Nucleotide substitution parameters were calculated for the coding nucleotide alignments independently for each gene [see Additional file 1]. We then tested for differences between the three groups of genes (Hox, Hox-derived and non-Hox) (top section of Table 1) [see Additional file 2]. Our results showed that Hox-derived genes are evolving much faster and with less functional constraint than Hox and non-Hox genes. Differences among the three groups are significant for the number of nonsynonymous substitutions per nonsynonymous site, d_N (P = 0.022), and the level of functional constraint, ω (P = 0.000) (see Methods). The gene zen2 is the main gene responsible for the high values of nucleotide substitutions (both synonymous and nonsynonymous) in its group [see Additional file 1]. On the contrary, Hox and non-Hox genes have a similar number of nucleotide substitutions, t (P > 0.1). However the level of functional constraint is even higher (lower ω) in non-Hox genes than in Hox genes (ω = 0.04156 versus ω = 0.06094, respectively), although differences are only marginally significant (P = 0.063). Therefore, Hox genes do not seem to be evolving more slowly than other non-homeotic genes, despite their essential function in early development.

Then, we plotted d_N and ω in sliding windows along the coding sequences of Hox and Hox-derived genes to see whether or not these parameters behave homogeneously along the sequence. Figure 1 shows that, in all genes except zen2, there is a substantial decrease of both d_N and ω near the homeobox. zen2 contains a rapidly evolving homeobox with high ω values. Contrarily, we have observed that peaks of d_N tend to lie within repetitive regions (data not shown).

To control for a possible effect on the overall nucleotide evolution of both the homeobox and the repetitive regions (see Methods) of these Hox and Hox-derived genes, we tested again for differences among the three groups of genes excluding these regions. Removing the homeobox in Hox and Hox-derived coding sequences (second section of Table 1) elevated the number of nucleotide substitutions in these two groups, and decreased further their level of functional constraint. Again, differences among groups were significant for d_N (P = 0.005) and ω (P = 0.000), and the same tendency of the previous analysis with complete coding sequences was observed. In contrast, removing repetitive regions (third section of Table 1) decreased the number of nucleotide substitutions, especially in Hox genes, where all the genes in the group contain this type of region. Therefore, the elimination of repetitive regions slightly increases the difference between Hox and non-Hox genes in terms of nucleotide substitutions, and reduces the difference in functional constraint. Once more, differences among groups were significant for d_N (P = 0.030) and ω (P = 0.001). Excluding both the homeobox and the repetitive regions (bottom section of Table 1) gave intermediate results. Therefore, we can conclude that: (1)Hox and non-Hox genes are evolving similarly in terms of nucleotide substitutions, (2) Hox-derived genes are evolving much faster and with less functional constraint than the other two groups of genes, and (3) neither the homeobox nor the repetitive regions alter the estimates significantly, and thus are not entirely responsible for the two previous conclusions.

An excess of nonsynonymous over synonymous substitutions is a robust indicator of positive selection at the molecular level. Therefore, we searched for values of nonsynonymous/synonymous rate ratio (d_N/d_S = ω) greater than 1 to investigate whether Darwinian selection has been acting on any of the coding sequences analyzed in this study. However, no evidence of positive selection in any coding sequence or region of it was found.

We used the protein alignments to calculate the proportion of amino acid differences and indels. In the first case (Table 2, Figure 2), differences among the three groups – Hox, Hox-derived and non-Hox – were not significant (P = 0.101). However, the proportion of amino acid differences was substantially higher for Hox-derived genes (40.43%) than for Hox and non-Hox genes (22.80% and 23.77%, respectively). This result is in full agreement with our previous estimates of d_N (Table 1), which showed high values of this parameter for Hox-derived genes, but very similar values for Hox and non-Hox genes.

Second, we analyzed the proportion of indels in the alignments (Table 3, Figure 2). In this case, differences among the three groups of genes were highly significant (P = 0.000). Surprisingly, differences were due to the low indel proportion in non-Hox genes (8.73%) compared to the high values for Hox and Hox-derived genes (25.77% and 37.53%, respectively). Furthermore, we tested for differences in indel length using a nested ANOVA. The results indicated that, although the variation in indel length between genes within groups is significant (P = 0.021), the difference between groups is even more significant (P = 0.001). Mean indel length for Hox, Hox-derived and non-Hox genes is 4.22, 5.99 and 3.55 amino acids, respectively. Non-Hox genes not only have on average shorter indels, but also their longest indel is only 23 amino acids, in comparison with 43 and 40 amino acids for Hox and Hox-derived genes, respectively. In all groups, the indel length distribution follows a negative exponential curve: short indels are common and their abundance declines as length increases (data not shown).

Finally, we tested whether the proportions of amino acid differences and indels are correlated. The Pearson correlation indicated that these two variables are positively but not significantly correlated (r_Pearson = 0.307, P = 0.175). Therefore, genes with a high proportion of indels do not necessarily have a high proportion of amino acid substitutions. This probably points to different causal mechanisms for amino acid substitutions and indels.

Most Hox and Hox-derived proteins contain large repetitive regions present throughout the protein except the region near the homeobox and other highly conserved regions (see for instance the amino acid sequence of ABD-A in Figure 3). Predominant repetitions are poly-glutamine (poly-Q), poly-alanine (poly-A) and serine-rich regions (S-rich). These repetitive regions seem to include most of the indels and amino acid differences, and therefore they might be responsible for the surprisingly high evolutionary rate of Hox and Hox-derived proteins.

To test this hypothesis, we repeated the analyses of amino acid differences and indels inside and outside these repetitive regions (see Methods), and compared these two kinds of sequences (repetitive and unique). In the case of amino acid differences (Table 2), the percentage of aligned, non-conserved amino acids is higher in repetitive regions than in unique sequence in all the three groups. The T-test for paired samples (unique versus repetitive) on proteins having both types of regions showed significant differences between unique and repetitive sequences (P = 0.001), the mean of repetitive sequences being more than twice that for unique sequences (51.01% versus 23.19%, respectively). Despite this higher percentage of amino acid differences in repetitive than in unique sequence, the three groups of genes behave in a similar manner in both types of regions (note that the ranking is the same in both unique and repetitive regions).

Finally, we wanted to determine whether or not repetitive regions accumulate a larger number of indels than unique sequence (Table 3). The results show that in all the three groups, the percentage of indels in repetitive regions is much higher than that in unique sequence. These differences are significant (P = 0.006) according to a T-test for paired samples, giving an average value of 42.32% in repetitive regions versus 15.53% in unique sequence. Nevertheless, the ANOVA computed after removing repetitive regions remained highly significant (P = 0.003). Thus repetitive regions are not entirely responsible for the high percentage of indels in Hox and Hox-derived proteins. Therefore, Hox and Hox-derived genes have a tendency to accumulate indels even outside of repetitive regions, which does not seem to be allowed in non-Hox genes.

This study shows that Hox genes seem to be evolving differently from other essential genes expressed in early development, with complex expression patterns or with long introns rich in cis-regulatory elements. Both the number of nonsynonymous substitutions and the degree of functional constraint are not significantly different between Hox and non-Hox genes, and this remains true even when the most peculiar regions (the homeobox and the repetitive regions) are excluded (Table 1). Therefore, Hox genes do not seem to be evolving more slowly than other non-homeotic genes, despite their essential function in the early development and even though their interchangeability between species has been proven to be functional in some cases 2223242526.

Differences in the evolutionary rate among the three groups of genes (Hox, Hox-derived and non-Hox) could be mediated by some properties of genes that are correlated with the number of nucleotide substitutions (t). One possibility is that Hox and Hox-derived genes experience similar background rates of mutation that are different from those of non-Hox genes. We can use the number of synonymous substitutions per synonymous site (d_S) as a measure of the mutation rate of a gene. This variable is not significantly different among the three groups of genes (P = 0.530), and thus we can consider that mutation rate is constant across groups [see Additional file 2]. Another possibility is that genes within a group may have correlated levels of synonymous codon bias. Given that genes with higher codon bias tend to evolve more slowly 2843, codon bias may contribute to spurious differences in the rates of protein evolution among groups. We have measured codon bias for each gene using the Effective Number of Codons, N_C 44. There are no significant differences in the codon bias among groups, and the average N_C value for non-Hox genes is the lowest among the three groups (the highest codon bias) [see Additional files 1 and 2].

Some Hox and Hox-derived genes considered here have been included in previous studies 2941. Davis et al. 29 showed that the strongest negative relationship between expression profile and evolutionary rate occurs at a late stage in embryonic development, soon after the main burst of expression of segment polarity and Hox genes. However, they also show that the most constrained transcription factors and signal transducers, the functional class that contains many developmentally essential genes, are expressed precisely at the same time as the segment polarity and Hox genes. One of the two Hox genes included in their study has also been analyzed here (abd-A), and it is incidentally the gene with the lowest number of nonsynonymous substitutions and the one that is most constrained in our sample of Hox genes. On the other hand, bcd, although being one of the first genes acting in Drosophila development, was reported in the same study as an exceptional case of a gene acting in the earliest stages of development but evolving surprisingly fast 29.

Furthermore, Hox genes depart from a negative correlation found in previous studies between evolutionary rate at the protein level and intron size, number of conserved noncoding sequences within introns, or regulatory complexity 32. In this respect, all Hox genes used in this study contain a total intron size >10 Kb [see Additional file 3], which corresponds to the longest intron size category used in 32. Therefore, Hox genes are expected to evolve slowly as they contain long intronic sequences. Both Hox-derived and non-Hox genes contain shorter intron lengths than Hox genes [see Additional file 3], and thus would be expected to evolve faster.

The percentages of amino acid differences and of indels in Hox proteins also depart from the initial expectations. While the percentage of amino acid differences is not significantly different among the three groups compared (Table 2), the percentages of indels in Hox and Hox-derived proteins are much higher than that in non-Hox proteins (Table 3). Therefore, Hox proteins are as divergent as non-Hox proteins in terms of amino acid changes, but they are much more divergent in terms of indels. A visual inspection of the alignments pointed out a possible explanation to these results (Figure 3). Hox and some Hox-derived proteins contain large repetitive regions, mostly homopeptides, present all along the protein except the region near the homeodomain and other highly conserved regions. It is within these repetitive regions where most indels and amino acid differences seem to accumulate, in some cases resulting in poor alignment, and therefore they could be responsible for the surprisingly high amino acid and indel evolution of Hox and Hox-derived proteins.

Although repetitive regions have been shown to be richer in amino acid differences and indels than unique sequence, they do not fully explain the high variation found in Hox and Hox-derived proteins. Even excluding repetitive regions, Hox and Hox-derived genes contain many more indels than non-Hox genes, although the percentage of amino acid substitutions is not significantly different between Hox and non-Hox genes. Therefore, taking amino acid differences and indels altogether we can state that the overall rate of evolution of Hox and Hox-derived genes is faster than that of non-Hox genes. The percentage of the alignment that has changed is 43.39% in Hox proteins, 64.73% in Hox-derived proteins and 30.97% in non-Hox proteins (the percentage of amino acid differences has been recalculated before being added to the percentage of indels to account for the total number of sites, both gapped and non-gapped, in order to make both percentages comparable). Finally, a lack of correlation between the proportion of indels and amino acid differences in the set of genes used in this study highlights the different evolutionary mechanisms that regulate both types of changes.

Multiple long homopeptides are found in 7% of Drosophila proteins, most of which are essential developmental proteins expressed in the nervous system and involved in transcriptional regulation 3445. What is the role of these homopeptides? They could be tolerated, non-essential insertions that may play a role as transcriptional activity modulators. Some examples have been described in Hox and Hox-derived proteins 15 that illustrate the acquisition of new functions in the insect lineage while maintaining their homeotic role. In these examples, selection against coding changes might have been relaxed because of functional redundancy among Hox paralogs. These sequence differences could be involved in the functional divergence of Hox proteins and the evolutionary diversification of animals 15.

The large effects of Hox genes on morphology suggest that they regulate, directly or indirectly, a large number of genes. It would be expected that such pleiotropic proteins would be constrained in their sequence variation and, hence, their contribution to morphological variation. However, it has been shown that microsatellite sequences in developmental genes are a source of variation in natural populations, affecting visible traits by expanding or contracting at very high rates 46. One intrinsic characteristic of microsatellites is their hypervariability, resulting from a balance between slippage events and point mutations 3536. Their mutation rate has been estimated to be 1.5 × 10^-6 per locus per generation in the case of trinucleotide repeats in D. melanogaster 47, and is even greater in the case of dinucleotides. These values contrast with the general mutation rate of ~10^-8 per site per generation of base pair substitutions 48. These repeats typically generate regions in the alignment with high variability in sequence and length, and that are difficult to align.

A potential role for homopeptides is to serve as spacer elements between functional domains, to provide flexibility to the three-dimensional conformation, and fine-tuning domain orientation of the protein in its interactions with DNA and other proteins. To that effect, changes in nucleotide distances between target binding sites might be accompanied by complementary changes in the sequences spacing the binding domains of transcription factors (mostly homopeptides). This would produce a coordinated evolution between transcription factors and their target binding sites. Excessive expansions of homopeptides, however, have often been associated with disease in humans 49505152. Amazingly, essential developmental proteins like homeotic proteins that apparently need such homopeptides for their correct functioning have to suffer the consequences of their quick and apparently unpredictable evolution, and sacrifice in this way the conservation that would be expected in proteins of this type.

Among non-Hox genes, the cluster of cuticular genes (Ccp84Ac, Ccp84Ae, Ccp84Af and Ccp84Ag) behave similarly to Hox and Hox-derived genes and account for the vast majority of indels in their group (Figure 2). These short proteins share a conserved C-terminal section 53 and include a 35–36 amino acid motif known as the R&R consensus, present in many insect cuticle proteins, an extended form of which has been shown to bind chitin (chitin-bind 4; PF00379) 54. Outside these conserved domains, cuticular proteins share hydrophobic regions dominated by tetrapeptide repeats (A-A-P-A/V), which are presumed to be functionally important 5556 and are responsible for the high percentage of indels found in these proteins. These repeats are usually complex repeats that are not annotated in UniProt, nor detected as runs of identical amino acid repetitions (see Methods), and thus contribute to the percentage of indels in unique sequence in non-Hox genes (Table 3). When complex repeats were annotated and considered as repetitive sequence (see Methods), the percentage of indels in the unique portion of all classes of genes decreased substantially, but especially in non-Hox genes [see Additional file 4]. The elimination of complex repeats in cuticular genes was crucial in this reduction, and further increased the differences among groups.

Therefore, our results show that long repetitive sequences are not enough to explain all the differences found between Hox or Hox-derived genes and non-Hox genes. Hox and Hox-derived genes have a tendency to accumulate indels outside these repetitive regions that is not observed in non-Hox genes. We propose that spontaneous deletions between short repeated sequences could be the mechanism responsible for this difference 57. Such deletions have been described in phages 5859, Escherichia coli 606162636465 and humans 6667, and predominate between short sequence similarities of as few as 5–8 base pairs 68. Two different models can explain the generation of spontaneous deletions: slipped mispairing during DNA synthesis, and recombination events mediated by enzymes that recognize these sequence similarities. In either case, the repetitive and compositionally biased nature of several regions within Hox and Hox-derived sequences might explain the major incidence of indels in these two groups. This would also explain the large differences in protein lengths among species that have been observed in some Hox proteins 8. This higher probability of mutation would presumably be accompanied by a higher tolerance to indels of Hox and Hox-derived proteins outside their binding domains.

For a correct interpretation of our results, the set of non-Hox genes should be an unbiased sample of genes, both in terms of protein expression and structure. We have gathered this information from the literature, and verified that our non-Hox sample comprises a variable group of genes that are expressed through the fly life cycle (from young embryo to adult) and contains a wide variety of protein domains [see Additional file 5]. Therefore, we assume that, although small, it represents an unbiased sample of all non-Hox genes in the genome, and that results presented here are reliable.

The three Hox-derived genes used in this study (bcd, zen and zen2) originated from two consecutive duplications of the ancestral Hox3 gene. Seemingly, bcd and zen have specialized and perform separate functions in the establishment of the embryo's body plan 111213. This is supported by our data, as these two genes have a moderate evolutionary rate but low level of functional constraint (high d_N/d_S rate ratio). However, the finding of Barker et al. 42 that genes with a maternal effect experience relaxed selective constraint resulting from sex-limited expression is not supported by our data. Our results show that bcd and zen are evolving at very similar rates in the Drosophila lineage, and bcd is even more constrained than zen [see Additional file 1].

The function of zen2 is unclear. It has the same expression pattern as zen and, despite its high divergence across species, it has been maintained for more than 60 Myr 19. Conservation of two paralogous genes maintaining the same function is unlikely, and could only be explained under some peculiar conditions (e.g. two strongly expressed genes whose products are in high demand 40). It could be that this gene is experiencing a process of pseudogenization, supported by the fact that the evolutionary rate of zen2 is more than twice that of bcd and zen, and that it has also the highest percentage of the alignment represented by indels. If so, we would expect to see a relaxation of the functional constraint. However, the relatively high level of functional constraint of zen2 (ω = 0.09144) rather indicates a process of neofunctionalization, even though positive selection was not detected. The fact that this gene does not show an explicit pattern of variation of ω along its sequence (Figure 1) further supports the progressive loss of its original homeotic function and the acquisition of new functions.

Compared to the other two groups (Hox and non-Hox genes), Hox-derived genes are evolving significantly much faster and with less functional constraint. It is also the group with the highest proportion of amino acid differences and indels. These results reflect their relatively recent origin by duplication, which was followed by extensive changes in their role during the development of insects.

All the completely sequenced genes in D. buzzatii with a clear ortholog in D. melanogaster and D. pseudoobscura (23) were included in our analysis: abd-A, lab, pb, bcd, zen, zen2, Dbuz\Ccp3 (ortholog of Dmel\Ccp84Ac), Dbuz\Ccp6 (ortholog of Dmel\Ccp84Ae), Dbuz\Ccp7 (ortholog of Dmel\Ccp84Af), Dbuz\Ccp8 (ortholog of Dmel\Ccp84Ag), CG1288, CG14290, CG14609, CG14899, CG17836, CG2520 and CG31363 from Negre et al. 19; Adh-related (Adhr) from Betran and Ashburner 69; α-Esterase-2 (α-Est2) and α-Esterase-3 (α-Est3) from Robin et al. 70; CG13617 from Puig, Caceres, and Ruiz 71; and Larval serum protein 1 β (Lsp1β) and Lsp1γ from Gonzalez, Casals and Ruiz 72. Sequences of D. melanogaster orthologs were collected from Flybase 7374, and those of D. pseudoobscura were annotated on the scaffolds from the whole genome shotgun sequencing project 7576. We identified the D. pseudoobscura orthologs by using the alignment of this species with the D. melanogaster genome generated by the Berkeley Genome Pipeline 77, and annotated the target sequences with the aid of ARTEMIS v. 7 78 and BIOEDIT v. 7.0.4.1 79. A complete list of all genes, accession numbers (from Genbank or Flybase) and chromosomal locations is provided [see Additional file 3]. The longest transcript of each gene was used for the analyses. Genes were classified into three categories: 1) Hox genes (abd-A, lab and pb); 2) Hox-derived genes (bcd, zen and zen2); and 3) non-Hox genes (the remaining 17 genes). Results in each group were produced by calculating the average of all the genes within the group.

A set of Perl scripts, together with modules from PDA v. 1.4 80 and BIOPERL v. 1.2.3 81, were used to automatically check sequence annotations, extract the coding sequences (CDSs) of the selected transcripts and calculate basic gene structure and base composition parameters (gene and protein lengths; codon bias measured by the Effective Number of Codons (N_C); and G+C content in second, third and all codon positions) [see Additional file 1]. Differences among the three groups of genes were tested with one-way ANOVAs and pairwise contrast tests 82, assuming homogeneity of variances for those variables that gave non-significant P values for the Levene test 83 [see Additional file 2]. Orthologous coding sequences in D. buzzatii, D. melanogaster and D. pseudoobscura were aligned according to their translation to protein using RevTrans 1.3 Server 84 with some manual editing using BIOEDIT v. 7.0.4.1 79. Two non-Hox genes of the initial sample (CG1288 and CG17836) showed a doubtful alignment, containing many gaps and few residue matches, and thus were excluded from the analyses to avoid unreliable estimates. A total of 15 non-Hox genes were therefore used in this study.

The numbers of synonymous and nonsynonymous substitutions per site (d_S and d_N, respectively) were estimated on the nucleotide alignments of each gene using maximum likelihood methods with the program codeml of the PAML v. 3.14 package 85 [see Additional file 1]. We used an unrooted tree and the codon equilibrium frequencies (π_i) estimated from the nucleotide frequencies of the three codon sites (F3X4 option of codeml). Differences among the three groups of genes were tested using one-way ANOVAs and pairwise contrast tests as before. Furthermore, we visualized differences along the genes by plotting d_N and ω in sliding windows of 240 nucleotides and a step size of three nucleotides (one codon).

We measured the proportion of amino acid differences and indels in the protein alignments (translated from the previous nucleotide alignments) using in-house Perl scripts. The methodology was based on measuring the number of non-conserved positions due to either amino acid differences (point changes) or indels (structural changes) in the protein multiple alignments (e.g. the minimum indel length is one amino acid, corresponding to three nucleotides in the nucleotide sequence). We can estimate in this way the percentage of the protein which has been changed in our set of species. We think that this is a simple (yet somewhat rough) measure to estimate the degree of constraint relaxation of proteins.

Specifically, the number of amino acid differences was computed as the number of non-gapped positions with non-identical amino acids in the three species. All percentages are given in relation to the total number of aligned amino acids (non-gapped positions). Similarly, the number of indels was computed as the number of different indels (gaps affecting different positions) in the complete alignment (gapped and non-gapped sites). Therefore, an indel shared by two species was considered a single indel, while overlapping gaps were considered separately. Indel lengths were taken into account to calculate the percentage of the alignment affected by indels. In this case, all percentages are given in relation to the total length of the alignment (gapped and non-gapped positions).

We used one-way ANOVAs to test for differences between Hox, Hox-derived and non-Hox proteins in both parameters: the proportion of amino acid differences and the proportion of indels. We also used the Pearson correlation coefficient to test for a correlation between the two measures (e.g. to test whether proteins with a high proportion of amino acid differences also have a high proportion of indels), and a nested ANOVA 82 to test for differences in indel length among the three groups, taking into account the variation within groups.

In order to test the effect of the homeobox and the repetitive regions in our estimates of nucleotide substitutions, we repeated the previous analyses excluding one or both types of sequence. Repetitive regions were identified in three different ways. First, we searched in the UniProt Knowledgebase Release 8.6 (Swiss-Prot Release 50.6 + TrEMBL Release 33.6) 86 for annotated compositionally biased regions (defined in the feature table as COMPBIAS) in the protein sequences encoded by Hox, Hox-derived and non-Hox genes [see Additional file 3]. In the case of Hox genes, all three genes in the group contained at least one annotated repetitive region, while for Hox-derived and non-Hox genes only one entry of each group (bcd and CG2520, respectively) contained annotated repetitive regions. Note that only repeats in D. melanogaster are identified by using this methodology. Second, we identified simple repeats as those runs of 5 or more identical amino acids (e.g. QQQQQ), or at least 4 identical repetitions of 2 or more amino acids (e.g. GVGVGVGV), in any of the three species. By using this second approach, we extended the number of proteins with repetitive sequences in both the Hox-derived and non-Hox groups. Finally, we tried to visually annotate complex repeats as those imperfect runs of amino acid repetitions or compositionally biased regions in the protein (e.g. regions in the protein with a high content of Q, S, A, P, H, G, V, etc.). Data was analyzed using a combination of the three approaches as follows: (1) using UniProt only; (2) using UniProt + Simple repeats; and (3) using UniProt + Simple repeats + Complex repeats. Because the identification of complex repeats is somewhat subjective, we present in the main text the results obtained by identifying repeats using the second combination (UniProt + Simple repeats). However, results do not differ significantly among the three combinations [see Additional file 4].

We also calculated the proportion of amino acid differences and indels in repetitive and non-repetitive (unique) sequence in the three groups, and tested for differences between these two types of regions using a T-test for paired samples 82 on those proteins having both types of regions.