Future studies of akirin genes would benefit from a common nomenclature system to aid the dissemination of results between different research groups. Of the current names utilised, we suggest that the naming system employed by Goto et al. 1 is used in future submissions, since it is derived from the most detailed functional analysis and suitably describes the evolutionary relationships of different orthologues and paralogues. The designation 'FBI1' 2 (i.e. for akirin2) is also founded on important functional data, but is very similar to a gene named factor binding IST protein 1 (FBI-1: NP_056982) and as with the name 'Mighty' 3 (i.e. akirin1), does not account for evolutionary relationships within the gene family.

An akirin proto-orthologue was not identified in the Bacterial or Archeal genomes examined, which either excludes a prokaryotic origin, or means that insufficient phylogenetic signal remains to identify these ancient orthologues. The earliest eukaryotes in which an akirin gene was retrieved were the protist taxa Alveolata (in Guillardia theta) and Heterolobosea (in Naegleria gruberi) (fig. 1). The Alveolata and the Euglenozoa (which are phylogenetically closely related to the Heterolobosea, 4) are thought to have arisen close to 2 billion years ago 5. This places an origin for akirin prior to the split of animal/plant/fungi lineages. However, an akirin gene was not identified in any plant or fungal genome as previously noted 1, although it was represented in the Amoebozoa (in Dictyostelium discoideum) and the choanoflagellate Monosiga brevicollis (fig. 1), which branches as the closest known outgroup to metazoans 6. The presence of an akirin gene was an exception rather than a rule in non-animal eukaryotic genomes. This is consistent with the independent non-functionalization of akirin in many eukaryotic lineages, before it obtained a fundamental function in the metazoans.

In D. discoideum and N. gruberi, akirin comprises a single exon, whereas M. brevicollis has conserved a 3-exon gene (fig. 2a). Choanoflagellates are the closest known living relatives to metazoans and its genes are comparably rich in intronic sequences 6. Thus, relative to these more basal eukaryotes, it might be expected that certain features of the M. brevicollis akirin proto-orthologue would be conserved with animal akirins. While the two exon-exon boundaries of the choanoflagellate gene are not conserved with any metazoans (not shown), a string of 4 codons (ATG-GCC-TGC-GCG) 38 nucleotides into the second exon code the signature residues Met-Ala-Cys-Ala, conserved at the start region of all invertebrate metazoan Akirins (fig. 2a). Additionally, in close downstream proximity to this motif, a nuclear localization signal (NLS) (Pro-Val-Lys-Arg-Arg) is present (fig. 2a), which is conserved in metazoan sequences. These findings suggest that the metazoan akirin1/2 coding sequence was derived from exons 2 and 3 of a single gene in a common ancestor to metazoans and choanoflagellates.

A comparison of the genomic organisation of metazoan akirins provides insight into their evolutionary heritage (fig. 2). In all vertebrate species examined (mouse and zebrafish shown), akirin1 and akirin2 are organized as 5 exons of comparable size and 4 more variable introns (fig. 2a). In cephalochordates (Branchiostoma floridae), akirin also comprises 5 exons, although exon 5 is made up solely of untranslated nucleotides (fig. 2a, shaded in vertical lines). In fact, exon 4 of the B. floridae gene is equivalent to exon 5 of vertebrates (fig. 2a, evidenced by conserved position of stop codon) and the addition of exon 5 was probably a lineage specific acquisition. In Placozoans (Trichoplax adhaerens), akirin comprises 3 exons (fig. 2a). In Cnidarians (Nematostella vectensis), which represent a basal metazoan lineage that branched later than Placozoans 78, akirin comprises 4 exons. The boundary of exons 1/2 of Placozoan akirin is conserved with the boundary of exons 1/2 in all other metazoans examined (fig. 2b). Further, the boundary of exons 2/3 of Placozoan akirin is conserved with the boundaries of exons 2/3 in sea anemone/amphioxus and exons 3/4 in vertebrates (fig. 2b). Additionally, the boundary of exons 2/3 and 3/4 of the amphioxus/sea anemone genes are respectively conserved with the boundary of exons 3/4 and 4/5 in vertebrate akirin1/2 (fig. 2b). The most parsimonious evolutionary scenario to account for these distributions of conserved exon-exon boundaries is that firstly, an exon-gain event occurred in the akirin gene after the split of Placozoans with a common ancestor to Cnidarians and Bilatarians (fig. 2b). In support of this, exon 4 of the anemone/amphioxus proto-orthologue and vertebrate akirin1/2 genes starts with the last three residues of the protein (consensus sequence: Tyr-Val/Leu-Ser), which are conserved in all animals examined except Placozoans. Subsequent to the proposed exon gain event, an intron was seemingly inserted into exon 2 of akirin in a common deuterostome ancestor, after the split of cephalochordates and higher chordates, but before the event separating akirin1 and akirin2 (fig. 2a, b). We conclude that strong stabilising pressures have been enforced throughout metazoan evolution to maintain the comparable genomic organisation of present-day akirin genes across diverse taxa, in support of ancient patterns of gene regulation.

We performed an exhaustive search for akirins in animal genomes and transcriptomes employing a broad taxonomic sampling strategy. These results are summarised in fig. 1 and additional file 1. In virtually all diploid vertebrates examined, a single akirin1 and akirin2 gene was identified. Almost without exception, both genes were strongly represented among EST databases of model and non-model vertebrate species. In common with the gnathostomes (jawed vertebrates), Petromyzon marinus (marine lamprey) had two sequences with marked identity to akirin (fig. 1, additional file 1). However, one was an EST that could not be identified in the Ensembl 5.9X genome pre-assembly and was partial at the C-terminal. Further, a single akirin orthologue was retrieved in the Myxinid (hagfish) lineage (additional file 1). In the model Avian Gallus gallus (red jungle fowl), no akirin1 orthologue was present in the current Ensembl genome assembly. Further, it was not represented among ~600,000 Genbank G. gallus ESTs, despite the presence of multiple positive akirin2 hits. Likewise, in other model birds including zebra finch (Taeniopygia guttata) and turkey (Meleagris gallopavo), no akirin1 orthologues were retrieved in EST databases containing ~92,000/17,500 respective sequences. Thus the absence of akirin1 in the class Aves reflects the genuine loss of a gene family member, rather than repeated artefacts of insufficient sequencing resolution. This is consistent with a recent finding showing that the number of gene family members common to tetrapods/teleosts is markedly reduced in the class Aves 9. Interestingly, gene families, which, like akirins, had known roles in the immune system, were the most strongly affected 9.

In many invertebrate metazoans, a single gene was retrieved that shared significant identity to fly akirin and vertebrate akirin1 and akirin2 across its entire length (fig. 1, additional file 1), but had no clear identity to other characterised or uncharacterised genes. This included several bilaterian lineages with a strong representation of deuterostome and protostome taxa, plus more ancient phyla including Cnidarians and Placozoans, among the most ancient known animals 78. However, an orthologue was not retrieved in sponges. A notable invertebrate lineage lacking an akirin gene was the family Cionidae, which has a completed high-resolution genome sequence and an abundance of EST sequences. This is consistent with the observation that the compact genome of Ciona intestinalis (~150 Mb) has undergone significant gene loss compared to other deuterostomes 10. However, another Ascidian (Halocynthia roretzi) has retained an akirin orthologue.

Since a single akirin sequence was invariably retrieved in invertebrate/non-metazoan eukaryotes and diploid vertebrates had no more than two akirin homologues, we hypothesised that akirin was an ancestral proto-orthologue that gave rise to two akirin paralogues in a common vertebrate ancestor. This was tested by phylogenetic analysis using vertebrate Akirin1 and Akirin2 amino acid sequences and the single sequence representing four invertebrate deuterostome lineages as outgroups. The sequence alignment is provided in additional file 2. Maximum likelihood (ML) and Bayesian phylogenetic analyses were initially performed (fig. 3). Further, to test the robustness of the analysis to different methods of reconstruction, we also performed neighbour joining (NJ), minimum evolution (ME) and maximum parsimony (MP) analyses (fig. 4). Additionally, a NJ tree was constructed solely from unsaturated positions in the alignment 11, to test the hypothesis that mutational saturation at certain sites might influence tree topology. Very similar topologies were retrieved by all approaches (fig. 3 fig. 4) and a clear branching was apparent internal to invertebrate outgroups, separating Akirin1 and Akirin2 clades (fig. 3, fig. 4, 96/90/94/95/66 % bootstrap confidence values by ML/NJ/'unsaturated' NJ/ME/MP and 100% posterior probability in Bayesian analysis). In both Akirin clades, vertebrate sequences branched according to established taxonomic relationships, including the expected split of the Sarcopterygii and Actinopterygii (fig. 3, fig. 4). The included lamprey sequence (EC384219) branched as the most external node of the vertebrate Akirin1 clade in all trees (fig. 3, fig. 4). The partial lamprey sequence that was not included in the phylogenetic analysis, shares higher sequence identity to akirin2 (41%), than to akirin1 (30.5%) and could represent a lamprey akirin2 orthologue. The single hagfish orthologue was not used in the final phylogenetic analysis, since in preliminary reconstructions, its position was unstable and it formed either a weakly supported branch as the most external node of the Akirin2 clade or alternatively, branched with more basal invertebrate outgroups. Taken together, these results indicate that Akirin1 and Akirin2 are indeed paralogues that arose by gene duplication in a common ancestor to the vertebrate lineage, prior to the separation of gnathostome and agnathan lineages.

Next, we examined the genomic neighbourhood surrounding akirin1/2 in mammals, birds, amphibians and fishes (fig. 5, fig. 6). In all vertebrate genomes examined, akirin1 and akirin2, when both present, were positioned on different chromosomes, except in mice, where they are located on different ends of chromosome 4, separated by ~100 Mb. A comparison of genes in the neighbourhood of akirin1 and akirin2 (i.e. separately considering fig. 5. and fig. 6) demonstrates a strong level of conserved synteny across vertebrate classes, with limited intra-chromosomal rearrangements and few inter-chromosomal rearrangements in tetrapod species with known karyotypes. In teleosts, two tracts were present with conserved synteny relative to single akirin1/akirin2-containing regions of tetrapods (see following section). Consistent with BLAST homology screens (fig. 1, additional file 1), an akirin1 gene was absent from the G. gallus genome and gene order was disrupted around this region relative to mammalian and amphibian genomes examined (fig. 5). In summary, these results indicate that the chromosomal organisation of akirin-containing tracts of vertebrate genomes were conserved from a common ancestor prior to the speciation events separating the major vertebrate classes.

A single akirin1 gene was identified in all teleost species examined, whereas two akirin2 copies were retrieved from Acanthopterygian taxa i.e. pufferfishes, medaka, sticklebacks and sea bream. All methods of phylogenetic analysis separated teleost akirin2 sequences into two clades (fig. 3, fig. 4). The first was represented by one of the two sequences in species of the Acanthopterygii and the single Ostariophysi copy (i.e. zebrafish, Danio rerio and fathead minnow, Pimephales promelas) (fig. 3, fig. 4) The second clade was represented by the remaining Akirin2 sequences of Acanthopterygian species (fig. 3, fig. 4). Thus, each tree branches prior to the split of Acanthopterygian and Ostariophysian samples, which indicates that this duplication event occurred in a common teleost ancestor rather than in the Acanthopterygian lineage. However, statistical confidence in this branching was weak by all methods (fig. 3, fig. 4, 50/59/<50/68/<50% respective bootstrap support in the ML/NJ/'unsaturated' NJ/ME/MP analyses) excepting the Bayesian analysis (fig. 3, 100% posterior probability values). Bayesian phylogenetic reconstruction was shown under certain conditions to produce an overestimate of branch confidence 12. Thus, we also sought evidence to either provide support or refute this branching topology, using comparisons of conserved genomic synteny. The synteny map indicates that an expansive genomic region containing akirin2 duplicated in a common ancestor to zebrafish and stickleback (Gasterosteus aculeatus), since two orthologous chromosomal tracts exist in both species that retain common synteny to a single region in tetrapod genomes (fig. 6). Specifically, tetrapod genes are present in teleosts as either single orthologues interspersed between the two tracts (e.g. rars2, rragd, pnrc1, rngtt, orc3l, gjb7) or are present as duplicated co-orthologues on both regions (e.g. akirin2, gabrr1, gabbr2, znf292, syncrip) (fig. 6). A similar pattern of double conserved synteny is seen in teleosts relative to tetrapods on the akirin1 synteny map, although akirin1 is only retained on a single chromosome (fig. 5). These patterns of synteny may be the result of a genome tetraploidization event that occurred in a basal teleost ancestor after the split of the Actinopterygii and Sarcopterygii lineages 1314. However, this interpretation requires that one of the akirin1 paralogues from this event was non-functionalised either in a common teleost ancestor, or within individual lineages. Furthermore, one of the akirin2 paralogues must have been non-functionalised in an ancestor to the Ostariophysi lineage, since a single akirin2 gene is found in zebrafish and fathead minnow.

Duplicated genes from teleost species are generally annotated as either gene-1/gene-2 or gene-A/gene-B according to the order of their discovery. However, this nomenclature system is rarely based on phylogenetic premises and generally does not accommodate paralogues from distinct duplication events in different teleost lineages. For certain genes where teleost duplicates have been retained from both the teleost WGD and more recent lineage specific events, appropriate nomenclature systems have been proposed to simplify confusing existing naming systems (e.g. MyoD: 15). Due to the fact that akirins are uncharacterised in fishes, we have a rare opportunity to set out a logical nomenclature framework from the onset of their study. We recommend, as indicated in fig. 1 and additional file 1, that teleost akirin2 paralogues derived from the teleost whole-genome duplication event 1314 are named as either akirin2(1) or akirin2(2). Paralogues of these genes from more recent duplication events in certain teleost lineages e.g. salmonids 16 should be named akirin2(1a/1b) or akirin2(2a/2b). Similarly, if new teleost akirin1 paralogues are discovered in the future then an equivalent naming system should be employed.

If akirin1 and akirin2 arose from the duplication of a region of the genome in a common vertebrate ancestor (either through a segmental duplication or tetraploidization), then other genes or gene families would have been duplicated at this time. Several vertebrate gene families are located in syntenic chromosomal regions where different members are located proximally to both akirin1 and akirin2 or to tracts where copies of these genes have been lost (i.e. in certain regions of double conserved synteny in teleosts, fig. 5, fig. 6 and on the region of chicken chromosome 23 shown in fig. 5). Several form small gene families, with just two members in diploid vertebrates (as observed for the akirins) and where one member is found on each chromosomal tract containing either akirin1 or akirin2. For example, in several vertebrate classes, genes for rragc, nt5c1a, cnr2, pnrc2, hnrnp-r, and maneal are found in the akirin1 genomic neighbourhood and have a single putative paralogue on the akirin2 chromosomal tract (respectively: rragd, nt5c1b, cnr1, pnrc2,hnrnp-q [aka syncrip] and manea) (compare fig. 5 and fig. 6). Comprehensive phylogenetic reconstructions of these gene families were performed with amino acid alignments using NJ, Bayesian, ML and MP approaches (fig. 7, alignments provided in additional file 3). For 5 of these 6 gene families, a single putative orthologue was identified in C. intestinalis/B. floridae genomes and employed as an outgroup. For Ras-related GTP-binding protein (Rrag), Heterogeneous nuclear ribonucleoprotein (Hnrnp), Cytosolic 5'-nucleotidase 1 (Nt5c1) and Proline-rich nuclear receptor coactivator (Pnrc) protein families, tree topologies were very similar by all methods of reconstruction (fig. 7a–d). In each case the tree branched into two clades containing family member orthologues from different vertebrate classes, closely reflecting expected taxonomic relationships (fig. 7a–d). Each of these branching topologies is compatible with a duplication event at the base of the vertebrate lineage, as shown for akirins (fig. 3, fig. 4).

Phylogenetic analysis of the Cannabinoid receptor (Cnr) family was sensitive to the reconstruction method and only the NJ analysis split the tree into two clades of Cnr1 and Cnr2 orthologues (fig. 7e). Other methods strongly supported a single Cnr1 clade, but did not resolve Cnr2 sequences into a single clade, when teleost sequences were included (not shown). However, it is noteworthy that previous phylogenetic studies have suggested that Cnr1 and Cnr2 (also known respectively as CB1 and CB2) duplicated from a single proto-orthologue in the vertebrate stem of the chordate lineage 1718.

For the glycoprotein endo-alpha-1,2-mannosidase family, all four methods of reconstruction produced similar topologies in which the tree did not branch into separate Manea and Maneal clades due to the inclusion of teleost Manea sequences as the external branch of a clade containing solely other vertebrate Maneal sequences (not shown). We tested the hypothesis that tree topology was being influenced by mutational saturation at a proportion of sites in the alignment. When saturated positions were removed from the analysis, a NJ topology was obtained splitting the tree into separate vertebrate Manea and Maneal clades (fig. 7f). Therefore, it is possible that mutational saturation caused an aberrant branching of teleost Manea sequences and that the corrected tree again reflects a duplication event of a Manea/Maneal proto-orthologue in a common vertebrate ancestor.

In most vertebrate classes, two members of the potassium voltage-gated channel family (kcnq4 and kcnq5) were located in the respective genomic neighbourhood of akirin1 and akirin2 (fig. 5, fig. 6). This gene family contains up to five members in diploid vertebrates and 2 members in the C. intestinalis genome. All methods of phylogenetic analysis produced near identical topologies with a clade including vertebrate and C. intestinalis Kcnq1 orthologues that branched externally to remaining family members (fig. 7g). Internal to this clade, the other C. intestinalis Kcnq sequence branched externally to the remaining four vertebrate Kcnq sequences, which split into two well-supported clades containing Kcnq2/3 and Kcnq4/5 sequences respectively (fig. 7g). These clades split into sub-clades containing individual Kcnq2 and 3 orthologues and Kcnq4 and 5 orthologues (fig. 7g). This branching pattern can be explained by two duplication events in the vertebrate lineage, where a single proto-orthologue to Kcnq2/3/4/5, duplicated to create two ancestor genes to Kcnq2/3 and Kcnq4/5 which both duplicated again to produce Kcnq2, kcnq3, kcnq4 and kcnq5 genes as conserved in current vertebrate genomes.

The branching patterns of these gene families, are therefore generally consistent, not only with at least one duplication event in a common ancestor to mammals, birds, frogs and fishes, but in the case of the highlighted members, often reflect their respective chromosomal proximity to akirin1 or akirin2. In other words, when orthologues from a gene family (i.e. one clade in the tree) were located in the genomic neighbourhood of either akirin1 or 2, paralogues from that family (in the other clade) tended to be proximal to, or at least on the same chromosome as the other akirin copy. A parsimonious explanation for these findings is that a duplication event occurred in the vertebrate stem of the chordates that affected a chromosomal region containing both proto-orthologues to akirin and to components of neighbouring gene families. Two-rounds of genome polyploidisation in vertebrates has been long been proposed [e.g. 19] and support for this hypothesis has been obtained by comparing vertebrate genome organisation, with deuterostome relatives with unduplicated genomes, including urochordates 2021 and recently cephalochordates 10. For example, Putnam et al. showed that Gnathostome genomes share quadruple conserved synteny with the Branchiostoma floridae genome providing 'conclusive evidence for two rounds of duplication on the jawed vertebrate stem' 10. However, this idea has been historically controversial and certain studies using phylogenetic analysis of vertebrate gene families found a lack of supporting statistical evidence e.g. 2223, while others found results compatible with the hypothesis e.g. 24.

It is widely accepted that gene duplication can create opportunities for functional divergence in paralogues. Divergence is thought to occur where one duplicate retains the original protein function and the other accumulates changes, (either through redundancy or by positive selection) or alternatively, through the partitioning of the functions of an unduplicated ancestor protein [reviewed in 25]. Whatever the mechanism, if functional divergence has occurred between duplicated genes, then it should be observable as changes within their coding regions, since functionally important and non-functionally important residues should evolve under different constraints.

It is known that Akirin1 and Akirin2 differ in at least one function 1. The branch length leading to the Akirin1 clade is extended relative to Akirin2 in all phylogenies, (fig. 3, fig. 4). This suggests that after the akirin duplication, Akirin1 evolved at a faster rate than Akirin2. This result was confirmed by significant relative rate test results for several vertebrate lineages (result not shown). To examine whether this difference in evolutionary rate was accompanied by altered selective constraints, we examined pairwise rates of synonymous (dS) and non-synonymous (dN) substitutions between Akirin1 and 2 for several vertebrate lineages. Two approaches were implemented: firstly, the likelihood method of Goldman and Yang 26 and secondly, the Nei-Gojobori approach 27. Both results were comparable and low dN/dS ratios (<<1) were estimated when different vertebrate lineages were compared for Akirin1 and Akirin2 (additional file 4). Specifically, dN/dS ratios averaged from both methods, were ~0.14 for Akirin1 and ~0.09 for Akirin2. Thus, Akirin1 and Akirin2 proteins, as a whole, have evolved under comparably strict purifying selection.

It is known from both a large-scale protein-protein interaction study 30 and specific studies 23132 that the ancestor protein to Akirin1 and Akirin2 has many binding partners (>20 are currently known) and several of these are conserved in equivalent vertebrate pathways (see section: 'a synthesis of information on akirin function'). Thus, there are many potential residues of functional significance in Akirin (i.e. within putative binding sites) that may have evolved at different rates between Akirin1 and 2. Furthermore, binding sites for 14-3-3 proteins have been biochemically confirmed in Akirin2 2. Statistical methods have been proposed to predict whether functional divergence of related members of a protein family has occurred (e.g. 28) and identify the most likely residues involved (e.g. 2829). Initially, we used a likelihood ratio test (LRT) implemented in the program DIVERGE 33 to test the hypothesis that functional divergence of Akirin1 and 2 was a reflection of a shift in evolutionary rates at certain amino acid sites between paralogues (known as type-I divergence, 28). The model underlying this approach was described in 28. To summarise, we tested whether the coefficient of functional divergence (θ) between Akirin1 and 2 clusters in the phylogenetic tree topology provided (as in fig. 3), was >0. The LRT predicted that functional divergence had occurred between Akirin1 and 2 (θ = 0.37 +/- 0.06, statistically significant 2833). Next, DIVERGE was used to establish the posterior probability of type-I divergence at each site in the alignment (fig. 8a). Employing a cut-off posterior probability value of 0.6, ~20 sites were predicted as the most likely candidate sites for type-I functional divergence. For comparative purposes, the same alignment and phylogeny was submitted to a ML LRT, which, like the Bayesian method provides a statistical framework where evolutionary rate shifts at particular protein positions can be established 29. The statistically most likely positions predicted to underlie functional divergence were comparable by both methods, particularly for the highest-ranking candidates (fig. 8a, b). High scoring positions were mapped onto an amino acid alignment of the Akirin gene family, in relation to known functional motifs (fig. 9). A PDF output of the ML analysis, in its original format, is also provided (additional file 5).

The extreme N-terminus (first 30 residues) and C-terminus (last ~70 residues) of Akirin proteins are clearly under strong purifying selection based on the near absence of fast-evolving sites (additional file 5) and the presence of many sites that have evolved at a significantly slower rate than the average of all positions (fig. 9, additional file 5). Further, in these N and C-terminal regions, very few sites (respectively none and two) are predicted to contribute to functional divergence between Akirin1 and 2 (fig. 8, fig. 9). Of the last 65 sites in Akirins, 20% are conserved from basal metazoans to vertebrates and ~55% code for isofunctional replacements (not shown). Additionally, it is only the ~70 most C-terminal residues that share significant identify with the basal Amoebozoan and protist orthologues (not shown). Therefore these conserved regions must perform essential functions common to Akirins and are obvious candidates for experimental characterisation.

A known functional motif found in Akirins, is a highly conserved N-terminal NLS 1 (fig. 9). As expected, sites within this motif have evolved significantly slower than the average in all Akirins (fig. 9, additional file 5), in support of its necessity for nuclear localisation as demonstrated in insect and mammalian Akirins 1. Further, another NLS was predicted in PSORT2 34 to be present in Akirin of invertebrate deuterostomes (plus several other invertebrates, dating back to Placozoans, not shown) and Akirin2, but not Akirin1 (fig. 9). However, rate shifts at these sites were not predicted to contribute to functional divergence between paralogues. Interestingly, Akirin1 was detected in both the nucleus and cytoplasm of C2C12 myoblasts 3. Further experimental tests will be needed to examine whether this second NLS augments the nuclear import of Akirin and Akirin2 proteins relative to Akirin1, which would have important implications for the sub-cellular context of the vertebrate paralogues.

Almost all of the highest scoring candidate positions for functional divergence between Akirin paralogues are found in the middle region of the protein (positions 30–130 in our alignment), which also has numerous sites that evolved at a significantly higher rate in both Akirin1 and 2 compared to the average of all positions (additional file 5). The highest scoring site for functional divergence in both the Bayesian analysis and ML LRT (site 122) corresponds to a proline conserved in all Akirin2 orthologues, two invertebrate Akirin orthologues but not in Akirin1 proteins (fig. 9). In all tetrapod and most teleost Akirin2 orthologues, as well as hemichordate Akirin, this site is the final residue of a putative 14-3-3-recognition site, biochemically validated in rodent Akirin 2 (consensus: serine/threonine -X-proline in rat Akirin2 2). Further, two other high scoring positions fall either on putative 14-3-3 binding sites (site 52) or are just upstream of a 14-3-3 binding site conserved in both Akirin1 and Akirin2 (sites 111 and 113–114). It is feasible that these sites have contributed to altered 14-3-3 binding properties of Akirin1 and 2. Another region that is a strong candidate for type-I divergence between Akirin1 and Akirin2 is found at sites 58–67. In this region, 5/10 positions have evolved at a significantly slower rate in Akirin2 than Akirin1 (fig. 8) and are among the highest scoring candidate residues for type-I functional divergence (fig. 9). This region may be a binding site that is functional in the invertebrate Akirins and Akirin2, but not in Akirin1.

Of the five 14-3-3 protein-binding sites identified in rat Akirin2 2, four are conserved across amniote orthologues (not shown), and fewer in teleost orthologues (fig. 9). Akirin1 has between one and four putative 14-3-3 binding sites across a broad phylogenetic range of vertebrates, generally in regions conserved with at least one Akirin2 protein. Deuterostome invertebrate Akirins generally have two to four 14-3-3 binding sites, usually in regions aligning with vertebrate Akirins, but rarely with other invertebrate Akirins (fig. 9). The M. brevicollis, D. discoideum and N. gruberi orthologues have a single putative 14-3-3 binding site whereas G. theta has none (not shown). Therefore, the number of potential 14-3-3 binding sites in Akirin proteins increased rapidly at the base of metazoan evolution. However, sites are absent or greatly reduced in certain metazoan lineages, including D. melanogaster (0 sites), Anopheles gambiae (0 sites), Lumbricus rubellus (0 sites) and Caenorhabditis elegans (1 site) (not shown). The preferred binding motifs of 14-3-3 proteins are Arg-Ser-x-Ser-x-Pro and Arg-x-x-x-Ser-x-Pro, although functional variations in these motifs are tolerated 35. Almost invariably, sites in Akirin proteins have the consensus-binding site Ser/Thr-x-Pro or Ser-x-Ser/Thr-x-Pro (fig. 9). The single exception is the sea squirt sequence, which has a perfect site (Arg-Ser-Pro-Pro-Ser-Ser-Pro) (fig. 9). Unsurprisingly, multiple sites were needed for the formation of the Akirin2–14-3-3 complex 2. Considering the variability in the number (sometimes none) and physical locations of 14-3-3 sites, it is likely that the binding affinity for 14-3-3 proteins will vary considerably between Akirin1 and Akirin2 paralogues within vertebrate species and between orthologues from different lineages.

In this section, we combine the findings of this study with available literature on the known roles of akirin genes in order to provide novel insight into their biochemical functions. We hope that this will prompt the sharing of akirin literature between researchers from different fields and open up new avenues of investigation.

Consistent with the embryonic lethal knockdown of akirin and akirin2 in flies and mice respectively 1, the ablation of akirin in the embryos of the nematode C. elegans by RNAi knockdown was also lethal (http://www.wormbase.org/ search term: E01A2.6). Further, RNAi knockdown of akirin in ticks (i.e. subolesin, previously named 'protective antigen 4D8',36) dramatically affected the growth and fertility phenotype, with enormous associated reductions in survival, weight and oviposition, as well as developmental abnormalities in several different tissues 37. These findings support the idea that akirin is an essential developmental gene across a broad phylogenetic range of metazoans. Another conserved feature of Akirins in metazoans is their nuclear localisation (fly: 1 and Flybase: http://flybase.bio.indiana.edu/; mammals: 123 and broad or near-ubiquitous expression patterns in embryonic and adult tissues (fly: 1, nematodes: http://www.wormbase.org/, search term: E01A2.6; ticks: 36; zebrafish: 38; mammals: 12. These basic comparisons indicate that akirins function in a wide range of processes, through direct or indirect regulation of gene transcription, consistent with current literature 12331.

In vertebrates, akirin1 is not essential for embryonic development, and has even been lost in the class Aves. Thus, relative to Akirin and Akirin2, Akirin1 has diverged in at least one essential function (i.e. in innate immunity, although other functions of Akirin1 in this system could be masked by functional redundancy 1). This is supported by significantly faster rates of evolution in multiple sites of Akirin1 compared to its paralogue (fig. 7, fig. 8). However, there were also several sites that have evolved faster in Akirin2 than Akirin1, and could represent regions where a function has been conserved in Akirin1 but was lost in Akirin2. It is known that akirin1 (aka mighty) has a role in regulating vertebrate myogenesis, as it was identified in mice from a suppression subtraction hybridization cDNA library produced using myostatin-null mice as the 'tester' material 3. Myostatin (aka GDF-8) is a potent negative regulator of mammalian myogenesis and mice lacking a functional copy have a double-muscled phenotype 39. akirin1 was reportedly upregulated in the muscles of myostatin^-/- mice 3. Mstn protein was also shown to inhibit the transcription of the akirin1 proximal promoter 3. Interestingly, akirin1 also functions in myogenesis in flies. Specifically, Akirin (as Bhringi) bound the bHLH factor Twist and this interaction was necessary for the normal expression of Twist target proteins 32, representing another example of Akirins as co-regulators of transcription. Fly mutants lacking akirin had considerable defects in muscle mass and morphology 32. This is a strikingly opposite phenotype to that induced by the overexpression of akirin1 in mdx mice, where muscle mass, fibre size and structural integrity was markedly increased 3. Thus, the role of mammalian akirin1 in regulating muscle growth may be conserved from the akirin proto-orthologue. If the function of akirin1 in amniote muscle growth is essential, then its absence in birds, where muscle physiology is strongly conserved with mammals, particularly in terms of the functions of key genes (e.g. myostatin), could only be fulfilled by akirin2.

akirin2 (as FBI1) was also shown to promote carcinogenesis by interacting with the phosphoserine-threonine-binding protein 14-3-3β 2. 14-3-3 proteins are highly conserved in eukaryotes and regulate many cellular activities including the cell cycle, intracellular signalling, apoptosis and malignant transformation (reviewed by 3540). The 14-3-3β isoform had previously been shown to regulate tumour formation and was upregulated in several cancer cell lines 41 acting through the mitogen-activated protein kinase (MAPK) pathway 42. akirin2 was also upregulated in tumour cell lines and its mRNA downregulation reduced tumour metastasis by inducing the expression of MAP kinase phosphotase 1 (MKP1), which reduced the activation of the extracellular-signal regulated kinases (ERKs), ERK1 and ERK2 2. Specifically, the akirin2-14-3-3β complex functioned as a transcriptional repressor of the MKP-1 promoter 2. Based solely on the presence of a comparable repertoire of 14-3-3 protein-binding sites, redundancy of this carcinogenic-promoting function with akirin1 cannot be excluded. However, distinct evolutionary rates in positions within, or adjacent to 14-3-3 binding sites in Akirin1 and Akirin2 are probably important explanatory variables underlying their functional divergence (fig. 9). Interestingly, there also exists evidence to suggest that akirin1, like akirin2, indeedfunctions as part of the ERK signalling pathway. It is established that the inhibitory effect of Myostatin on myogenesis is mediated through activation of components of the MAPK/ERK signalling pathway 34344. akirin1 transcription was inhibited by treatment with Myostatin protein and conversely was upregulated by chemical inhibition of MEK1/ERK signalling 3. Thus, it was suggested that Myostatin signals to akirin1 through ERK signalling 3.

In vertebrate immune response signalling pathways, akirin2 functions at a level close to, or downstream of NF-κB to selectively regulate some of its target genes 1. Since a direct interaction of fly Akirin and NF-κB was not demonstrated, it was suggested that Akirins interact with intermediary components 1. 14-3-3 proteins are potential candidates, since they are known to regulate the nuclear localisation of transcription factors, are found in many transcriptional complexes, can bind to histones and can regulate histone acetylation 3540. Importantly, a 14-3-3-Akirin2 complex bound to and regulated promoter activity 2. 14-3-3 proteins regulate NF-κB activity by binding both IκB and the p65 subunit of NF-kB 45. IκB is known to inhibit NF-kB by sequestering p65 in the cytoplasm 46 and further, the IκBα isoform also facilitates its nuclear export 47. TNFα treatment induced the nuclear localisation of 14-3-3 proteins and the disruption of 14-3-3-protein function caused the nuclear localisation of both IκB and p65 45. Furthermore, following TNFα treatment, both IκB and 14-3-3β/γ proteins bound to the promoter regions of IL-6 and RANTES, presumably disrupting the interaction of p65 and chromatin 45. It was suggested that 14-3-3 proteins formed a complex with IκB and p65 that was efficiently exported from the nucleus 45. Interestingly, these same NF-KB transcriptional targets (IL-6, RANTES) were strongly repressed in akirin2 knockout mice following TLR, IL-1β and TNFα treatment 1. Therefore, an interesting line of investigation will be to examine whether the transcriptional repression of NF-kB targets in akirin2 knockout mice is accounted for by altered 14-3-3-protein activity. In addition to a predicted interaction with 14-3-3 proteins to regulate chromatin, fly Akirin (as Bhringi) was shown to bind Bap60 30, a DNA binding protein that forms part of the SWI/SNF-like chromatin remodelling complex 48 which is highly conserved in eukaryotes. Akirin also interacts with the GATA-transcriptional activator Pannier 49 and with TDP45 30, (TAR DNA binding protein 43), a highly conserved RNA binding protein with roles in transcriptional repression 50 and in regulating exon skipping 51. It is also noteworthy, that fly Akirin physically interacts with CG1473 30, a protein with high homology to a E2 Ubiquitin-conjugating enzyme. The ubiquitin-conjugating enzyme UBC13 forms part of the ubiquitin-conjugating complex important in the activation of IKK (and thus activation of NF-κB transcriptional activity) through TRAF6 52. CG1473, like Akirin, also binds to the chromatin remodelling protein Bap60 30, indicating a wider protein-interaction network.

14-3-3 proteins are also known to regulate insulin-like growth factor signalling, a pathway activated by Akirin1 overexpression 3. The 14-3-3ε isoform binds to phosphorylated forms of both the IGF-I receptor (IGF-I R) and the insulin receptor substrate-I (IRS-I) 53 while the 14-3-3β-isoform binds to activated IRS-I reducing its ability to activate PI(3) kinase (PI(3)K) 54. During myogenesis, a feed-forward cascade occurs, where IGF-II secreted during early myoblast differentiation, binds to and activates the IGF-IR, in turn activating IRS-1, and the PI(3)k-Akt phosphorylation pathway, which then promotes efficient transcriptional activation of muscle differentiation genes through a MyoD-E-protein complex and several known co-factors 55. In myoblasts overexpressing Akirin1, differentiation was accelerated, with a concurrent increase in MyoD, Myogenin and IGF-II protein expression, activated Akt expression and a massive increase in the transcription of IGF-II mRNA 3. These results suggest that Akirin1 can stimulate IGF-II-PI(3)K-Akt signalling, culminating in the transcription of muscle differentiation genes. Akirin1 has several low affinity 14-3-3 binding sites (fig. 8) and was detected in the cytoplasm 3. It is therefore possible that the positive effect of Akirin1 on the IGF-II signalling pathway is mediated through binding 14-3-3 proteins in the cytoplasm, sequestering them and effectively stimulating the activation of the IGF1-R and IRS-1 and downstream components of the pathway.

BLASTp searches of the NCBI http://www.ncbi.nlm.nih.gov/ non-redundant protein collection using D. melanogaster Akirin and M. musculus Akirin1/Akirin2 sequences as in silico probes, revealed homologues of these proteins in multiple metazoan taxa. Subsequently, manual screening of Ensembl release 50 genome assemblies was performed http://www.ensembl.org using the orthologue and paralogue prediction function with fly akirin as a reference point. Ensembl genome assemblies screened included Chordates (from the taxa Ascidiacea, Actinopterygii, Amphibia, Aves, Petromyzontiformes and Mammalia), Arthropods (Aedes aegypti, A.gambiae and D.melanogaster), nematodes (C. elegans) and Fungi (Saccharomyces cerevisiae).

To identify akirin1/akirin2 orthologues in a broader range of metazoans, directed tBLASTn searches of NCBI nucleotide and EST databases were performed for the following taxa: Acoelomorpha, Annelida, Arthropoda, Brachiopoda, Bryozoa, Chaeognatha, Chordata (classes: Ascidiacea, Aves, Cephalaspidomorphi, Cephalochordata and Myxini), Cnidaria, Ctenophora, Echinodermata, Entoprocta, Hemichordata, Mollusca, Nematoda, Nematomorpha, Nemertea, Onychophora, Placozoa, Platyhelminthes, Porifera, Rotifera, Tardigrada and Xenoturbellida. Non-metazoan eukaryotes were also screened by the same approach, including the following taxa: Amoebozoa, Choanoflagellata, Chromalveolata, Fungi, and Plantae. Finally, genome databases at the DOE Joint Genome Institute http://www.jgi.doe.gov/, Welcome Trust Sanger Institute http://www.sanger.ac.uk/, Arabidopsis Genome Initiative http://www.arabidopsis.org/ and TIGR Rice Genome Annotation http://www.tigr.org/tdb/e2k1/osa1/index.shtml were BLAST screened for akirin orthologues for the following taxa: Amoebozoa (D. discoideum, Entamoeba histolytica), Archea (Methanococcoides burtonii, Sulfolobus islandicus), Bacteria (Mycobacterium sp., Enterobacter sp. Escherichia coli, Staphylococcus aureus), Choanozoa (Monosiga brevicollis), Chromalveolata (Emiliania huxleyi, Thalassiosira pseudonana, Aureococcus anophagefferens), Excavata (N. gruberi, Trypanosoma brucei, Trichomonas vaginalis, Giardia lamblia), Fungi (Aspergillus niger, Candida albicans), Placozoa (T. adhaerens) and Plantae (Chlamydomonas reinhardtii, Selaginella moellendorffii, Sorghum bicolour, Oryza sativa).

Synteny maps for the genomic neighbourhoods surrounding akirin1 and akirin2 were constructed using data manually obtained from release 50–52 Ensembl genome assemblies for H. sapiens, M. musculus, G. gallus, X. tropicalis, D. rerio and G. aculeatus. The genomic neighbourhoods surrounding H. sapiens akirin1/akirin2 were used as a starting reference. The intron-exon organisation of eukaryotic akirin orthologues was established by loading genomic and corresponding cDNA sequences into Spidey 56. PSORTII 34 was used to predict NLSs.

27 full coding amino acid sequences of Akirin were used for phylogenetic analysis. This included Akirin1/Akirin2 sequences spanning broad vertebrate taxa as well as deuterostome outgroups representing the single invertebrate gene related to both vertebrate akirin1/akirin2 in Urochordates (H. roretzi), Cephalochordates (B. floridae), Hemichordates (Saccoglossus kowalevskii) and Echinoderms (Strongylocentrotus purpuratus). Sequence alignment was performed using PROMALS 57 at http://prodata.swmed.edu/promals/. The first output was improved by removing indels and low scoring regions of the alignment as well as manual checking of alignment quality. ML was performed using PhyML 58 at http://atgc.lirmm.fr/phyml/. The JTT substitution model was utilised with concurrent estimation of the gamma distribution parameter. 1000 bootstrap replicates were sampled to obtain a measure of branch confidence. The Bayesian approach was implemented in MrBayes3.12 59 with estimation of the substitution rate model, and gamma distribution of among site rate variation. 2 runs were used, each with a single chain of 20 million generations, sampled every 10,000 generations. Convergence was assessed by comparing the standard deviation of split frequencies between runs. 1000 trees were excluded from a total sample of 2001 trees in each run. The independence of the remaining samples was then assessed by analysing autocorrelation in tree log-likelihood values implemented using the ACF function of Minitab 13.2 (Minitab, Inc.). Sample independence was confirmed as no significant increase in log-likelihoods was observed after the burnin phase. Additionally, NJ, ME and MP analyses were performed in Mega 4.0 60, in each case obtaining branch confidence values by bootstrapping with 1000 iterations. For NJ and ME analyses, the JTT model was used with a gamma distribution parameter estimated by PhyML (α = 0.91). Finally, ASATURA was used to remove saturated amino acid positions from the alignment prior to NJ tree reconstruction 11 using the JTT model.

Phylogenetic analysis was performed on seven vertebrate gene families, where members were represented on both akirin1 and akirin2 containing chromosomal tracts in at least two vertebrate classes (further details are provided in the results and fig. 7). High quality amino-acid translations were obtained from Ensembl release 52 genome databases for representatives of four vertebrate taxa (mammalia, aves, amphibia and Actinopterygii). Outgroup sequences were obtained either through orthologue screening of Ensembl databases for C. intestinalis, or non chordate invertebrates, or by BLAST screening of NCBI C. intestinalis or B. floridae protein databases. Sequence alignment was performed with Promals 57 followed by manual checking and submission to Gblocks at http://molevol.cmima.csic.es/castresana/Gblocks_server.html to remove poorly aligned and divergent regions 61. Bayesian phylogenetic reconstruction was performed as for the Akirins, except with different sampling parameters for each gene family. Briefly, 5 million generations were performed with sampling every 2500 generations for the ras-related GTP-binding, heterogeneous nuclear ribonucleoprotein, cytosolic 5'-nucleotidase 1, proline-rich nuclear receptor coactivator, glycoprotein endo-alpha-1,2-mannosidase families. For the cannabinoid receptor and potassium voltage-gated channel families, 10 million generations were performed with sampling every 5000 generations. In each analysis, runs had converged (i.e. the standard deviation of split frequencies between runs was <0.005) before half of the final number of generations were reached. 1000 trees were excluded from a total sample of 2001 trees in each run before consensus phylogenies were reconstructed. ML, NJ and MP analysis were performed essentially as described for the Akirin dataset.

Estimates of synonymous and non-synonymous substitution rates for Akirin1 and Akirin2 were performed using codon-alignments obtained by loading aligned amino acid and corresponding nucleotide sequences into PAL2NAL 62. Akirin1 orthologues from H. sapiens, M. musculus, X. tropicalis and D. rerio were compared. Akirin2 orthologues from H. sapiens, M. musculus, G. gallus, X. tropicalis and D. rerio were compared. PAL2NAL was set to automatically calculate synonymous and non-synonymous substitution rates, for each pairwise comparison using a model 26 normally implemented in codeml of PAML 63. Additionally, two codon alignments were produced separately for the Akirin1 and 2 orthologues described above and loaded into Mega 4.0 60. Pairwise estimates of the number of synonymous and non-synonymous substitutions between different orthologues were then calculated using the Nei-Gojobori method 27, with the P-distance model.

To examine potential shifts in evolutionary rates between Akirin paralogues, an amino acid alignment with 14 Akirin2 orthologues, 9 Akirin1 orthologues and 4 Akirin orthologues from invertebrate deuterostomes (additional file 2) was loaded into DIVERGE 33 with a corresponding phylogenetic tree in Newick format, that had the topology obtained by ML (fig. 3). The Akirin1 and Akirin2 clades were defined as separate clusters and the coefficient of functional divergence and posterior probability for functional divergence at each site in the alignment were estimated using the Gu99 algorithm 28. Additionally, the same alignment was loaded into the rate shift analysis server at http://www.daimi.au.dk/~compbio/rateshift/29 along with the same Newick file. Akirin1, Akirin2 and Akirin (outgroup) clusters were defined and the JTT model was employed.