The hindbrain is the most evolutionarily ancient part of the vertebrate brain (Jackman et al., 2000), and its basic organization into a series of seven or eight segments termed rhombomeres is similar across vertebrate species. This conservation is detectable at the morphologic, neuroanatomic, and molecular levels (Gilland and Baker, 1993; Fig. 1). For example, krox-20, which encodes a zinc-finger transcription factor with an important role in hindbrain patterning, is expressed in the third and fifth rhombomeres in all vertebrate embryos that have been examined, including the zebrafish (Fig. 1; Oxtoby and Jowett, 1993). Thus, insights gleaned from genetic, cellular, and molecular analyses of hindbrain patterning in the zebrafish are immediately relevant to our understanding of this process in vertebrates in general.
HINDBRAIN ORGANIZATION: ADULT FUNCTION FROM EMBRYONIC PATTERN
In the adult, the hindbrain is composed of the posteriorly located medulla oblongata, the ventroanterior pons, and the dorsoanterior cerebellum. The primary role of the cerebellum is to coordinate movement, and it thus receives information from many sources while transmitting information primarily to motor cortical areas. The medulla and pons form the fourth ventricle of the brain, and together with the midbrain, are often referred to as the “brainstem.” The brainstem contains a conglomerate of cell groups that form a complex network termed the reticular formation, which is involved in higher order behaviors such as respiration, circulation, and wakefulness. The reticulospinal neurons of the brainstem, considered in more detail ahead, provide the major route through which the brain communicates with the spinal cord to control locomotion.
The hindbrain also contributes 8 of the 12 pairs of cranial nerves, numbers V through XII. The early segmental organization of these cranial nerves is rhombomere dependent, as described in more detail below. Cranial nerves can have both motor and sensory components, and those derived from the hindbrain are responsible for a wide variety of behaviors, including taste, hearing, balance, mastication, facial expressions, some eye movements, and, in terrestrial vertebrates, the secretion of tears and saliva. For example, the VIIIth (vestibulo-acoustic) nerve lies in the most posterior part of the pons, adjacent to the ear, and controls both balance and hearing.
Hindbrain-derived cranial nerves also innervate adjacent pharyngeal arches, which lie ventral to the surface of the head and, like the hindbrain itself, show a segmental organization. The segmentation and patterning of hindbrain and pharyngeal arches are intimately linked, with migratory neural crest cells that make up the bones, cartilages, and connective tissues of the pharyngeal arches deriving from specific rhombomeres (Lumsden et al., 1991). Thus in the zebrafish, neural crest cells that arise dorsally in r1-r3 migrate into the first pharyngeal arch to form jaw structures (Schilling and Kimmel, 1994). These same structures are innervated by the Vth (trigeminal) cranial branchiomotor nerve, which has its cell bodies in r1-r3 (Higashijima et al., 2000). A similar registry of segments in the hindbrain and head periphery is reiterated in the second and third pharyngeal arches.
Hindbrain Organization During Development
The morphological segmentation of the zebrafish hindbrain is transiently visible at the 18 somite stage (18–20 hr postfertilization [hpf]; Kimmel et al., 1995), when five prominent bulges along the anterior-posterior extent of the hindbrain, rhombomeres (r)2-r6, are detectable in the vicinity of the developing otic vesicle, which lies lateral to r5 (Fig. 2A, Hanneman et al., 1988). Confocal time-lapse analysis of Bodipy-ceramide–stained embryos allows the detection of forming rhombomere boundaries during early somite stages (Moens et al., 1998; Fig. 2B), and later boundary cells begin to acquire distinct identities as indicated by the expression of boundary-specific genes (Fig. 2C).
Rhombomere boundaries do not arise in a straightforward anterior to posterior order as somite boundaries do in the trunk, but rather in a seemingly erratic order that nevertheless hints at the underlying genetic events. The order of boundary formation is stereotypical for embryos of a given species: in the zebrafish, the r4 territory is defined first with the appearance of the boundary between r3 and r4 (r3/4) and then the r4/5 boundary, followed by the r1/2, r2/3, and r6/7 boundaries. That r4 appears first may reflect its importance as an early signalling center in the hindbrain (see below). That the r5/6 boundary forms last may reflect that r5 and r6 are subdivided from a common precursor (Moens et al., 1996, 1998; see below). No r7/8 boundary is detectable at any stage, either morphologically or by RNA in situ hybridization with boundary- or rhombomere-specific markers. This finding is consistent with the observation, both in chick (Clarke and Lumsden, 1993) and in zebrafish (Kimmel et al., 1985; Hanneman et al., 1988), that the neuronal organization of the posterior hindbrain is different and less overtly segmental than that of the rest of the hindbrain.
At the anterior end of the hindbrain, r1 is usually described as a large segment extending from the mid-hindbrain junction to the r1/2 boundary (reviewed by Lumsden, 1990). Fate mapping experiments in the chick, mouse, and zebrafish have shown that cells in this region contribute to the cerebellum (Zinyk et al., 1998; Wingate and Hatten, 1999; Koster and Fraser, 2001). The organization of r1 is complex, with distinct classes of neurons differentiating along its anterior-posterior axis. Morphological descriptions of the chick hindbrain (Vaage, 1969) suggested that this large “r1” is in fact composed of two distinct domains, “r0,” or the so-called “isthmic rhombomere” anteriorly, and a narrower r1 posteriorly. This nomenclature has fallen into disuse, because no molecular markers of this r0/r1 boundary have been identified in the chick. However in the zebrafish, molecular markers have been identified that are expressed in a narrow domain immediately anterior to r2 that may correspond to Vaage's r1 (see Fig. 5A). The anterior limit of this domain corresponds with the lateral angle of the fourth ventricle, which separates the upper and lower rhombic lip domains (Wingate and Hatten, 1999; Koster and Fraser, 2001). Evidence from the zebrafish suggests that anterior and posterior r1 are patterned independently (see below) and may have distinct fates within the ventral cerebellum (Koster and Fraser, 2001). In this review, we adopt Vaage's usage, referring to the anterior part of this large r1 segment as r0, and the posterior part as r1.
Principles of Hindbrain Development
Neuronal organization of the hindbrain corresponds with the rhombomeres.
Hindbrain neuroanatomy corresponds with morphological segmentation (Fig. 3; Hanneman et al., 1988; Lumsden and Keynes, 1989; Trevarrow et al., 1990). This correspondence, although true of many neuronal types, is most dramatically illustrated by the individually identified reticulospinal interneurons of the zebrafish. The reticulospinal neurons differentiate early during hindbrain development in a reiterated pattern that persists long after the morphological segmentation of the hindbrain has become obscured (Fig. 3A; Metcalfe et al., 1986; Mendelson, 1986a; Hanneman et al., 1988). Motor neurons, commissural neurons, and other, later-differentiating neuronal types also conform to this segmental pattern (Fig. 3B; Gilland and Baker, 1993; Chandrasekhar et al., 1997). Although the overtly segmental organization of these neurons and of the hindbrain as a whole is lost during subsequent growth and morphogenesis, the functional connections made between the hindbrain and the head periphery during the segmented period persist throughout adult life. Morphological and neuroanatomical segmentation are linked genetically, because mutations that disrupt morphological segmentation also disrupt the segmental organization of hindbrain neurons (Moens et al., 1996; Waskiewicz et al., 2001). This causative relationship between hindbrain segmentation and hindbrain neuronal organization underlines the developmental importance of the rhombomeres.
Rhombomeres have a two-segment periodicity.
A second principle of hindbrain patterning is that the rhombomeres have an alternating two-segment periodicity. This principle is also well illustrated by the zebrafish reticulospinal neurons, which lie laterally in even-numbered rhombomeres and more medially in odd-numbered rhombomeres (Fig. 3A; Mendelson, 1986a,b; Metcalfe et al., 1986). In the chick, several lines of evidence point toward a two-segment periodicity within the hindbrain: the branchiomotor nerves that innervate the pharyngeal arches exit the hindbrain from even-numbered rhombomeres (Lumsden and Keynes, 1989); the timing of neuronal differentiation is delayed in odd-numbered rhombomeres (Lumsden and Keynes, 1989); cranial neural crest cells that contribute to the pharyngeal arches migrate from even-numbered rhombomeres (Lumsden et al., 1991); and, finally, even- and odd-numbered rhombomeres exhibit different cell adhesive properties (Guthrie and Lumsden, 1991; Guthrie et al., 1993; Wizenmann and Lumsden, 1997). Many of these observations have also been made in the zebrafish, where cranial motor nerves similarly exit from even-numbered rhombomeres, where neuronal differentiation is similarly delayed in odd-numbered rhombomeres (Bally-Cuif et al., 1998; Higashijima et al., 2000) and where cranial neural crest similarly migrates from even-numbered rhombomeres (Schilling and Kimmel, 1994). A two-segment periodicity in cell adhesive properties has also been observed, and recent work in the zebrafish has uncovered a molecular basis for these adhesive differences (see below; reviewed in Klein, 1999).
Each rhombomere is unique.
Differences in the size, number, and projections of the reticulospinal neurons in each segment illustrate another principle of hindbrain organization, which is that, overlying the basic re-iterated pattern, are clear segment-specific differences. For example, the prominent Mauthner neuron in rhombomere 4 has a much larger cell body and axon than do its segmental homologs in other rhombomeres (Fig. 3A). This picture of a basic segmental plan overlain by segment-specific differences is reminiscent of the body plan of the fly, where homeotic selector genes, the Hox genes, specify differences between initially similar segments. As we will discuss further below, the vertebrate homologs of the fly Hox genes are expressed in rhombomere-restricted domains, and work in several vertebrate systems has demonstrated that, although the segmenting mechanisms appear to be different in flies and vertebrates, the Hox genes play conserved roles in specifying segment identities.
PATTERNING THE ANTERIOR-POSTERIOR AXIS OF THE ZEBRAFISH CENTRAL NERVOUS SYSTEM
Long before the hindbrain becomes segmented and hindbrain neuroanatomy is elaborated, the presumptive neurectoderm is broadly patterned along its anterior-posterior axis into forebrain, midbrain, hindbrain, and spinal cord. Indeed, as early as the beginning of gastrulation, largely nonoverlapping regions of the presumptive neurectoderm are fated to give rise to distinct anterior-posterior identities (Fig. 4A; Kimmel et al., 1990; Woo and Fraser, 1995). The early gastrula resembles a cap pulled halfway over a round head (the yolk), with a thickened edge—the margin or germ ring—where the first mesendodermal precursors have begun involuting. The dorsal side of the embryo is marked by the embryonic shield, a thickening that is the equivalent of the organizer of the Xenopus embryo and the node of amniote embryos, which gives rise to midline, or axial, mesendoderm (including notochord and prechordal plate). The cells in the lateral and ventral germ ring are fated to become nonaxial mesoderm (including somites and blood) and endoderm. Cells further from the margin (closer to the animal pole) give rise to ectoderm, with the neurectoderm mapping to the dorsal side (Kimmel et al., 1990). A more fine-tuned fate map of the presumptive neurectoderm at this stage showed that cells furthest from the margin have forebrain fate, whereas cells with hindbrain fate lie closer to the germ ring, lateral to the shield (Woo and Fraser, 1995; Fig. 4A). The reproducible nature of this fate map suggests that patterning interactions may already be well under way at these early developmental stages.
Signals From the Lateral Germ Ring Can Specify Hindbrain Fates
What are the signals that specify hindbrain fates, and where do they come from? Classic embryology experiments in amphibians led to the proposal of the two-step “activation-transformation” model for neural patterning (Nieuwkoop, 1952). The two-step model suggests that, during gastrulation, all neural tissue is initially formed with an anterior (forebrain) character and that a second “posteriorizing” signal modifies this anterior character in a graded manner to give hindbrain and spinal cord fates (reviewed by Sasai and De Robertis, 1997). Recently, fate mapping experiments in the zebrafish have substantiated this model by showing directly that cells that express a forebrain marker (otx2) in the early gastrula down-regulate it by midgastrula stages and instead express mid- and hindbrain markers (Erter et al., 2001). The timing of this transformation corresponds with when the presumptive hindbrain is committed to its fate as determined by transplantation experiments (Woo and Fraser, 1998).
The source of the first (“activation”) signal is the embryonic organizer (the dorsal lip of the amphibian blastopore), which is required for specification of neural tissue as well as for dorsalization of mesoderm and endoderm through production of BMP antagonists (reviewed by De Robertis et al., 2000; Schier, 2001). The source and molecular identity of the second (“posteriorizing”) signal is more controversial. In the frog, as gastrulation proceeds, the cell population within the organizer is continuously changing, and the “late” organizer emits a signal that is sufficient to respecify anterior neurectoderm to hindbrain fates (Nieuwkoop and Albers, 1985; Poznanski and Keller, 1997). However, grafting and co-culture experiments in the zebrafish, chick, and mouse have suggested that nonaxial mesendoderm is also a source of posteriorizing signals that specify hindbrain fates. Woo and Fraser (1997) observed that transplantation of mesendodermal tissue from the lateral or ventral germ ring into the animal pole of an early gastrula nonautonomously induced differentiation of hindbrain structures and expression of the hindbrain-specific gene krox-20 in the forebrain of host embryos (Fig. 4B).
Transplantation of the embryonic shield to the animal pole failed to have the same effect, suggesting that in the zebrafish, the second “posteriorizing” signal in the two-step model originates not from the organizer but from the lateral germ ring. Indeed, ablation of the zebrafish embryonic shield, either genetically (Fekany et al., 1999) or mechanically (Shih and Fraser, 1996; Saude et al., 2000), has little effect on hindbrain patterning, supporting the idea that in the zebrafish the organizer plays a relatively minor role in posteriorizing the central nervous system (CNS). The predominant role of lateral germ ring over the organizer in posteriorizing the CNS may be a general principle of neural patterning in vertebrates, because explant co-culture studies in chick and mouse specifically implicate the paraxial mesoderm in posteriorization of the CNS (Muhr et al., 1997; Ang and Rossant, 1993).
Role of Wnt Signalling and Its Repression in Brain Patterning
The molecular nature of the posteriorizing signals involved in setting up the hindbrain territory is complex. Several lines of evidence have implicated wnt signals in the specification of posterior neural fates. Wnt8 is expressed in the lateral germ ring at the onset of gastrulation, and blocking wnt8 function, either genetically, with dominant-negative forms of the protein, or by using antisense “morpholino” oligonucleotides that prevent translation of the cognate mRNA (Nasevicius and Ekker, 2000) causes expansion of forebrain markers and loss of hindbrain markers in zebrafish and Xenopus embryos (McGrew et al., 1997; Lekven et al., 2001; Erter et al., 2001). The presence of a posteriorizing wnt signal in the embryo is also strongly suggested by the observation, in several vertebrate systems, that wnt antagonists are required in the anterior of the embryos to specify forebrain development. Dickkopf-1, a secreted wnt antagonist expressed in the prechordal plate, has strong head-inducing activity in Xenopus embryos (Glinka et al., 1998) and is essential for forebrain development in the mouse (Mukhopadhyay et al., 2001). In zebrafish, headless/tcf3, which encodes a DNA-binding protein that represses wnt targets, is similarly essential for development of head structures anterior to the midbrain-hindbrain junction (Kim et al., 2000). Also in zebrafish, masterblind/axin1 encodes a component of the intracellular complex that antagonizes wnt signalling through phosphorylation of β-catenin, and mbl/axin1 mutants lack anterior forebrain (telencephalon and eyes; Heisenberg et al., 2001).
The picture that emerges is one in which anterior-posterior fates in the central nervous system are established by layers of mutual repression: posteriorizing signals are required to inhibit anterior identities, but these signals must in turn be antagonized anteriorly to limit the spread of posterior identities. These interactions are likely to be initiated in the gastrula, but continue into neural plate stages, when new sources of wnt signals within the CNS (at the midbrain-hindbrain junction, see below) function locally to subdivide the brain into ever finer domains.
Not surprisingly, if signals from the lateral germ ring are required for posteriorizing neural fates, mutations that disrupt the specification of lateral mesendoderm also abolish these signals. Nodal signals are essential for mesoderm induction in zebrafish (reviewed in Schier and Shen, 2000). Progressively eliminating nodal signalling, in single or double mutants of cyclops and squint or through ectopic expression of Antivin (a nodal antagonist), progressively eliminates posterior fates. The strength of the effect on anterior-posterior patterning corresponds with the degree to which paraxial mesendoderm markers, including wnt8, are eliminated (Feldman et al., 2000; Thisse et al., 2000; Erter et al., 2001).
Although forebrain and midbrain fates are strongly affected by manipulating wnt signalling, hindbrain fates are more resilient to these perturbations, suggesting that other factors are involved in specifying more posterior fates. Moreover, unpatterned expression of the appropriate wnt antagonists can rescue the hdl/tcf3 and mbl/axin phenotypes without causing anteriorized phenotypes (Kim et al., 2000; Heisenberg et al., 2001; van de Water et al., 2001), again indicating the presence of underlying anterior-posterior pattern in the absence of wnt signals. Fgfs and retinoic acid are other signalling molecules that can affect anterior-posterior patterning. Antagonizing Fgf signalling by expression of dominant negative receptors prevents trunk and tail development in Xenopus and zebrafish (Amaya et al., 1991; Griffin et al., 1995), which suggested that Fgfs may posteriorize the embryo. However, Fgf beads do not mimic the posteriorizing effects of the lateral germ ring grafts of Woo and Fraser (1997). Recent work in the chick has suggested that, rather than playing a primary role in anterior-posterior patterning, Fgfs may primarily be required to maintain a proliferative stem cell population in the posterior of the embryo (Mathis et al., 2001).
Paraxial Mesoderm Is a Source of Retinoic Acid That Induces Posterior CNS Fates
A large body of evidence from the chick, quail, rat, frog, and mouse has shown that retinoic acid signalling is also essential for anterior-posterior patterning (reviewed in Gavalas and Krumlauf, 2000). Reducing retinoic acid signalling, whether by mutation of biosynthetic enzymes or receptors, by dietary intervention, dominant negative receptor expression, or pharmacological inhibition of receptor function, results in hindbrain patterning defects ranging from partial transformations of hindbrain rhombomere identity to a severe loss of posterior hindbrain and anterior spinal cord. These phenotypes reflect the direct role that retinoic acid plays in regulating Hox gene expression in the hindbrain, because the enhancers of several Hox genes contain retinoic acid response elements that are essential for the onset of expression within the mouse hindbrain (reviewed in Gavalas and Krumlauf, 2000).
The zebrafish neckless mutant, in which the ultimate enzyme in the retinoic acid biosynthetic pathway, raldh2, is disrupted, supports a posteriorizing role for RA (Begemann et al., 2001). In nkl/raldh2 mutants, the hindbrain posterior to r6 is truncated and hox gene expression in this region is delayed and abnormal. What the zebrafish nkl/raldh2 mutant provides, aside from confirmation that retinoic acid is important for hindbrain patterning, is the opportunity to use genetic mosaic analysis to identify the source of this retinoic acid signal in vivo. Begemann et al. discovered that by transplanting wild-type cells into the paraxial mesoderm of nkl/raldh2 mutant embryos, they could rescue some of the patterning defects in the posterior hindbrain, including expression of the retinoic acid responsive Hox gene, hoxb4. The conclusion drawn from this experiment is that nkl/raldh2 functions within the paraxial mesoderm to produce retinoic acid, which is required for the normal patterning of the adjacent central nervous system. This interpretation is consistent with the results of explant co-culture experiments in the mouse in which a retinoic acid-responsive reporter could be activated by a somite-derived retinoid signal (Gould et al., 1998).
Raldh2, like wnt8, is expressed in the lateral and ventral germ ring, raising the possibility that retinoic acid, wnts, and possibly other germ ring–derived signals function in a partially redundant manner to specify fates along the anterior-posterior axis. That disrupting wnt signalling primarily affects forebrain and midbrain fates, whereas disrupting retinoic acid signalling primarily affects hindbrain and anterior spinal cord fates, suggests that, at different anterior-posterior levels, different signalling systems may predominate.
In an effort to find molecules important for early neurectodermal regionalization, subtractive cloning methods have been used to identify genes differentially expressed in anterior versus posterior domains of the zebrafish gastrula (Sagerstrom et al., 2001). This screen identified eight genes expressed in the posterior half of the gastrula, including several novel clones whose function has yet to be determined. This type of unbiased approach, which does not rely on analysis of previously characterized candidate genes, may ultimately improve our understanding of the earliest phases of hindbrain development.
REGIONALIZING THE HINDBRAIN
Very shortly after the end of gastrulation (10 hpf), the zebrafish hindbrain is already clearly subdivided into distinct molecular territories that presage the rhombomeres. Presumptive rhombomere markers include krox-20 in r3 and r5, valentino/mafB in r5 and r6, and the hox genes (Fig. 7A). The hox genes encode evolutionarily conserved homeodomain-containing transcription factors that specify segment identity, and are discussed in greater detail below. Mutational analyses in the mouse and fish have identified some of the molecules that are responsible for the finer-scale patterning that occurs within the hindbrain after its initial specification during gastrulation.
Signals From the Mid-Hindbrain Junction Pattern the Anterior Hindbrain
At the anterior end of the hindbrain lies a prominent constriction, the isthmus, which plays an important role in patterning both the midbrain and the hindbrain (Liu and Joyner, 2001). The zebrafish ace/fgf8 gene is expressed in a narrow domain within the isthmus and ace/fgf8 mutants lack this structure (Brand et al., 1996; Fig. 5D). The role of isthmic Fgf8 in patterning the adjacent midbrain of zebrafish, chick, and mouse embryos has been well described (reviewed in Rhinn and Brand, 2001; Liu and Joyner, 2001). However Fgf8 also patterns the anterior hindbrain. In ace/fgf8 mutants, cell types characteristic of the anterior part of r0, which include the neurons of the locus coeruleus, are absent (Guo et al., 1999). At the same time, the expression domain of fgfr3, a marker of r1, is expanded anteriorly (Sleptsova-Friedrich et al., 2001; Fig. 5A,B). Thus ace/fgf8 is involved in specifying r0 and in setting the boundary between r0 and r1.
Fgf8 may in fact influence regional identity much more broadly in the hindbrain. In chick embryos in which Fgf8 signalling at the midbrain-hindbrain junction was locally disrupted by using blocking antibodies, the anterior limit of r2 was shifted anteriorly (Irving and Mason, 2000). Recently, L. Maves and C. Kimmel have shown that reducing function of both ace/fgf8 and another fgf expressed in the hindbrain, fgf3, prevents the specification of r5 and r6; however, eliminating fgf3 alone has very mild effects (personal communication). Finally, work in the chick has recently shown that eliminating fgf signalling with a pharmacological antagonist also disrupts specification of r3, r5, and r6 (Marin and Charnay, 2000).
One difficulty with interpreting the effects of loss of ace/fgf8 function on zebrafish hindbrain development is the fact that ace/fgf8 expression is not limited to the mid-hindbrain junction. Early in development, ace/fgf8 is expressed in the germ ring, and before becoming restricted to the isthmus, ace/fgf8 is expressed throughout the anterior hindbrain, including in r4 where its expression overlaps with that of fgf3 (Reifers et al., 1998; Phillips et al., 2001; Fig. 5C). Although it is likely that ace/fgf8 is required at the mid-hindbrain junction to pattern the anterior-most hindbrain, it is also likely that its broader effects on hindbrain patterning result from more short-range signalling within the hindbrain itself. Indeed, Maves and Kimmel have shown that the r5 and r6 defects caused by reducing Fgf3 and Fgf8 signalling can be rescued by wild-type cells transplanted into the hindbrain, suggesting that an Fgf signalling center(s) within the hindbrain itself acts to pattern the surrounding rhombomeres (personal communication).
The zebrafish spiel-ohne-grenzen (spg) mutant phenotype resembles that of ace/fgf8 mutants in that both lack an isthmus and have hindbrain patterning defects. In spg mutants, r1 is expanded anteriorly and r3 and r5 are reduced and abnormally shaped (Schier et al., 1996; Belting et al., 2001; Burgess et al., 2002; Hauptmann et al., in press). Spg encodes Pou2, a homeodomain transcription factor that is expressed during gastrulation in the anterior hindbrain and mid-hindbrain boundary primordia and then becomes restricted to r2 and r4 at around the time when other segment-restricted gene expression is initiated (Belting et al., 2001; Burgess et al., 2002). Ubiquitous mis-expression of spg/pou2 is sufficient to rescue the mutant hindbrain phenotype, suggesting that the spg/pou2 gene product is providing a permissive factor that is required for the embryo to respond to another, instructive determinant involved in the specification of hindbrain rhombomeres (Belting et al., 2001). That this determinant might be an Fgf is suggested by the observation that overexpression of fgf8 or implantation of Fgf8-coated beads have no effect on marker gene expression or morphogenesis in the spg/pou2 mutant neuroepithelium (Reim and Brand, 2002). Furthermore, expression of a known Fgf target, sprouty4, is strongly reduced both in the hindbrain and at the mid-hindbrain boundary of spg/pou2 mutants. Thus spg/pou2 may act as a regional competence factor for Fgf signalling in the hindbrain primordium and then, during segmentation stages, it may further limit Fgf responsiveness to specific rhombomeres (Reim and Brand, 2002).
Valentino Functions Cell-Autonomously to Specify r5 and r6 Identity
Confocal time-lapse imaging of the developing zebrafish hindbrain demonstrates that some segment boundaries form later than others (Moens et al., 1998). In the zebrafish, the presumptive r5-r6 domain, which expresses the bZip transcription factor Valentino (Val)/MafB initially appears as a broad domain that is subsequently subdivided into the definitive rhombomeres (Moens et al., 1998). The val/mafB mutant lacks r5 and r6 and consequently lacks rhombomere boundaries between r4 and r7 (Moens et al., 1996). Mosaic analysis of val/mafB demonstrated that this gene is required autonomously for cells to take on either r5 or r6 identity: cells lacking val/mafB function are gradually excluded from r5 and r6 of a wild-type host embryo (Moens et al., 1996, 1998; Fig. 6). Thus, val/mafB functions in a manner similar to a Drosophila gap gene, in the absence of which blocks of segments are deleted.
Analysis of a classic mouse mutant, Kreisler, which carries a mutation in the mouse MafB gene (Cordes and Barsh, 1994), has suggested that this gene functions differently in the mouse than in the zebrafish. In the Kr/MafB mouse, r5 is absent, whereas some markers of r6 are still expressed. Furthermore, cells from the abnormal r5/6 region of a Kr/MafB embryo transplanted into a wild-type embryo are excluded from r5 but not from r6. These experiments have led to the conclusion that r6 is normal in Kr/MafB mice (Manzanares et al., 1999b, but see McKay et al., 1994). Differences between val/mafB function in the zebrafish and Kr/MafB function in the mouse correlate with differences in both gene expression and neuroanatomy between fish and mice. For example, the mouse hoxb3 gene, which is a direct target of Kr/MafB (see below), is expressed only in r5, whereas the zebrafish hoxb3 is expressed in both r5 and r6, and in the mouse, the motor neurons of the sixth cranial nerve differentiate in r5, whereas in the zebrafish, they differentiate in r5 and r6. The putative functional differences between val/mafB and Kr/MafB do not correlate with differences in their regulation, because both genes are expressed throughout r5 and r6, but may result from differences in the expression of essential cofactors (Manzanares et al., 1997).
val/mafB and Kr/MafB both function upstream of hox genes in r5 and r6. Analysis of the regulatory sequences of the mouse Hoxb3 and Hoxa3 genes has identified essential Kr/MafB binding sites required for expression in r5 and r5 and r6, respectively (Manzanares et al., 1997, 1999a). In the zebrafish, the hoxa3 and hoxb3 genes are normally both up-regulated throughout r5 and r6, and this up-regulation fails to occur in val/mafB mutants (Prince et al., 1998). By extrapolation from the mouse work, the positive effect of val/mafB on hoxb3 expression in r5 and r6 in the zebrafish is likely to be direct. val/mafB and Kr/MafB may also negatively regulate hox gene expression in r5 and r6. Expression of hoxb1a, which is normally restricted to r4, expands posteriorly in val/mafB mutants (Prince et al., 1998). The negative effects of val/mafB and kreisler on hoxb1a expression are poorly understood, either in the fish or in the mouse where effects of Kr/MafB on hoxb1 expression remain controversial (Frohman et al., 1993; McKay et al., 1994; Manzanares et al., 1999b). It is possible that posteriorly expanded hoxb1a expression in val/mafB mutants results from loss of direct or indirect inhibition of hoxb1a expression. Alternatively, hoxb1a-expressing cells from r4 may spread posteriorly because of the absence of an r4/5 boundary. Lineage analysis of individual rhombomeres in val/mafB mutants should help to address this question.
How is val/mafB itself regulated? Recently a new player with a key role in specifying segment identity in the r5/r6 region has been identified through the isolation and cloning of an insertional mutant in the zebrafish (Sun and Hopkins, 2001). Zebrafish embryos lacking the homeobox gene vhnf1 resemble valentino mutants; however vhnf1 is genetically upstream of valentino, because mutants never express valentino in the r5/6 region. vhnf1 is expressed in the gut and pronephros where it has other essential functions, but its earliest expression is within the central nervous system with a sharp anterior limit at the r4/5 boundary. How vhnf1 itself is regulated is as yet unknown, but is clearly an important question. Further study of the relationship between vhnf1, val/mafB, and the hox genes will fill out our picture of the hierarchy of events involved in setting up pattern in the posterior hindbrain.
HOX GENE FUNCTION: SPECIFYING RHOMBOMERE IDENTITY AND SEGMENTATION
Thus far we have discussed the events that lead to the onset of Hox expression in the hindbrain. In this section, we address the role that Hox genes play in specifying segment identity. Vertebrate Hox genes in paralog groups 1 through 4 (PG1–PG4) are expressed in the developing hindbrain and migratory cranial neural crest cells. The Hox PG1–PG4 gene expression domains prefigure and respect rhombomeric territories, with each rhombomere expressing a different combination of Hox genes (Fig. 7A,B), thus providing a mechanism for specifying unique segment identities.
The roles of many Hox genes in this process have been determined in mice, where most of the 11 PG1–4 genes have been knocked out singly or in combination (reviewed in Lumsden and Krumlauf, 1996). The classic phenotype associated with Hox loss-of-function mutations in Drosophila is a homeotic transformation to a more anterior segment identity (Lewis, 1978). Some PG1–4 mutant mice do exhibit subtle transformations of hindbrain segment identity. For example, in the Hoxb1 mutant r4 is partially transformed to r2 identity (Studer et al., 1996). Other mutants exhibit anterior transformations of structures derived from hindbrain neural crest, as has been shown most clearly for the Hoxa2 mutant in which the second pharyngeal arch is transformed to first arch identity (Gendron-Maguire et al., 1993; Rijli et al., 1993). However other Hox PG1–4 mutant phenotypes are very subtle indeed. This is likely due to partial redundancy between Hox paralogs, as has been shown most clearly for the PG3 genes (Greer et al., 2000). It also explains why no Hox mutants have been identified in zebrafish mutant screens to date: with the possible exception of the segmental defects associated with the mouse Hoxa1 mutant (reviewed in Morrison, 1998), no single Hox loss-of-function phenotype is likely to be severe enough to have been isolated in the morphology- or even marker-based screens performed in the zebrafish thus far. Nevertheless, gain-of-function approaches and knock-down approaches by using modified antisense oligonucleotides (Nasevicius and Ekker, 2000), as well as analysis of zebrafish mutants lacking essential Hox partners (see below) have given us insights into Hox gene function in zebrafish in particular and in vertebrates in general.
Hox Cluster Organization in Zebrafish: The Zebrafish Has Extra Hox Genes
Defining features of Hox genes include their conserved Antenappedia-class homeobox sequences and their organization into gene clusters (reviewed by McGinnis and Krumlauf, 1992). All vertebrates have multiple clusters of Hox genes, which are believed to reflect ancient genome duplication events in the vertebrate lineage (Sidow, 1996). In mouse and human, there are four Hox clusters (A–D), each lying on a separate chromosome, and available data from chick and frog suggest this four-cluster arrangement is conserved amongst the tetrapods. The genes fall into 13 paralog groups, but no individual cluster has representatives of all 13 paralogs because there have been multiple gene losses during evolution; the total number of Hox genes for mouse and human is 39. The teleost fishes have undergone another major duplication event in their lineage, which has resulted in a seven Hox cluster arrangement for zebrafish (Amores et al., 1998; Fig. 7C). Zebrafish Hox organization also shows the consequences of multiple gene losses during evolution, including loss of an entire D cluster plus many individual genes to produce a total of 48 Hox genes. Although this number is far from being twice the 39 genes present in the mammals, it nevertheless reflects retention of several duplicates—it has been suggested that availability of such duplicates could have helped to facilitate the radiation of the teleosts (Amores et al., 1998; Taylor et al., 2001). Studies on zebrafish Hox genes can help us to understand the general mechanisms that underlie loss or retention of Hox gene duplicates in the vertebrates.
Although there are a larger number of zebrafish Hox genes in the PG1–4 class (13 in the zebrafish compared with 11 in the mouse; Fig. 7C; Amores et al., 1998), overall they show very similar expression patterns to their murine or chick homologs. For example, the two members of PG2, hoxa2b and hoxb2a, have almost identical expression patterns to mouse Hoxa2 and Hoxb2, with anterior limits of expression reaching to the r1/2 boundary and the r2/3 boundary, respectively, and expression in the second pharyngeal arch (Fig. 7A,B, Prince et al., 1998).
Within PG1, there are two hoxb1 duplicates in addition to a hoxa1 and a hoxc1 gene. Some clues as to why both hoxb1 duplicates have been retained are provided by their expression patterns (McClintock et al., 2001). The hoxb1a gene has a high-level expression domain in a “stripe” in r4 that is maintained well into development (Fig. 7A,B), very similar to the pattern for mouse Hoxb1. However, the expression pattern of hoxb1b is curiously similar to that of mouse Hoxa1—so much so that when the gene was first isolated, and in the absence of linkage data to allow its unambiguous assignment to a specific cluster, it was named zebrafish hoxa1 (Fig. 7C,D; Alexandre et al., 1996). Nevertheless, there is a true zebrafish Hoxa1 ortholog, hoxa1a, which has been clearly assigned to the Hoxaa cluster based on both linkage analysis and sequence comparison. Yet zebrafish hoxa1a is not expressed in broad territories of the developing hindbrain but rather in the ventral midbrain in a subclass of interneurons (the nucleus of the MLF; McClintock et al., 2001; Shih et al., 2001). These results suggest that the hoxb1b duplicate has taken on the role played in mouse by Hoxa1, freeing the hoxa1a gene to function in the midbrain. This kind of “function shuffling” as a result of gene duplication events may turn out to be a common theme in the vertebrates, which is particularly easy to reveal in the zebrafish.
Gain-of-function experiments have revealed some gene-specific differences in functional capacity of the four zebrafish Hox PG1 genes (McClintock et al., 2001). However, it is knock-down analyses by using stabilized antisense morpholinos that have confirmed the model that hoxb1a and hoxb1b are functional equivalents of mouse Hoxb1 and Hoxa1, respectively (McClintock et al., in press). In the future, the new morpholino knock-down technology (Nasevicius and Ekker, 2000) will allow rapid tests of function for individual Hox genes, as well as for gene combinations.
Pbx Genes: Revealing Conserved Roles for Hox Genes in Specifying Segment Identity
Hox proteins do not carry out their functions as transcriptional regulators alone, and elimination of function of their DNA binding partners has provided further insights into the broad patterning role of Hox genes in the hindbrain. The vertebrate Pbx and Meis genes encode divergent homeodomain-containing proteins of the TALE (three amino acid Loop extension) class, which are essential DNA binding partners for the Hox proteins (reviewed by Mann and Affolter, 1998). Two zebrafish pbx genes and five zebrafish meis genes are expressed in the presumptive hindbrain during the period when Hox genes are specifying segment identity (Pöpperl et al., 2000; Vlachakis et al., 2000; Waskiewicz et al., 2001; Zerucha and Prince, 2001). Eliminating the zygotic function of one of these, Pbx4, as occurs in the lazarus (lzr)/pbx4 mutant, mimics the phenotypes associated with loss of several different Hox genes in the mouse (Pöpperl et al., 2000; K.L. Cooper, V.E. Prince, and C.B. Moens, unpublished results). For example, in lzr/pbx4 mutants the motor neurons of the seventh cranial nerve do not migrate posteriorly from r4 (a phenotype identical to that of both hoxb1 mutant mice and zebrafish embryos injected with a hoxb1a morpholino oligonucleotide); and motor neurons of the fifth cranial nerve project aberrantly into the second pharyngeal arch (identical to the hoxa2 mutant mouse). Direct evidence that lzr/pbx4 is indeed broadly required for Hox protein function in zebrafish is the observation that lzr/pbx4 function is required for the gain-of-function phenotypes associated with hoxb1a, hoxb1b, hoxa2b, and hoxb2a overexpression (Pöpperl et al., 2000; K.L. Cooper, V.E. Prince, and C.B. Moens, unpublished results). Applying the technique of genetic mosaic analysis that was so informative in the analysis of the nkl/raldh2 mutant to the lzr/pbx4 mutant will elucidate where and when Pbx proteins, together with their Hox partners, function to control aspects of segment identity such as segment-specific neuronal behavior and neural crest migration.
Meis Genes: Multifunctional Regulators of Hox Activity
The other class of Hox partners is the TALE-homeobox containing Meis proteins, which have more tightly regulated and dynamic expression during hindbrain development (Vlachakis et al., 2000; Waskiewicz et al., 2001; Zerucha and Prince, 2001; Sagerström et al., 2001; Fig. 8B). Hox gain-of-function phenotypes in the zebrafish can be significantly exacerbated when ectopic meis is also supplied, suggesting that Meis protein can be limiting for Hox function (Vlachakis et al., 2001). Conversely, expression of mutant forms of Meis that prevent its DNA binding results in dominant phenotypes similar to those of lzr/pbx4 mutants (Waskiewicz et al., 2001), supporting the idea that trimeric Hox/Pbx/Meis complexes are required for normal hindbrain patterning.
Positive regulatory interactions between Pbx and Meis have been identified that work both at the transcriptional and posttranscriptional levels and that are likely to contribute to Hox activity in the hindbrain. meis mRNA expression closely parallels Hox expression during zebrafish hindbrain development (Sagerstrom et al., 2001; Waskiewicz et al., 2001), and this meis expression is likely to be dependent on normal Hox function, because it is lost in embryos lacking Pbx function (C.B. Moens and A. Waskiewicz, unpublished data). Thus meis genes are targets of Hox/Pbx (whether direct or indirect is not known). Furthermore, Meis function is controlled posttranslationally by Pbx at the level of nuclear localization. A dominant negative form of Pbx4 that binds Meis but not Hox proteins and lacks a nuclear localization motif causes coexpressed Meis to be localized to the cytoplasm and thus inactivated (Choe et al., 2002). Zebrafish embryos expressing this dominant-negative Pbx4 resemble embryos expressing dominant negative Meis (Waskiewicz et al., 2001; Choe et al., 2002) and lzr/pbx4 mutants, confirming that Meis is likely to be required for most Pbx-dependent functions in the developing hindbrain. Although the above data demonstrate that Pbx controls Meis function, the reverse is also true: Meis is able to positively regulate Pbx, and thus Hox function, in this case at the posttranslational level. Although lzr/pbx4 mRNA is expressed ubiquitously, Pbx protein accumulates at higher levels posterior to the r1/2 boundary (Fig. 8A; Pöpperl et al., 2000). The r1/2 boundary is also the anterior limit of both meis and hox expression within the hindbrain (Fig. 8B,C; Prince et al., 1998; Waskiewicz et al., 2001), suggesting that Hox and/or Meis may contribute to Pbx stability. Indeed, ectopically expressed Meis protein can stabilize Pbx protein in vivo, and Meis expression can partially rescue the phenotype caused by loss of zygotic lzr/pbx4 function, presumably by stabilizing maternal stores of Lzr/Pbx4 protein (Waskiewicz et al., 2001). Taken together, these data suggest that a positive regulatory loop operating between Hox genes and their partners functions to enhance Hox function during hindbrain development.
SEGMENTING THE HINDBRAIN
High-resolution analysis of rhombomere-restricted gene expression patterns in chick and zebrafish embryos has shown that initially ragged expression boundaries are gradually transformed into razor sharp boundaries over a period of less than 8 hr (Fig. 9). Over the same period, morphological boundaries appear, followed by the expression of boundary-specific markers (Fig. 2). How this gradual sharpening of expression boundaries occurs, and how boundaries are subsequently maintained, are important questions broadly relevant to our understanding of developmental regionalization. Two mechanisms can be proposed: one in which cells on the “wrong” side of a presumptive boundary (such as the krox-20-expressing cells arrowed in Fig. 9A) move to the other side of that boundary, the other in which cells on the wrong side of a presumptive boundary change their identity to match that of their neighbors. Either mechanism could result in the sharpening of gene expression boundaries and the eventual appearance and maintenance of morphologic boundaries. The two mechanisms need not be mutually exclusive, and indeed, work in the zebrafish has shown that both may contribute to rhombomere boundary formation and maintenance (reviewed Cooke and Moens, 2002).
Repulsive Interactions Between Ephs and ephrins Mediate Cell Sorting in the Hindbrain
Support for the first model, where cells move across boundaries to join the appropriate territory, has come from examining the role of repulsive interactions between Eph receptors and their ligands, the ephrins, in cell sorting at rhombomere boundaries (reviewed in Klein, 1999). Several Ephs and ephrins are expressed in complementary rhombomere-restricted patterns in vertebrate embryos. In the mouse, a simple two-segment periodicity of receptor and ligand expression has been described, with an Eph receptor, EphA4, being a target of krox-20 in r3 and r5, and a ligand, ephrin-B2, being expressed in r2, 4, and 6 (Flenniken et al., 1996; Theil et al., 1998). Importantly, signalling between these classes of Eph and ephrin is cell-contact dependent and bidirectional, with downstream signalling being activated in both ligand and receptor expressing cells (Holland et al., 1996; Bruckner et al., 1997). In the embryo shown in Figure 9A, activation of EphA4 signalling in the misplaced krox-20-expressing cells in r2, and activation of ephrin signalling in the cells immediately surrounding them is hypothesized to cause those cells to move until they are safely within r3, amongst other ephA4-expressing cells and away from the influence of ephrinB2-expressing cells.
Disrupting Eph–ephrin interactions by using a dominant negative form of the receptor disrupts the normal segment-restricted expression of krox-20 in the zebrafish hindbrain (Xu et al., 1995). More recently, Xu et al. (1999) showed that cells ectopically expressing ephrin-B2 sort out of r3 and r5 in mosaic embryos. This behavior is independent of retrograde signalling by the ligand-expressing cells, demonstrating that activation of Eph signalling in neighboring cells is sufficient to expel ligand-expressing cells from Eph-expressing rhombomeres in a nonautonomous manner. Nevertheless, ephrin signalling can contribute to cell sorting, because cells ectopically expressing a truncated form of the Eph-A4 receptor, which can activate signalling in ligand-expressing cells but which blocks signalling in Eph-expressing cells, are similarly excluded from ephrinB2 expressing even-numbered rhombomeres (r2, r4, and r6).
Recently, interactions between zebrafish Ephs and ephrins have been directly implicated in the establishment of the r4/5 and r6/7 boundaries (Cooke et al., 2001). In the valentino mutant, rhombomere boundaries in the posterior hindbrain are absent, and this finding correlates with the loss of EphB4a and ephrinB2a expression interfaces. Specifically, in val-/- embryos, r5- and r6-restricted EphB4a expression is lost, whereas ephrinB2a expression is expanded homogeneously from r4 to r7. In genetic mosaics, val-/- cells, which are unable to appropriately down-regulate ephrinB2a and up-regulate EphB4a in r5 and r6, are excluded from r5 and r6 of a wild-type host embryo (Fig. 6, Moens et al., 1996, 1998; Cooke et al., 2001). Cooke et al. found that by blocking bidirectional Eph-ephrin signalling between val-/- cells and surrounding wild-type cells in these genetic mosaics, they could partially prevent the sorting-out of wild-type and val-/- cells. This result suggested that Eph–ephrin interactions between wild-type and mutant cells cause the mutant cells to be excluded, and more generally that Eph-ephrin interactions in the wild-type embryo could be responsible for the finer-scale cell sorting hypothesized to occur at forming rhombomere boundaries.
Boundary Formation by the Dynamic Regulation of Selector Gene Expression: A Second Mechanism?
The inference from this work is that cells that find themselves in the wrong segment (the val-cells in r5 or r6 of a wild-type host embryo being an exaggerated model for the krox-20-expressing cells “arrowed” in the wild-type embryo shown in Fig. 9A), respond by moving or being pushed into the “correct” rhombomere in an Eph–ephrin dependent manner. However, several lines of evidence suggest that sorting is not the only process that controls the sharpening of rhombomere boundaries, and that plasticity—the ability of a cell to modulate segment identity in response to its surroundings—may also play a role. Lineage marking experiments in the zebrafish hindbrain have demonstrated very little of the anterior-posterior movement that would be predicted if cell sorting contributed exclusively to boundary formation (T. Schilling and J. Clarke, personal communication). Furthermore, transplantation experiments in the fish and mouse have shown that individual cells can readily change their anterior-posterior identity after transplantation to a different rhombomere (Trainor and Krumlauf, 2000; Schilling et al., 2001). Indeed, cells that are known to cross rhombomere boundaries in directed migrations, such as the motor neurons of the facial (nVII) nerve, readily regulate their gene expression patterns (Garel et al., 2000), suggesting that at least some cells within the hindbrain exhibit plasticity as a normal part of their development. What the signalling process is that drives cells in the wrong segment to change their identity remains to be determined. However, it seems clear that the establishment and maintenance of rhombomere boundaries is likely to be a complex process, involving both cell sorting and plasticity.
The zebrafish embryo is tractable to a range of experimental approaches, and in recent years, work in this system has provided significant new insights into vertebrate hindbrain development. Nevertheless, details remain to be uncovered for all of the major questions we have posed: How is the hindbrain territory initially specified? How is it subsequently regionalized? How do the rhombomeres attain all the various aspects of their identity? How is segmentation itself achieved? For instance, we know that specification of hindbrain fates involves signals from the lateral germ ring, but the molecular nature of these signals remains obscure. Similarly, although the importance of local signalling centers in hindbrain regionalization has now been recognized, the precise mechanisms by which these signals are transmitted and interpreted have not been elucidated. We know that Eph–ephrin interactions can control cell sorting between rhombomeres, but the relative importance of cell sorting and cell plasticity in establishing rhombomere boundaries has yet to be determined. Finally, the very earliest steps in setting up region-restricted expression of important transcriptional regulators such as vhnf1 are still unknown.
The tools available for zebrafish studies continue to improve and expand. Our understanding of the anatomy of the embryo has grown hand in hand with increasingly detailed genetic maps, and the complete genome is currently being sequenced. Transgenic lines of fish are providing the realistic prospect of a wide range of tissue and cell-specific markers that are visible in the live embryo, including the hindbrain. Future studies will continue to make use of classic methodologies but with new twists. Cell lineage tracing studies have been simplified by the availability of fluorescent dye uncaging technology (Kozlowski et al., 1997). Forward genetic screens can be targeted to allow isolation of mutants that disrupt very specific processes. New reagents also continue to become available, such as morpholinos to knock-down gene function, and caged RNAs or inducible promoters to allow greater control of ectopic gene expression experiments (Nasevicius and Ekker, 2000; Halloran et al., 2000; Ando et al., 2001). One of the most useful techniques applicable to the zebrafish system is the ability to generate genetic mosaics, and this approach will also continue to provide new information, for example regarding precisely how and when Hox proteins together with their cofactors act to confer regional identity to the hindbrain. Overall, the zebrafish system promises to continue to help us unravel the details of the hindbrain patterning network.
We thank W. Dreiver, G. Hauptmann, J. Clarke, T. Schilling, C. Kimmel, and L. Maves for communicating their unpublished results. We also thank C. Kimmel, L. Maves, J. Clarke, and the members of our laboratories for their critical reading of the manuscript, and David Wilkinson for providing the images of krox20 expression in chick, frog, and mouse embryos. V.E.P. is supported by grants from The Brain Research Foundation, the Whitehall Foundation, the March of Dimes and the NSF. C.B.M. is supported by grants from the March of Dimes, the NSF, and the NIH, and is an Assistant Investigator with the Howard Hughes Medical Institute.