To determine the prevalence of uORFs most likely to be translated, Drosophila melanogaster release 4.3 transcript sequences (19,389) were used to identify the largest, or major, ORF (mORF). Of these, 13,746 contain unique Flybase gene numbers, 5,851 of which contain one or more AUGs upstream of the mORF. This suggests that 43% of Drosophila mORF proteins could be affected in their expression level by translated uORFs. Our calculated percentage is slightly lower than previously reported Drosophila uAUG frequencies 2, but this discrepancy can be explained by the smaller dataset used in the previous study.

Putative dipteran homologs were found by comparing D. melanogaster mORFs to 666,153 NCBI ESTs using tBLASTn. Many of the EST sequences contained truncated uORF and mORF sequences, therefore the search was limited to species that diverged from D. melanogaster more than 15 Mya (non-melanogaster group species; AAA: 12 Drosophila Genomes Website) 2829, to increase detection of negative selection acting on short protein sequences. For each pair of homologs, global alignment of uORFs identified candidate CPuORFs and K_a/K_s ratios were used to further verify evolutionary conservation of the uORF amino acid sequence. In addition, Flybase transcript annotations were used to discard any genes in which the putative CPuORF was fused to the mORF in any given transcript splice variant.

K_a/K_s ratios < 1 indicate that a sequence is under negative selection, K_a/K_s ratios close to 1 imply that the sequence is undergoing drift, and K_a/K_sratios > 1 suggest that the sequence is under positive selection. We found a total of 44 CPuORFs with a K_a/K_sratio significantly less than one (Table 1; Additional File 1). Importantly, our K_a/K_s ratio analysis distinguishes between high-scoring amino acid alignments that reflect conservation of nucleotide sequences versus alignments that reflect true evolutionary conservation of the amino acid sequence, and therefore are good indicators of translation.

Another indicator of translation is start codon context. Based on nucleotide frequencies of sequences surrounding mORFs, it is predicted that the Drosophila optimal consensus sequence is CAaa

AUG

AUG

A number of uORF-mORF pairs were used as positive controls for the modified uORF-Finder program. In a previous study, CG9865 was shown to contain a putative uORF-mORF pair that has been conserved among distantly related insect species 19. This gene was identified by our analysis, therefore validating our approach. Drosophila Tat-like (DTL), a gene containing a uORF with amino acid similarity in D. melanogaster and D. pseudoobscura 24 was also found by the uORF-Finder program. A third gene identified by our analysis, CG10238, is a bicistronic transcript encoding the small and large subunit of Molybdopterin synthase 2 (MOCS2) 23. It is well conserved across distantly related eukaryotic species (see Additional File 2). In addition, 5 of the 31 dicistronic genes described by Misra et al 22 were shown to contain CPuORFs (Table 2; denoted by Misra and colleagues as CG33071ORFA-CG33071ORFB, Tim9b-CG12788, CG33009ORFA-CG33009ORFB, CG33005ORFA-CG33005ORFB, and snapin-CG9960, but subsequently renamed CG33713-CG33714, CG12788-CG17767, CG33671-CG33672, CG33786-CG33785, and CG9960-CG9958, respectively). Many of the dicistronic transcripts identified by Misra et al 22 are transcripts with ORF pairs that are not well conserved among the Drosophila species. For example, the mei217-mei218uAUG is not conserved in any of the 11 other sequenced Drosophila genomes (UCSC D. melanogaster genome browser) 32, therefore it is not surprising that a number of the dicistronic genes were not identified by the uORF-Finder program. Additionally, it is likely that neither the D. melanogaster annotations nor the dipteran ESTs are representative of the complete transcript population within each species due to the incomplete annotation of 5' transcription start sites 33, and incomplete coverage of the genomes by ESTs.

Initially, 41 genes and 43 uORFs showed evidence of mild to strong purifying selection (K_a/K_s ratio significantly < 1), and an additional gene with one uORF was detected during subsequent duplication analysis (see below). The proportion of genes in the Drosophila genome showing evidence of CPuORFs is approximately 0.3% (42 genes out of 14,040 genes), which is similar to the frequency predicted for the Arabidopsis genome (0.4–0.5%) 19. The present study likely underestimates the prevalence of CPuORFs due to incomplete EST resources and potentially misannotated 5' regions in D. melanogaster.

Consistent with calculated K_a/K_svalues, the majority of CPuORFs with a low K_a/K_s ratio are present in lineages beyond the Drosophilidae (Table 1) and therefore have been conserved more than 40 My (Assembly/Alignment/Annotation of 12 Drosophila species) 2829. Those uORFs that exhibit a low K_a/K_s ratio but are only found within Drosophila species may represent uORFs that have recently emerged within the Drosophila lineage but are nonetheless under mild to strong selection pressures.

Two studies have shown that the length of a uORF can influence the ability of a ribosome to reinitiate scanning and translation initiation at a mORF 3435. The plant and mammalian cell systems used in these studies show that reinitiation at a downstream AUG is generally more efficient in the presence of shorter uORFs, and in plant protoplasts reinitiation drops sharply in constructs containing uORFs longer than 34 amino acids. Both studies were carried out using viral components, and as such it is not clear whether these observations extend to mRNAs in a native eukaryotic cellular environment. Nonetheless, uORF length could play an important role in the regulation of mORFs, therefore we analyzed Drosophila CPuORFs in terms of their amino acid lengths. Initial characterization of the 44 putative CPuORFs under negative selection reveals a wide distribution of lengths, ranging from 15 to 179 amino acids (Table 2, Figure 1A).

To date, most, if not all, functionally characterized uORFs are smaller than 100 amino acids, but more than one fourth (12/44) of D. melanogaster CPuORFs are above this size. In general, Drosophila CPuORFs seem to be larger than those found in plants. While 83% of Arabidopsis CPuORFs are between 21 and 60 amino acids in length (mean of 40 amino acids ± 16 standard deviation; Figure 1B), the Drosophila uORF length distribution peaks between 41 and 80 amino acids (mean of 76 amino acids ± 44; Figure 1A). These plant and insect datasets were not generated by comparing species with the same evolutionary distance but a more convincing comparison can be made by analyzing uORFs that have been conserved over more than 200 My: between Arabidopsis and rice, and between Drosophila and non-Brachycera lineages (e.g. Anopheles). The Arabidopsis distribution peak remains essentially unchanged under these restrictions (mean of 39 amino acids ± 13), whereas the distribution of Drosophila uORFs peaks at an even greater length, 81–100 amino acids (mean of 92 amino acids ± 29; Figure 1C). Longer uORF lengths in Drosophila may reflect a need for stronger suppression of mORF translation than in plants, consistent with the observations of the above-mentioned cell culture studies. Alternatively, insect cells may exhibit more efficient reinitiation resulting in a requirement for longer uORFs to attenuate mORF translation.

In insect and mammalian genomes, clusters of closely related genes can sometimes occur, such as the Hox gene clusters 36. To determine whether genes with uORFs cluster in certain parts of the genome, the 44 uORFs were placed on the D. melanogaster cytological map (Table 1) and compared to a random distribution (Methods). uORF frequencies were not statistically different from a randomly generated dataset except for a cluster of 6 uORFs residing on band 57 (p-value = 10^-5), five of which fall on a much smaller segment of the chromosome, band 57F. Upon closer examination, some of these uORFs may have arisen as a result of tandem duplications; one uORF found in the CG30290 transcript as well as two uORFs found in the CG9865 transcript (uORF1 and uORF3) all contain twin CX₉C motifs. Interestingly, the observed clustering is not dependent upon the putative duplication events of CX₉C motif-containing uORFs. Eliminating the duplication bias by collapsing CX₉C-containing uORFs to one representative, clustering is still statistically significant, with 4 uORFs on cytological band 57 (p-value = 0.004) and 3 uORFs on band 57F (p-value = 0.0002). Therefore, the data suggest that there is a preponderance of both clustering and duplicate retention of uORFs on band 57. Clustering at this region could be an indicator of chromatin interactions at this site that could mediate CPuORF regulation.

The twin CX₉C motif is an integral part of coiled-coil helix, coiled-coil helix (CHCH) domains, a domain previously implicated in uORF-mORF associations in group 8 plant uORFs 19. In fact, the group 8-like Drosophila uORF member described in the plant study is uORF3 of CG9865. It is interesting to note that the plant group 8 uORF has consistently lost its duplicate copy during both recent and ancient polyploidy events whereas the Drosophila group 8 putative homologue may be retaining its duplicates. Different duplication retention histories could indicate that twin CX₉C motif-containing ORFs play different roles in plants and animals.

To determine whether there has been retention of uORF-mORF pair duplicates within the Drosophila genome itself, the 41 mORFs with strongly conserved uORFs were compared to the D. melanogaster transcriptome. A single gene, CG17325 showed evidence of a duplicate copy, CG10570, in which the uORF-mORF pair is conserved (See Additional File 3). CG10570 was not detected by our program due to the short length of its mORF (< 100 amino acids), therefore this gene was added to our list of CPuORFs following our duplication analysis (Tables 1 and 2). CG17325 and CG10570 reside adjacent to one another on chromosome 2, band 37A4-A5, and are transcribed on opposite strands away from one another. The close proximity of the genes suggests a segmental duplication gave rise to the two genes, both of which are conserved throughout the Drosophila lineage and exhibit a K_a/K_s ratio < 0.28 (Table 1). This duplication presumably occurred more than 40 Mya since both loci are present in D. melanogaster, D. grimshawi, and D. virilis. Unlike the extensive uORF-mORF duplication retention history of the Arabidopsis genome, CG17325 and CG10570 were the only example of gene duplicate retention in Drosophila.

Further differences between plant and insect CPuORFs were observed following gene ontology (GO) term analysis. GO term frequencies in the D. melanogaster genome were compared to frequencies in the insect uORF dataset to look for overrepresentation of terms. P-values were determined using the Bonferroni correction method, a method that accounts for multiple comparisons and calculates a conservative p-value. Also, the recent tandem duplicate (see above) was not included in the analysis to eliminate bias from recent duplication events. Because GO terms have been assigned to all ORFs found in bicistronic transcripts, GO terms were extracted for both uORF and mORF gene identifiers, designated hereafter as the uORF dataset (41 mORFs and 7 uORFs). This analysis differs from previous analyses in plants; it not only identifies 1) classes of mORF proteins that tend to associate with CPuORFs, but it also identifies 2) ORFs that preferentially associate with other ORFs on a single transcript. In plants, a large proportion of CPuORFs associate with mORFs encoding transcription factors, however this trend was not observed in insects. Instead, mORF proteins showing evidence of N-methyltransferase activity (GO term for CG9666 and CG9960 mORFs; Table 3) tend to associate with CPuORFs (p = 0.02). This methyltransferase activity may act on DNA or RNA, since both types of Interpro domains are overrepresented in these two genes.

Additionally, overrepresentation of GO term 'protein import into the mitochondrial inner membrane' is driven by two proteins in the Drosophila uORF dataset, CG9878 (Translocase of inner membrane 10, Tim10) and CG17767 (Tim9b), which contain the Interpro Zn-finger Tim10/DDP-type domain (p = 0.01). Unlike the overrepresented methyltransferase domain, the Tim10/DDP-type domain is not limited to the mORFs, but appears in either the uORF or mORF, demonstrating that these ORFs show a preference for associating with other ORFs in a transcript. Specifically, Tim10 is encoded by the mORF of its transcript while Tim9b is encoded by the uORF. This does not imply that Tim9b does not act as a regulatory uORF, however. Tim9b may act both as a chaperone in the intermembrane space, as well as a regulatory element controlling the translation of its associated mORF.

In support of a model in which mitochondrial proteins preferentially associate with other ORFs on a single transcript, a further connection to the mitochondrial inner membrane is found when examining other genes in the uORF dataset. The CG8026 mORF encodes a putative mitochondrial folate transport protein 3738 (Table 4). Interestingly, this trend may extend to the mammalian lineage, exemplified by the human Uncoupling protein 2 (UCP2) mORF, a putative inner mitochondrial membrane transporter. The UCP2 mORF is not only associated with what appears to be a CPuORF, but it is regulated by its uORF in a glutamine-dependent manner 39. B-cell lymphoma 2 (BCL-2) is another mammalian oncogene that produces a protein from its mORF, BCL-2, which is localized to mitochondria 40 and is associated with a functional uORF 15.

Other Drosophila genes also have potential links to the mitochondrion, such as CG18624, a putative NADH dehydrogenase that is predicted to act in mitochondrial electron transport (Table 4). Also, uORF1 of CG9865 is a putative homolog of p8Mature T-Cell Proliferation 1 (p8MTCP1), an ORF that is transcribed on the same mRNA as p13MTCP1, is targeted to mitochondria 41, and may play a role in oncogenesis 4243. CG9865 uORF1 has a twin CX₉C motif, as do p8MTCP1 and other proteins targeted to mitochondria, namely yeast proteins Mitochondrial Ribosomal Protein 10 (Mrp10p) 44, Cytochrome Oxidase 19 (Cox19p) 45, Cytochrome Oxidase 17 (Cox17p) 46, and Mitochondrial intermembrane space Import and Assembly 40 (Mia40p) 47. In humans, the twin CX₉C motif found in Mia40p is required for import and stable accumulation of Mia40 in the intermembrane space 48. Several genes in the uORF dataset contain ORFs with CX₉C motifs, such as uORFs 1 and 3 of CG9865, the uORFs of CG30290 and CG9288, and the mORF of CG7950 (See Additional File 2). These open reading frames could be interacting with other ORFs on the same transcript to target them to the mitochondria or to form a stabilizing protein complex.

It is possible that these ORF associations are vestiges of ancient prokaryotic operons that originated in the mitochondrion and were transferred to the nuclear genome over time. This hypothesis runs counter to the prevailing thought that mitochondrial proteins involved in transport are generally of eukayotic origin 49. Regardless of their origin, nuclear ORFs coding for mitochondrial proteins may maintain an association with other ORFs on a single transcript over long periods of evolutionary time for several reasons. Both ORFs may be co-regulated at the transcriptional level and be required at similar times in development, thus providing more efficient transcription of DNA. Alternatively, the uORF may be regulating expression of the mORF with important biological consequences. These possibilities are not mutually exclusive and further experimentation will be required to determine whether this energy-producing organelle is influenced by the translational regulation of uORF-mORF pairs on single transcripts.

Interestingly, the trend in animal mitochondrial ORFs was not observed in plants. Instead, plant uORFs tend to associate with mORFs encoding transcription factors 19. Perhaps these unique characteristics reflect fundamental differences in the two eukaryotic lineages. Despite their differences, plants and animals both seem to contain uORF-mORF pairs involved in a wide range of biochemical and regulatory pathways (Table 4). There is some evidence in the literature that transcripts with uORFs can occur in similar biochemical pathways, such as genes affecting the polyamine biochemical pathway 50, but this is the exception rather than the rule and no additional examples have been born out by our analyses. To facilitate future studies of these elements, all CPuORF annotations will be submitted to Flybase.

Drosophila melanogaster transcript sequences, release 4.3 (19,389 sequences) were downloaded from Flybase 51, Anopheles gambiae transcript sequences, build 3.4 (14,127 sequences) were downloaded from Ensembl 52, and dipteran expressed sequence tags (ESTs) (666,153) were downloaded from NCBI 53 December 15, 2006. Because the melanogaster group members (includes D. simulans, D. yakuba, D. erecta, and D. ananassae) diverged from D. melanogaster relatively recently 2829, their transcript sequences are of limited use in detecting strong negative selection over short sequence lengths due to the accumulation of few synonymous and non-synonymous substitutions. Therefore these species were excluded from this first comparison, as were D. melanogaster ESTs.

Comparative analysis of D. melanogaster and A. gambiae sequences was performed using uORF-Finder 19, a program that identifies the longest open reading frame of a transcript in the first species (defined as the mORF), finds the putative homolog in the second species, and aligns all open reading frames upstream of these homologs to identify putatively conserved uORFs. uORF-Finder was designed to compare full-length cDNA sequences from two species, therefore to accommodate a D. melanogaster full-length transcript-to-dipteran EST comparison, the program was modified and putative homologs in the ESTs were identified using the first 100 amino acids of the D. melanogaster mORFs. uORF size was also limited to 200 amino acids (no additional uORFs were found when uORF size was limited to 300 a.a.).

The presence of putative CPuORFs was established in at least three different species by either extracting the first 100 amino acids of the D. melanogaster mORF sequence and searching the NCBI EST database using tBLASTn for putative homologs with conserved uORF sequences, or by scanning the UCSC D. melanogaster genome browser and inspecting other Drosophila genomes for conservation of uORF start and stop codons 32. Any putative uORF sequences that showed evidence of in-frame fusion with the mORF on the UCSC browser (in an alternative splice form, for example) were not included in the final list of CPuORF-containing transcripts.

The K_a/K_s ratio was determined using pairwise_kaks.PLS (version 1.7) 54 and is derived from the highest scoring BLAST homolog in the D. melanogaster-dipteran high scoring pairs. Both the approximate method (option -kaks yn00) and the maximum likelihood method (-kaks codeml) were used. Only the approximate method calculation is reported in Table 1 due to the typically short evolutionary distance between the organisms found in the highest scoring BLAST pairs. The Nei-Gojobori p-distance model was used to test for purifying selection (Null hypothesis K_a = K_s; alternate hypothesis K_a <K_s). MEGA4 default settings were used to run codon-based Z-test analyses 55 on highest scoring BLAST homologs.

To determine whether the 44 uORFs were randomly distributed along the Drosophila chromosomes relative to annotated transcript positions, a perl script was written to generate a random distribution of 44 positions along the chromosomes. Cytological positions for each CG gene identifier were extracted from D. melanogaster release 4.3 gene annotations 51, from which 44 positions were randomly chosen. This ensured that clustering would not simply reflect gene rich regions. The number of 'hits' within a given cytological band were tallied, and the entire process was iterated 30,000 times, providing a random distribution of 'hits' at any given band when 44 positions were picked across the entire genome. The random distributions were then used to provide a p-value for the observed number of uORFs within a given cytological band.

Over- and under-representation of Gene Ontology (GO) terms in the uORF dataset (41 mORFs and 7 uORFs with associated GO terms) versus the D. melanogaster genome was determined using Genemerge v.1.2 56, a program which provides a Bonferroni-corrected p-value. Association files were derived from Gene Ontology website files (D. melanogaster annotation received from Flybase March 13, 2007) 57, and from the BioMart website 58 (Ensembl Gene ID, Pfam ID, and Interpro ID numbers obtained; downloaded files are based on D. melanogaster genome release 4.3). Description files were derived from GO term files 59 (gene_ontology.obo.zip), and from Interpro files 60.