|Home | About | Journals | Submit | Contact Us | Français|
The vitamin A derivative retinoic acid performs many functions in vertebrate development and is thought to act as a diffusible morphogen that patterns the anterior-posterior axis of the hindbrain. Recent work in several systems has led to insights into how the spatial distribution of retinoic acid is regulated. These have shown local control of synthesis and degradation, and computational models suggest that degradation by the Cyp26 enzymes plays a critical role in the formation of a morphogen gradient as well as its ability to compensate for fluctuations in RA levels.
Segmentation along the anterior-posterior (A-P) axis of the vertebrate embryo creates a reiterated series of rhombomeres in the hindbrain, which form the blueprint for the adult brainstem. The hindbrain consists of up to eight rhombomeres, each with a distinct segmental identity according to its A-P position (Kiecker and Lumsden, 2005), as well as distinct types of neurons and patterns of gene expression (Box 1 – Segmental organization of the hindbrain and its early neuronal derivatives). This segmental framework allows for the specialization of neural and glial sub-types along the A–P axis and correlates with the migration patterns of cranial neural crest cells (NCCs). Motor neurons that contribute to each cranial nerve are derived from clusters of progenitors that form in specific rhombomeres and send out axons along stereotypical pathways to coordinate patterning in the hindbrain with the peripheral tissues that it innervates (Box 1).
Rhombomeres are specified during gastrulation under the control of signals from the posterior of the embryo, including the vitamin A derivative, retinoic acid (RA; Schilling and Knight, 2001; Gavalas, 2002; Maden, 2002). RA is thought to act as a graded “morphogen”-promoting segment-specific expression of homeotic (Hox) genes and patterning rhombomeres in a concentration-dependent manner. However, the mechanisms that establish an RA signaling gradient and its interactions with other signaling pathways that control A–P patterning largely remain a mystery.
Retinoids are potent modifiers of cell fate and behavior. Embryos treated with exogenous RA have severe defects in the hindbrain, arches, limbs, heart, gut and many other organs (Durston et al., 1989; Marshall et al., 1992; Godsave et al., 1998). RA synthesis requires sufficient dietary vitamin A and vitamin A-deficiency (VAD) causes severe developmental defects (Fig. 1, 2A,B) (McCaffery et al., 2003). In the hindbrain, VAD embryos lack posterior rhombomeres (r4–7) and more anterior rhombomeres expand posteriorly (Gale et al., 1999; White et al., 2000). Similar “anteriorization” of the hindbrain occurs in: 1) loss-of-function mutations in the gene encoding the major enzyme responsible for RA synthesis in the embryo, aldh1a2 (aldehyde dehydrogenase 1a2, also known as raldh2), in both mice and zebrafish (Niederreither et al., 1999; Niederreither et al., 2000; Begemann et al., 2001; Grandel et al., 2002), 2) loss-of-function mutants in RA receptors (RARs and RXRs) in mice, 3) embryos injected with dominant-negative RARs, or 4) treated with pharmacological inhibitors of both RA production and receptor activity in a variety of species (Daam et al., 1993; Blumberg et al., 1997; Dupé et al., 1999; Dupe and Lumsden, 2001; Begemann et al., 2004). Taken together, these results support a role for RA as a diffusible morphogen in the hindbrain. However, animals outside the laboratory encounter dramatic fluctuations in dietary vitamin A and yet their hindbrains develop normally. Thus, the question remains how such a gradient forms correctly in the face of varying RA levels.
A major new insight into the regulation of RA signaling came with the identification of the Cyp26s, cytochrome p450 enzymes that degrade RA. These are found in all deuterostomes that have been examined, and constitute the initial step in RA removal (White et al., 1996; Fujii et al., 1997; White et al., 1997; Hollemann et al., 1998; Swindell et al., 1999; Nagatomo and Fujiwara, 2003; Canestro et al., 2006). Cyp26s are required for segmental patterning of the hindbrain itself, and appear to constitute a negative feedback loop required to modulate RA signaling. Recently, new data have emerged on the roles of Cyp26s and RA degradation in hindbrain patterning (Maves and Kimmel, 2005; Sirbu et al., 2005; Uehara et al., 2006; Hernandez et al., 2007; White et al., 2007). In this review, we discuss the various models drawn from these studies, with particular emphasis on the role of degradation in the formation of an RA morphogen gradient. We focus on three main points: 1) Cyp26-mediated degradation is essential for embryogenesis, 2) RA-inducible degradation by Cyp26s makes RA signaling robust (able to compensate for changes in RA levels), and 3) other RA-inducible components of the RA pathway contribute to this robustness.
As illustrated in Fig. 1, RA levels are tightly regulated by a combination of synthesis and degradation, and depend on the availability of its precursors. In turn, a number of proteins that bind RA modulate its distribution and control its movement from one cellular compartment to another. Multiple feedback loops, recently identified at many steps in the RA signaling pathway add to this complexity (Fig. 1; green and red arrows). RA can negatively regulate its own production directly by repressing expression of aldh1a2 (Niederreither et al., 1997; Dobbs-McAuliffe et al., 2004) and lecithin:retinol acyltransferase (Matsuura and Ross, 1993; Zolfaghari and Ross, 2002). RA also positively regulates enzymes that degrade it, such as Cyp26a1 (White et al., 1996; Dobbs-McAuliffe et al., 2004; White et al., 2007) and induces expression of RARs (Rara, Rarb, Rarg) and binding proteins, Crabps (I & II) (Giguere et al., 1990; Leroy et al., 1991a; Leroy et al., 1991b; Smith et al., 1991; Kamei et al., 1993; Serpente et al., 2005). We argue that such positive and negative feedback in the pathway allows RA signaling to compensate for both varying RA levels as well as changes in the size and shape of the embryo during development.
All animals require some form of vitamin A (retinol) in their diet. This comes in the form of beta-carotene from plants and retinyl esters from animal sources (Fig. 1). The major storage form in embryos varies from species to species (reviewed in Simoes-Costa et al., 2008). For example, in some marine fishes retinal predominates, whereas embryos of freshwater fishes that have been examined contain a higher proportion of retinol/retinyl esters (Plack, 1964; Irie and Seki, 2002). The trend in higher vertebrates is toward increased use of retinol and retinyl esters (Plack and Kon, 1961), with some exceptions. For example, retinol predominates in chicken embryos, while embryos of ducks and geese prefer retinal (Plack, 1964). In placental mammals, retinol and retinyl esters are provided to the embryo and fetus through the maternal circulation (Quadro et al., 2004; Quadro et al., 2005). Retinol is transported in serum bound to retinol binding protein (RBP, described later), but cannot cross the placenta (Quadro et al., 2004). Instead it dissociates from RBP and diffuses across the yolk sac, the site of expression of fetal RBP (Johansson et al., 1997) for delivery to embryonic and fetal tissues.
β-carotene-15,15′-oxygenase (bcox) breaks down beta-carotene into two molecules of retinal (Fig. 1), and recent results in zebrafish have demonstrated that bcox is required for embryogenesis (Lampert et al., 2003) (Fig. 2A, C). Antisense morpholino oligonucleotides (MOs) that deplete bcox in embryos cause defects in formation of the pharyngeal arches and pectoral fins. In the hindbrain, bcox MO-injected (morphant) embryos show decreased expression of a Hox gene (hoxb4) in rhombomere 7 (r7). All of these defects appear to result indirectly from a requirement for bcox in endodermal cells that convert beta-carotene stored in the egg yolk (Lampert et al., 2003). In contrast, retinyl esters are stored in the liver and synthesized from retinol by lecithin:retinol acyltransferase (lrat; Fig. 1), an enzyme also recently shown to be required for zebrafish embryogenesis (Liu and Gudas, 2005; Isken et al., 2007; Kim et al., 2007). Loss of lratb function leads to embryos with elevated RA levels and reduced retinyl esters, resulting in a reduction in the gap between r5 and the spinal cord (r6 and r7), a phenotype reminiscent of RA-treated embryos (Fig. 2D; Isken et al., 2007). Both bcox and lrat are regulated by RA to adjust the amounts of retinol and thus RA (Matsuura and Ross, 1993; Zolfaghari and Ross, 2002; Lampert et al., 2003; Liu and Gudas, 2005; Isken et al., 2007). Retinyl ester stores are mobilized by retinyl ester hydrolases (Fig. 1).
Conversion of retinol to RA involves two steps, catalysed by different sets of dehydrogenases (reviewed in Duester, 2000). Some cytosolic alcohol dehydrogenases (Adhs) and microsomal retinol dehydrogenases (Rdhs) convert retinol to retinal, and in turn aldehyde dehydrogenases (Aldh1as) convert retinal to RA (Fig. 1). Recently, a mutation in mouse Rdh10 was reported to have defects in the forelimb, frontonasal process, pharyngeal arches and cranial ganglia. Rdh10 is expressed embryonically in the ventral hindbrain and anterior-most somites (Cammas et al., 2007; Sandell et al., 2007). This provides the first evidence that the conversion of retinol to retinal is also spatially restricted, and that Rdh10 is the major contributor to this conversion in the early embryo.
Some of the first genetic evidence supporting a role for endogenously-synthesized RA in embryos came with the publication of loss-of-function mutations in the RA-synthesizing enzyme, Aldh1a2, in both fish and mice (Niederreither et al., 1999; Niederreither et al., 2000; Begemann et al., 2001; Grandel et al., 2002). These papers demonstrated that Aldh1a2 (of the three Aldh1as in vertebrates) is critical for early embryogenesis, and that an inability to synthesize RA leads to a loss of posterior and expansion of anterior rhombomeres (Fig. 2E). Hindbrain defects correlate with a loss of expression of transgenic reporters driven by consensus DNA binding sites for RARs known as RA-response elements (RAREs), confirming a loss or reduction of RA signaling. However, the unique and redundant functions of different Aldh1as in RA signaling remain unclear. Loss of Aldh1a2 function does not phenocopy all aspects of VAD (compare Fig. 2B and 2E), perhaps due to other sources of RA.
Surprisingly, some reports suggest that a CYP1 enzyme (another member of the cytochrome p450 family) can also contribute to RA synthesis from both retinol and retinal (Chen et al., 2000b; Zhang et al., 2000; Choudhary et al., 2004; Chambers et al., 2007). Treating embryos with an inhibitor of CYP1B1 leads to a slight down-regulation of Hoxb1 expression in r4, indicative of reduced RA signaling. Adult Cyp1b1−/− mice also have eye abnormalities reminiscent of human congenital glaucoma, which could reflect defects in the well-known roles of RA signaling in eye development (Libby et al., 2003).
RA exerts most of its effects through a set of nuclear hormone receptors, the retinoic acid receptors (RARα, β,γ) and retinoid X receptors (RXRα, β, γ) (Fig. 1). RARs and RXRs form heterodimers, which regulate transcription through binding to RAREs. An exhaustive set of knock-out studies in mice has shown distinct, but overlapping functions for these receptors in embryogenesis (Li et al., 1993; Lohnes et al., 1993; Lufkin et al., 1993; Kastner et al., 1994; Lohnes et al., 1994; Mendelsohn et al., 1994a; Mendelsohn et al., 1994b; Sucov et al., 1994; Lohnes et al., 1995; Luo et al., 1995; Subbarayan et al., 1997; Dupé et al., 1999; Wendling et al., 2001). Single mutants are viable, although some less so than wild-types postnatally (Li et al., 1993; Lohnes et al., 1993; Lufkin et al., 1993; Mendelsohn et al., 1994b; Luo et al., 1995). Compound RAR/RXR mutants exhibit more severe embryonic defects (Fig. 2F, G). In the hindbrain, RAR α/βdouble mutants have an enlarged r5–7 region, whereas r5–7 are lost and r3–4 expand posteriorly in RAR α/βdouble mutants (Dupé et al., 1999; Wendling et al., 2001). This suggests that the receptors act combinatorially and in a semi-redundant manner to transduce the RA signal.
The situation is further complicated by the fact that in the absence of ligand, RARs and RXRs recruit corepressors to repress transcription of RA target genes, whereas ligand-bound complexes contain coactivators and activate these same targets (Horlein et al., 1995; Chen et al., 1996; Koide et al., 2001). This means that even in areas devoid of RA, RARs can effect gene expression. Indeed, the forebrain and midbrain require low RA levels to develop properly (Koide et al., 2001). A truncated co-repressor that acts as a dominant-negative blocks repression by endogenous cofactors, but not activation by RARs, and leads to a failure of anterior development. This demonstrates that repression is required for head development (Koide et al., 2001, not shown). In addition, forcing repression by treating embryos with an inverse agonist of RARs (AGN193109) upregulates anterior markers. The equilibrium between ligand-bound and unbound receptor dimers that determines the output of any RA-target gene is an attractive mechanism for establishing very sharp thresholds of activation, a common feature of many RA targets.
Other methods of interfering with receptor function have included over-expression of dominant-negative receptors and treatments of embryos with chemical receptor antagonists (one an inverse agonist that stabilizes the co-repressor complex). These have given broadly consistent data in the hindbrain, pharyngeal arches, gut and limbs. However, there are important differences between each of these interventions. For example, inverse agonists of RARs such as AGN193109 (Agarwal et al., 1996) increase repression, whereas loss-of-function mutations or MO injections remove receptor proteins, blocking both activation and repression. Thus, despite an exhaustive set of functional studies of receptors, future studies are needed to distinguish between positive and negative effects.
There is also evidence that RA binds and activates molecules other than RARs and RXRs, including PKCα (Radominska-Pandya et al., 2000; Ochoa et al., 2003) and PPARβ/δ (Shaw et al., 2003; Schug et al., 2007). This implicates RA in other processes such as lipid metabolism and obesity, which are both interesting avenues of research in the RA field for the future.
Several proteins bind either retinol or RA and regulate their cellular movements (Fig. 1). Crucial roles were recently demonstrated for an extracellular Retinol Binding Protein (Rbp) receptor called Stra6 (Kawaguchi et al., 2007), human mutations in which cause eye, heart and lung defects reminiscent of a reduction in RA signaling (Pasutto et al., 2007). These results suggest that Stra6 positively regulates RA production by promoting retinol uptake. Intriguingly, RA also binds an extracellular fatty acid binding protein with high affinity, the functions of which are unclear. One such protein purified from rat epididymis binds RA, but not retinol (Ong and Chytil, 1988; Newcomer, 1993; Sundaram et al., 1998). There are no data, however, on its expression or function during embryonic development.
Intracellularly, the cellular retinol-binding (Crbps)(Ross, 1993) and RA-binding proteins (Crabps; Ong and Chytil, 1978; Napoli, 1993; Donovan et al., 1995) modulate signaling (Fig. 1). These have extremely high affinities for their substrates, suggesting that RA remains bound to such proteins most of the time. This facilitates the movement of a lipophilic molecule such as RA in an aqueous environment. Where and how the bound RA moves, however, remains unclear.
Three functions have been suggested for Crabps: 1) transporting RA to the nucleus to interact with RARs, 2) sequestering RA in the cytoplasm away from RARs, and 3) transporting RA to Cyp26 enzymes for degradation. These functions may differ for each Crabp subtype, and are not mutually exclusive.
Evidence to support the first model, that Crabps (at least CRABPII) help transport RA to RARs in the nucleus, has come from many angles. For example, overexpression of xCrabp in Xenopus causes a loss of anterior structures, similar to RA treatments (Dekker et al., 1994), suggesting that this binding protein potentiates RA signaling, a function which is more consistent with the first hypothesis. In vitro, overexpression of CRABPII in breast cancer cell lines (Jing et al., 1997), as well as in COS-7 cells, enhances RAR-mediated activation of a reporter gene via a direct interaction between CRABPII and RARs (Dong et al., 1999). Indeed, binding of RA to CRABPII produces a change in conformation that exposes a putative nuclear localization signal. This results in the import of the RA-CRABPII complex into the nucleus (Sessler and Noy, 2005).
In contrast, in vitro studies with CRABPI suggest a negative role in signaling. Overexpression of CRABPI in F9 teratocarcinoma cells decreases their sensitivity to RA treatments, and this correlates with an increased production of RA metabolites, while depletion of CRABPI expression increases sensitivity (Boylan and Gudas, 1991; Boylan and Gudas, 1992). In addition, rat testes microsomes metabolize RA bound to CRABPI much more rapidly than RA alone (Fiorella and Napoli, 1991). These data suggest that, unlike CRABPII, CRABPI regulates the rate of RA metabolism, possibly by sequestering it in the cytoplasm, or by presenting RA to degrading enzymes, or both.
Despite their effects in vitro, mice carrying deletions of CRABPI or CRABPII individually or in combination show relatively mild developmental defects (de Bruijn et al., 1994; Gorry et al., 1994; Fawcett et al., 1995; Lampron et al., 1995), suggesting that RA signaling can occur without these proteins. This has been interpreted to mean that the CRABPs are not essential, but this is only in a laboratory setting where vitamin A levels are presumably sufficient. Such proteins may play a major role modulating signaling in response to environmental fluctuations in RA or its precursors.
RA is removed from tissues by oxidation to polar metabolites that are more easily excreted. Oxidation is initially catalyzed by the Cyp26 enzymes (White et al., 1996; Fujii et al., 1997; White et al., 1997; Hollemann et al., 1998; Swindell et al., 1999), of which there are three in mammals, Cyp26a1, Cyp26b1 and Cyp26c1. These ER-associated enzymes oxidize RA with the help of cytochrome p450 reductase (Cpr/Por). Loss-of-function mutations in Cyp26a1 in both mouse and zebrafish (Abu-Abed et al., 2001; Sakai et al., 2001; Emoto et al., 2005) as well as in Cyp26b1 and Cyp26c1 in mouse (Yashiro et al., 2004; Uehara et al., 2007) and Cpr/Por (Otto et al., 2003; Ribes et al., 2007b) all show defects consistent with an increase in RA signaling (Fig. 2H–L).
Products of oxidation by the Cyp26 enzymes include 4-oxo-RA, 4-hydroxy-RA and 5,6 or 5,8-epoxy-RA (Fiorella and Napoli, 1994; White et al., 1996; Fujii et al., 1997; Swindell et al., 1999; White et al., 2000). These metabolites themselves activate RARs, raising the possibility that the Cyp26 enzymes do not simply antagonize RA signaling (Pijnappel et al., 1993). Another subtype of Cyp enzyme (CYP3A7) has been reported to hydroxylate RA in human fetal liver where it may potentiate signaling (Chen et al., 2000a). Further conversion of RA metabolites to more polar molecules by, for example, UDP-glucuronosyltransferases, leads to their eventual excretion (Little and Radominska, 1997; Samokyszyn et al., 2000).
Since the more polar products of Cyp26s are rapidly excreted, the primary function of Cyp26s is likely to be as catabolic enzymes that eliminate RA and inhibit signaling (Abu-Abed et al., 2001; Sakai et al., 2001; Abu-Abed et al., 2003), consistent with loss-of-function data. Both RA-sensitive and insensitive regions express Cyp26s and RA activates expression of some of these enzymes (e.g. Cyp26a1 in the nervous system) and not others (see Table 1). Where RA induces Cyp26, it creates a negative feedback loop that regulates RA levels and compensates for changes in RA levels (as discussed below).
To date, Cyp26 enzymes have been cloned and analyzed from mouse (Fujii et al., 1997; de Roos et al., 1999; MacLean et al., 2001; Tahayato et al., 2003), chick (Swindell et al., 1999; Blentic et al., 2003; Reijntjes et al., 2003; Reijntjes et al., 2004), quail (Reijntjes et al., 2005), Xenopus (Hollemann et al., 1998; de Roos et al., 1999) and zebrafish (Kudoh et al., 2002; Dobbs-McAuliffe et al., 2004; Emoto et al., 2005; Gu et al., 2005; Zhao et al., 2005; Hernandez et al., 2007). The recent identification of a Cyp26 in hemichordates, echinoderms and some urochordates (Canestro et al., 2006), including Ciona intestinalis (Nagatomo and Fujiwara, 2003), suggests that they were present in the common ancestor of all deuterostomes. Cyp26s in vertebrates have distinct expression domains (Fig. 3) – some in RA-sensitive regions, others in RA-independent regions. Here we focus on expression and function in the hindbrain in early zebrafish, chick and mouse embryos (Fig. 2, ,4).4). Table 2 summarizes their expression patterns in the hindbrain and elsewhere. Expression features at late developmental stages are not reviewed here (for details see MacLean et al., 2001; Abu-Abed et al., 2002; Tahayato et al., 2003; Sakai et al., 2004).
In all three species Cyp26a1 is initially expressed in the anterior neural plate during gastrulation, in the future forebrain and midbrain (blue in Fig. 3A, D, G) (Fujii et al., 1997; Hollemann et al., 1998; Swindell et al., 1999; Kudoh et al., 2002). Between this domain and expression of Aldh1a2, in the trunk paraxial mesoderm, a gap of several hundred micrometers defines the presumptive hindbrain region (Swindell et al., 1999) and the source and sink of RA (Fig. 4A). The posterior boundary of Cyp26a1 expression differs slightly in mice (r2/3 boundary) and zebrafish (r3/4 boundary) (Kudoh et al., 2002; Sirbu et al., 2005; Hernandez et al., 2007). Our recent results in zebrafish, however, show that while this boundary defines a region of strong cyp26a1 expression, many cells within the presumptive hindbrain region (r3–7) express cyp26a1 at gastrula stages at a much lower level which we show is crucial for RA-dependent patterning (Fig. 3A) (White et al., 2007). In contrast to expression of cyp26a1 anterior to r2/3, this low-level hindbrain expression is induced by and requires RA, indicating that it modulates levels of RA signaling through negative feedback (Fig. 4E).
Cyp26a1 is essential for embryonic development (Fig. 2H). In Cyp26a1−/− mutant mice both the hindbrain and vertebrae are posteriorized and there is a severe caudal truncation with spina bifida and occasionally sirenomelia, a mermaid-like deformity with hindlimbs fused at the midline and no tail (Abu-Abed et al., 2001; Sakai et al., 2001). All homozygous mutant embryos die by postnatal day 1.
In the hindbrains of Cyp26a1−/− mutants, r3 is reduced and cells with an r4 identity expand as far anteriorly as r2 (Fig. 2H) (Abu-Abed et al., 2001; Sakai et al., 2001; Kudoh et al., 2002; Uehara et al., 2006) (Emoto et al., 2005; Hernandez et al., 2007). This eventually disrupts cranial nerves emerging from these rhombomeres. Sakai et al. (2001) showed using an RA-responsive lacZ transgene that RA signaling expands anteriorly into prospective r2 and r3 in Cyp26a1−/− mutants. Thus Cyp26a1 acts as a local antagonist of RA signaling. Consistent with this, haploinsufficiency for Aldh1a2+/− (Niederreither et al., 2002), or deletion of RARγ (Abu-Abed et al., 2003), rescues the Cyp26a1−/− phenotype. Recently, it was reported that approximately 30% of Cyp26a1−/− embryos have more severe hindbrain defects than previously appreciated, including a dramatic expansion of RA signaling based on RARE-lacZ (Uehara et al., 2006).
Interestingly, Cyp26a1−/− mutants are more sensitive to RA than wild types. Expansion of the posterior hindbrain caused by low doses of exogenous RA (5nM) in mutants resembles wild-type embryos treated with much higher doses (Hernandez et al., 2007). This suggests that Cyp26a1 protects the embryo from elevated RA. Consistent with this idea, injection of high levels of retinal, to artificially increase the rate of RA production, posteriorizes embryos lacking Cyp26a1 expression. Cyp26a1 is also required for exogenous RA treatments to rescue RA-deficient embryos (Hernandez et al., 2007). Thus Cyp26a1 seems to have a key role in the hindbrain, distinct from that of Cyp26b1 or Cyp26c1 (which are not induced in the nervous system by RA) in compensating for changes in RA levels (White et al., 2007).
Cyp26a1-mediated degradation also contributes to spatial dynamics of RA signaling in many other tissues and organs. Table 2 reviews domains of cyp26a1 expression outside the hindbrain. These include cranial mesoderm in fish (blue in Fig. 3B) (Dobbs-McAuliffe et al., 2004) and mouse (Fig. 3I), limb bud ectoderm and dorsal spinal cord in chick (Swindell et al., 1999; Blentic et al., 2003) somite boundaries in both Xenopus and zebrafish at later segmentation stages (Hollemann et al., 1998; de Roos et al., 1999; Dobbs-McAuliffe et al., 2004) as well as the tailbud, neural retina and pharyngeal arches. Cyp26a1, like Cyp26b1 and Cyp26c1 (see below) may also play a role in later patterning within the hindbrain both in chick and mouse, since it is expressed in a stripe of cells within the hindbrain, reported as either r2 or r3 (Fig. 3F, H) (Fujii et al., 1997; Swindell et al., 1999; Blentic et al., 2003). Outside the hindbrain, Cyp26a1 likely also acts within the vascular system in mouse (MacLean et al., 2001; Ribes et al., 2007a) and in many other tissue in which it is expressed.
In contrast to Cyp26a1, Cyp26b1 expression appears later and in a more dynamic pattern in the hindbrain (yellow in Fig. 3; Table 2). The pattern differs slightly in each case: 1) in zebrafish first in r3/4 (Fig. 3B) and r2–6 by early-somite stages (Fig. 3C)(Zhao et al., 2005), 2) in chick first in r4 expanding to r1–6 (Fig. 3F) (Reijntjes et al., 2003), 3) in mice initially in r3 and r5 and later in r2–6 (Fig. 3H, I)(MacLean et al., 2001). These patterns suggest that Cyp26b1 creates a new sink for RA within the central hindbrain (r3–5) at the end of gastrulation that eventually covers all but the most posterior rhombomeres.
Direct evidence that cyp26b1 contributes to hindbrain development comes from studies in zebrafish. MOs that inhibit Cyp26b1 protein production cause no defects in hindbrain patterning on their own, but when injected into cyp26a1 mutants they cause r4 to expand and the r6/7 boundary to shift anteriorly (Fig. 2I) (Hernandez et al., 2007; see below)), and at later stages can cause defects in craniofacial structures (Reijntjes et al., 2007). These data suggest that cyp26a1 and cyp26b1 play partially redundant roles in inhibiting RA signaling in the hindbrain.
Loss-of-function Cyp26b1−/− mutations in mice cause abnormalities in the limbs and craniofacial skeleton as well as a loss of germ cells from the testis (Yashiro et al., 2004; MacLean et al., 2007), but no major hindbrain defects. Cyp26b1 is expressed distally in limb buds, and both the forelimbs and hindlimbs are truncated in mutants, with proximal fates expanded at the expense of distal as well as elevated expression of RA-responsive reporters (Yashiro et al., 2004; MacLean et al., 2007). In addition, germ cells enter meiosis prematurely and die by apoptosis in these mutants (MacLean et al., 2007). Wild-type testes treated with a synthetic retinoid that cannot be metabolized by CYP26B1 show the same defects. These results suggest that an increase in RA signaling, rather than a loss of CYP26B1-derived metabolites, causes these phenotypes.
Table 2 summarizes Cyp26b1 expression in regions outside of the hindbrain. These include the tailbud (not shown) and cranial mesoderm in chick (Fig. 3E) (Reijntjes et al., 2003), which resembles the expression of cyp26a1 in presumptive cranial mesoderm in zebrafish and mouse (Dobbs-McAuliffe et al., 2004) (Fig. 3B, I).
Cyp26c1 is also expressed in the hindbrain and the pattern varies dramatically between species: 1) in zebrafish, expression begins at blastula stages, becomes restricted to r2–4 (red in Fig. 3A), and eventually expands throughout r2–6 (Fig. 3B) (11h; Gu et al., 2005; Hernandez et al., 2007) and the dorsal hindbrain, 2) in chick, expression begins during gastrulation in anterior mesoderm and presumptive r2/3 (Fig. 3D, E), and expands to r4–6 (Fig. 3F) and the roof-plate (Reijntjes et al., 2004), 3) in mouse Cyp26c1 expression initially appears in the head mesenchyme at E7.5 (Fig. 3G; Uehara et al., 2007) and is then expressed after gastrulation in r2 and r4 (Fig. 3H) and is subsequently lost in r4 (Fig. 3I) (Tahayato et al., 2003). These patterns suggest that Cyp26c1, like Cyp26b1, forms a sink for RA within the central rhombomeres (r2–6) of the hindbrain that both reduces RA within cells that express it and helps shape gradients of RA in adjacent cells.
Cyp26c1-deficient mice are viable and have no overt anatomical abnormalities (Uehara et al., 2006). However, Cyp26a1−/− Cyp26c1−/− double mutants have severely posteriorized hindbrains and die during embryogenesis (Fig. 2J). The heads and eyes of mutants are reduced and the neural tube fails to close. Hoxb1 and Fgf8 expression in the hindbrain expands anteriorly, as does expression of RARE-lacZ (Uehara et al., 2006) confirming an increase in RA signaling. Krox-20 (Egr2b, which normally marks r3 and r5) is expressed in a single broad stripe in double mutants (Fig. 2J). In addition, loss of Aldh1a2 function in a Cyp26a1−/− Cyp26c1−/− double mutant partially rescues the phenotype, confirming that it is due to elevated levels of RA (Uehara et al., 2006).
Hernandez et al. (2007) also depleted cyp26c1 alone and in combination with other Cyp26s in zebrafish and found their requirements to be partially redundant. Injection of cyp26c1-MO alone or together with cyp26b1-MO only slightly shortened the hindbrain. However, injection of the cyp26c1-MO into cyp26a1−/− embryos caused a loss of r3, anterior expansion of r4, shortening of r5/6 and an anterior shift of the r6/7 boundary (Fig. 2K). Depletion of all three Cyp26 enzymes in zebrafish gives the most dramatic phenotype – a massive posteriorization of r2–7 in which the entire hindbrain appears to acquire an r7-like identity (Fig. 2L). In triple morphants, r3 and r5 are lost, r4 abuts the cerebellum (r1) and the r6/7 boundary appears to lie just posterior to r4 (Hernandez et al., 2007). Similar transformations were produced by treating embryos with a chemical inhibitor of Cyp26 activity and this could be reversed by co-treating embryos with DEAB (a compound that blocks synthesis of RA). Taken together, these results provide strong evidence that 1) all three Cyp26s are required to pattern the A–P axis of the hindbrain and 2) that defects caused by removing their functions are due to an excess of RA signaling, rather than a lack of Cyp26-dependent metabolites.
Table 2 summarizes other sites of Cyp26C1 expression across species including pharyngeal arches, head mesenchyme (presumptive mesoderm) near the ear and a small patch of the retina in zebrafish (Gu et al., 2005).
Given all this new information on RA degradation, how do we now envisage RA signaling along the A-P axis of the hindbrain? RA is synthesized by Aldh1a2 in the paraxial mesoderm in the trunk (Niederreither et al., 1997), and degraded by Cyp26a1 in the anterior neural plate in the head (Hale et al., 2006; Tallafuss et al., 2006). This has led to a model in which RA forms an anteriorly-declining concentration gradient that acts as a morphogen to pattern the hindbrain field (Fig. 4A) (Berggren et al., 1999; Swindell et al., 1999). However, a gradient of RA has never been directly visualized, largely due to technical reasons. We review the evidence for and against this “morphogen model” for RA in hindbrain development.
Initial evidence for the model came from experiments on frog and chick embryos in which exogenous RA treatments caused concentration-dependent effects on hindbrain patterning (Durston et al., 1989; Sive et al., 1990; Marshall et al., 1992; Godsave et al., 1998). Notably for the model, excess RA caused posterior rhombomeres to expand, apparently at the expense of anterior segments. In addition, treatments with chemical antagonists of the RARs (such as BMS493), demonstrated that rhombomeres require RA signaling in both a concentration- and time-dependent manner (Dupe and Lumsden, 2001). In the absence of RA signaling, anterior rhombomeres expand and replace posterior segments, consistent with the gradient model. Posterior rhombomeres also require higher concentrations of RA to develop, and different concentrations of a pharmacological blocker of Aldh1as to inhibit RA synthesis also cause a graded anteriorization of the hindbrain (Begemann et al., 2004; Maves and Kimmel, 2005).
One major argument against the morphogen model has been the clear time-dependent aspect to rhombomere formation (Dupe and Lumsden, 2001; Maves and Kimmel, 2005; Sirbu et al., 2005). Cells may not respond to the concentration of RA that they are exposed to, but rather detect the length of time of exposure (Gavalas, 2002; Maden, 2002; Sirbu et al., 2005). Several studies have proposed two phases of RA signaling in the hindbrain, 1) one initially extending up to the r2/3 boundary, and 2) a later one restricted posteriorly to the r4/5 boundary due to degradation by Cyp26c1 (Fig. 4B). In this model of “shifting boundaries” of degradation, r5/6 are specified because they are exposed to RA signaling for longer than r3/4. This, however, does not explicitly preclude concentration-dependent RA effects. In fact, Sirbu et al. (2005) invoke concentration-dependence to explain the expression patterns of hoxb1 and vHNF1 in their model. Cyp26c1 mutant mice also show no hindbrain phenotype, arguing against the later phase of the model (Uehara et al., 2006).
Maves and Kimmel (2005) argue against a timing model based on their work in the zebrafish hindbrain. They showed that RA target genes (e.g. hox genes) require RA signaling only at the onset of their expression. In addition, sufficiently high concentrations induce posterior target genes such as hoxd4 much earlier than their normal onset (Maves and Kimmel, 2005). Thus, posterior genes do not require a longer exposure to RA. Instead they may require higher levels of RA than anteriorly-expressed genes and may become activated later because RA synthesis increases at later stages (Fig. 4C) (Maves and Kimmel, 2005). This would suggest that the RA gradient grows with time and that the high concentrations needed for posteriorly-expressed genes are only reached later, giving an apparent time-dependence.
It remains possible that cells somehow integrate the strength and duration of exposure to the RA signal, and that high concentration equals long exposure. A similar model has been proposed for the mechanism of action of Shh in patterning the Dorsal-Ventral (D-V) axis of the vertebrate neural tube (Dessaud et al., 2007).
Hernandez et al. (2007) propose a different model to incorporate timing, in which the degradation enzymes also play a central role (Fig. 4D). This is based on an elegant set of loss-of-function studies of the different Cyp26s in zebrafish. In their model RA acts permissively, rather than instructively, in the hindbrain. A sufficient level of RA signaling overall, rather than a concentration gradient, allows gene expression at different times and places, under the control of some other A-P patterning mechanism. Signaling occurs in regions of the hindbrain not expressing any Cyp26 enzyme, which permits expression of the appropriate RA target gene, similar to the model of Sirbu et al. (2005). Developmental time dictates which genes can be expressed at any given step. With these shifting boundaries of degradation, RA first sets the anterior border of hoxb1 expression and subsequently sets boundaries of progressively more posterior genes.
Like Sirbu et al. (2005), these studies show that both the spatial and temporal activity of RA is controlled by Cyp26-mediated degradation. Both models imply that some other A-P patterning information governs Cyp26 expression. Thus, an important next step is to identify the nature of this information.
Through our own experimental work in zebrafish embryos, combined with computational analyses of RA signaling, we recently proposed yet another model that we think can reconcile most features of those discussed above (Fig. 4E) (White et al., 2007). In contrast to many previous studies, we have shown that cyp26a1 is expressed more broadly than just the anterior neural plate. During gastrulation the presumptive hindbrain itself also expresses cyp26a1, though at very low levels, and this domain of expression requires RA signaling. These results suggest a more subtle role for cyp26a1 than simply acting as an anterior sink. We propose that cyp26a1 modulates levels of RA signaling dynamically and compensates for fluctuations in RA synthesis rate. Thus, if RA production drops, Cyp26a1-mediated degradation drops to compensate, and vice-versa.
Fgf signaling from the posterior of the embryo also inhibits cyp26a1 expression. Fgf forms an anteriorly-declining gradient that lies in parallel to RA and promotes posterior development (Kudoh et al., 2002; White et al., 2007). Therefore, opposing actions of RA and Fgf regulate Cyp26a1 levels, with RA promoting and Fgf suppressing expression, respectively. We propose that these opposing forces create an extremely robust system (i.e. one that is resistant to fluctuations in vitamin A availability and RA synthesis, for example). As we noted earlier, the RA signaling pathway includes many feedback and feed-forward loops, as does the Fgf signaling pathway, and there are many points of interaction between the two systems (Shiotsugu et al., 2004; White et al., 2007). Thus, we argue that the two putative morphogen gradients that arise in the posterior of the embryo, RA and Fgf, form an integrated system.
This model helps resolve the classic view of RA as a morphogen with at least two seemingly contradictory results: 1) that treatment of RA-deficient embryos with a uniform concentration of exogenous RA fully rescues hindbrain patterning within a wide concentration range (Fig. 5), and 2) that rhombomeres do not all arise at the same stage – genes specifying posterior segments are expressed later than those that specify anterior segments.
Counters to the first criticism come from our computational approaches to the morphogen model. Interestingly, one consequence of having cyp26a1 induced by RA and inhibited by Fgf is that any uniform extracellular concentration of RA becomes quickly converted into an intracellular concentration gradient. Furthermore, the induction of cyp26a1 by RA also makes this gradient robust to changes in RA levels. Consistent with this idea, a functional Cyp26a1 (and not cyp26b1 or cyp26c1) is required for exogenous RA to rescue hindbrain patterning (Hernandez et al., 2007). While the rescue of hindbrain patterning in RA-deficient embryos occurs over a 20-fold concentration range (20-fold; Hernandez et al., 2007), robustness has its limits. Treatments outside the upper limit of this range posteriorize the hindbrain (Fig. 5F), while treatments below the lower limit (0.25 nM) lead to incomplete rescue (Fig. 5C).
A counter to the second criticism regarding timing comes from work by Maves and Kimmel (2005), which shows that expression of posterior RA-target genes in the hindbrain does not require prolonged exposure to RA – high levels of RA can activate them prematurely. This suggests that the RA gradient grows over time. Growth could occur by increasing the rate of RA synthesis gradually, or by reducing levels of RA degradation. Based on our computational model, we propose that the latter is more likely - inhibition of cyp26a1 by Fgf decreases as the embryo enlarges. This results in an increase in the RA gradient that remains tightly coupled to gastrulation movements controlled, in part, by Fgfs (White et al., 2007).
However, to date our model only incorporates RA and Fgf interactions explicitly, and it is restricted to gastrula stages, when these signals are exerting their first effects on A-P patterning and hindbrain development. Different relationships may exist at later stages. For example, Fgfs become expressed at the midbrain-hindbrain boundary and within the hindbrain itself by the end of gastrulation (e.g. r4), and both cyp26b1 and cyp26c1 are expressed in specific rhombomeres (see Fig. 3). Thus only the very early events in A-P patterning of the hindbrain fit our model involving regulation of cyp26a1. Later degradation by cyp26b1 and cyp26c1 protect subsets of cells in the hindbrain from RA in a similar manner to that previously proposed (Sirbu et al., 2005; Hernandez et al., 2007).
The discovery of the Cyp26 enzymes and recent investigations into their roles have led to new insights into how cells establish and maintain appropriate levels of RA signaling. Degradation shapes the distribution of RA and is highly regulated, both in vertebrate embryos and presumably later in adults, in the various tissues and organs that RA is known to effect. However, many unanswered questions remain, such as how the complex expression patterns of Cyp26s are regulated. Our work has touched on two of these in the case of Cyp26a1, Fgf signaling and RA itself, but the network of interactions is no doubt much more complex.
It is also unclear if Cyp26-mediated degradation acts primarily in clearing RA from cells completely or more in subtle modulation of RA levels. Previous studies have implied that Cyp26a1 expressed in the anterior neurectoderm creates a perfect sink, such that essentially all RA is removed. With our discovery of low-level cyp26a1 expression in the hindbrain, a tissue known to require RA, it seems that cells expressing cyp26 can still respond to RA. We have begun to address this question by transplanting small numbers of cells either overexpressing or deficient in Cyp26a1 and examining their responsiveness to RA (White et al., 2007).
One huge challenge that remains is visualizing the endogenous RA gradient, something that is currently technically impossible. Previous attempts to demonstrate a gradient have included measuring the amounts of RA using HPLC in the anterior and posterior limb bud and implanting beads soaked in radiolabeled retinoids into limbs to measure the distribution produced (Eichele and Thaller, 1987; Thaller and Eichele, 1987). These approaches have shown that artificial RA gradients can form and that some of the differences in RA concentrations one might expect of a morphogen do exist endogenously, but definitive proof is still absent. One approach would be to fluorescently tag RA and introduce it, either on implanted beads or as a caged-molecule, which can be uncaged locally using a laser. This would allow for real-time visualization of an ectopic gradient. Such information is crucial for understanding how differences in the concentration of this small, lipophilic vitamin derivative lead to such specific effects on cell fates, such as the formation of rhombomeres and segment boundaries in the vertebrate hindbrain.
The hindbrain is subdivided into 8 rhombomeres (r1–8), which are all molecularly distinct and form different neuronal cell types. (A) Diagram of the early vertebrate (zebrafish) hindbrain; lateral view, anterior to the left. Boxes above the hindbrain represent the expression patterns of various transcription factors and cell adhesion molecules: green - hoxd4; yellow - MafB; orange - hoxb1; light blue - krox20 and epha4; dark blue - ephrin b2. Stippled boxes denote genes that require RA for their expression. Neural crest cells (NCCs; red arrows) migrate to different pharyngeal arches depending on their rhombomeric origin (midbrain & r1–3 - arch 1; r4–5 - arch 2; r6–8 -arches 3–7). (B) Dorsal view (anterior to the left) of a later hindbrain showing neurons within the hindbrain that require RA signaling for proper differentiation, independent of its role in rhombomere identity. purple; trigeminal (V), orange; facial (VII), green; vagus (X), yellow; hindbrain noradrenergic (NA) neurons, stippled boxes denote neurons that require RA signaling for differentiation.
We would like to thank members of the Schilling lab for comments on the manuscript. This work was supported by the NIH (NS-41353).