As a first step in determining when neural crest genes evolved, we filtered mouse proteins through a sequential blast pipeline. All 23,658 known mouse protein sequences (EnsEMBL v31) were consecutively blasted against available genomes grouped into seven different evolutionary categories (prokaryota, eukaryota, metazoa, deuterostomia, chordata, vertebrata, and mammalia) using a relaxed threshold of E = 10^-4, as established in similar studies 2627. Proteins exhibiting homology when blasted against the prokaryotic genomes were classified as ancient. The remaining genes were subsequently blasted against eukaryotic genomes and the procedure was repeated until all genes were classified (Figure 1a). According to our definition, 'new genes' in each category are those encoding proteins that did not exhibit any significant homology in previous categories, either because they have diverged extensively from a former protein or because they have evolved de novo.

A direct comparison of the percentage of genes appearing in each category with an estimation of their respective age in millions of years 28 indicated that the frequency of gene emergence is higher for late categories (specifically, metazoans to mammals; Figure 1b,c). This higher frequency of innovation correlates with the reported observation that the rate of evolution for proteins (calculated as the ratio between nonsynonymous and synonymous amino acid substitutions) is also higher for more recent categories 26.

To elucidate whether 'new proteins', because of their divergent amino acid sequences, correlate with the emergence of novel molecular functions, we performed a GO analysis 29. For each evolutionary category we identified the GO terms that are statistically over-represented compared with all of the known mouse proteins. The 10 most significantly over-represented GO terms for each of the seven different categories are listed in Table 1 (also see Additional data file 1 for a full list of over-represented GO terms). Our analysis shows that, within a large evolutionary window, innovations are associated with the emergence of 'new genes'. Although the first category, prokaryota, is enriched in genes that are involved in general cell metabolism, GO terms of genes appearing first in eukaryotes demonstrate their function in the newly evolved subcellular organelles. In metazoans we find the GO terms 'cell communication', 'signal transduction', and 'receptor activity' to be highly over-represented, which is in accordance with a de novo requirement for cell-cell communication and tissue subspecialization in the context of multicellularity. Interestingly, the collection of genes appearing first in vertebrates and mammals is enriched in terms such as 'hormone activity', 'receptor binding', 'extracellular space', and 'cytokine response', suggesting that diversification of receptor ligands is linked to vertebrate evolution. In summary, our sequential blast pipeline reliably classifies genes according to their first appearance within the phylogenetic tree.

In order to investigate when neural crest genes arose during evolution, it was necessary to build a comprehensive list of genes involved in the development of this tissue. A large number of studies, in particular the phenotypic analysis of mutations in mice, generated by either mutagenesis or genetic engineering, have led to the identification of many genes that are involved in neural crest development 7. The Mammalian Phenotype Browser, at MGI 23, provides a comprehensive resource of phenotypic information derived from mouse mutant studies 22. Because phenotypic analysis annotations offer the most reliable read out of gene function, we took advantage of this large collection of mouse mutants in our study. The collection includes more than 14,000 genotype records associated with a total of 6,442 genes (27% of the total mouse transcriptome), and furthermore it includes the majority of the genes demonstrated to play a bona fide role in neural crest development. In the MGI database each mutation is annotated by a controlled vocabulary of phenotypic terms that describe the effect of a genetic variation on different tissues, organs, or systems. We selected the Mammalian Phenotype Ontology for terms associated with mutations affecting both neural crest precursors and its derivative cell types and tissues.

At the Mammalian Phenotype Browser the ontology term 'abnormal neural crest cells' (MP:0002949:) is reserved for phenotypes that affect the early migration of neural crest cells. Because of this stringent definition, only eight genes are included in this definition. However, when we took phenotypes associated with the development of neural crest derivatives into account, we retrieved a comprehensive list of 615 genes. In our analysis we considered three main groups of neural crest derivatives: pigmented cells, skeletal components, and elements of the peripheral nervous system. The 'pigmentation derivatives phenotype' is completely covered by a single term, namely 'pigmentation phenotype' (MP:0001186). The 'bone derivatives phenotype' terms consist of 'craniofacial phenotype' (MP:0005382) and 'skeleton phenotype' (MP:0005390). At this point, it could be argued that vertebrate neural crest cells only give rise to cranial skeleton and teeth, whereas the axial skeleton has a mesodermal origin. As already mentioned, however, paleontologic records indicate that skeletal elements evolved within the context of the neural crest and only later was this genetic program co-opted by the sclerotome 17. The 'peripheral nervous system derivatives phenotype' consists of 'abnormal autonomic nervous system morphology' (MP:0002751), 'abnormal peripheral nervous system glia' (MP:0001105), 'abnormal somatic sensory system morphology' (MP:0000959), and 'peripheral nervous system degeneration' (MP:0000958). We grouped these three categories under the general term 'neural crest derivatives phenotype'.

The sequential blast pipeline provides a list of genes that emerge along the evolutionary tree in each of the seven defined categories, whereas the phenotypic annotation provides a functional link for each of these genes. Combining both, we determined in which category each of the 615 neural crest genes emerged (see Additional data file 2 for the full dataset). Previous studies had promoted the idea that gene co-option was the driving force for neural crest invention 6. Our data strongly support this view because the majority (91%) of genes involved in neural crest development was already present in basal metazoans or even before. Thus, key transcription factors acting as both 'neural plate border specifiers' (such as Pax3, Dlx5, Zic, and Msx1/2) and 'neural crest specifiers' (such as FoxD, Snail/Slug, Sox9/10, Twist, and AP-2) can be traced back to our category 'metazoans' or 'eukaryotes'. Similarly, the Fgf, Wnt, and Bmp signaling pathways involved in induction of the neural plate border are ancestral. Although their corresponding ligands can be traced back to basal metazoans, the kinase activity of their receptors was already present in prokaryotes. Altogether, these data confirm the idea that gene recruitment played an important role during neural crest evolution.

However, we found that a substantial percentage of the genes (9%, listed in Table 2) involved in neural crest development evolved in deuterostomes during the past 550 million years. To determine, within this evolutionary window, how the rate of gene emergence in the neural crest relates to the rate of innovation in other tissues, we plotted the cumulative number of genes appearing in each category. In these graphs, the tissue-specific evolutionary profile of gene emergence is depicted (Figure 2). In order to quantify the profile of the graphs we calculated 'gene emergence rate' (ger) values, as a numeric representation of the gene innovation rate from an earlier category to a later one (see Materials and methods for a description of the formula). A ger value of 1 indicates a constant profile of gene innovation. Higher ger values indicate increased appearance of new genes in a particular tissue.

For each of the tissue-specific gene programs studied, we ordered the ger values at the chordate-vertebrate transition (Figure 2a). Notably, tissues/systems ontogenetically derived from ventral mesoderm, and hence considered modern vertebrate innovations 2173031, such as the hematopoietic, immune, or renal/urinary system, exhibit graphs that peak at the chordate-vertebrate transition (Figure 2b). In contrast, other tissues already present in all chordates, namely the epidermis or endodermal derivatives such as liver, respiratory, and digestive systems, have a flat profile, with lower ger values (Figure 2b). Both the profile of the neural crest gene emergence plot (Figure 3) and its ger value (3.1) indicate that the neural crest is among the most innovative vertebrate tissues (Figure 2a). This concept can be extended to each individual neural crest lineage, in particular to pigmented or bone derivatives, as deduced from their respective gene emergence plots (Figure 3). Interestingly, compared with the other crest derivatives, the ger value of the gene set associated with the peripheral nervous system derivatives is lower (1.6). This may best be explained by co-option from the ancestral program of neural development. In summary, our gene emergence plots that reliably reflect evolutionary innovation highlight the novelty of neural crest as a tissue.

The notion of neural crest as a tissue with a high rate of gene innovation apparently contradicts our finding that all known neural crest specifiers can be traced back at least to metazoans. To further address this point, we focused on the collection of neural crest 'new genes' to gain insight into their molecular nature and function.

Neural crest has been postulated as a fourth germ layer 32. This concept builds on neural crest pluripotency and the fact that in vertebrates it gives rise to novel cell types such as the skeletal derivatives or the specialized melanocytes 11. Consistently, in the collection of vertebrate/mammalian new genes, we found molecules defining the physiology of these novel cell types. This is the case for the genes Ru (Hermansky-Pudlak syndrome 6) and silver, which encode components of the specialized melanocyte lysosomes, the melanosomes. Similarly, several new genes encode extracellular proteins that constitute part of the bone matrix (for example, bone gla protein and the phosphoglycoprotein mepe) and enamel, the outermost covering of teeth and the hardest tissue in the body (for example, ameloblastin and amelogenin).

Strikingly, 50% of neural crest genes appearing first in vertebrates encode extracellular ligands. This remarkable enrichment (confirmed by exploring GO term frequency; see Additional data file 3) is in accordance with our previous whole-transcriptome GO analysis (Table 1). It suggests that diversification of receptor ligands played an important role during vertebrate evolution in general and neural crest evolution in particular. Individual analysis of the function of these peptides during the development of the neural crest demonstrates that they control the commitment of precursors to the different lineages.

Conserved signaling pathways have an early influence on the phenotypic diversification of premigratory neural crest cells 13. Bmp2/4 can directly induce autonomic neurogenesis 3334, while Wnt signaling participates in melanocyte specification 35. Superimposed on this, a second network of 'modern' vertebrate specific cytokines, produced locally, acts not only in neural crest cell fate specification but also in the migratory behavior and survival of all neural crest lineages 12. Melanocyte specification and survival depend on soluble proteins such as steel factor (kit ligand), endothelin-3, α-melanocyte stimulating hormone, and nonagouti 36; gliogenesis in the peripheral nervous system is controlled by neuregulins and endothelin-3 3738; the development of autonomic and sensory neurons is controlled by neurothropins (brain-derived neurotropic factor, neurothropin-3, and neurothropin-4) and GDNF family members (GDNF and neurturin) 3940; and, finally, the differentiation of the skeletal lineage is specified by endothelin-1 41. Our sequential blast pipeline analysis shows that the vast majority (9/11) of the above-mentioned cell fate specification ligands emerged in vertebrates or, to a lesser extent (steel factor and nonagouti), in mammals.

Interestingly, the blast pipeline uncovered a positive hit in the echinoderm Strongylocentrotus purpuratus genome for the neurotropin family members brain-derived neurotropic factor and neurothropin-3. Because it has been proposed that neurotropins constitute a vertebrate innovation 42, we performed a ClustalX alignment 43 of mouse neurotropins against the echinoderm sequence ( Additional data file 4). This revealed that the particular array of cysteines conserved in all neurotropins, the so-called 'cysteine knot' 44, is also present in the echinoderm sequence and therefore identifies it as a putative growth factor. However, the limited amino acid identity (33%) and the lack of conservation in critical residues required for neurotropin binding to Trk receptors indicate that the echinoderm neurotropin-related protein cannot be considered a bona fide neurotropin. This suggests that neurotropins evolved from divergent ligands present in ancestral chordates. In fact, the example of neurotropins may be just part of a more general mechanism because other 'new cytokines' can be related to pre-existing growth factors. Supporting this view, GDNF and neurturin are divergent members of the TGF-β superfamily of ligands, as indicated by their particular cysteine knot and hence folding 44. Similarly, despite their limited homology, neuregulins belong to the epidermal growth factor superfamily of ligands 45.

Taken together, our data show that the cytokine network acting in neural crest cell fate specification is mainly a vertebrate innovation (Figure 4). Furthermore, these analyses indicate that an important proportion of the 'new ligands' are derived from fast evolving growth factors.

The comparative analysis of gene emergence plots highlights a high rate of gene innovation for the neural crest during vertebrate evolution. In fact, there are reasons to believe that our estimation on the rate of gene emergence may be conservative. In the sequential blast pipeline analysis, the presence of an ancestral conserved domain will shadow the appearance of evolutionarily more recent domains within the same molecule. This may be particularly relevant in the case of large multidomain proteins such as receptors.

To overcome this constraint and to complement our studies, we conducted a phylogenetic analysis of the Pfam motifs (defined by multiple alignment of proteins 46) occurring in the collection of 615 neural crest genes. From a total of 8,183 Pfam domains annotated in EnsEMBL, 499 are present in the set of 615 neural crest genes. We screened for these motifs in the seven different categories, detecting homology through two different approaches: blasting Pfam consensus sequences (threshold of E = 10^-4) and searching for hidden Markov models (HMMs) using HMMER software with standard parameters 46. We compiled a table including all neural crest genes with their Pfam domains and when they occur first in the defined seven temporal classes, as detected using either of the methods (Additional data file 5). A list including only those genes that contain a Pfam domain emerging in vertebrates is compiled in Table 3. Pfam domain detection supports and refines our sequential blast pipeline results. Thus, GDNF and neurturin were identified as divergent members of the TGF-β superfamily, and the kit-ligand and nonagouti domains were detected as vertebrate novelties (previously detected as mammalian innovations; Table 2). Furthermore, the analysis also confirmed the ClustalX alignments demonstrating that the neurotropin domain (nerve growth factor; Table 3) is indeed a vertebrate innovation. In summary, our domain-based approach (more sensitive and accurate, but limited to annotated Pfam domains) complements the sequential blast analysis (Table 2), providing independent confirmation of the emergence in vertebrates of growth factors that are involved in the specification/survival of the neural crest cells (Table 3).