In order to construct a comprehensive phylogenetic tree, GenBank was searched for all possible Aurora kinases. In addition to previously published Aurora kinase sequences, further chordate and urochordate Aurora homologs were found by using mammalian Aurora-A, Aurora-B and Aurora-C protein sequences as queries in BLASTP or TBLASTN 24 searches of the genomes of the pufferfish, Takifugu rubripes 25, the zebrafish, Danio rerio 26, and the ascidian, Ciona intestinalis 27.

Multiple sequence alignments show that the Aurora kinase family is highly conserved among species (Fig. 1). Pairwise sequence comparisons estimate that the mean proportion of similar amino acids (based on the Blosum62 matrix) is much higher among all the different families of Aurora-A, Aurora-B and Aurora-C of vertebrates (0.84 ± 0.5) than within the same family (Aurora-A or Aurora-B) between vertebrates and invertebrates species (0.69 ± 0.3 for both families). This would suggest a recent evolutionary radiation of Aurora families within vertebrates.

Phylogenetic trees constructed using four methodologies, all rooted using polo-like kinases type 4 (PLK4), show that all vertebrate Auroras form a clade distinct from those of invertebrates (Fig. 2). The phylogenetic tree constructed by the neighbour-joining distance method shows moderate boot-strap support (67%) for the evolution of all vertebrate Auroras from a urochordate ancestor, represented by the ascidian, C. intestinalis. The use of alternative kinase families, other than PLK4, to root the tree did not alter the internal topology of the Aurora clade. Although its genome sequence is incomplete, C. intestinalis likely has only a single Aurora homolog since other probable kinase open reading frames associated with the next top five BLASTP 24 hits did not cluster with Auroras from other species in phylogenetic trees.

Among true vertebrates, our phylogenetic tree shows that the Aurora kinases underwent two major gene duplication events. The first split in cold-blooded vertebrates lead to the formation of two Aurora subfamilies. One branch encompasses all known vertebrate Aurora-A sequences in a single orthologous lineage that includes fishes, amphibians and mammals. This family includes previously identified Aurora-A kinases in Xenopus laevis, rodents and humans as well as a new putative ortholog in T. rubripes.

The second family, previously known as Aurora-B 228 consists of cold-blooded vertebrate and mammalian Aurora-B as well as mammalian Aurora-C. Mammalian Aurora-B and Aurora-C are similarly related to the cold-blooded vertebrate Aurora presently known as "Aurora-B" in amphibians (X. laevis) and fish (D. rerio and T. rubripes). Searches of T. rubripes and D. rerio protein and DNA sequence databases detected several other putative serine/threonine kinase homologs but none were Auroras according to phylogenetic analyses. Thus, cold-blooded vertebrates appear to have only a single Aurora-A ortholog, a single Aurora-B-like homolog, and lack an Aurora-C ortholog. Conversely, Aurora-A, Aurora-B and Aurora-C appear to be ubiquitous to mammals (at least placentals) where they are encoded by separate chromosomal loci. It would appear that mammalian Aurora-B and Aurora-C evolved from a duplication event involving the ancestral Aurora-B found in cold-blooded vertebrates. This depiction of the evolutionary relationships of vertebrate Auroras was consistently determined by four different phylogenetic methods with high bootstrap or Bayesian posterior probability values (Fig. 2).

Comparisons of human Aurora-B and Aurora-C sequences to the resolved 3D structure of human Aurora-A 29 lends further support to the evolutionary scenario that vertebrate Aurora-B and Aurora-C are closely related paralogs (Fig. 3a). Of the 26 residues lining the ATP-binding active site, only three vary among the different human Aurora kinases; Leu215, Thr217 and R220 (numbering and residue identity based on Aurora-A), and all of these variants were specific to Aurora-A (Fig. 3b). Aurora-B and Aurora-C did not vary in their active site residues. Furthermore, all three Auroras have a carboxy-terminal destruction box (D-box) but only Aurora-A has the necessary amino-terminal A-box (also known as the D-box activating-domain) for its functional activation 3031. Collectively, these comparisons of structure and motifs support the phylogeny depicting an early divergence of Aurora-A from an Aurora-B / Aurora-C clade.

The Aurora kinases of plants and invertebrates are all outgroup lineages to chordates / urochordates (Fig. 2). Although all phylogenetic methods strongly support the monophyly of chordate Aurora kinases, the exact ordering among nodes leading to the various plant and invertebrate clusters were not resolved with similarly high bootstrap or probability values. Placement of plant Aurora kinases between chordates and invertebrates might be an artifact of tree construction methods. (Plant, protist, fungal and invertebrate lineages were all highly diverged from vertebrate Aurora kinases as witnessed by their longer branch lengths.) The earliest lineages of the Aurora tree are those fungal model organisms with a single Aurora-like homolog S. cerevisiae (Ipl1) and S. pombe (Ark1). Other basal branches are the amitochondrial fungi, Encephalitozoon cuniculi, and the kinetoplast protist, Leishmania major 32.

Invertebrate Aurora kinases, including those of the model organisms C. elegans and D. melanogaster, occupy separate early branches and are not, as their current names suggest, orthologs to either Aurora-A or Aurora-B of vertebrates. An unrooted phylogenetic tree with only model organism species shows the same topology of vertebrate Auroras as the more species-rich tree rooted by PLK4 kinases (Fig. 4). However, similar kinases from C. elegans and D. melanogaster now cluster together. The unrooted tree suggests that the invertebrate Aurora-B kinase family evolved prior to the invertebrate Aurora-A kinase family although further examples from other species are desirable to confirm this hypothesis. The consensus scenario in both rooted and unrooted trees is that vertebrate Aurora kinases are paralogous, rather than orthologous, to their invertebrate counterparts.

Aurora-B and Aurora-C, as specific innovations in mammals, might have distinct protein-binding partners and cellular functions from those of Aurora-B kinases in amphibians. The perturbation of Aurora-B function in different systems suggests variable kinetochore-microtubule interactions 33. Transfection of normal rat kidney cells with a kinase-inactive, dominant negative form of Aurora-B caused multiple defects in mitosis 34 while an Aurora-B kinase inactivating antibody seemed to have milder effects in Xenopus tissue culture cells 35. Xenopus Aurora-A functions in the extrusion of the first polar body 36 while in C. elegans Aurora B plays a similar role 5. Also, C. elegans Aurora-B binds to a protein CSC-1 which has no homolog in other studied systems 37. While these studies used different experimental methods, the lack of direct orthology among vertebrate and invertebrate Aurora-A and Aurora-B might also account for functional differences in these systems.

The evolutionary analysis presented here also suggests revisiting the present Aurora nomenclature. Adams et al. 28 proposed a naming scheme where, irrespective of species, the original Aurora is known as Aurora-A (also called AIRK1, Aurora, Aurora-2, AIK, BTAK, human STK15, mouse STK6 and others), followed by Aurora-B (also known as AIRK-2, IAL, Aurora-1, AIK2, STK12 and others) and Aurora-C (or STK13). However, the proposed nomenclature fails to reflect evolutionary, and possibly functional, relationships among the Auroras. We suggest that Aurora-A be retained as the name for all orthologs in mammals and cold-blooded vertebrates. While Aurora-B and Aurora-C seem appropriate for mammalian versions, the ancestral cold-blooded vertebrate "Aurora-B" might be renamed "Aurora-BC". As for invertebrates, the so-called Aurora-A or Aurora-B genes are clearly not orthologs to their respective vertebrate counterparts. However, introducing a new nomenclature here might simply add further confusion to the field.

There have been several recent reports of Aurora kinase inhibitors that are under development by pharmaceutical or biotechnology companies for cancer treatment. The compounds Hesperadin (Boehringer Ingelheim 21) and ZM447439 (AstraZeneca 22) are suggested to be targeted to Aurora-B. While both studies show lesser levels of compound inhibition of Aurora-A as well as several other kinases, neither report included Aurora-C in their kinase profile. Selective inactivation of multiple kinases is not an undesirable pharmaceutical profile for a small molecule inhibitor and, in fact, could be the best strategy to achieve maximal clinical efficacy of an anti-cancer agent 38. Indeed, an intense area of anti-cancer research is the development of small molecular ATP analogues that generally target the kinase domain of protein kinases 39. For example, Gleevec (also known as imatinib and made by Novartis) for chronic myelogenous leukemia, is a small-molecule inhibitor that targets BCR-ABL, c-Kit and platelet-derived growth factor receptor kinases 40. Recently, a selective inhibitor of all three Aurora kinases, VX-680 (made by Vertex Pharmaceuticals), was reported to inhibit cell-cycle progression and induce apoptosis in various human tumor cell types and in vivo xenograft models 23. Interestingly, although VX-680 is a potent inhibitor of all three Aurora kinases, its apparent inhibition constant is much lower for Aurora-A (0.6 nM) than for either Aurora-B (18 nM) or Aurora-C (4.6 nM). Again, the compound's greater affinity for Aurora-A, relative to Aurora-B and Aurora-C is compatible with the proposed evolutionary scenario of mammalian Auroras.

All Aurora kinase orthologs and paralogs were initially collected from GenBank nonredundant protein database by performing separate searches using BLASTP 24 with human Aurora-A, Aurora-B and Aurora-C proteins as query sequences and a cut-off E-value of 1.0e-10. Since this dataset included additional kinases to the Auroras, preliminary multiple sequence alignments and phylogenetic analysis using program CLUSTALW v1.7 41 served to identify the clade of all known Aurora kinases. Takifugu rubripes, Danio rerios and Ciona intestinalis homologs were obtained by BLASTP and TBLASTN 24 of species-specific protein and DNA sequence databases, respectively. The top five homologs from each species retrieved from separate searches with human Aurora-A, Aurora-B and Aurora-C were entered into a preliminary phylogenetic analysis using all retrieved Aurora kinases from Genbank. These analyses revealed that T. rubripes and D. rerios had orthologs to X. laevis Aurora-A and Aurora-B but not mammalian Aurora-C. C. intestinalis had a single Aurora-like kinase.

PLK4 kinases were selected as the outgroup for phylogenetic analyses because they were the most similar non-Aurora kinases to either human Aurora-A, Aurora-B or Aurora-C in multiple BLASTP 24 searches of the non-redundant protein database of GenBank. Using alternative kinases as outgroups made no difference to the topology of the Aurora clade. Initial multiple sequence alignments were performed using the program CLUSTALW v1.7 41 with default settings and subsequently, refined manually using the program SEQLAB of the GCG Wisconsin Package v11.0 software package (Accelrys, San Diego, CA, USA). We removed regions with residues that could not be unambiguously aligned or that contained insertions or deletions. The final multiple sequence alignment was 240 amino acids in length. Pairwise comparisons for the proportion of similar residues were estimated from the length of the shortest sequence without gaps and the Blosum62 weighting matrix as implemented in the program OLDDISTANCES in GCG.

We constructed phylogenetic trees using distance neighbor-joining (NJ), maximum parsimony (MP), maximum likelihood quartet puzzling (QP), and Bayesian posterior probabilities (BP). NJ trees were based on pair wise distances between amino acid sequences using the programs NEIGHBOR and PROTDIST (Dayhoff option) of the PHYLIP 3.6 package 42. The programs SEQBOOT and CONSENSE were used to estimate the confidence limits of branching points from 1000 bootstrap replications. ML tree topologies were constructed using the software PUZZLE 4.0 43, employing 1000 puzzling steps, the JTT substitution matrix, estimation of rate heterogeneity using the gamma distribution model with eight rate categories, and the gamma-parameter estimation from the dataset. MP analysis was performed using PAUP4.0b5 software 44 where the number and lengths of minimal trees were estimated from 100 random sequence additions, while confidence limits of branch points were estimated by 1000 bootstrap replications. BP trees were constructed using the software MrBayes v3.0B4 4546. Bayesian analysis used the mixed model of sequence evolution with random starting trees. Markov chains were run for 10⁶ generations, burn-in values were set for 10⁴ generations, and trees sampled every 100 generations. All trees were visualized using the program TREEVIEW v1.6.6 47.

For the Aurora kinase phylogeny rooted with PLK4 kinases shown in Fig. 2, the log likelihood of the final ML tree was -8059.78. Four minimal length MP trees were recovered, 1522 steps in length with a consistency index (CI) of 0.5802 and a retention index (RI) of 0.6016. The variable branch arrangements were terminal nodes (human, pig and cow Aurora-B) which did not affect the central findings.

For the unrooted phylogeny of Aurora kinases of model organisms shown in Fig. 3, the log likelihood of the final ML tree was -4469.20. A single minimal length MP trees were recovered, 779 steps in length with a consistency index (CI) of 0.7214 and a retention index (RI) of 0.2786.

The SwissPDBviewer program 48 was used to obtain the surface representation of human Aurora-A kinase (PDB ID 1muoA). The active site residues, defined as being within 5A of the ADP cofactor, were identified using the program CAST 49. The multiple sequence alignment for the three human Aurora kinase proteins was obtained using CLUSTALW 41. Multiple sequence alignment and sequence GenBank accession numbers are available as Supplementary Information [see additional file 1 and 2].