To systematically identify gene losses in vertebrate and insect representatives of Bilaterian species, we delineated orthologous groups based on all-against-all Smith-Waterman comparisons using the official gene sets of five vertebrates (human, mouse, opossum, chicken and pufferfish) and five insects (fruitfly, malaria mosquito, dengue/yellow fever mosquito, honeybee and red flour beetle) (see details in Materials and methods). The species choice aimed to maximize the phylogenetic coverage with similar lineage radiation times in both deuterostomia and protostomia. The fraction of genes with recognized orthology among these species represents about 70-80% of their predicted gene pools. The comparative analysis of the shared content of gene repertoires across these species is discussed in the study presenting the analysis of the first beetle genome, that of Tribolium castaneum 10, and here we focus on the analysis of losses of orthologous genes. According to their phyletic distribution and gene copy-number in each of the species, we considered the following types of orthologous groups reflecting different selection pressures: the universal single-copy orthologs (U); the universal multiple-copy orthologs (N); patchy orthologs (P) that are present in both phyla in at least three species, in single or multiple copies; and insect- or vertebrate-specific orthologs (I and V, respectively). While universal single-copy orthologs (U) are evolving under a distinct pressure for copy-number control, the number of universal multiple-copy orthologs (N) under similarly strict copy-number control is extremely low (only six groups have equal multiple-copy gene number in at least nine species). Although U, N and P orthologs must all have been present in the last common ancestor of insects and vertebrates, the Urbilateria, losses in these fractions occur at different rates.

Figure 1b shows the size distribution of the orthologous fractions and the number of losses in each species and ortholog category. The tree shown in Figure 1a illustrates the species phylogeny, which allowed us to infer the number of losses on the internal branches, assuming evolutionary parsimony, which minimizes the number of events required to explain the phyletic gene distribution in each orthologous group. The phylogenetic tree was reconstructed using maximum-likelihood analysis of the concatenated alignment of 1,150 universal single-copy orthologs 10 where the lengths of the branches are proportional to the number of accumulated mutations, allowing us to compare the gene loss rates with the rates of lineage divergence (measured as the rate of protein substitutions). For the branches closest to the root, the numbers of gene losses cannot be inferred without an additional outgroup.

The analysis identified hundreds of differentially lost Urbilaterian genes of U, N and P orthologs in each of the species (see table of Figure 1). Overall, about 40% of ancient orthologous groups have been lost in at least one (out of the ten) species, illustrating the extent of the evolutionary flexibility of gene pools. Moreover, there are dozens of genes lost in each of the species that otherwise appear as universal single-copy orthologs. Koonin et al. 14 noted that nearly all pan-eukaryotic single-copy orthologs are subunits of known protein complexes; nevertheless, the observed losses indicate that even seemingly tightly constrained genes are, to a certain degree, dispensable in evolution.

The inferred distribution of losses over the internal branches of the species phylogeny allowed us to correlate them with the estimated geological time of species radiations and the molecular evolutionary rate in each of the lineages. The molecular rates of evolution were quantified using genome-wide maximum-likelihood analysis of amino acid substitutions in the well aligned regions of single-copy orthologs (see Materials and methods). Figure 2a displays the correlation of the number of losses of the different types of orthologs plotted versus the protein sequence divergence. Losses of U and N orthologs (Figure 2b) occur only at the terminal branches as the fraction definition requires presence of the orthologous genes in at least nine species, whereas losses of P orthologs (Figure 2c) occur at both internal and terminal branches. Correlations are statistically significant for all categories (see Figure 2 legend for details), and there is no distinction between insects and vertebrates. Moreover, although the absolute numbers of insect- and vertebrate-specific losses appear different, they in fact follow the same trend when normalized to the total number of the phylum-specific orthologous groups (Figure 2d; see Additional data file 1 for a graph with absolute numbers). The different slopes of the regression lines reflect the varying stringency of evolutionary constraints that differ between the postulated types of orthologs. Despite the difference in the absolute numbers of lost U and N orthologs, their rates of loss are indistinguishable when normalized to the number of such orthologous groups, indicating the same level of purifying selection (Figure 2b). The data show that P orthologs are about 8-fold less constrained than U and N orthologs; this roughly corresponds to about 20% of the common Bilateria gene pool evolving 8 times faster than the remaining, more constrained fraction. I and V orthologs appear to be about three-fold more constrained than P orthologs, which is not surprising as they may represent a similar mixture of a slower evolving fraction of 80% and a faster evolving minority. The level of correlation between the number of losses and the protein sequence divergence rates (Figure 2) is similar to that observed between other genome-wide measures of species divergence 18. Chicken was excluded from this and all following analyses as a clear outlier (see Discussion).

Protein sequence divergence is significantly larger between insects than between vertebrates (see the longer branch lengths in Figure 1; Mann-Whitney U test, p = 0.009). Similarly, this is reflected in the observation of significantly more frequent gene losses in insects than in vertebrates (Mann-Whitney U test: N orthologs, p = 0.016; P orthologs, p = 0.04). In comparison with vertebrates, the rate of evolution in bee and beetle is about two-fold higher and up to three-fold higher in Diptera. This especially high rate of evolution in Diptera, particularly at the base of the Dipteran radiation, has been noted previously 19.

Despite inherent dating uncertainties, the correlation between the number of lost orthologous groups and divergence times is significant for U and P orthologs (Spearman rank correlations: U orthologs, rs = 0.84, p = 0.007; N orthologs, rs = 0.58, p = 0.11; P orthologs, rs = 0.57, p = 0.03), indicating that losses of ancient genes occur in a roughly clock-like manner. The good correlation between the rate of losses with molecular rate and time indicates their stochastic nature. Projection of these trends as shown in Figure 3 to 600 million years ago (MYA), presumably dating the radiation of insects and vertebrates, suggests that over 1,000 (95% confidence interval 799-1,456) Urbilaterian genes have been lost from insects and only half this number (95% confidence interval 404-678) from vertebrates. This leads to the lower estimate of the number of Urbilaterian genes of just over 7,000 (remarkably, we obtained 7,114 orthologous groups with at least one insect and one vertebrate member). This estimate, however, does not take into account: genes that currently appear as insect- or vertebrate-specific, many of which could be of Urbilaterian origin; closely related Urbilateria paralogs that remain unresolved and are likely grouped together in N groups; as well as fractions of fast diverging genes that escaped our orthology classification.

Protein sets were retrieved from Ensembl for Drosophila, Anopheles and all vertebrates as of 4 August 2006. Tribolium and Apis proteins were retrieved from Baylor College of Medecine and Aedes proteins from VectorBase. The assignment to orthologous groups was performed as described earlier 95960. Namely, we retained the longest open reading frame per locus and performed all-against-all comparisons using the Smith-Waterman algorithm as implemented in ParAlign 61 with default parameters. The orthologous groups were then assembled from the best reciprocal hits (BRHs; reciprocally best maching genes in between-genome-comparisons) applying a COG-like 62 procedure to join BRHs across three or more species, going from the best scoring ones until an E-value cut-off of 10^-6, and keeping single BRH pairs only with E-values less than 10^-10. Furthermore, the orthologous groups were expanded by genes that are more similar to each other within a proteome than to any gene in any of the other species, and by very similar copies that share over 97% sequence identity, which were identified initially using CD-hit 63. All proteins in a group were required to have aligned regions overlapping by at least 20 residues to avoid the 'domain walking' effect.

A maximum likelihood species tree was calculated using the concatenated multiple alignments of 1,150 orthologs present in exactly one copy in all the organisms studied here. Multiple protein alignments were produced using muscle 64 and confidently aligned regions were extracted using Gblocks 65 with default settings. Individual protein alignments were concatenated into a 336,069 amino acid superalignment that was then subjected to maximum-likelihood analysis using the JTT model (G4+I+F) as implemented in PhyML 66 and we used Tree-Puzzle 67 to join separate bootstrap analyses. All shown branchings have at least 99% bootstrap support estimated from 500 replicates.

Species radiation dates were taken from the literature: for insects from 18 and for vertebrates from 6869. We used MatLab version 7.2 (MathWorks, Natick MA, USA) for statistical analysis and data plotting. Regression lines were required to cross the origin. For each category of orthologs, the slopes of the regression lines for insects and vertebrates were compared based on a Student's t-distribution and were found not to be significantly different. Because traits were not normally distributed, we used non-parametric Spearman's correlation coefficients and Mann-Whitney U tests. Chicken data were excluded from graphs and statistical tests (see Discussion).

Selected orthologous groups were examined manually as follows. The absence of Tribolium proteins was verified by screening the Tribolium proteome, genome (assembled and single reads) and expressed sequence tags using the Baylor College of Medicine blast server 70. All sufficiently similar sequences, including members of other orthologous groups, were aligned using muscle v3.6 64 with default settings and all positions containing gaps were trimmed from conserved blocks using Gblocks 65. Phylogenetic trees were constructed using maximum likelihood as implemented in PhyML 66 using the JTT model of amino acid substitution, a gamma distribution of rates over four rate categories and 100 bootstraps.

We used pathway annotations from the KEGG database 71, mapping genes to Biocarta and co-citation analysis using Webgestalt 29 web interface. For data mining we used Ensembl 72, Swiss-prot/UniProt 73, Flybase 74, the Interactive Fly 22 and Online Mendelian Inheritance in Man 75 annotations.