For over a century, developmental biologists have noted an ontogenic pattern among evolutionary relationships: earlier developmental stages are morphologically more similar across species than later stages; this is also known as Von Baer's third law [1–4]. While more recent studies in vertebrates have determined that the very earliest stages of ontogeny (for example, gastrulation) may be subject to substantial variation even among closely related species, upon reaching the tailbud stage, embryos begin to share more similarity in appearance that gradually declines with subsequent development . This 'hourglass' model of developmental similarity among vertebrates suggests that, while certain stages of development undergo substantial change over evolutionary time, there exists a significant conservation of the mechanisms underlying development across vertebrates [6–9]. Darwin originally interpreted Von Baer's observations via a selectionist framework [10, 11]. He suggested that divergence should be greatest during ontogenic stages in which organisms experienced the most varying 'conditions of existence' and, as a result, occasioned opportunity for differential selection . Embryos of varied species are therefore more similar than adults due to exposure to very similar fetal environments. Furthermore, he noted that derived features rarely appeared in an organism before the stage when they were used, indicating that the effect of selection was also specific to the stage where selection pressure actually occurred. This observation was important to his overall hypothesis, as selection pressures occurring during one stage that selected for traits expressed in other stages would be inconsistent with Von Baer's observations. Using secondary sexual traits as a primary example, Darwin compiled a large number of observations indicating that male-specific structures known to be highly divergent even among closely related species rarely developed until reproductive maturity was reached [10, 12].
Modern interpretations of Von Baer's third law have focused on another, non-mutually exclusive mechanism: genes implicated in early aspects of development are more likely to regulate a large number of downstream effectors via hierarchical regulatory cascades, and are thus more evolutionarily constrained due to the large deleterious pleiotropic effects of mutations. This is known as the developmental constraint hypothesis [3, 13, 14]. The complex hierarchical nature of gene regulatory networks has become a focus of major interest in the field of organismal development [15, 16] with special attention being paid in particular to those network modules critical to early development and conserved over broad evolutionary distances . For instance, the well known homeotic genes involved in establishing the anterior/posterior axis in the early development of most metazoans provide a striking example of highly conserved genes whose mutations are known to have extensive pleiotropic consequences [18–20]. These transcription factors are also known to act as master regulatory switches in cascades involved in regulating the proper expression of many downstream, developmentally important effectors . Another example is the gene regulatory feedback loop required for endoderm specification in echinoderms, which encodes several transcription factors whose inactivation has catastrophic effects on the entire body plan [17, 22]. These instances highlight the strength of purifying selection acting on specific genes known to be involved in complex developmental regulatory networks; however a more recent interest has concerned the broader evolutionary patterns of the genome with respect to ontogeny.
The evolutionary dynamics of genes expressed over the course of development have recently been examined at the genomic level in the case of flies and nematodes, using microarray-based information about the developmental timing of gene expression [23–25]. Castillo-Davis and Hartl  used previously published, developmental stage-specific microarray data  in order to compare the rates of coding sequence divergence of a relatively small number of genes (224) betweenCaenorhabditis elegans andC. briggsae (20 million to 120 million years diverged (MYD)). Genes in their dataset were classified either as 'non-modulated' genes (that is, invariant in expression level over development), early-expressed genes (that is, embryonic), or late-expressed genes (that is, larval and adult) based on the developmental stage at which their peak level of expression occurred. The authors found no significant difference in the rates of protein evolution among the three categories, though the early-expressed genes showed a higher rate of synonymous substitution as well as a lower codon usage bias (CUB) than late-expressed genes. The analysis of the same 2 species was subsequently refined by Cutter and Ward  using a larger dataset of 7,281 genes and a larger source of developmental expression data [27, 28]. Their results support some theoretical predictions of both the developmental constraint as well as Darwin's 'selection opportunity' hypothesis: when genes were classified based on the stage at which their peak expression level occurred, adult genes were found to be evolving more rapidly than those in the earlier, larval stage. Expression level in the larval stage, relative to the adult, was also found to be negatively correlated with sequence divergence, while the opposite was observed for expression in adults. However, the authors noted no unidirectional trend in evolutionary rates in genes expressed over the course of embryogenesis, as would be predicted by the developmental constraint hypothesis, leading them to suggest that constraint may not explain the evolutionary rates of proteins expressed during embryonic development in these species. Furthermore, when examining the tissue specificity of genes expressed in adult nematodes, the authors found that the majority, though not all, of the acceleration in evolutionary rate observed in this stage was explained by genes expressed primarily in the male gametes, providing evidence of a significant effect of sexual selection, presumably acting through sperm competition between males and hermaphrodites or antagonistic coevolution between genes expressed in sperm and oocytes .
Daviset al.  used the results of a microarray study of the expression levels of 4,028 genes over the course ofDrosophila melanogaster ontogeny  and examined their rates of sequence divergence betweenD. melanogaster andD. pseudoobscura (25 to 55 MYD). They noted that gene expression level in the late embryo relative to later stages was negatively correlated with sequence divergence, while the opposite was observed in the case of adult males. However, the authors noted no significant correlation between expression levels and sequence divergence for the many of the sampled developmental stages. Unfortunately the species pairs used in both of these studies were quite distantly diverged and thus interpretation of these data is limited due to the saturation of synonymous site divergence (d
S), which largely prevents investigation of questions regarding evidence of selection [30, 31]. Furthermore, comparisons at such evolutionary distances allow the possibility that expression patterns (for example, time of expression, sex bias, and so on) have diverged between species, questioning whether similar selective pressures are acting along both lineages at the level of individual genes .
Holometabolic insects such asDrosophila provide an excellent model for studying gene evolution over ontogeny as they pass through four separate, unambiguous developmental stages (embryo, larva, pupa, and adult). A large body of information about the evolutionary dynamics of the genomes of drosophilids has accumulated, aided significantly by the recent release and analysis of the complete genomes of 12Drosophila species . However, the relationship between development and genomic evolution remains largely unexplored. Here, we analyze a larger dataset than was previously available, using information generated from publicly available developmental stage-specific expressed sequence tag (EST) libraries to assign genes to specific developmental stages and determine their evolutionary patterns within theD. melanogaster group, allowing more reliable estimates of divergence parameters as well as reducing the caveats associated with comparing distantly related species . We report a gradient of increasing mean evolutionary rate in genes expressed in subsequent stages of fly development, culminating in exaggerated gene sequence divergence specifically in adult males. When comparing genes expressed specifically in the gonads of embryos to adults, we found that the increased rate of divergence observed in adults is explained entirely by those genes expressed in the testis. No such pattern of accelerated gene divergence is observed in the embryonic gonads, supporting Darwin's expectations that selection pressures should act predominantly in the stage where the opportunity for selection occurs . Finally, when classifying genes into specific developmental stages using a series of increasing stage-specificity thresholds, we found a significant correlation between specificity of temporal stage of expression and evolutionary rate. We also reanalyzed the dataset used by Daviset al.  using our methods in order to refine their estimates of divergence and test the generality of their results (Additional files 1, 2, and 3). Taken together, our results support both developmental constraint acting to limit the divergence of early expressed, developmentally important genes [5, 8], as well as the notion that accelerated divergence in adults is primarily due to increased selection pressures occurring during this stage.