Fibroblast Growth Factors (FGFs) are polypeptides with diverse biological activities in multiple developmental and metabolic processes. The human/mouse Fgf gene family comprises 22 members. Most FGFs mediate their biological responses as extracellular proteins by binding to and activating cell surface tyrosine kinase FGF receptors (FGFRs) (Ornitz and Itoh,2001; Itoh and Ornitz,2004; Thisse and Thisse,2005). However, FGFs 11–14 function as intracellular proteins, hereafter referred to as iFGFs, and act in an FGF Receptor (FGFR)-independent manner (Goldfarb,2005).
The genes encoding FGFs have been identified in multicellular organisms ranging from Caenorhabditis elegans to humans (Itoh and Ornitz,2004). Two Fgf genes were found in C. elegans, whereas 22 Fgf genes have been identified in humans and mice, indicating that the Fgf gene family greatly expanded during the evolution of primitive metazoa to vertebrates. The Fgf family expanded in two phases (Itoh and Ornitz,2004; Popovici et al.,2005). In the first phase, during early metazoan evolution, Fgfs expanded from two or three to six genes by gene duplications. In the second phase, during the evolution of early vertebrates, the Fgf family expanded via two large-scale genome duplications. Phylogenetic and gene location analyses of the human Fgf family indicate that it comprises several subfamilies. Ancestors of the subfamilies were generated in the first phase. In the second phase, the subfamilies expanded to contain three or four members. However, their detailed expansion history remains mostly unclear.
According to the “exon theory of genes,” introns are expected to be ancient relics of primordial genes (Rogozin et al.,2005; de Roos,2007; Csuros et al.,2007), indicating that exon/intron organizations suggest the potential evolutionary history of a gene family. In addition, as ontogeny is expected to correlate with phylogeny (Seitz et al.,2006; Madsen,2007), functional roles of genes in development also provide useful clues for elucidating the evolutionary history of a gene family. As exon/intron organizations of vertebrate Fgf genes and their roles in development have been most studied in mice, we have examined the potential evolutionary history of the mouse Fgf gene family based on their gene organization and function in development along with phylogenetic and gene location analyses. In this review, we propose a model for the evolutionary history of the Fgf family, using the mouse Fgf family as a model.
IDENTIFICATION OF THE MOUSE FgF GENE FAMILY
Two FGFs, Acidic FGF (FGF1) and Basic FGF (FGF2), were originally isolated from the brain and pituitary gland as growth factors for fibroblasts (Gospodarowicz,1974; Gospodarowicz and Morgan,1974; Gospodarowicz et al.,1975). Thereafter, seven Fgfs, Fgf3–Fgf9, were also identified as oncogenes or isolated as growth factors for cultured cells (Dickson et al.,1991; Yoshida et al.,1991; Goldfarb et al.,1991; Coulier et al.,1991; Aaronson et al.,1991; Tanaka et al.,1992; Miyamoto et al.,1993). FGFs 1–9 range in size from ∼150 to 260 amino acid residues and have a conserved ∼120-amino acid residue core with ∼30 to 70% sequence identity (Ornitz and Itoh,2001; Itoh and Ornitz,2004). Based on their conserved amino acid sequences, Fgfs 10–14 and 16–23 were isolated by conducting a homology-based polymerase chain reaction or identified by homology-based searches in nucleotide sequence databases (Yamasaki et al.,1996; Smallwood et al.,1996; Miyake et al.,1998; Hoshikawa et al.,1998; Ohbayashi et al.,1998; Nishimura et al.,1999,2000; Ohmachi et al.,2000; Nakatake et al.,2001; Yamashita et al.,2000). In addition, Fgf15 was identified as a downstream target of the chimeric homeodomain oncoprotein E2A-Pbx (McWhirter et al.,1997). Fgfs 1–23 have been identified in humans and mice. However, FGF19 is a human ortholog of mouse Fgf15. Thus, the mouse/human Fgf gene family comprises 22 members (Ornitz and Itoh,2001; Itoh and Ornitz,2004). FGFs 10–23 range in size from ∼160 to 250 amino acid residues and have a conserved ∼120-amino acid residue core (Ornitz and Itoh,2001; Itoh and Ornitz,2004).
Except for the iFGFs (FGF11–14) and hFGFs (FGF15/19, 21, 23), all other FGFs activate FGFRs with high affinity and different degrees of specificity (Ornitz et al.,1996; Zhang et al.,2006). We refer to these FGFs as canonical FGFs. Canonical FGFs are secreted (extracellular proteins) that bind to FGFRs and induce their dimerization and the phosphorylation of specific cytoplasmic tyrosine residues (Eswarakumar et al.,2005). Four Fgfr genes, Fgfr1–Fgfr4, have been identified in humans and mice. These genes encode receptor tyrosine kinases (∼800 amino acids) that contain an extracellular ligand-binding domain with three immunoglobulin domains (I, II, and III), a transmembrane domain, and a split intracellular tyrosine kinase domain. Fgfr1–Fgfr3 encode two different versions of immunoglobin-like domain III (IIIb and IIIc) generated by alternative mRNA splicing that utilizes one of two unique exons. The immunoglobulin-like domain III is an essential determinant of ligand-binding specificity (Johnson and Williams,1993). Thus, seven FGFR proteins (FGFRs 1b, 1c, 2b, 2c, 3b, 3c, and 4) differing in ligand-binding specificity are generated from four Fgfr genes in vertebrates (Ornitz et al.,1996; Zhang et al.,2006). In contrast, although iFGFs bear strong sequence similarity to canonical FGFs, their biochemical and functional properties are largely unrelated to those of canonical FGFs. Current data suggest that iFGFs act in an FGFR-independent manner. They have been shown to interact with intracellular domains of voltage-gate sodium channels and with the neuronal mitogen-activated protein kinase scaffold protein, islet-brain-2 (reviewed in Goldfarb,2005).
Canonical FGFs and the hFGF subfamily can be further characterized based on the mechanism by which they are released from cells. FGFs 3–8, 10, 15, 17, 18, 21, 22, and 23 are secreted proteins with cleavable amino terminal signal peptides (Ornitz and Itoh,2001). FGFs 9, 16, and 20 are also secreted proteins, but contain uncleavable bipartite signal sequences (Miyakawa et al,1999; Revest et al.,2000). By contrast, FGFs 1 and 2 do not have identifiable signal sequences but nevertheless can be found in extracellular locations. FGFs 1 and 2 might be released from damaged cells or by an exocytotic mechanism that is independent of the endoplasmic reticulum-Golgi pathway (Mignatti et al.,1992). It has also been reported that exogenously added FGF2 can be translocated to the nucleus where it can interact with and activate nuclear targets (Bonnet et al.,1996; Bailly et al.,2000). However, the physiological significance of FGF2 action in the nucleus remains unclear.
The canonical FGFs all have binding sites for acidic glycosaminoglycans including heparin and heparan sulfate (Ornitz,2000). In the presence of heparan sulfate, FGFs stably bind to FGFRs, which lead to the formation of 2:2:2 FGF-FGFR-heparan sulfate dimers (Mohammadi et al.,2005). In addition, acidic glycosaminoglycans in the form of heparan sulfate proteoglycans function to retain these FGFs in the vicinity of FGF-producing sites, such that they primarily act in a paracrine manner. By contrast, the hFGFs have low-affinity heparin-binding sites and have been found to act in an endocrine manner (Tomlinson et al.,2002; Lundasen et al.,2006; Kharitonenkov et al.,2005,2007; Fukumoto and Yamashita,2007; Liu and Quarles,2007).
PHYOLOGENETIC AND GENE LOCATION ANALYSES OF THE MOUSE FgF GENE FAMILY
Phylogenetic analysis of the mouse Fgf gene family identifies seven subfamilies: Fgf1 (1,2), Fgf4 (4,5,6), Fgf7 (7,10,22), Fgf8 (8,17,18), Fgf9 (9,16,20), iFgfs (11,12,13,14), and hFgfs (15, 21, 23) (Fig. 1) that are essentially consistent with those of the human FGF family (Itoh and Ornitz,2004). However, phylogenetic analysis alone is not sufficient to determine all evolutionary relationships (Horton et al.,2003). Analysis of gene loci on chromosomes also indicates potential evolutionary relationships within a gene family. We have examined mouse Fgf gene loci and conserved chromosomal gene location (synteny). In contrast to the phylogenetic analysis, the conserved gene location analysis indicates that the mouse Fgf gene family may more meaningfully be divided into six subfamilies: Fgf1/2/5, Fgf3/4/6/15/21/23, Fgf7/10/22, Fgf8/17/18, Fgf9/16/20, and iFgfs (Fgf11/12/13/14) (Fig. 2). The mouse Fgf subfamilies derived from the gene location analysis are also essentially consistent with the human Fgf subfamilies (Itoh and Ornitz,2004). Members of the Fgf7, Fgf8, Fgf9, and iFgf subfamilies from the gene location analysis are consistent with those of the Fgf7/10/22, Fgf8/17/18, Fgf9/16/20, and iFgf subfamilies from the phylogenetic analysis. Fgf5, a member of the Fgf4 subfamily in the phylogenetic analysis (Fig. 1), is however more closely linked to Fgf2 by gene location analysis. Both Fgf5 and Fgf2 are closely linked to Annexin A3 (Anxa3) and Annexin 5 (Anxa5), respectively (Fig. 2). This indicates that Fgf1, Fgf2, and Fgf5 are members of a common subfamily, FGF1/2/5.
Conserved gene location is observed among Fgf4, Fgf6, Fgf15, Fgf21, and Fgf23, indicating that these Fgfs belong to the same subfamily (Fig. 2). In addition, although Fgf3 is a member of the Fgf7 subfamily according to the phylogenetic and functional analysis, the gene location analysis indicates that Fgf3 is linked to Fgf4 and Fgf6, indicating that these are members of the same subfamily. However, the phylogenetic analysis and consideration of the biochemical and functional properties described above indicate that the hFgfs (Fgf15, Fgf21, and Fgf23) are distinct from Fgf3, Fgf4, and Fgf6. Therefore, these Fgfs should be divided into the Fgf3/4/6 and hFgf subfamilies. In summary, we propose that the secreted Fgfs should be divided into six subfamilies with three members in each; the canonical Fgfs (Fgf1/2/5, Fgf3/4/6, Fgf7/10/22, Fgf8/17/18, Fgf9/16/20) and the hFgfs (Fgf/15/21/23) (Fig. 2).
EXON/INTRON ORGANIZATIONS OF MOUSE FgF GENES
It is proposed that introns are relics of primordial genes (Rogozin et al.,2005; de Roos,2007; Csuros et al.,2007). This leads to the hypothesis that exon/intron organization can be used to infer the evolutionary history of a gene family. To this end, we have examined the exon/intron organizations of mouse Fgf genes. The conserved FGF core domain in the coding region of Fgf4 is divided by two introns (Fig. 3). The intron locations are highly conserved in the core regions of all canonical Fgfs and the hFgfs (data not shown). These two introns are also conserved in the core regions of the iFgf subfamily, indicating that iFgfs, hFgfs, and canonical Fgfs are evolutionarily related to each other (data not shown). However, two additional introns are also located outside of the core region of Fgf13 within sequences encoding the amino and carboxy terminal domains of the protein (Fig. 3). The positions of these four introns are also highly conserved among the other iFGFs (data not shown).
Several Fgfs have acquired alternatively spliced amino terminal ends. Alternative spliced variants of Fgf8 and Fgf17 are generated from distinct splice sites that divide the “first” exon into four (in humans) sub-exons (MacArthur et al.,1995; Gemel et al.,1996; Xu et al.,1999; Olsen et al.,2006). Similarly, the iFgfs have acquired alternatively spliced amino terminal variants that are generated by alternative utilization of distinct first exons (reviewed in Goldfarb,2005). Importantly, none of these alternative splicing events affects the sequences of the conserved FGF core domain.
PHENOTYPES OF FgF KNOCKOUT MICE
Biological function of mouse Fgfs in development and physiology also provide useful clues to the evolutionary history of the Fgf gene family. Most Fgf genes have been disrupted by homologous recombination in mice. Phenotypes range from early embryonic lethality to subtle changes in adult physiology (Table 1 and Fig. 4). Fgf4 and Fgf8 knockout mice die at early embryonic stages. Fgf4 and Fgf8 have essential roles in blastocyst formation and gastrulation, respectively (Feldman et al.,1995; Sun et al.,1999). Fgf9, Fgf10, and Fgf18 knockout mice died shortly after birth. Fgf10 is critical for epithelial–mesenchymal interactions necessary for the development of epithelial components of multiple organs (Min et al.,1998; Sekine et al.,1999; Ohuchi et al.,2000; Sakaue et al.,2002). Fgf9 and Fgf18 have essential roles in the development of mesenchymal components of multiple organs (Colvin et al.,2001a,2001b, Ohbayashi et al.,2002; Liu et al.,2002; Usui et al.,2004). In contrast, Fgf15 knockout mice die at variable times during embryonic and postnatal stages of development. FGF15 functions in the development of the cardiac outflow tract (Vincentz et al.,2005) but also acts as an endocrine hormone in postnatal life (Inagaki et al.,2005). Other Fgf knockout mice either are viable or die during postnatal stages (Table 1). Conditional knockouts of Fgfs that have essential roles early in development have also revealed important functions for these Fgfs at later times of development (Moon et al.,2000; Moon and Capecchi2000; Sun et al.,2000; Chi et al.,2003; Lewandoski et al.,2000).
Table 1. Phenotypes in Fgf Knockout Mice
Phenotypes of most Fgf knockout mice have been published (reviewed in Ornitz and Itoh,2001; Itoh and Ornitz,2004). Fgf16 and Fgf21 knockout mice are viable (N. Itoh et al., unpublished data). Fgf22 knockout mice are also viable (H. Umemori, personal communication). Phenotypes of Fgf11, Fgf13, and Fgf20 knockout mice remain unclear. E, embryonic day; PD, postnatal day; PW, postnatal week. Asterisks indicate potential prototypes of the Fgf subfamilies. —, indicates unknown phenotypes.
Inner ear, tail and CNS development
Subtle, muscle regeneration
Cardiovascular, skeletal, and neuronal development
Skeletal and lung development
Lung, heart, vascular, GI tract, and testis development
Subtle, muscle regeneration
Multiple organ development
Heart development and bile acid metabolism
Phosphate and vitamin D metabolism
EVOLUTIONARY HISTORY OF THE MOUSE FgF GENE FAMILY
The coding regions of the iFgfs are divided by four introns, while the coding regions of canonical Fgfs and hFgfs are divided by two introns. Consistent with the idea that introns are relics of primordial genes (Rogozin et al.,2005; de Roos,2007), these organizations indicate that the iFgf subfamily is the likely ancestor of the canonical Fgf families.
Numerous examples exist in which phylogenetic relationships appear to correspond with gene function at very early developmental stages (Seitz et al.,2006; Madsen,2007). This model suggests that FGF family members with very early functions in development may represent the prototype genes for their subfamily. By examining early developmental phenotypes of Fgf knockout mice as a guide, one can therefore make predictions of prototype members of Fgf subfamilies. The iFgf subfamily comprises Fgf11–Fgf14. Fgf12 and Fgf14 knockout mice are viable and have neurological phenotypes (Wang et al.,2002; Xiao et al.,2007; Goldfarb et al.,2007; Laezza et al.,2007). Although Fgf11 and Fgf13 knockout mice have not been reported, the expression of these genes in the central and peripheral nervous system suggests that these genes will also be found to function in neurodevelopment or neurophysiology. Experiments with Xenopus embryos indicate that Fgf13 is essential for neural differentiation in early embryonic development (Nishimoto and Nishida,2007). In addition, the C. intestinalis Fgf11/12/13/14, an ancestor of the vertebrate iFgf subfamily (Satou et al.,2002), is most homologous (57% amino acid identity in the core region) to Fgf13, although phylogenetic analysis between the mouse iFgf subfamily and the C. intestinalis Fgf11/12/13/14 has not clearly indicated their relationship (Fig. 1). Thus, Fgf13 appears most similar to an ancestral member of the iFgf subfamily.
All canonical Fgf subfamilies and the hFgfs comprise three members. Phenotypes of knockout mice for all Fgf3/4/6, Fgf7/10/22, Fgf8/17/18, and hFgf subfamily members have been reported. Knockout studies indicate that Fgf4, Fgf10, Fgf8, and Fgf15 have the most severe phenotypes at early developmental stages within their corresponding families and can thus be considered as the ancestral members of their subfamilies (Fig. 4). Of the members of the Fgf9/16/20 subfamily, Fgf9 is essential during embryonic stages, whereas Fgf16 knockout mice are viable (N. Itoh et al., unpublished data). Although phenotypes of Fgf20 knockout mice have not been reported, we tentatively have assigned an Fgf9-like gene as the ancestor of the Fgf9/16/20 subfamily. Knockout mice for the Fgf1/2/5 subfamily are all viable with relatively subtle phenotypes (Miller et al.,2000; Hebert et al.,1994). FGF1 and FGF2 lack signal peptides while FGF5 functions as a typical secreted protein. As the ancestral members of other canonical Fgf subfamilies are typical secreted proteins, we assign an Fgf5-like gene as the ancestor of the Fgf1/2/5 subfamily.
No conserved gene order was observed among ancestors of the Fgf subfamilies, indicating that the ancestors were generated by gene duplications, not by genome duplications. The ancestral gene of the Fgf family, which is tentatively assigned to Fgf13-like, encodes an intracellular protein without a secretory signal sequence, whereas the ancestral members of the other Fgf subfamilies encode secreted proteins with identifiable secretory signal sequences. In addition, although Fgf13 has four introns within its coding regions, all other ancestral Fgfs have only two introns. These results indicate that the ancestral members of the canonical Fgf subfamilies and hFgfs were derived from an Fgf13-like ancestral gene accompanied by loss of two introns. Of the canonical Fgfs, Fgf4 is required at the earliest stages of development, indicating that an Fgf4-like gene may be the ancestor of the canonical Fgf subfamilies (Fig. 4). From these results, we propose a model for the evolutionary history of the Fgf family (Figs. 5 and 6) with an Fgf13-like gene as the ancestral Fgf. An Fgf4-like gene was then generated from Fgf13-like by a gene duplication event followed by gene translocation. During this evolution, Fgf4-like lost two introns in the first and third exons of its coding region and acquired a cleavable amino terminal signal sequence in the first exon. Fgf5, Fgf8, Fgf9, and Fgf10 were then generated from Fgf4 by gene duplications followed by gene translocations. Cleavable secreted signal sequences and the intron positions were conserved in Fgf4, Fgf5, Fgf8, and Fgf10 expansions. The Fgf9/16/20 subfamily appears to have formed next with the evolution of an uncleaved bipartite signal sequence in an Fgf9-like ancestral gene.
All canonical FGFs have high-affinity heparin-binding sites and act in a paracrine manner. However, the hFGFs have low-affinity heparin-binding sites. We thus propose that an Fgf15-like gene was derived from Fgf4-like by a cis-gene duplication accompanied by loss of its high-affinity heparin-binding capacity, thus allowing it to function as an endocrine molecule. The hFGF family also acquired affinity for alternative co-factors, Klotho and βKlotho, required for signaling in target tissues (see below).
As most ancestors of the Fgf subfamilies have been identified in the ascidian, C. intestinalis, Fgf family expansion must have occurred before the evolution of vertebrates (Itoh and Ornitz,2004). Ancestors of the canonical Fgf subfamilies have distinct receptor-binding specificity (Zhang et al.,2006). However, the Fgfr family expanded during the evolution of early vertebrates (Itoh and Ornitz,2004), indicating that canonical FGFs acquired diversity in receptor specificity during the evolution of early vertebrates.
Conserved gene orders are observed among members of each Fgf subfamily, indicating that each subfamily expanded into three or four members via two large-scale genome duplications during the evolution of early vertebrates (Itoh and Ornitz,2004). Although each canonical FGF has distinct receptor binding specificity, each member of an FGF subfamily has similar receptor specificity (Zhang et al.,2006). The most notable example of this is the ability of FGF9, FGF16, and FGF20 to bind FGFR3b in addition to c splice forms of FGFRs 1–3. In contrast, other canonical FGF subfamilies bind either b or c splice forms of FGFRs but not both.
The human/mouse Fgf gene family comprises 22 members with high-sequence similarity (Itoh and Ornitz,2004). However, their roles and mechanisms of action are highly diverse. FGF11–FGF14 (iFGFs) function as intracellular proteins in an FGFR-independent manner (Wang et al,2000; Goldfarb,2005). In contrast, canonical FGFs function as extracellular proteins in an FGFR-dependent manner.
Canonical FGFs mediate their biological responses by binding to and activating FGFRs. Seven FGFR proteins (Fgfrs1b, 1c, 2b, 2c, 3b, 3c, and 4) with differing ligand-binding specificity are generated from four Fgfr genes in vertebrates (Ornitz et al.,1996; Zhang et al.,2006). Canonical FGFs have heparin-binding sites that bind to extracellular acidic glycosaminoglycans. Acidic glycosaminoglycans function as FGF cofactors to facilitate efficient activation of FGFRs (Ornitz,2000). In addition, acidic glycosaminoglycans limit the diffusion of FGFs, localizing their activity to the vicinity of FGF-producing cells (Flaumenhaft et al.,1990). In contrast, hFGFs have poor heparin-binding affinity and act on target cells far from their site of production in an endocrine manner (Tomlinson et al.,2002; Lundasen et al.,2006; Kharitonenkov et al.,2005,2007; Fukumoto et al.,2007; Liu et al.,2007). In addition, FGF15, FGF21, and FGF23 require co-receptors, Klotho or βKlotho, to activate FGFRs (Ogawa et al.,2007; Urakawa et al.,2006). These co-receptors are specifically expressed in target cells and are required for the specific actions of these FGFs in target tissues. These findings indicate that hFGFs have evolved as novel metabolic regulators along with corresponding novel mechanisms of regulation.
Loss-of-function studies in mice have identified many other functions of FGFs. Most members of the Fgf family have been targeted in mice. Phenotypes range from early embryonic lethality to subtle changes in the physiology of adult mice, to no identifiable phenotype (Fgf1). Most canonical FGFs have essential roles as proliferation or differentiation factors in the development of various organs and tissues. In contrast, hFGFs have important roles as hormones that regulate bile acid metabolism, phosphate and vitamin D metabolism, and energy metabolism at postnatal stages (Shimada et al.,2004; Inagaki et al.,2005; Inagaki et al.,2007; Badman et al.,2007). Several of the Fgfs that exhibit early embryonic lethality as null mutations have also been studied later in development by conditional gene targeting and have been found to have many additional activities. Additionally, redundancy between FGFs, both within subfamilies and across subfamilies, has been identified. For example, redundancy between Fgf4 and Fgf8 in limb bud development was revealed by conditional, limb bud–specific, targeting of Fgf4 and Fgf8 (Moon et al.,2000; Moon and Capecchi,2000; Sun et al.,2000,2002; Lewandoski et al.,2000; Boulet et al.,2004). Redundancy between Fgf3 and Fgf8 occurs in otic vesicle and hindbrain development (Maroon et al.,2002; Walshe et al.,2002).
The studies of human diseases also have identified many potential functions of FGFs. Human hereditary disorders in FGF signaling result in diverse diseases consistent with the wide range of action of FGFs. Autosomal dominant hypophosphataemic rickets are caused by mutations in FGF23 that stabilize the FGF23 protein (ADHR Consortium,2000; White et al.,2001). FGF23 also functions as a humoral phosphaturic factor responsible for tumor-induced osteomalacia (Shimada et al.,2001). Both aplasia of lacrimal and salivary glands and lacrimo-auriculo-dento-digital syndrome are caused by mutations in FGF10 (Entesarian et al.,2005,2007; Milunsky et al.,2006; Rohmann et al.,2006). Michel aplasia is caused by mutations in FGF3 (Tekin et al.,2007). FGF20 is a potential risk factor for Parkinson's disease (van der Walt et al.,2004; Satake et al.,2007). FGF13 is a candidate gene for Börjeson-Forssman-Lehmann syndrome (Gecz et al.,1999) and a hereditary spinocerebellar ataxia syndrome (SCA27) is caused by mutations in FGF14 (van Swieten et al.,2003; Dalski et al,2005).
Since all human FGF gene loci have been identified, it is likely that mapping studies will uncover other genetic diseases that involve additional members of the FGF family. Additionally, an abundance of mutations in Fgfrs result in a variety of skeletal dysplasia syndromes (reviewed in Ornitz and Marie,2002), suggesting that mutations in Fgfs will also be found to affect skeletal development or homeostasis.
As described above, FGF signaling is critical for many developmental and metabolic processes in vertebrates, and disorders in Fgf signaling result in various human diseases. The proposed functional evolutionary history of the Fgf family will be useful to elucidate additional functions, redundancies, and networks in development and metabolism.
This work was supported in part by grants 18659021 and 19390021 from the Ministry of Education, Culture, Sports, Science and Technology, Japan (to N.I.), the Takeda Science Foundation, Japan (to N.I.), the Mitsubishi Foundation, Japan (to N.I.), the National Institutes of Health grants HL07666401 and HD049808 (to D.M.O.), and the March of Dimes Foundation (to D.M.O.).