regulation of a variety of biological functions, including the control of copy number, conjugation and post-segregational killing in plasmids, lysis/lysogeny switches in phages, and transposition frequency in transposons 4. In eukaryotes, a very detailed characterization of the molecular role of specific NATs has only been achieved for a few examples.

NAT-mediated interference with splicing is illustrated by the alternative processing of mRNAs of the gene for the thyroid hormone receptor ErbAα, which is regulated by an antisense transcript 5. Overlapping genes can share a bidirectional poly(A) region as demonstrated for the human genes ABHD1 and Sec12 6. Several examples document the fact that antisense transcripts can increase sense transcript stability, when dsRNA regions cover the 3' untranslated region (UTR) and possibly mask out target sequences for RNA cleavage 7. Alternatively, RNA duplex formation can increase transcript sensitivity and induce site-specific cleavage, as shown for the human TYMS mRNA and TRS antisense transcripts 8.

An example of RNA interference (RNAi)-based regulation of an endogenous gene via NATs is the repression of the testis-expressed Stellate gene in Drosophila by paralogous Su(Ste) tandem repeats 9. Both strands of repressor Su(Ste) repeats are transcribed, producing sense and antisense RNA, most probably as part of a dsRNA-based silencing mechanism, as Stellate silencing is associated with the presence of short Su(Ste) RNAs. Antisense expression can also affect translation, as illustrated by the influence of an antisense transcript on the translation of different isoforms of fibroblast growth factor-2 (FGF2) 10. In the nucleus, dsRNA can be edited by dsRNA-dependent adenosine deaminases, which convert about 50% of adenosine residues into inosines, leading to the unwinding of the RNA duplex 11. Inosine-containing RNAs are not translated as they are retained in the nucleus 12.

In mice about 35% of overlapping genes transcribe noncoding RNA. Overlapping genes are scattered around the genome with no apparent bias. Overlaps range from 20 to 3,400 base-pairs (bp) with an average of 372 bp, as far as the quality of transcript annotation allows such predictions. There is some evidence for an over-representation of overlapping genes among specific functional categories, that is, imprinted genes and DNA repair genes 1. Twenty-two out of 58 known imprinted murine genes are transcribed from both strands. Frequently, one partner transcribes a noncoding RNA. Antisense transcripts may regulate imprinting states of the sense promoter (Kcnq1/Kcnq1ot1) or may induce dsRNA-based gene silencing as proposed for Ifd2R/Air. About 20% of known human DNA repair genes overlap either convergently or divergently in an antiparallel arrangement 1.

Mammalian mRNAs that form sense-antisense pairs frequently exhibit reciprocal expression patterns, but permanent coexpression of sense and antisense transcripts can also occur in some tissues, although it is difficult to prove that both genes are transcribed in the same cell. Coexistence of sense and antisense transcripts may indicate a stabilizing effect of dsRNA, or it may depict cases where RNA duplex formation is impaired as a result of secondary structures, or because sense and antisense transcripts or the enzymes required for duplex formation are separated by compartmentalization 13.

To gain an insight into the existence and role of overlapping antisense pairs in plants, we have screened the Arabidopsis thaliana genome for COPs with sense and antisense genes that encode a protein, and have compared the expression profile of the associated genes.

We have characterised the organization and expression profiles of 956 convergent overlapping gene pairs of A. thaliana to assess the potential molecular mechanisms associated with this unusual gene organization, which provided the opportunity for dsRNA formation as a result of the annealing of a sense and antisense transcript.

In animal genomes especially, a number of different mechanisms have been described that involve dsRNA formation. dsRNA formation can interfere with biological activities that require binding of RNA or proteins to the transcript 13. This may include processes such as RNA splicing, editing, transport, degradation or translation. Alternatively, dsRNA could function as a trigger for an RNAi process, providing a target for Dicer enzymes, and be specifically degraded to small interfering RNA (siRNA) molecules 18. The latter mechanism would lead to mutual destruction of the sense and antisense transcripts, whereas antisense transcript-mediated effects on sense RNA processing would not necessarily alter the primary transcript levels, although they could still influence the potential for primary transcript expression.

As in other species 119, the majority of overlapping gene pairs in A. thaliana are arranged in a convergent orientation. Only a very small group of 13 out of 956 COPs show an overlap between the ORFs of the sense and antisense transcripts, which probably reflects the associated evolutionary stress of such an arrangement, as any mutation in the overlapping region would affect both genes. For 50.1% of COPs the sense-antisense overlap does not include any protein-coding region, which makes it unlikely that in this subgroup of COPs the antisense transcript plays a role in regulating the coding region of the sense gene. Antisense transcripts for the members of this group are more likely to jointly use bidirectional poly(A) signals 6 or to regulate transcript stability 7.

We do observe a very high level of joint expression of sense and antisense transcript from overlapping gene pairs. This is in marked contrast to data from Plasmodium falciparum, where only 5% of sense-antisense loci show joint expression 20 and thus support models for a direct regulation of sense transcript by antisense expression via dsRNA degradation. In contrast, the relatively high expression frequency of COPs in Arabidopsis, and the joint presence of sense and antisense transcripts in the same tissue, do not support a dsRNA degradation model. Even a detailed analysis of 1,437 microarrays does not imply that under any conditions or for any specific tissue the COPs gene pool is significantly depleted. While dsRNA-based transcript degradation may occur for some COPs, our data suggest that for the majority of COPs, antisense expression is not linked to transcript degradation pathways.

An interesting observation, which may hint at an alternative interference mechanism between sense and antisense transcripts, is the significant bias for COPs to be spliced, and the enrichment among COPs of alternatively spliced transcripts. These features may indicate a role for antisense transcripts in alternative splicing. Such a mechanism would resemble the effect of antisense expression for the thyroid hormone receptor gene, erbAα, which leads to alternative RNA processing 5. The bias for antisense transcripts to terminate close to the final intron-exon boundary remains a mystery. One could assume that the termination of the antisense transcript near the final sense intron-exon boundary might reflect a selection for antisense transcripts that interfere with splicing of the last intron. However, we could not observe any positive bias for such events among this COPs gene group.

The assumption that antisense transcripts can interfere with splicing events is further supported by the observation that overlaps between antisense transcripts and a sense intron region generate a bias for alternative splicing and also for polyadenylation variation. This may reflect a linkage between these two mechanisms, which has been demonstrated for animal systems where polyadenylation and the splicing of the final intron especially can be coupled 21.

During the review process of this paper, a similar study by Wang and co-workers 22 was published. The main differences between the two studies are in the sets of overlapping genes considered, and the nature of the experimental evidence of gene expression. While we consider only gene pairs where both partners show evidence of protein-coding capacity, Wang and co-workers also considered cases where at least one transcript may not be protein coding. We base our studies of expression on two large microarray datasets, while Wang and co-workers use data from a massively parallel signature sequencing (MPSS) study. While we find no evidence in the microarray data of exclusive transcription relationships for COPs gene pairs, the MPSS evidence of exclusive transcription of the gene pairs in the Wang study is clear. This apparent contradiction may be explained by differences in the gene sets studied, particularly as expression data is only available for a subset of the genes in each study, or by differences in the quality of the expression data. Nevertheless, the two studies taken together give evidence of various significant biological consequences of gene overlaps, including effects on sense gene splicing or polyadenylation (our study) and coexpression of gene pairs 22.

It is important to remember that in our study, we have concentrated on a specific subgroup of convergently overlapping genes, with both sense and antisense transcripts encoding an ORF. Among this group there may be an over-representation of gene pairs for which sense and antisense transcription jointly regulate the production efficiency of both proteins, for example via the common use of a bidirectional polyadenylation region, or by co-editing of both strands associated as dsRNA. Such mechanisms would require the joint transcription of both genes in the same tissue, and our data do suggest that sense and antisense transcripts are frequently coexpressed. On the other hand, one would assume that co-regulation would preferentially be used for genes encoding proteins that are linked in their biological role. One would therefore expect a high degree of conservation for both proteins among homologous COPs, whereas our data show that COPs with homologous proteins encoded by their sense genes do not show the same conservation for the proteins encoded by their antisense genes.

The selection of sense/antisense transcripts with coding capacity may also be the reason for the lack of an indication of dsRNA-based degradation of COPs. A considerable proportion of overlapping antisense genes are noncoding RNAs 23 or are trans-NATs transcribed from different genetic loci 3. These overlapping transcript types may contain a much higher proportion of genes regulated by transcript degradation than the COPs analyzed in this study. A final confirmation of the role of dsRNA for individual genes will require a more detailed experimental analysis. Our analysis should, however, provide a useful first step in defining distinct groups of COP genes as a basis for a more detailed molecular characterization.

The Arabidopsis genome contains 956 COPs with coding capacity that have the potential to form dsRNA. In contrast to data from other species, a comparative expression analysis indicates that sense and antisense transcripts of COPs loci can coexist in the same tissue at the same frequency as the transcripts of any other unlinked genes, with no indication of specific degradation of such sense-antisense transcript pairs. This observation does not exclude the presence of dsRNA degradation pathways for individual loci, but it refocuses the attention on alternative roles for natural antisense transcripts in plants, preferentially those that do not lead to an overall change in transcript levels but rather affect transcript processing or localization. In line with this view, we observe a high proportion of intron-containing genes in COPs, in both Arabidopsis and rice, and an enrichment for genes with alternatively spliced transcripts, indicative of a role for some COP antisense transcripts in splicing modification. In addition, we detect a potential link between alternative splicing and poly(A) site variation. This work provides a set of databases for COPs, based on the degree of sense-antisense overlap and expression, which should provide a basis for the selection of individual candidate loci for a detailed molecular analysis of the different dsRNA pathways.

The following additional files are available with the online version of this paper. Additional data file 1 is a supplement to Figure 1 listing all overlapping genes with ID, annotation and size of overlapping region. Additional data file 2 is a supplement to Figure 2 classifying 1,912 COPs genes according to their overlapping regions. Additional data file 3 is a supplement to Figure 3, calculating the antisense transcript end position in relation to the sense intron-exon boundary for the 956 COPs pairs. Additional data file 4 is a supplement A to Table 2, listing 242 COPs member genes of 89 COPs families. Additional data file 5 is a supplement B to Table 2, comparing homology among 1 kb promoter regions of COPs family members. Additional data file 6 is a supplement C to Table 2, comparing homology among sense and antisense encoded proteins for members of 89 COPs families. Additional data file 7 is a supplement to Table 4, showing expression analysis of the total gene pool and the COPs gene pool for 1,437 microarray experiments. Additional data file 8 is a supplement to Table 6, including a correlation analysis of alternative splicing, TSS variation and polyadenylation variation for COPs with respect to the termination of the antisense transcript in relation to the sense intron-exon boundary.