Order and coherence in the fate map of the zebrafish nervous system

The vertebrate central nervous system is the product of inductive interactions between the ectoderm and mesoderm. The patterning of the neurectoderm is thought to be a progressive process. At the end of gastrulation, anteroposterior and dorsoventral axes of the neurectoderm are organized in broad domains that can be defined morphologically and molecularly. These are further refined as neurulation proceeds, as evident, for example, in the formation of rhombomeres (Keynes and Lumsden, 1990), and the acquisition of dorsal and ventral neural tube fates (Ruiz i Altaba and Jessel, 1993). Any detailed investigation of the regional patterning of the neural plate requires knowledge of the positions of various primordia of the CNS before, during and after the process. Such information is contained in fate maps. These depict, at a specific developmental stage, what the different regions of an embryo will become during normal development (Slack, 1991). Fate maps of different vertebrates have been invaluable in aiding the design of grafting and explant experiments, as well as the interpretation of molecular markers (for example, chick: Couly and Le Douarin, 1988; mouse: Lawson et al., 1991; Xenopus: Eagleson and Harris, 1990; fish: Oppenheimer, 1947; Ballard, 1973; Kimmel et al., 1990). A series of fate maps for different stages may also provide information about the cell and tissue rearrangements of gastrulation and neurulation (Keller, 1975; Keller, 1976; Lawson et al., 1991; Selleck and Stern, 1991).

The zebrafish (Danio rerio; formerly known as Brachydanio rerio) provides several unique characteristics suitable for the study of vertebrate neural-induction and -patterning mechanisms. Its optical transparency permits direct observations of tissue movements, cell behaviors, and cell interactions during neurogenesis (Warga and Kimmel, 1990; Kimmel et al., 1994). Its amenability to genetic analysis has already resulted in the isolation of a variety of mutants (Mullins et al., 1994; Driever et al., 1994) that will be powerful tools for examining developmental mechanisms. At present, the results of experimental studies and developmental mutants must be interpreted with reference to Xenopus, in which most modern studies of neural induction and axis formation have been performed (reviewed in Kessel and Melton, 1994; Donaich, 1993; Ruiz-i-Altaba, 1993; Slack and Tannahill, 1992). While the expression patterns of a variety of gene homologues suggest similarities in the components of the neural-patterning machinery of Xenopus, zebrafish, and other vertebrates, it is unclear if the analogy can be extended to the workings of this machinery. For example, there are clear differences in the early cellular dynamics between species: in Xenopus, blastomeres change neighbors slowly during cleavage and blastula stages (Wetts and Fraser, 1989); in zebrafish, cell mixing has an earlier onset and is more dramatic (Kimmel and Law, 1985; Warga and Kimmel, 1990; Helde et al., 1994). Such differences might require the operation of distinct signaling mechanisms to pattern the embryo. Furthermore, significant differences exist in tissue distribution within their respective organizer regions. For example, in Xenopus, the organizer (dorsal lip) contains precursors of only mesodermal (notochord, somite) and endodermal tissues (Keller, 1975, 1976); in zebrafish, recent evidence suggests that the organizer (shield) encompasses an intermixed group of mesodermal, endodermal and neural precursors (Shih and Fraser, 1995). Thus, the results of experimental studies and the phenotypes of mutants cannot be interpreted solely by analogy with data from Xenopus; instead, a detailed description of the zebrafish neural fate map is required.

In this study, we present detailed regional neural fate maps of the zebrafish at the beginning and end of gastrulation (6 hours and 10 hours of development, respectively, both key stages in zebrafish neurogenesis), and use video time-lapse microscopy to observe directly the cell rearrangements that transform one fate map to the other. Gastrulation begins at 50% epiboly (about 5.2 hours); by 6 hours, the embryonic shield has become morphologically pronounced, enabling unambiguous identification of ‘dorsal’ (staging table in Westerfield, 1994). At 10 hours, the epiblast of the zebrafish embryo completely covers the yolk, marking the end of gastrulation (Westerfield, 1994). In the period between 6 and 10 hours of development, several genes implicated in embryonic patterning and signaling, such as goosecoid (Stachel et al., 1993; Schulte-Merker et al., 1994), axial (Strahle et al., 1993), hedgehog (Krauss et al., 1993) and the rtk genes (Xu et al., 1994), initiate their expression in the dorsal side of the embryo. Although the presumptive neurectoderm appears outwardly homogeneous at these stages, genes believed to play roles in anteroposterior patterning including pax6 (Krauss et al., 1991; Puschel et al., 1992), engrailed (Hatta et al., 1991a), and Krox20 (Oxtoby and Jowett, 1993) display regionally restricted expression patterns. As such findings suggest that events critical to the patterning of the neurectoderm take place between 6 and 10 hours, a knowledge of the neural fate maps and relevant cell movements at these times is a critical first step towards understanding the formation of the zebrafish nervous system.

Zebrafish (Danio rerio) were maintained and handled essentially according to procedures described in the Zebrafish Book (Westerfield, 1994). Embryos were obtained by natural spawning. Staging was performed by morphology (Westerfield, 1994). Stages were expressed as hours postfertilization at 28.5°C (h). In general, cell labeling was performed at either 6 or 10 hours (h), and scoring of regional fate between 19-36 hours.

For the 6h fate map, we labeled cells between the ‘early-shield stage,’ defined as the earliest time at which the shield is visible, about 5.5h (Shih and Fraser, 1995), and the ‘shield stage’ (at 6h). The positions of the labeled cells were assessed at 6h. Between 5.5h and 6h, cell movement is minimal in the outer layer of the ectoderm (Shih and Fraser, 1995). Embryos were dechorionated with watchmaker forceps at the germ-ring stage, approximately 5.2h. Using a hair loop, embryos were oriented (typically animal pole up) on a thin layer of 3% methyl-cellulose in 30% Danieau solution (full-strength Danieau solution is a modified Niu-Twitty solution with double the normal CaCl₂ concentration; Shih and Fraser, 1995).

Single cells were injected iontophoretically with either 3×10³M_r fluorescein dextran or 3×10³M_r rhodamine dextran (Molecular Probes D-3306 and D-3308). To reduce toxicity, these dyes were cleaned of impurities with a 3×10³M_r size-exclusion HPLC column. Iontophoretic injections were performed essentially as described previously (Wetts and Fraser, 1988). Micropipettes, pulled from AlSi thinwalled capillaries (A-M systems), were filled at their tips with a fluorescent dextran solution (about 100 mg/ml) and then back-filled with 1.2 M LiCl. Injection was accomplished by passing a positive current of 4-5 nA for 3-10 seconds. Each injection was immediately verified by viewing the embryo under epi-fluorescence optics.

The dye injections were made into the cells that the pipette first encountered in the blastoderm, as judged by the membrane potential recorded through the pipette. Thus, even though the deep cell layer of the 6h embryo is between 3-4 cells thick at the animal pole (5-6 cells thick close to the germ ring), most injections were within its outermost 1-2 cells. The injections were intended to label single cells; however, occasionally 2 or 3 closely placed cells were labeled after a single injection, presumably due to cytoplasmic bridges among cells. At the time of initial scoring, animals containing florescent debris or labeled periderm were discarded. Those containing 1-3 distinctly labeled cells were kept for later analysis.

The 10h fate map data were collected using two different approaches: in the first, we targeted our injections into particular regions at 6h, scored the positions of labeled cells immediately, and then re-scored the same embryo at 10h. This enabled us to compare the relative positions of a labeled cell at 6h and its progeny at 10h. In the second, we labeled cells at 10h and scored cell position immediately. These two approaches gave essentially identical fate maps, so they were combined in the final map presented here.

Since injections were performed on either the left or right sides of the embryos, we examined the assumption of left-right symmetry in the 6h embryo. We calculated the percentage of each tissue-fate obtained from injections to either the left or the right side. The two sides displayed nearly identical tissue-fate distributions. (Data not shown.) This assures us that the assumption of left-right symmetry is valid for this study; therefore, all fate maps are presented by reflecting each point across the midline.

Observation and documentation were performed with a Zeiss Axioplan epifluorescence microscope equipped with a light-intensifying camera (Hamamatsu SIT C2400), and an image processor (Imaging Technology 151) controlled by the VidIm software package (Belford, Stollberg and Fraser, unpublished data). Light exposure to the labeled cells was minimized by using computer-controlled electronic shutters and neutral-density filters in the epi-illumination pathway (OD = 1.0 to 2.5). Each recorded image was collected by averaging 8-32 frames. All data were recorded onto optical disc (Panasonic optical disc recorder, OMDR 3038). Adobe Photoshop was used to enhance contrast of and apply labels to images presented in this paper.

Immediately after each injection session, the positions of the injected cells were recorded. Because the accuracy of the fate maps is critically dependent on the precision with which we can identify the location of each injected cell, we took particular care to ensure that the embryos were properly oriented before being documented by video microscopy. For the 6h scoring, we first recorded an animalpole view. The embryo was oriented so that the germ ring appeared to have the same ‘thickness’ around its circumference; this ensured that the animal pole was uppermost (see Fig. 1A). We then documented the positions of the cells in profile by rotating the embryo 90° about its dorsal-ventral axis until half of the shield was visible on one side. An acceptable profile view was attained when the edge of the blastoderm was oriented perpendicular to the microscope slide; this can be assessed by focusing at the upper and lower surfaces of the embryo. An example of a typical profile view is shown in Fig. 1B.

For the 10h scoring, we again recorded the animal-pole view first. Since no significant structure existed in the epiblast at this stage, we utilized features in the underlying hypoblast for orientation. We centered the dorsal midline by locating the crescent shape formed by the advancing hypoblast (white arc in Fig. 1C) and the thickening of the forming notochord (arrowheads in Fig. 1C). Thereafter, for the profile-view documentation, we rotated the embryo about its d-v axis until the base of the embryo appeared straight, and half of the neural axis was visible on one side (for example, see Fig. 1D).

After initial documentation, embryos were raised individually in multi-well culture plates until 19-36h of development, when each embryo was examined in several views (side, ventral, top and headon, using objectives up to 40×) to ensure correct identification of the positions of all labeled progeny. Occasionally, when a cell was in a particularly interesting region or was located at or between a boundary of tissues, a second screening was performed at a later time to better define the labeled cell’s location.

The positions of injected cells are represented according to the following conventions.

As illustrated in Fig. 1A, maps were drawn onto a polar coordinate grid, with each concentric circle marking 10° in latitude and each straight line 10° in longitude. The center of the embryonic shield was defined as 0° longitude, the animal pole defined as 90° latitude. In Fig. 1A, the cells were located between 60°-70° longitude, and 30°-35° latitude. (Due to embryo curvature, 0° -10° latitude is not distinguishable from the top; therefore the outermost line = 0°-10° latitude.)

In the profile maps (Fig. 1B), each horizontal line represented 10° of latitude, and each curved line 10° of longitude. 0° longitude was at the right of each map (the location of the shield), and 0° latitude was the edge of the blastoderm. The cell in Fig. 1B was at 10°-20° longitude and 30° latitude. (Again, due to embryo curvature, 0°-10° and 170°-180° longitude are represented by the same line.)

We defined our coordinate axes by drawing (1) a straight-line tangent to the crescent shape made by the advancing hypoblast (highlighted by the white arc), and (2) a line centered on the thickening of the forming notochord (marked by arrows) (Fig. 1C). The intersection of these two lines defined the origin (O) of the coordinate system used to catalog each cell’s position. As raw data, we measured Δ, the angle from the dorsal midline (0°); and r, the distance from the origin, which was subsequently scaled to the average diameter derived from 40 randomly selected embryos. Scaling the data in this way does not affect the spatial relationships among the data points, but permits the data for all cases to be presented on a single diagram. The distance between the origin and the edge of the embryo was divided into 10 equal units; these are represented by the arcs in Fig. 1C. (It is important to note that these lines do not represent latitude measurements, and are not equivalent to those in the 6h map.)

The coordinate of each labeled cell is defined by h (distance from the base of the embryo), and w (distance from the dorsal midline) (Fig. 1D). To standardize the data from embryos of various profile shapes, we defined a ‘model embryo profile,’ computed using a random set of 40 embryos, and plotted the scaled data from each case on this standard profile.

Both top and profile views of the fate maps are presented, since together these two-dimensional views give a more complete representation of the three-dimensional embryo. The 6h top view provides good resolution in a large portion of the blastula cap, where most of the data points are located. It allows for a better appreciation of medial-lateral differences; however, its resolution below 30° latitude is limited, due mainly to the curvature of the embryo. The profile view (Fig. 1B) complements the top view by providing better resolution around the equator region (0°-30° latitude) and accentuating animalvegetal differences. Its resolution at the pole and its accuracy near the axis are limited.

We present the top view of the 10h fate map with the origin off center (Fig. 1C), since this view permits a larger fraction of the neural axis to be represented with good resolution than is possible from a polar view. Beyond seven units from the center, the curvature of the embryo made positions that were in different focal planes appear to be at the same point on the map, thus affecting the resolution in this area. The 10h profile view (Fig. 1D), while giving restricted resolution at the pole, emphasizes animal-vegetal separations between cells, and allows visualization of cells located below mid-embryo that are otherwise unseen from the top.

In the final data set, we tolerated deviations from ‘ideal’ alignment of up to 5° in profile views, and 5° in top views . During the course of data analysis, we rejected more than 50 embryos due to bad alignment in the initial documentation. With these constraints in mind, we estimated the maximum error in all the position measurements to be less than 10% of the embryo’s radius. For example, in Fig. 1B, if the embryo were to be rotated +5° around its animal-vegetal axis from its ideal profile position, then the longitude of the labeled cell would be misjudged by about 5°; this translates to about 7% of the embryo’s radius.

Time-lapse video recordings were made using a SIT camera, as described above, at 26-28°C. All films began at shield stage. Epiblast cells were labeled by 1-4 injections as above, at either 30% epiboly or shield stage. Embryos injected at 30% epiboly contained 1-4 small groups of labeled cells; those labeled at shield stage had several individual marked cells. A single embryo was placed in a pyramid-shaped well precut to a thin layer of Sylgard (Corning) at the bottom of a Petri dish filled with 30% Danieau solution. This arrangement kept the embryo stable, yet permitted it to move freely during morphogenesis while we recorded the relative movement of labeled cells. Bright-field and fluorescent images were captured simultaneously, usually once every four minutes. Neutral-density filters and the short light exposure were used to minimize possible photodamage to labeled cells. We re-focused periodically, and reoriented the embryo only when absolutely necessary, since we wanted to observe continuous cell movements.

The goal of this study was to investigate in detail the organization of the zebrafish blastoderm at 6h and 10h of development. To construct the fate maps, we divided the descendants of cells labeled at these times into six categories (Fig. 2): telencephalon (T), including the olfactory placode; diencephalon (D), excluding the neural retina; retina (R); midbrain (M); hindbrain (H); and other (O), including spinal cord neurons and such non-neural-tube derivatives as somite, neural crest, and epidermal (including placodal) derivatives. We identified putative neural crest cells by their distinctive morphology of migrating cells, and by their locations, such as within a branchial arch. We included the olfactory placode in the telencephalon category because it is too closely associated with the brain proper at the time of scoring to be distinguished reliably.

Table 1 shows the composition of our data. A total of 224 individual embryos were used to construct the 6h fate map; an overlapping set of 137 embryos was included for the 10h map. As discussed in Materials and Methods, although we aimed to inject only one cell per embryo, in some instances two or more neighboring cells were labeled. Since the goal of this study was a fate map rather than a clonal analysis, we included embryos bearing two or more closely placed labeled cells. It does not appear that the inclusion of these cases blurs the fate map borders. In support of this, Table 2 lists the percentage of embryos with labeled descendants in a single brain region, analyzed with respect to the number of cells initially labeled. The majority of the embryos have labeled cells in only one region, even when 3 or more cells were injected. At 10h, the number of injections yielding a single regional fate appears to be slightly increased. The data in Table 2 suggest that there are regions of fate homogeneity (i.e., single-fate domains) within the dorsal blastoderm as early as 6h, because neighboring cells tend to give rise to the same regional fate; and that these regions undergo some refinement in organization during gastrulation. The fate maps presented below explore the topo-graphic nature of this blastoderm organization.

At 6h of development, the dorsal side becomes morphologically distinct with the appearance of the embryonic shield. Figs 3 and 4 show the neural fate map at 6h, in animal-pole and profile views, respectively. The progenitors of the different brain regions, represented by different colors, appear somewhat intermixed, especially in profile view. For example, progenitors of telencephalon (red), diencephalon (black), retina (green), and midbrain (orange) can all be found within the area at 55°-65° latitude and 30°-70° longitude (Figs 3 and 4, top). On closer examination, a significant order becomes apparent in the form of domains that give rise to only one brain region. For example, in the top view (Fig. 3), the domain between ±20° longitude and 65°-80° latitude is almost entirely occupied by retinal precursors (green); whereas, in the region further from the pole, at 40°-50° latitude, only diencephalic precursors (black) are seen. In the profile map (Fig 4), between 70°-90° longitude, a pure hindbrain domain is located at 20°-40° latitude. At the dorsal midline, the map shows orderly overlaps between telencephalic and retinal precursors, diencephalic and retinal precursors, and diencephalic and midbrain precursors near 80°, 60°, and 30° latitude, respectively (Fig. 3, top). Overlaps of more than two sub-regions (more than two colors) appear to be confined to the region of 40°-70° longitude.

Plotting the progenitors of each brain region separately (Figs 3 and 4, bottom panels) permits a more detailed evaluation of the fate map. From the profile views (Fig. 4), one can see that forebrain precursors are located closer to the animal pole, while midbrain and hindbrain precursors are found progressively closer to the margin, as reported previously (Kimmel et al., 1990). However, the coherence of each brain sub-region has not been appreciated previously (Figs 3 and 4, bottom panel). For example, in the top view, 90% of telencephalic progenitors are found within 20° latitude from the animal pole. The retinal precursors are gathered into a single coherent domain at 50°-90° latitude and ±90° longitude. The hindbrain precursors appear split into two domains, at 30°-110° longitude and 10°-50° latitude on either side of the midline; however, because we did not map the ±30° longitude region below 15° latitude, we cannot rule out a midline connection between these two domains. Interestingly, the midbrain domain appears to bracket the diencephalic domain at its lateral boundary (50°-70° longitude above 60° latitude), and its posterior boundary (±40° longitude at 40° latitude). In short, a considerable amount of predictable organization is seen in the presumptive neurectoderm at 6h.

The presumptive ectoderm shows some predictable order as well. Cells that give rise to ectodermal derivatives (yellow) are located throughout the ventral half of the blastoderm at 6h. In the region 70°-110° from the midline (that is, the more dorsal portion of the ectodermal domain), we mainly find cranial neural crest, head epidermis, and placodal derivatives. Further lateral (beyond 110° longitude), we find trunk neural crest, trunk epidermis and tail epidermis. An animal-vegetal order is seen, such that cranial neural crest and epidermis are found closer to the animal pole (data not shown). Since we have focused on the more anterior brain regions, only a few spinal cord precursors (light blue) were labeled. From these few cases we can deduce that the spinal cord domain is likely to reside below 30° latitude at 6h. We cannot determine the degree of overlap between spinal cord cells (light blue) and somite progenitors (purple), although the limited data suggests that they may be somewhat intermixed.

Our second fate map was made at 10h, the transition point between gastrulation and neurulation. Gastrulation and epiboly end with the complete coverage of the yolk cell by the epiblast cells (Warga and Kimmel, 1990). A dorsal thickening (the future neural axis) can be seen in profile (Fig. 1D), with the tail bud forming at the posterior tip. In the top view (Fig. 1C), two features of the hypoblast can be used as landmarks: the forming notochord, and a crescent shape composed of advancing axial hypoblast cells. This same shape is marked by, for example, goosecoid (Stachel et al., 1993) and rtk3 (Xu et al., 1994) in 10h embryos. Figs 5 and 6 show the 10h fate map in top and profile views.

Comparison of 6h and 10h fate maps shows a refinement in the ordering of the progenitor groups. At 10h, the entire neurectoderm has narrowed from 6h, placing the neural/epidermal border 40°-70° from the midline, consistent with previous observations in the anterior spinal region at the two-somite stage (Schmitz et al., 1993). The amount of convergence from 6h to 10h varies with axial level. The lateral limit of the retinal domain narrows 30° during this time interval, while the hindbrain domain narrows by 60°. These movements have brought the CNS regions into roughly the same order as that found later in the neural tube. Beginning at the front of the axis and moving caudally, one sees the telencephalic progenitors at 0-4 units, retina at 1-5 units, diencephalon at 3-7 units, midbrain at 5-8 units, and hindbrain at 7-10 units from the origin (Fig. 5). In the profile view (Fig. 6), the resolution of the midbrain/hindbrain region is improved. Here, these two domains appear even more distinct. Regions of overlap exist at the border of each domain, particularly at the lateral edges. For example, retina, diencephalon and midbrain progenitors overlap in the region 40° from the midline at 5-6 units; however, domains of single fate are more readily seen, such as between 5-6 units and 7.5-9.0 units, where only diencephalon and midbrain progenitors are located, respectively.

The shapes of three domains have changed significantly between 6h and 10h. The telencephalic domain, mainly rectangular in shape at 6h, has developed a bend in the center by 10h (compare Figs 3T and 5T). Initially somewhat split across the midline at 6h, both the hindbrain and the midbrain domains have coalesced into single, coherent masses directly posterior to the forebrain. (Compare Figs 3M and 3H with Figs 5M and 5H, and see below.) These changes in the progenitors’ topography suggest a predictable pattern of morphogenetic movements, which we shall explore further below.

To explore any further subdivision within each domain, we analyzed the location of the precursors contributing to anterior vs. posterior portions of each brain region. The criteria for anterior (red) and posterior (green) division is schematized in Fig. 7 (top). A clear AP polarity is seen in the forebrain (Fig. 7A): the precursors of the telencephalon are clustered closer to the animal pole, thus more anterior in the future neural axis, than those of the diencephalon at both 6 and 10h. Within the midbrain (Fig. 7B), anterior progenitors are found both closer to the animal pole (at 30°-90° longitudinal region) and closer to the midline (between ±20° longitude) than posterior ones; by 10h the anterior progenitors in the midbrain are uniformly anterior to the posterior progenitors. This predicts that the 6h anterior progenitors at the midline advance rapidly toward the animal pole, permitting more progenitors to join behind. In the hindbrain at 6h, anterior progenitors are, in general, closer to the animal pole (Fig. 7C), and the order is refined significantly by 10h. The organization of anterior and posterior progenitors in the forebrain appears to be more advanced, as the AP order is largely the same at 6 and 10h. However, as with the midbrain, there is some indication that cells nearer the dorsal midline advance more rapidly.

The fate maps show significant segregation between dorsal (dark blue) and ventral (orange) neural tube precursors (Fig. 8). Within the forebrain (Fig. 8A), midbrain and hindbrain (Fig. 8B), the ventral progenitors are positioned closer to the midline than are the dorsal progenitors (Fig. 8). This arrangement of progenitors in the dorsal blastoderm agrees with observations in a variety of species, ranging from those of the neural area of Salmo gairdneri (Ballard, 1973), to the more recent results in zebrafish anterior spinal cord (Papan and Campos-Ortega, 1994). Interestingly, those ventral forebrain progenitors (Fig. 8A) close to the margin and at the midline at 6h (especially below 60° latitude) appear to have advanced significantly towards the pole when compared to their laterally placed neighbors (dorsal progenitors at 20°-40° longitude). A similar but less-pronounced change is also seen in the midbrain and hindbrain regions (Fig. 8B, 6h). This rearrangement is consistent with that seen in the AP analysis (Fig. 7), and suggests regional differences in cell movements that will be explored below.

Patterning in the retinal domain was analyzed separately. Nasal and temporal retina halves were defined by the axis demarcated by the choroidal fissure and posterior groove (Schmitt and Dowling, 1994); dorsal and ventral divisions were defined at 90° to this (Fig. 9 schematic). In most embryos, single progenitors gave rise to cells that remained clustered in the retina. The few cases in which labeled cells were found in both anterior and posterior, or both dorsal and ventral, halves of the retina were excluded from this analysis (less than 10 cases each). At both 6h and 10h, the nasal (anterior) and temporal (posterior) progenitors of the retina are distinct (Fig. 9A), and aligned with the AP axis of the forebrain (Fig. 7A). However, the arrangement of DV progenitors is quite different (Fig. 9B). Dorsal progenitors (dark blue) are not exclusively lateral to ventral progenitors (orange). At 10h, the clustering of the dark blue cells within the field of orange cells may suggest that these progenitors are arranged in a V-D-V pattern, aligned with the AP axis of the neural axis (Fig. 9B). In contrast to the rest of the neural axis, there is no difference in the predicted motions of cells close to the midline versus those more lateral to it.

The surprising degree of organization revealed by the above analyses suggests that neural precursors undergo slow mixing and/or orderly cell movements throughout gastrulation. In our attempt to visualize the rearrangement that transforms one map to the other, we estimated the movements of labeled cells between 6h and 10h. This was enabled by the fact that for a portion of the sampled embryos, we had scored the position of labeled cells at both time points. A representative set of these trajectories is shown in Fig. 10A. Each arrow represents the movement of a cell: the blunt end of each arrow marks the position of the cell at 6h; the arrowhead points to its position at 10h. The entire arrow ‘field’ thus represents the predicted flow of movement of the epiblast during gastrulation. Note that cells closer to the midline tend to advance more rapidly in the anterior direction; cells located further from the midline tend to first move posteriorly, and then undergo a flanking movement to join in the medial anterior flow (Fig. 10A). Although local order is observed, as expected in the refinement of the brain-region fate map with developmental time (discussed above) the cells appear to move as individuals. This is most readily seen near the midline (top view), where neighboring trajectories both converge and diverge.

To investigate the dynamics of individual cells predicted by Fig. 10A, we performed time-lapse recording of embryos (n > 20); in these, several cells in each embryos were marked, and their relative positions were followed directly using low-light-level video microscopy. Because the entire blastoderm is in motion during these stages of morphogenesis, there are no fixed reference points within the embryo. The presence of two or more labeled cells therefore permits us to follow their relative changes in cell position over time.

The time-lapse sequences confirm the general pattern of cell movement deduced both from the static trajectory analysis and from the arrangement of the progenitors examined above. In Fig. 10B, of the two cells at the same latitude, the more medial cell (m) advanced to the animal pole sooner and maintained a more-anterior position within the neural axis than the lateral cell (l). From a profile perspective (Fig. 10C), this same relative motion of axially and laterally located cell groups can be followed. As epiboly proceeds, the axial group (three cells) advances towards the animal pole, while the lateral group as a whole (four cells, initially closer to the yolk cell) slide to a more posterior position relative to the medial group. By 10h, these two groups are aligned in a straight line, with the axial group at the anterior portion of the line, as predicted by the AP arrangements of hindbrain progenitors (Fig. 7).

Time-lapse sequences offer a feel for the dynamics of cell movement not revealed by static analyses. For example, it is possible to observe that laterally located cells, which move in a broad, sweeping movement caudally towards the midline, initiate active convergence movements only after they are above or near the rapidly advancing hypoblast. (Note relation between labeled cells and hypoblast lateral boundary in Fig. 10B3 and B4.) Furthermore, in films in which we recorded small groups of cells, AP order was generally maintained. For example, note that the four cells in the lateral group (Fig. 10C) maintained their relative AP order throughout morphogenesis. The cells near the midline, perhaps under the influence of the underlying hypoblast, underwent different morphogenetic movement than cells more laterally located. From the 6h fate map (Fig. 3), this midline population corresponds predominantly to forebrain progenitors. A smooth transition between the 6h and 10h fate maps can thus be envisioned: during gastrulation, the forebrain progenitors advance first to maintain their anterior positions, whereas the midbrain/hindbrain/ spinal cord progenitors sweep posteriorly and then medially to fill in the more posterior positions. Both the DV organization of the neural axis and the AP order of the forebrain are maintained as the midbrain and hindbrain domains ‘zipper’ up at the midline and refine their AP organization (Fig. 11).

Judging from the relative distance between labeled axial cells and the limit of the hypoblast crescent underneath, it appears that the axial epiblast cells do not advance much beyond the axial hypoblast. (Data not shown.) Instead, as extension continues, the anterior tip of the neural region thickens, resulting in a structure that resembles the cephalic neural fold of other vertebrates at the beginning of neurulation.

In this paper, we present a detailed fate map of the anterior nervous system of zebrafish at two stages: the beginning (6h) and end (10h) of gastrulation (summarized in Fig. 11). The 6h fate map (Fig. 11A) demonstrates that at early gastrulation, discernible domains in the dorsal blastoderm can be assigned to major subdivisions of the brain. These domains become progressively distinct as gastrulation concludes at 10h (Fig. 11B). Using time-lapse video microscopy, we observe an orderly transition between the two fate maps that is mediated by a predictable rearrangement of the dorsal blastoderm cells (Fig. 10). This rearrangement consists of two broad components: a continuous anterior flow of the forebrain progenitor population (especially those nearer the dorsal midline, overlying the advancing hypoblast), and a concurrent translocation/rotation of the midbrain-hindbrain-spinal cord domains from their lateral positions to a location posterior to the forebrain domain (Fig. 10).

By sampling several-fold more cells than a previous study (Kimmel et al., 1990), we found a significant degree of order and predictability in the neuronal fate map. Our ability to outline the approximate extent of each brain subdivision at 6h by no means indicates that these subdivisions are completely distinct and separate. (Note shaded regions of overlap in Fig. 11.) As an indicator of the order, those progenitor cells in the 6h blastoderm that contributed to more than one brain subdivision were only found in adjacent subdivisions; a cell having descendants in both telencephalon and hindbrain, for instance, was never observed. Furthermore, dual-fate progenitors were invariably located at the border of adjacent domains on the fate map. Another indicator of order in the fate map was that the majority of the embryos had labeled cells in only one brain domain, regardless of whether the embryo had initially had one or more closely spaced cells labeled (Table 2). It is therefore very likely that our 6h fate map overestimates the extent of overlap between different brain regions, as embryo variability and measurement error would blur the distinctions. Closer examination at the bordering regions and grafting between regions will be needed before any definitive conclusions can be drawn about the relative roles of limited cell mixing and clonal restrictions at region boundaries in generating such an ordered fate map.

The organization of the zebrafish gastrula map appears to differ from that of another teleost, Salmo gairdneri (Ballard, 1973). In Salmo, the brain regions are extensively overlapped at the onset of gastrulation, such that forebrain, midbrain, and hindbrain, as well as mesodermal tissues, appear to be arranged in layers (Ballard, 1973). In the zebrafish early gastrula, the brain regions and other mesodermal derivatives are reasonably well separated laterally across the blastoderm (Fig. 11A; Kimmel et al., 1990). As we have only mapped the outer 2-3 cells of the blastoderm, we may have missed any layered organization in the deeper layers. However, morphological differences between the two species may be a bigger factor. In zebrafish, the entire blastula cap contributes to the embryo proper; in Salmo, the dorsal half of the cap contains the entire embryo. Furthermore, the zebrafish gastrulates at 50% epiboly, when the epiblast has thinned to about 2-3 cells thick at the pole and 5-6 cells at the margin; Salmo, much like Fundulus (Trinkaus, 1984), begins to gastrulate around 30% epiboly, when the epiblast is several layers thick. Thus, it is not surprising that the different primordia of the zebrafish brain are more spread out than those of Salmo when gastrulation begins. Perhaps fate mapping the zebrafish at 30% epiboly, if the embryo can be properly oriented, would reveal any layered organization that might be present.

There are both similarities and differences between the zebrafish 6h fate map and the Xenopus gastrula fate map. This frog fate map was derived by backtracing individual cells in time-lapse recording from the mid-neurula map (Keller et al., 1992a). In both maps, the forebrain region occupies about two-thirds the area of the presumptive neural plate. Furthermore, the axial brain regions are compressed animal-vegetally such that the forebrain is relatively close to the dorsal lip region in Xenopus, and almost next to the shield in zebrafish. In contrast to Xenopus, in which the spinal cord progenitors stretch laterally from the midline, those in zebrafish are positioned almost exclusively laterally (with the exception of those few located within the shield; Shih and Fraser, 1995). The 10h zebrafish neural fate map resembles the neurula fate map of Xenopus (Eagleson and Harris, 1990) and chick (Couly and Le Douarin, 1988), in that the brain regions are aligned in the expected anteroposterior order of the eventual neural tube. A notable difference is the shape of the zebrafish telencephalic domain, which is not a pair of split ‘lateral’ structures, as in chick and Xenopus. In fish, the diencephalon does not lie most anterior. It is important to note that, strictly speaking, a 10h zebrafish is not a neurula, since ‘neural plate’ is not seen until 1 hr later (Schmitz et al., 1993). It may be that, given the forward-flow movement observed in the axial epiblast cells (forebrain progenitors), the telencephalic anlage is pushed aside by diencephalic progenitors that adopt an anterior midline position (Fig. 10B; and unpublished observations).

In analyzing possible patterning within each brain subdivision, we find that AP ordering is pronounced in the forebrain region at 6h, while it is gradually refined in the midbrain and hindbrain region, in agreement with the observed morpho-genetic movements. AP-neighbor relationships would be expected to be maintained among the forebrain progenitors, because they move uniformly forward, beginning at 6h. Among the midbrain-hindbrain progenitors, however, greater neighbor exchanges are expected, since they must first converge towards the midline before aligning with the orientation of the axis (Fig. 10). In contrast, DV patterning is evident throughout the axis from the start of gastrulation: dorsal progenitors of each region are always located lateral to ventral progenitors. Although this patterning is temptingly suggestive of a primary neurulation mechanism (Schmitz et al., 1993), we cannot at present determine the neurulation process of the zebrafish head region based on this pattern alone. The head region of the neural keel could be formed by an orderly infolding of cells, with the lateral cells giving rise to the dorsal brain structures. Another possibility is that lateral cells, moving as individuals, are limited to taking dorsal positions simply because the more-ventral ones are already occupied. Clearly, in vivo time-lapse analysis, as well as ultrastructural studies, are needed to resolve this issue.

An interesting aspect of the fate map is that both neural retinas map to a single coherent region at the anteromedial portion of the forebrain (Figs 3R, 5R). In other teleosts, such as Salmo, a single retinal domain is also seen in the gastrula (Ballard, 1973). In our study, a single cell can give rise to progeny in both the ipsilateral and contralateral retinas (data not shown), indicating that both retinas originate from this singular domain. Similar crossing has been observed in Xenopus (Jacobson and Hirose, 1978), implicating some coherence of the retinal domains at an early developmental stage, despite the split domains seen in the neurula fate map (Eagleson and Harris, 1990). Our data also shows that both the AP and the DV axes of the retinal domains are oriented parallel to the neural axis. In the Xenopus neurula, the DV axis of the retina is arranged in the expected lateral-to-medial fashion (Eagleson and Harris, 1990). The difference in axis arrangement may reflect dissimilarities in the morphogenetic mechanisms that drive the optic cup formation in amphibians and teleosts (Schmitt and Dowling, 1994; Reyer, 1977). At a later stage, the retinas have been shown to undergo rotations with respect to the neural axis (Schmitt and Dowling, 1994). It will be of interest to investigate if, and to what extent, the cellular interactions that pattern the AP and DV axis of the neural tube occur within the retina.

The fate maps reveal interesting tissue juxtapositions and movements that may have implications in the interpretation of mutant phenotypes. For example, cyclopia, a notable defect in the cyclops mutant, is believed to result from a massive deletion of the ventral forebrain (Hatta et al., 1994), perhaps from impaired cell proliferation (Hatta et al., 1991b; Strahle et al., 1993). Because both retinas originate from a single contiguous domain in our 6h fate map, the anlage must be split during morphogenesis to make two retinas. It appears that the ventral diencephalic progenitors physically split the retinal domain in two. Thus, it seems possible that some mutations leading to cyclopia might act to impair motility of the ventral diencephalic progenitors during gastrulation, resulting in a failure of the retinal anlage to be bifurcated. This scenario does not preclude a change in the cell-state specification of diencephalic progenitors in some mutants (Hatta et al., 1994), but suggests an alternative hypothesis, focused on cell motility, that can now be tested.

The fate maps presented here provide details of the progenitor topology, thus permitting dissection of possible neural patterning mechanisms in the zebrafish. In vertebrates, it is generally believed that the organizer plays a pivotal role in neural induction and patterning (Spemann, 1938; Kessler and Melton, 1994, and references therein). Based on the expression patterns of such markers as goosecoid (Stachel et al., 1993; Schulte-Merker et al., 1994), the embryonic shield is believed to be the organizer of the zebrafish, although its ability to organize a complete axis remains to be demonstrated (Oppenheimer, 1936; J. Shih, personal communications); nonetheless, it may be instructive to speculate about possible patterning scenarios using the ‘constraints’ imposed by the fate maps reported here. At present, our fate map is compatible with both planar and vertical patterning mechanisms (Donaich, 1993; Ruiz-i-Altaba, 1993). For instance, given the close proximity of the forebrain, midbrain, and hindbrain to the shield, one can envision the involvement of one or more short-range planar signals from the organizer (Fig. 11A). Similar schemes, based mainly on fate-mapping and cell-movement data, have been proposed as part of the neural-induction mechanism in Xenopus (Keller et al., 1992a,b). However, vertical inductive interactions between axial hypoblast and epiblast could also be a significant patterning influence. It is interesting to note that the timing of initial forward motility exhibited by laterally located cells seems to coincide with their becoming close to the axial hypoblast (Panel 10B). Whether or not this change in the behavior of the epiblast cells is a result of some inductive interactions with the hypoblast remains to be investigated.

The organization displayed in the 6h fate map could also be taken as support for the existence of regional identities in the blastoderm prior to gastrulation. The slow, orderly cell mixing in the zebrafish epiblast layer (Fig. 10; see Wetts and Fraser, 1989) could, in effect, establish a default ‘pattern’, with cells closer to the animal pole generally becoming forebrain cells, and cells closer to the germ ring generally becoming hindbrain cells, by virtue of their locations, independent of the time of the actual inductive events (Wilson et al., 1993). Alternatively, this ‘prepattern’ might have a specific molecular basis. For example, zebrafish homologs of maternal wnt gene may be expressed in parts of the epiblast, and act as competence modifiers of mesoderm and neural inductions, as has been proposed for Xenopus (Moon and Christian, 1992). The pre-gastrula appearance of goosecoid (Stachel et al., 1993; Schulte-Merker et al., 1994) and axial (Strahle et al., 1993) may indicate such early organization and inductive events. At present, little is known about the regional specification or commitment states of zebrafish epiblast cells at any time during gastrulation. By exploiting the fate maps presented here, future experiments will permit more direct tests of the nature and timing of these early events that establish and pattern the nervous system of the zebrafish.

We would like to thank Drs D. Anderson, B. Condron, A. Collarzo, J. Shih, P. Sternberg, B. Trevarrow, B. Wold, and K. Zinn for critical reading of the manuscript, as well as two anonymous reviewers for helpful comments. We thank Dr J. Shih in particular for stimulating discussions and suggestions for data presentation. We are grateful to Ms Dian De Sha for her editorial expertise. This work is supported by the Evelyn Sharp Graduate Fellowship (to K. W.) and grants from NIH and NIMH (to S. E. F.).