Development 108,569-580(1990)Printed in Great Britain © The Company of Biologists Limited 1990

569

Cell movements during epiboly and gastrulation in zebrafish

RACHEL M. WARGA and CHARLES B. KIMMEL

Institute of Neuroscience, University of Oregon, Eugene OR 97403, USA

Summary

Beginning during the late blastula stage in zebrafish,cells located beneath a surface epithelial layer of theblastoderm undergo rearrangements that accompanymajor changes in shape of the embryo. We describethree distinctive kinds of cell rearrangements. (1) Radialcell intercalations during epiboly mix cells located deeplyin the blastoderm among more superficial ones. Theserearrangements thoroughly stir the positions of deepcells, as the blastoderm thins and spreads across the yolkcell. (2) Involution at or near the blastoderm marginoccurs during gastrulation. This movement folds theblastoderm into two cellular layers, the epiblast andhypoblast, within a ring (the germ ring) around its entirecircumference. Involuting cells move anteriorwards inthe hypoblast relative to cells that remain in the epiblast;the movement shears the positions of cells that were

neighbors before gastrulation. Involuting cells eventu-ally form endoderm and mesoderm, in an anterior-posterior sequence according to the time of involution.The epiblast is equivalent to embryonic ectoderm.(3) Mediolateral cell intercalations in both the epiblastand hypoblast mediate convergence and extension move-ments towards the dorsal side of the gastrula. By thisrearrangement, cells that were initially neighboring oneanother become dispersed along the anterior-posterioraxis of the embryo. Epiboly, involution and convergentextension in zebrafish involve the same kinds of cellularrearrangements as in amphibians, and they occur dur-ing comparable stages of embryogenesis.

Key words: blastula, gastrula, morphogenetic movements,involution, clonal analysis, cell lineage.

Introduction

In the zebrafish embryo, after an early developmentalperiod of rapid cleavages, morphogenetic movementsoccur that rapidly produce major changes in the appear-ance and organization of the blastoderm. During epi-boly (Trinkaus, 1984a; 19846), beginning at the lateblastula stage about 4h after fertilization, the blasto-derm thins and spreads to completely cover the yolk cellduring the course of 6h. Gastrulation begins about anhour after epiboly is underway. The blastoderm, asingle multilayer of cells, rearranges into a two-layeredstructure consisting of a more superficial epiblast, andan inner hypoblast (Wilson, 1891). Shortly after gastru-lation begins, the embryonic axis appears and lengthensalong one side of the embryo (the dorsal side), as cellsaccumulate and line up specifically at that location. Therearrangements that occur among the cells of theblastoderm during early morphogenesis, particularlywith respect to their lineal relationships and their futurefates, are not well understood.

For example, several early embryologists concludedthat during gastrulation the hypoblast originates by cellinvolution, a streaming of cells lying at the blastodermmargin inward and underneath their neighbors (Wilson,1891; Morgan, 1895; Pasteels, 1936). Later, Ballard(I966a,b,c) concluded that the movement was in the

opposite direction: deep-lying blastoderm cells spreadoutward towards the margin to form the hypoblast.Ballard's view has been generally accepted, but veryrecently, involution was observed directly in the smallembryo of a teleost, the rosy barb (Wood and Timmer-mans, 1988).

During the course of cell-lineage analyses, we havefollowed cell movements during epiboly and gastru-lation in zebrafish. We observed cell rearrangementsthat seemed nonsensical if considered only in terms ofthe eventual fates that the lineages produced. First, inthe late blastula, cells scatter chaotically (Kimmel andLaw, 1985b; Kimmel and Warga, 1986). Second, in thegastrula, neighboring cells at the blastoderm marginundergo anterior-posterior inversions in their positions(Kimmel and Warga, 1987a). Finally, cells in eitherectodermal (Kimmel and Warga, 1986) or mesodermal(Kimmel and Warga, 1987a) lineages disperse along theanterior-posterior axis of the embryo.

We now show that each of these cellular rearrange-ments are understandable if they are considered inrelation to the changes in form of the blastoderm thatoccur at the same time. Studies done mostly in Xenopussuggest that cells undergo specific rearrangements tomediate the changes in form (Keller, 1987). We find thesame rearrangements occur in zebrafish at the compar-able stages of development.

570 R. M. Warga and C. B. Kimmel

Materials and methods

Embryos and stagesZebrafish embryos were obtained from natural spawnings andstaged by cell number during early cleavage. They weredechorionated with watchmaker's forceps and kept at 28.5°Cin an incubation medium of 14mM NaCl, 0.6mM KC1, 1.3mMCaCl2, lmM MgSO4, and 0.07mM sodium-potassium phos-phate buffer (pH7.2). In some experiments, we used embryoshomozygous for the gol-1 (golden) mutation (Streisinger et al.1981), because they are lightly pigmented relative to the wildtype, and fluorescently labeled cells in their bodies can beobserved more clearly in whole-mount preparations afterpigment cells differentiate.

Developmental time usually was determined from themorphological features of the embryo, and Table 1 gives astaging series for the period of development of interest, frommidblastula period until somites begin to form. We use theletter h to mean hours after fertilization at 28.5°C. Apreviously published series, although less complete, includesuseful sets of photographs (Hisaoka and Battle, 1958;Hisaoka and Firlit, 1960). In our series, names in commonusage in embryology denote major periods of development(e.g. midblastula, gastrula), and the stages subdivide theseperiods. We name rather than number the stages, whichseems to help one to remember them, and is more flexible.

Blastomere injectionsSingle blastomeres were injected (Kimmel and Law, 1985a),in mid- and late blastula embryos with the lineage tracer dyetetramethylrhodamine-isothiocyanate dextran (MolecularProbes, Eugene, OR; lOxlO3 Afr, diluted to 5 % (wt./vol.) in0.2 M KC1). The second dye for double-label experiments wasfluorescein-dextran (Sigma), dissolved the same way. Injec-tions were made by pressure, usually over the course of a fewseconds, either into a cell in the surface enveloping layer(EVL), or, in other cases, into a cell in the deep layer (DEL)of the blastoderm. To inject a DEL cell, the injection pipette

was advanced through the intact EVL. It was technically moredifficult to specifically inject single DEL cells than EVL cells,even under visual control. As an aid, we monitored voltagethrough the injection pipette. We observed that successfulpassage of the pipette through the EVL was accompanied by arise in voltage of up to 40 mV; the extracellular spacesurrounding DEL cells is at a positive potential relative to thebath (Bennett and Trinkaus, 1970). Upon intracellular pen-etration of a DEL cell, we then observed the expected shift tonegative potential, reflecting the membrane potential of thecell.

Observations of fluorescent cells in live embryosFor short-term viewing of labeled cells, embryos were usuallypositioned as desired in a gel of 3 % methyl cellulose made inthe aqueous incubation medium described above and viewedwithout a coverglass. Alternatively, embryos in incubationmedium were sandwiched between two micro cover glassesthat were spaced apart with three pairs of cover glasses (each0.13-0.17mm thick). For longer term viewing and for time-lapse recordings, the embryos were held stationary in suchchambers in a gel of 0.1 % agarose made in the same medium,and the chamber was then sealed with Vaseline to preventevaporation. Observations were made using a Zeiss micro-scope with illumination from both a transmitted and an epi-light source (Zeiss filter set 48-77-14), which permittedsimultaneous imaging of labeled and unlabeled cells. Thefluorescent image was amplified with a Silicon-Intensified-Target (SIT) video camera (Dage) to prevent light-induceddamage to the labeled cells. In some experiments, the depthsof fluorescent cells were determined with a digital shaftencoder fitted to the fine-focus knob of the microscope.

For time-lapse recordings, single-frame images were takenwith a Gyre video recorder at 4 s intervals. The epi-lightsource was controlled by a shutter that illuminated theembryo for only 60 ms during each exposure, in order tominimize light-induced damage to the labeled cells. Frequentrefocusing of the image was required during the recording

Table 1. Series of normal stages for 3-10.5 h of development

Stage h1 HBb Description

1 k-cell

2k-cell

High

Oblong

3

3.2

3.5

3.7

Sphere

10

11

12

Dome30%-epiboly

50%-epibolyGerm-ringShield75%-epiboly100%-epiboly

Budl-Somite

4.34.7

5.25.5689.5

1010.5

1314

151617

Midblastula; yolk syncytial layer present; cell cycles of blastoderm cells fairlysynchronous, determined by presence or absence of interphase nuclei

Single row of yolk syncytial layer nuclei; cell cycles of blastoderm cells highlyasynchronous

Blastoderm perched high upon the yolk cell, giving the embryo a dumbbell shape; yolksyncytial layer nuclei in two rows

Flattening of the blastoderm over the yolk cell produces a single smooth contouredoutline, elongated along the animal-vegetal axis; multiple rows of yolk syncytial layernuclei

Late blastula; embryo has assumed a spherical shape; at a deep plane of focus the yolkcell-blastoderm interface is flat

Yolk cell bulging (doming) towards animal pole as blastoderm rapidly thins by epibolyBlastoderm shaped as an inverted cup of uniform thickness and covers 30% of the yolk

cellGastrula; 50% of the yolk cell is covered by the blastodermGerm ring visible from animal pole; 50%-epibolyEmbryonic shield visible from animal pole, 50%-epibolyThe blastoderm continues to spread over the yolk cell at a rate of 15 %-epiboly per hourYolk plug closed. Gastrulation movements nearly complete in the anterior parts of the

embryoTail bud prominent at the posterior end of the axisSegmentation; the first furrow appears in the paraxial (presomitic) mesoderm; about 2

somites are added per hour (Hanneman and Westerfield, 1989)

°h: hours after fertilization at 28.5°C.bHB: Approximate stage number in the zebrafish staging series described by Hisoaka and Battle (1958).

Eplboly and gastrulation In zebrafish 571

session, which began at sphere stage (4 h) and continued for atleast 6h, when epiboly is completed, and generally for a fewhours longer. Afterwards, the embryo was released from theviewing chamber and reexamined at 24-30 h, when many celltypes have begun to differentiate and can be distinguished bytheir morphologies and positions in identifiable tissues (Kim-mel and Warga, 1987«). Labeled cells that were still undiffer-

entiated were reexamined on the 2nd and/or the 3rd day ofdevelopment.

HistologyWe also studied a set of sectioned embryos. They were fixed atintervals between late blastula (4h) and midgastrula (7h)periods by immersion in Bouin's solution (Humason, 1962),

AP B

v

D

D V D

Fig. 1. Morphogenesis during zebrafish epiboly and gastrulation. Side views of living embryos with the animal pole (AP) tothe top. An outline of stage descriptions is given in Table 1. (A) Oblong stage, at the end of the midblastula period, 3.7 h.The blastoderm is a thick cap of cells occupying about a third of the volume of the blastula. The blastoderm margin (m)separates the blastoderm and the uncleaved yolk cell. At this time blastoderm cells are motile (D. A. Kane, in preparation),but major rearrangements among them have not yet occurred. (B) 50%-epiboly stage, onset of gastrula period, 5.2 h.Epiboly is well underway, and the blastoderm has thinned to take the form of a cup inverted over the yolk cell. Involutionand convergence movements appear to begin at this stage. (C) Shield stage, early gastrula, 6.Oh. Involution andconvergence movements have produced the embryonic shield, a pronounced accumulation of cells along the margin at thedorsal (D) side. Hence the blastoderm appears thicker here than ventrally (V), and the hypoblast has become prominent(arrow). To aid visualizing the hypoblast this embryo was slightly flattened between coverslips. Hence it appears a littlelarger than the others in this figure. (D) Bud stage, beginning of the tailbud period, 10h. Epiboly is completed; the yolkplug (YP) has just been enveloped by the blastoderm near the site of the original vegetal pole of the egg. The blastoderm isobviously thicker dorsally (D) than ventrally (V), due to the forming embryonic axis on the dorsal side. The tail bud (arrow)is present at the posterior end of the embryonic axis. Scale bar: 200 ^m.

572 R. M. Warga and C. B. Kimmel

dehydrated and embedded in Epon A12. Serial 5 fm\ sectionswere cut and stained with azure A, methylene blue and basicfuchsin (Humphrey and Pittman, 1974).

Results

Deep and shallow DEL blastomeres intercalate duringepibolyIn zebrafish, cleavages generate two populations ofdistinctive blastoderm cells; flattened epithelial cells ina surface enveloping layer (EVL), and rounded, moreloosely associated deep layer (DEL) cells lying beneaththe EVL. The EVL is a monolayer and the DEL amultilayer of cells. All of the movements we describe inthis paper pertain to the DEL: the EVL cells behaverelatively passively. Neighb6r exchanges occur withinthe EVL (Keller and Trinkaus, 1987), but they areinfrequent. We have not observed neighbor exchangesbetween the EVL and DEL.

B

Fig. 2. The double-label method used to distinguish themorphogenesis and fates of sibling clones located atdifferent depths in the blastoderm. (A) At the 32-cell stage,the blastomeres are often arranged in a single-layered 4x8array. One of the central rows of 8 cells is shown, and a cellin this row, adjacent to the animal pole, is injected withrhodamine-dextran (coarse stippling). (B) Following thenext (sixth) cleavage, which is horizontal, the upper of thetwo labeled sister cells is reinjected, now withfluorescein-dextran (fine stippling). The lower daughter,containing only rhodamine-dextran is adjacent to the yolkcell. (C) Following several more cleavage divisions the twosister clones are expected to remain coherent in themidblastula, since cell mixing is very limited before lateblastula stage. The rhodamine-labeled clone is expected tolie deep in the blastoderm, immediately beneath the doublylabeled clone, which extends to the blastoderm surface.

DEL cells become motile in the midblastula, after thetenth cleavage at 3h (D. A. Kane, unpublished obser-vations). The embryo flattens to take on a sphericalshape by 4h (late blastula; Fig. 1A), and during thenext hour of development, a rapid thinning of theblastoderm becomes evident, signifying epiboly isunderway. The first change observable is very deep inthe embryo, where the yolk cell begins to bulge or'dome' towards the animal pole (Fig. IB). The blasto-derm then rapidly takes on a cup-shaped appearance,and spreads to cover the yolk cell (Fig. 1C).

At the beginning of the late blastula stage, singleclones descended from a progenitor cell labeled earlierare coherent groups of cells. Later, during epiboly, theDEL cells in such clones rapidly spread apart, inter-spersing with unlabeled cells (Kimmel and Law, 1985b).In Xenopus, epiboly is known to occur by radial cellintercalations, in which cells at different depths in theblastoderm intercalate, thus producing its thinning(Keller, 1980). Such a rearrangement could produce thecell scattering we observed in zebrafish, and we haveexamined whether radial intercalations occur in thisspecies.

We took advantage of the pattern of cell divisionduring early cleavages to label, with two differentcolored dyes, two sibling blastomeres; one underlyingthe other at the 64-cell stage (Fig. 2). The deeper cellgenerated a clone located deep in the DEL of themidblastula, and immediately underlying the cloneoriginating from its superficial sib, as confirmed bydirect inspection (Fig. 3A). DEL cells of the two clonesbecame thoroughly intermixed by early gastrula stage(Fig. 3B; note that the intermixing does not extend intothe EVL). Subsequently, both sets of DEL cells gaverise to very similar sets of derivatives in the laterembryo. In this example, both clones developed headectodermal cell types (Fig. 3C). These results establishthat blastoderm cells intercalate along radii during earlyepiboly, and that such movements are confined to theDEL.

The hypoblast arises by involutionAt the time when the blastoderm half covers the yolkcell (5.2h; referred to as 50%-epiboly; Fig. 1C) newcell movements begin, including involution movementsthat form the hypoblast. These new movements markthe onset of gastrulation (see Discussion). Within about15 minutes, the blastoderm becomes noticeably thickerin a circumferential band at its margin. The band, orgerm ring, at first appears uniform in structure, andusing time-lapse video microscopy, we observed inviews from the animal pole (3 embryos) that it formsmore-or-less simultaneously, within about 15min,around the entire circumference of the blastoderm.

An analysis of involution is shown in Fig. 4, a casewhere we kept track of the depths of cells in theblastoderm as their rearrangements occurred. Here,minutes after the onset of gastrulation, a clone of 5labeled cells was located near the margin of the blasto-derm. The cells initially occupied a shallow positionwithin the DEL, indicated by blue color-coding in

Fig. 3. Deep cells in the DEL intercalate with superficial DEL cells, and then both populations exhibit similar fates. The double-labelingexperiment is explained in Fig. 2. The images shown here are computer-enhanced, with cells containing rhodamine alone shown as red,and cells doubly labeled with rhodamine and fluorescein shown as yellow. (A) A side view (as in Fig. 1A) at a deep plane of focus of themidblastula. As expected (Fig. 2C), the red clone lies underneath, deep to, the yellow one. The deepest cells in the red clone are adjacentto the yolk cell. The most superficial cells in the yellow clone are EVL cells. (B) Surface view (at a shallow plane of focus) of the earlygastrula. Red and yellow cells are intermingled within the DEL, with red cells now occupying very superficial positions in the DEL. Redcells are not present in the EVL. The only labeled cells within the EVL (arrow) are yellow, as expected from the labeling regime, andbecause DEL cells do not intercalate into the EVL. (C) view of clusters of labeled neuro-epithelial cells within the brain of the embryo at24 h (side view, with dorsal to the top). Both red and yellow cells have contributed to the CNS. The experiment was repeated 5 times, andthe mixing among DEL cells was always observed. Scale bars: 100,um (A & B), 50f*m (C).

Fig. 4. Changes in thepositions of cells in a cloneof DEL cells duringinvolution in the gastrula.Depth in the blastoderm iscolor-coded in thesecomputer-enhanced images:blue, superficially lyinglabeled DEL cells; red, mostdeeply positioned cells;green, intermediate. Theblastoderm margin isindicated with arrows, andthe orientation isapproximately the same inall the panels. A-D showsuccessive times duringgastrulation, at 6.6h (60%-epiboly),7.6h(70%-epiboly), 8h (75%-epiboly),and 9h (90%-epiboly)respectively. Eventually theprogeny of the involutingcells in this clone formedsomitic muscle. (A) All thelabeled cells occupy shallowpositions in the DEL. Thecells are moving towards theblastoderm margin. (B) Cellnumber in the clone hasincreased by cell division,and the cells nearest themargin (green) have begunto involute. (C) Twentyminutes later the involutingcells form a stack at themargin. The cells that hadbegun involution earliest (inB) are at the bottom of thestack. Those involuting lastare at the top. (D) An hourlater the first involuting cells(red and green pair) havemoved away from themargin, and are now in adeep location (in thehypoblast) beneath theirsuperficial relatives, which are still in the epiblast (blue). The second pair of involuting cells have moved to the deepest location at themargin by this time, and shortly later will also move away from the margin within the hypoblast. Scale bar: 25 ,um.

Fig. 5. The germ ring forms during involution. The panels (A-D) show selected frames from a time-lapse video tapefollowing cells during the earliest gastrulation movements. At the beginning of the experiment a single DEL cell,located beneath the EVL at the blastoderm margin in the blastula (3.1 h), was injected with rhodamine-dextran. (A) By4.5 h (30%-epiboly), before the onset of gastrulation, the labeled cell had divided, and its daughters are close neighborsat the margin (arrow). The color for this figure was computer-generated, and codes cell lineage (i.e. a red lineage and agreen lineage), so that the reader can keep track of the fate of these two daughter cells. (B) An hour later (5.5 h, 50%-epiboly), involution movements are just beginning. The two labeled cells have become separated by a single unlabeledone. (C) Six minutes later both cells are dividing, and are in the process of involution. (D) Forty minutes later (6.3 h,shield stage) the germ ring has formed. All of the cells of the clone have now moved well away from the margin and arepresent in the hypoblast within the confines of the germ ring. One cell in each of the two lineages is very close to theupper boundary of the germ ring (upper arrow). The lower boundary of the germ ring is the margin (lower arrow). Asin this example, we invariably observed that only DEL cells located near the blastoderm margin involute duringgastrulation. In all, in this study, involution was followed by time-lapse analysis in labeled clones in six embryos thatwere oriented such that we could clearly distinguish the borders of the hypoblast, and in each of these cases the labeledinvoluting DEL cells (31 in all) entered, and then remained in the hypoblast. Conversely, in three embryos with cloneslocated farther from the margin, 18 DEL cells were followed that never involuted and were later present in derivativesof the epiblast. A clone in one additional embryo was located about 80% of the way between the animal pole and themargin, and in this case 5 of the DEL cells involuted, and 6 of the DEL cells (initially farther from the margin) did notinvolute. In eight of the embryos the clones included labeled EVL cells as well as labeled DEL cells, and in none ofthese cases was an EVL cell observed to involute. Scale bar: 25 ^m.

Epiboly and gastrulation in zebrafish 573

Fig. 4A. The labeled cells moved towards the margin,apparently actively, since we observed blebbing andformation of filopodia, and two of them divided. Uponreaching the margin, each cell protruded processes andmoved deeper within the DEL (Fig. 4B; green), awayfrom the EVL and towards the surface of the yolk cell.Each marked cell then reversed its direction relative tothe margin, and proceeded away from the margin, nowlocated deeply in the blastoderm (Fig. 4C and D; red),within its new hypoblast layer.

These findings show that involuting DEL cells formthe hypoblast. The first ones to involute are thoselocated just at the blastoderm margin at the beginningof gastrulation (Fig. 5A), and as they migrate inwardsthe germ ring forms behind them (Fig. 5B and C).Analysis of sectioned embryos revealed that beforegerm ring formation there is no layering of cells withinthe DEL itself (Fig. 6A). However, after the germ ringforms, the DEL appears folded inwards at the margin,and is split, within the germ ring specifically, into theepiblast and hypoblast (Fig. 6B). As epiboly and gas-trulation continue, cells that initially were locateddistantly from the margin move towards it within the

V. a-

f

epiblast, and involute. Consequently the hypoblastincreases in area and extent beneath the epiblast,eventually spreading all the way from the margin to theanimal pole.

Fates of epiblast and hypoblast cellsSingle cells present during gastrulation generate clonesrestricted to single types of tissues (Kimmel and Warga,1986). Now we have shown that some of these cells, butnot others, involute during gastrulation, and we can askwhether involuting and noninvoluting cells have differ-ent fates. We kept track of cells in labeled clones duringgastrulation and determined the fates of their descend-ants at a later stage, as illustrated in Fig. 7A. From suchrecords, we reconstructed the cell lineage (Fig. 7B) andthe pathways of movement of the cells during gastru-lation (Fig. 7C). In this example, all of the labeled cellsinvoluted, inverting their relative positions as they didso. The progeny of one of the cells originally present(Fig. 7, black lineage) all formed somitic mesoderm,including differentiated muscle fibers. The progeny ofthe other cell (stippled) formed derivatives of two

B

Fig. 6. Involution produces layers of DEL cells within the blastoderm. Stained plastic 5;<m sections that show theblastoderm margin. (A) The onset of gastrulation (50%-epiboly; 5.2 h); (B) the midgastrula (75%-epiboly, 8 h). Bothmicrographs are oriented the same way with the blastoderm margin near the bottom, the EVL present as a very thin celllayer to the left, and the yolk to the right. In A the DEL is a homogeneous multilayer of cells. In B, except just at themargin where cells are involuting, the DEL is layered into the epiblast (underlying the EVL) and the hypoblast (overlyingthe YSL of the yolk cell). This interpretation exactly follows Wilson (1891), who did not have the benefit of followinglabeled cells in live embryos, but relied exclusively on studying fixed material such as shown in this figure. Scale bar: 50fan.

574 R. M. Warga and C. B. Kimmel

as h

5.6h

60h6.5h

ana *PPP ©aaa Oppp

B

4 -

5 -

6 -

7 -

8 -

9 -

to -

11 -

12 -

ZOh

mus.und

mes

78h

mus

mes

gut

end

30h

different germ layers; gut epithelium (endoderm) andsomitic muscle (mesoderm).

Summary lineage analyses for cells in other embryosanalyzed the same way are shown in Fig. 8. Withoutexception (and also for 5 additional clones not illus-trated), cells that involuted (arrows) later formedmesoderm and endoderm, and cells in lineages whereinvolution did not occur (no arrows) formed ectoderm.The lineage shown in Fig. 8B is noteworthy, for in thiscase two sibling subclones, separated at the first divisionshown in the diagram, developed differently from oneanother but still followed the general rule: None of thecells in one of the subclones involuted and they sub-sequently developed as ectoderm. The cells in thesecond subclone all involuted and then formed meso-derm. We conclude from these results that the epiblastis the rudiment of ectoderm and the hypoblast is therudiment of both mesoderm and endoderm.

The time of involution is related to fateCells that involute early during gastrulation usually

form endoderm, and cells that involute later formedmesoderm. This can be seen most clearly from singleclones that contributed to both germ layers (e.g. arrowsin Figs. 7B and 8D). Except for the early involution of acell that later formed heart tissue, a mesodermalderivative (Fig. 8D), this rule was generally followed(including those clones that contributed cells to only asingle germ layer; compare the times of involution ofcells in Fig. 8C and E).

Cells involuting at different times during gastrulationalso had different pathways of movement within thehypoblast, and later occupied different positions in theembryo. Cells that involuted soon after the beginning ofgastrulation sharply reversed their direction of move-ment as they involuted, turning towards the animal pole(Fig. 7C; stippled cell ppp). Cells that involuted some-what later turned less sharply (stippled cell aaa), andcells that involuted much later during epiboly (blackcell aaa) did not turn at all, but continued to movetowards the vegetal pole after entering the hypoblast.No matter whether a turn occurred or not, all involuting

Epiboly and gastrulation in zebrafish 575

Fig. 7. Analysis of movements, lineages and fates ofclonally related cells that involute. A single DEL celllocated near the margin of a blastula embryo was labeledwith rhodamine-dextran (2.3 h). Its subsequentdevelopment was recorded by time-lapse video duringepiboly and early tailbud stages (4-13 h). The fates of thecells in this clone were then determined by comparingpositions of the labeled cells at 13 h (after gastrulationmovements were over), and later, when they had begundifferentiation in the embryo at 30 h. (A) A series oftracings from selected frames during the first 4h from thetime-lapse recording (3.8-7.8 h), and, in the last drawing,the fates of the cells in the 1-day embryo. Anterior is to thetop, the blastoderm margin is to the bottom (except for thelast drawing, where there is no margin). The foundinglabeled DEL cell had divided a single time by the time therecording session began. Cells of the subclone generated byits more anterior daughter are coded black in thesedrawings, and cells of the subclone generated by its moreposterior daughter are stippled. Cell divisions are indicatedby v's, and arrowheads point to newly involuting cells.Initially the stippled cell is nearest the margin. Its progenyinvolute earliest, and some of them eventually differentiateas epithelial cells of the foregut (gut). Other cells in thissubclone, and cells of the black subclone differentiate assomitic muscle (mus). A few cells in the black subclone thatare present within one somite remain undifferentiated(und). Notice that the stippled cells are initially closer tothe margin, involute earlier, and then form more anteriorfates, including endoderm. (B) Cell lineage diagram for thesame two subclones for the period 3.8-13 h. Hours ofdevelopment are shown to the left. Horizontal lines in thediagram indicate divisions and vertical lines indicate cells;the more anterior daughter cells (and farther from theblastoderm margin) arising at each divisions are shown asleft-side branches in the diagram. Arrowheads indicatewhen individual cells involuted; in general cells towards theleft involute later because they are farther from margin.The endodermal (end) and mesodermal (mes) cell fates areindicated below the diagram. Notice the symmetrical(generative) form of the lineage: sibling cells tend to divideat the same time, as is also the case in the examples shownin Fig. 8. (C) Pathways of movements of selected cellspresent in the same two subclones. aaa refers to the mostanterior, and ppp to the most posterior cells in the lineages(i.e. the left-most and right-most branches in B). Duringthe recording session, the embryo itself was held stationaryin agarose, and serves as a fixed reference for themovements of these individual cells within it. The pathwayswere reconstructed directly from the video tapes. The starshows the starting position of the cell, and the arrowheadshows the time of involution. The orientation is as in A.Individual points along the pathways represent approximateone-half hour intervals. Notice that the cells that involuteearlier (in the stippled subclone in this case) turn anterioras they involute. Whether or not a turn occurs, cells recedefrom the margin after involution. Scale bars (in A & C):100 ^m.

cells recede from the margin of the blastoderm, whichcontinues to rapidly advance by epiboly towards thevegetal pole. We measured the rates that the involutingDEL cells moved (18 cells from 3 embryos), but nosignificant differences were found in the speed of

movement that correlated with whether a turn oc-curred, or with later fate (data not shown).

The animal pole of the gastrula develops into theanterior-most structures of the later embryo (Kimmeland Warga, 1987b), and accordingly, cells that turntowards the animal pole during gastrulation developmore anterior structures. It follows that the order inwhich cells of a clone involute corresponds to theirsubsequent order along the anterior-posterior axis ofthe embryo, as can be seen from the positions of theblack and stippled cells in Fig. 7. This relationship wasconsistent, as shown in Fig. 9 where data is collectedfrom a set of 5 embryos in which the labeled clones werepositioned at approximately the same lateral blasto-derm location at the beginning of gastrulation.

Convergent extension movements are mediated bymediolateral cell intercalationsConvergence is a third morphogenetic cell movementoccurring simultaneously with epiboly and involution.DEL cells move towards the dorsal side of the gastrula(Ballard, 1973; Kimmel and Warga 1987b), 'converging'there from their original locations in the blastoderm.The formation of the embryonic shield, a local dorsalthickening of the germ ring (Oppenheimer, 1937;Hisaoka and Battle, 1958), is a prominent effect of earlyconvergence movements. In Xenopus, convergencecannot be separated from extension, the elongation ofthe embryonic axis (Keller and Danilchik, 1988). Cellsmove from lateral positions dorsalwards by intercalat-ing between neighboring cells that lie more medially(i.e. towards the axis). Such 'mediolateral intercal-ations' (Keller and Tibbetts, 1989) produce both nar-rowing and elongation of the axis.

Convergent extension occurs both in the epiblast(Kimmel and Warga, 1986; and see below) and in thehypoblast. Convergence of hypoblast cells is illustratedin Fig. 7C as a drift in the pathways of the cells towardsthe right side of the figure, the direction towards theembryonic shield in this example. That intercalationsoccur during this movement is revealed by the separ-ation and spreading apart of the clonally-related cellsalong the anterior-posterior axis (Fig. 7A). During thisdispersion, the labeled cells are intercalating amongmore medial unlabeled neighbors.

The intercalations are more completely illustrated inFig. 10, an example following cells that remained in theepiblast. This case is instructive because intercalationsoccurred not only between labeled and unlabeled cells,but also between different labeled cells, and it is clearwhen they occurred. Four DEL cells (from a singleclone) were present at the beginning of gastrulation.Their descendants eventually formed a dispersed seriesof clusters in the hindbrain, distributed along the axis,and on both sides of the midline. The cells are shaded toindicate their lineal relationships. Eventually black andhatched cells were positioned between different whitecells. The intercalations effecting this intermixing oc-curred in the gastrula, beginning at about 6.8 h(Fig. 10). Shortly after both the black and hatched cells

576 R. M. Warga and C. B. Kimmel

divided, one of the two daughter cells from each of thedivisions inserted between the pair of white sister cells.

In contrast to the radial intercalations that we con-sidered above, the intercalations occurring during con-vergent extension do not scatter cells indiscriminately.For example, we have not observed DEL cells in thehypoblast and epiblast to mix with one another during

gastrulation, although mixing within each of theselayers is extensive.

Discussion

This work has revealed three distinctive cell movements

4 -

5 "

6 -

7 -

8 -

9 -

10 - nnlln:: f 1N/

23ner

eel

6oto. 12 pip 18 nor

ect ect

B

— 3 mus, 4 und 6 mus

4 rbn,4 ncr.l 1 ner 9 mus, 8 und

ect mes

D4 -

5 -

6 -

7 -

8 -

9 -

10 -

11 -

12 -

13 - nnnJ,

h10 gut

end

30 gut

end

6 mus-

mes

23 gut

end3hrtmes

Fig. 8. DEL cell lineage diagrams for the period of epiboly in clones from eight embryos. The presentation is as describedfor Fig. 7B, and the data were obtained the same way. The clones are grouped (A-E) according to whether involutionoccurred (arrowheads) or not, and whether the cells gave rise to ectodermal (ect), mesodermal (mes) or endodermal (end)derivatives. Thus for example, in A none of the DEL cells in clones in two embryos involuted (there are no arrowheads),and all of the cells formed ectoderm. Individual cell fates are abbreviated as follows: gut, gut epithelium; mus, somiticmuscle; ncr, neural crest; ner, nervous tissue; oto, otocyst (ear vesicle); pip, posterior lateral line placode; pnd, pronephricduct; rbn, Rohon-Beard (sensory) neuron; und, undifferentiated-appearing mesenchyme.

Epiboly and gastrulation in zebrafish 577

o

2•L «

<

23 .

20 .

15 .

10 .

5 .

1 .

3 .

1 .

1 .

«n .

i••

o1 1

aa

"i

—t—

0•

1

[

H

C

1

1

1 1 1 10 1 2 3

Hours after 50%-eplboly

that accompany, and appear to produce, the earlychanges in shape of the zebrafish embryo. Thesemovements are epiboly, involution and convergentextension. They are diagrammed respectively inFigs 11-13.

EpibolyRadial intercalations (Keller, 1980) among DEL blasto-meres occur first, in the late blastula (Fig. 11), and

8.7 h

5.0 h6.5h

6.8h

Fig. 9. The time during gastrulation when a cell involutescorrelates directly with the later anterior-posterior positionof its progeny. The five different symbols represent cells inclones from five separate embryos, from the set shown inFigs 7 & 8. These five were selected because in the earlygastrula they were all located in the same lateral position,as measured along the margin relative to the position of theembryonic shield (at the dorsal side; see Kimmel el al.1990). The horizontal axis indicates when involutionoccurred, relative to the onset of gastrulation at 50%-epiboly. The vertical axis indicates position along theanterior-posterior axis of the embryo at 24 h. Vertical linesabove the individual symbols indicate the extent ofanterior-posterior spread within the subclone derived fromthe involuting cell, in those cases where the spread wasextensive.

along with an expansion and change in shape of the yolkcell that occur simultaneously, these cell movementsappear to mediate the thinning of the blastoderm thatoccurs rapidly during this period of development.Intercalations thoroughly scatter DEL cells and areresponsible for marked dispersion of clonally relatedcells that we have described elsewhere (Kimmel andLaw, 1985b). It is interesting, however, that DEL cellsdo not intercalate outward into the EVL. The EVL, bythis stage, has acquired the form of a highly flattenedsquamous epithelium. In Fundulus at a comparablestage of development, junctional complexes are presentbetween cells of the EVL (Betachaku and Trinkaus,1978), and it may be that such junctions mediate highadhesivity among the EVL cell, such that the underly-ing DEL cells are unable to penetrate this layer.

25.0h

12.7h

9.2 h

Fig. 10. DEL cells intercalate during convergent extension in the gastrula. The drawings are selected from a larger series,taken from video taped records, and show the positions of labeled DEL cells that were descended from a blastomereinjected in the midblastula with rhodamine-dextran. Methods are as in Kimmel and Warga (1986). The cell lineage is shownin the summary diagram, and the shading of the cells codes their lineal relationships. EVL cells were present in this clone,but are ignored after the first drawing (5.Oh), where 2 are shown by dotted outlines. None of the cells in this cloneinvoluted, but remained in the epiblast and by 25 h had formed a periodic series of bilateral clusters in the hindbrain (seeKimmel and Warga, 1986). mid: brain midline. pia, brain pial or outermost surface; oto, otocyst. As individual cells moveapart from one another they are intercalating among invisible (i.e. unlabeled) cells. Intercalations among the labeled cellsalso occur, (e.g. at 6.8h) such that subclonal groups often do not remain together. Scale bar: 100^m.

578 R. M. Warga and C. B. Kimmel

AP

>• DEL

\ yolkcell

Fig. 11. Cut-away diagram of radial intercalationsbeginning in the. midblastula (A), and through the onset ofgastrulation (B). Deep DEL cells move outward (radially),intercalating among more superficial DEL cells but notamong EVL cells (See Fig. 3). This movement contributesto epiboly, thinning and spreading the blastoderm*

epiblast

hypoblast

Fig. 12. Cut-away diagram of involution. (A) Onset ofgastrulation. (B) Germ-ring stage, 20min later. Side views.A DEL cell first at the blastoderm margin (black) is at thefront of the wave of involuting cells (See Fig. 5). Thismovement generates the hypoblast. EVL cells do notinvolute.

blastoderm

M

Fig. 13. Diagram of mediolateral intercalations. (A) shieldstage, early gastrula. (B) 80%-epiboly, late gastrula. Dorsalviews. DEL cells converge towards the dorsal midline,moving from lateral to medial positions. This movement(convergent extension) lengthens the embryonic axis(dashed line).

InvolutionInvolution movements of DEL cells located near theblastoderm margin produce the hypoblast, an innerlayer of the blastoderm (Fig. 12). Involution (or insome animals its counterpart invagination; see Trin-kaus, 19846) is the singular morphogenetic movementthat characterizes gastrulation in many different typesof animals; hence we consider the beginning of gastru-lation in zebrafish as the time when involution begins.This is the stage when the blastoderm has advanced, byepiboly, to cover just one-half of the yolk cell. The cellsmove first towards the blastoderm margin, in thegeneral direction of the vegetal pole. When they reachthe margin they involute to take up a new, deeper,position. Afterwards they either reverse their directionof movement and move towards the animal pole, or, inthe case of cells that involute later during gastrulation,they continue moving towards the vegetal pole. Ineither case, once cells are in the hypoblast, they are leftbehind the leading edge of the blastoderm, whichcontinues to advance across the yolk cell by epibolyduring the gastrula period. We have obtained noevidence that cells can enter the hypoblast by any othermovement than involution, although we have not yetcarefully examined cell movements within the embry-onic shield (at the dorsal side of the embryo).

Wood and Timmermans (1988) recently also ob-served involution in the rosy barb, another teleost in thesame family as zebrafish. However, Ballard could notfind involution in his careful and extensive studies of avariety of other (and larger) teleost embryos(1966a,b,c; 1973; 1981; 1982). It may be that gastru-

Epiboly and gastrulation in zebrafish 579

lation is dramatically different in large and small teleostembryos, but we think it more likely that all teleostsgastrulate as the zebrafish and rosy barb do, and thatthe cell marking procedures available to Ballard wereinadequate to reveal all the cell movements that occurin the DEL.

Involution is special in teleost fish, as compared toother types of vertebrates, in that the EVL is notinvolved (see below). Moreover, involution doesn'tseem to be initiated first at the dorsal side of the embryo(as it does for example in amphibians). As judged fromthe time-course of appearance of the germ ring, invol-ution in the zebrafish begins more-or-less simul-taneously around the circumference of the blastoderm.

Convergent extensionDuring gastrulation DEL cells also undergo mediolat-eral intercalations (Fig. 13), producing a general dorsal-wards drift of the cells that has been described pre-viously in other teleosts (e.g. Ballard, 1973; 1982). Thecells accumulate dorsally to form the embryonic shield,and the subsequent narrowing (convergence) andlengthening (extension) of the shield produces a well-defined embryonic axis within about two hours after theshield first forms.

We have shown that cells in both the hypoblast andepiblast undergo convergent extension. The intercal-ations appear to be regulated such that extensive mixingoccurs among the cells within both of these layers, butnot between the layers. Moreover, the fact that mostgastrula lineages are tissue-restricted (Kimmel andWarga, 1986) shows that mixing among cells must occurwithin, but not between, the primordia of differenttissues. However, the boundaries of the primordia areinvisible in the gastrula, such that we could not hope toobserve distinctive cellular behaviors in their vicinities.Later in embryogenesis the boundaries become recog-nizable, and no mixing occurs across at least one ofthem - the boundary separating the axial (prospectivenotochord) and paraxial (prospective somite) meso-derm - as recently shown for Xenopus (Wilson et al.1989) and rosy barb (Thorogood and Wood, 1987).

The enveloping layerAll of the movements we have described appear toclosely resemble their counterparts that have beenthoroughly described in Xenopus (Keller, 1986). Radialintercalations, involution movements and mediolateralintercalations occur at the equivalent stages, relative togastrulation onset, in zebrafish and Xenopus and theyproduce equivalent changes in shape and organizationof the embryo. There is a single important difference,however; the outside layer of cells in the teleostblastoderm does not participate in any of them. DELcells do not enter the EVL during their radial intercal-ations, as we have shown in this study. We alsoconfirmed that EVL cells do not undergo involution, aswas first convincingly shown by Ballard for the trout(1966a). This finding was expected in zebrafish since theexclusive fate of the EVL is the periderm - an outer-most epithelial cell layer covering the embryo (Kimmel

and Warga, 1986; Kimmel et al. 1990). We showedearlier (Kimmel and Warga, 19876) that the EVL cellsdo not undergo convergence, at least in the sense usedhere to mean a specific dorsalwards movement.

The EVL may be a relatively passive participant inblastoderm epiboly; as revealed by studies in Fundulus,it seems to be pulled and stretched across the yolk cellby the yolk syncytial layer of the yolk cell itself(Betchaku and Trinkaus, 1978; Trinkaus, 1984a). How-ever, perhaps active rearrangements among EVL cellshave recently been shown to occur both in Fundulus(Keller and Trinkaus, 1987) and the medaka(Kageyama, 1982), where they serve to continuouslydecrease the diameter of the EVL as epiboly is com-pleted and the marginal ring of EVL cells closes at thevegetal pole of the yolk cell. It is likely that thisrearrangement also occurs in the zebrafish, for EVLcells in single clones do sometimes become dispersedfrom one another, rather than being present in a singlecoherent patch (e.g. Kimmel and Warga, 1987b). Thedispersion is, however, markedly less than that occur-ring in the DEL.

Control and patterning of cell movementsOur studies are descriptive, and do not reveal themechanisms that underlie these morphogenetic move-ments. However, the rearrangements appear to beactive ones, for DEL cells constantly change in shapeand they move relative both to neighboring cells and toa fixed point on the yolk cell. The yolk cell and EVLcells both participate in epiboly and, as we have shownhere, so do DEL cells. The gastrulation movements ofinvolution and convergence may also depend uponinteractions among cells of all three classes. RecentlySymes and Smith (1987) suggested that activation ofgastrulation movements in amphibians is an earlyconsequence of mesodermal induction. This mightinvolve the yolk cell; Long (1983) obtained evidencefrom transplantation experiments in the trout that theyolk cell can induce dorsoventral polarity of the blasto-derm.

Progress in understanding how such specific move-ments are produced may come through mutationalanalysis in zebrafish. We have recently described amutation, spt-1, that appears to selectively disruptconvergence of laterally positioned mesodermal cellsduring gastrulation (Kimmel et al. 1989). Convergenceof ectoderm is not disturbed, suggesting that differentgenes control dorsalwards movements of cells thatoccupy different germ layers. Furthermore, mosaicanalysis suggests that the wild-type gene is required inthe mesoderm specifically (Ho et al. 1989). The genecould code for, or regulate the expression of, a receptoror adhesion molecule required for the convergencemovements of a subset of mesodermal cells.

An important finding from our study is that cells thatinvolute to enter the hypoblast then give rise to endo-derm or mesoderm and, conversely, that the epiblast isthe equivalent of ectoderm in other vertebrates. Thisobservation is in accord with the interpretations of earlyinvestigators of teleost embryology (Wilson, 1891; Mor-

580 R. M. Warga and C. B. Kimmel

gan, 1895; Pasteels, 1936). We also show that whether acell in the hypoblast will form endoderm or mesoderm,and where its clonal descendants will come to lie alongthe anterior-posterior axis of the embryo is directlycorrelated with when it entered the hypoblast. Further-more, we detected no differences in direction or rate ofmovements of DEL cells as they approached themargin, before involution, that correlated with theirfuture fates. Together, these observations lead to thesuggestion that whether and when a cell involutes is adirect function of how far from the margin it waspositioned prior to the onset of gastrulation. We ad-dress this issue in the accompanying paper (Kimmel etal. 1990), examining in more detail how cell fate isrelated to cell position in the early gastrula.

We thank D. A. Kane, A. Felsenfeld and D. Frost for theircritical comments on early versions of this paper, and forstimulating discussion throughout the course of the study. C.Cogswell, P. Myers, H. Howard, and R. Kimmel providedtechnical assistance. The research was supported by NSFgrant BNS-8708638, NIH grant HD22486, and a grant fromthe Murdock Foundation.

References

BALLARD, W. W. (1966a). The role of the cellular envelope in themorphogenetic movements of teleost embryos. J. exp. Zool. 161,193-200.

BALLARD, W. W. (19666). Origin of the hypoblast in Salmo I. Doesthe blastodisc edge turn inward? J. exp. Zool. 161, 201-210.

BALLARD, W. W. (1966C). Origin of the hypoblast in SalmoII. Outward movement of deep central cells. J. exp. Zool. 161,211-220.

BALLARD, W. W. (1973). Morphogenetic movements in Salmogairdneri Richardson. J. exp. Zool. 184, 27—48.

BALLARD, W. W. (1981). Morphogenetic movements and fate mapsof vertebrates. Am. Zool. 21, 391-399.

BALLARD, W. W. (1982). Morphogenetic movements and fate mapof the Cypriniform teleost, Catostomus commersoni (Lacepede).J. exp. Zool. 219, 301-321.

BENNETT, M. V. L. AND TRINKAUS, J. P. (1970). Electrical couplingof embryonic cells by way of extracellular space and specializedjunctions. J. Cell Biol. 44, 592-610.

BETCHAKU, T. AND TRINKAUS, J. P. (1978). Contact relations,surface activity, and cortical microfilaments of marginal cells ofthe enveloping layer and of the yolk syncytial and yolkcytoplasmic layers of Fundulus before and during epiboly. ExplZool. 206, 381-425.

HANNEMAN, E. AND WESTERFIELD, M. (1989). Early expression ofacetylcholinesterase activity in functionally distinct neurons ofthe zebrafish. J. comp. Neurol. 284, 350-361.

HISAOKA, K. K. AND FIRLIT, C. F. (1960). Further studies on theembryonic development of the zebrafish, Brachydanio rerio(Hamilton-Buchanan). J. Morph. 107, 205-225.

HISOAKA, K. K. AND BATTLE, H. I. (1958). The normaldevelopmental stages of the zebrafish, Brachydanio rerio(Hamilton-Buchanan). J. Morph. 102, 311-323.

Ho, R. K., KANE, D. A. AND KIMMEL, C. B. (1989). Celltransplantation in the zebrafish embryo: Is the spt-1 mutation cellautonomous? Soc. Neurosci. Abstr. 15, 809.

HUMASON, G. L. (1962). Animal Tissue Techniques. 3rd edition. W.H. Freeman Co.: San Francisco.

HUMPHREY, C. D. AND PITTMAN, F. E. (1974). A simple methyleneblue-azure II-basic fuchsin stain for epoxy embedded tissuesections. Stain Technol. 42, 9-14.

KAGEYAMA, T. (1982). Cellular basis of epiboly of the envelopinglayer in the embryo of the Medaka, Oryzyas latipes. II. Evidencefor cell rearrangement. J. exp. Zool. 219, 241-256.

KELLER, R. E. (1980). The cellular basis of epiboly: an SEM studyof deep cell rearrangement during gastrulation in Xenopus laevis.J. Embryol. exp. Morph. 60, 201-234.

KELLER, R. E. (1986). The cellular basis of amphibian gastrulation.In Developmental Biology: A Comprehensive Synthesis. Vol. 2.The Cellular Basis of Morphogenesis (ed. L. Browder), pp.241-327.

KELLER, R. E. (1987). Cell rearrangement in morphogenesis. Zool.Sci. 4, 763-779.

KELLER, R. E. AND DANILCHIK, M. (1988). Regional expression,pattern and timing of convergence and extension duringgastrulation of Xenopus laevis. Development 103, 193-209.

KELLER, R. E. AND TIBBETS, P. (1989). Mediolateral cellintercalation in the dorsal, axial mesoderm of Xenopus laevis.Devi Biol. 131, 539-549.

KELLER, R. E. AND TRINKAUS, J. P. (1987). Rearrangement ofenveloping layer cells without disruption of the epithelialpermeability barrier as a factor in Fundulus epiboly. Devi Biol.120, 12-24.

KIMMEL, C. B., KANE, D. A., WALKER, C , WARGA, R. M. AND

ROTHMAN, M. B. (1989). A mutation that changes cell movementand cell fate in the zebrafish embryo. Nature 337, 358-362.

KIMMEL, C. B. AND LAW, R. D. (1985a). Cell lineage of zebrafishblastomeres I. Cleavage pattern and cytoplasmic bridges betweencells. Devi Biol. 108, 78-85.

KIMMEL, C. B. AND LAW, R. D. (19856). Cell lineage of zebrafishblastomeres III. Clonal analysis of the blastula and gastrulastages. Devi Biol. 108, 94-101.

KIMMEL, C. B. AND WARGA, R. M. (1986). Tissue-specific celllineages originate in the gastrula of the zebrafish. Science 231,365-368.

KIMMEL, C. B. AND WARGA, R. M. (1987a). Cell lineagesgenerating axial muscle in the zebrafish embryo. Nature 327,234-237.

KIMMEL, C. B. AND WARGA, R. M. (19876). Indeterminate celllineage of the zebrafish embryo. Devi Biol. 124, 269-280.

KIMMEL, C. B., WARGA, R. M. AND SCHILLING, T. F. (1990). Origin

and organization of the zebrafish fate map. Development 108,581-594.

LONG, W. L. (1983). The role of the yolk syncytial layer indetermination of the plane of bilateral symmetry in the rainbowtrout, Salmo gairdneri Richardson. J. exp. Zool. 22&, 91-97.

OPPENHEIMER, J. M. (1837). The normal stages of Fundulusheteroclitus. Anat. Rec. 68, 1-15.

MORGAN, T. H. (1895). The formation of the fish embryo. J.Morph. 10, 419-472.

PASTEELS, J. (1936). Etudes sur la gastrulation des vertebresmeroblastiques. /. Teleosteens. Archives de Biologie 47, 206-308.

STREISINGER, G., WALKER, C , DOWER, N., KNAUBER, D. AND

SINGER, F. (1981). Production of clones of homozygous diploidzebra fish (Brachydanio rerio). Nature, Lond. 291, 293-296.

SYMES, K. AND SMITH, J. C. (1987). Gastrulation movementsprovide an early marker of mesoderm induction in Xenopuslaevis. Development 101, 339, 349.

THOROGOOD, P. AND WOOD, A. (1987). Analysis of in vivo cellmovement using transparent tissue systems. J. Cell Sci. Suppl. 8,395-413.

TRINKAUS, J. P. (1984a). Mechanism of Fundulus epiboly-a currentview. Am. Zool. 24, 673-688.

TRINKAUS, J. P. (19846). 'Cells into organs. The forces that shapethe embryo' New Jersey: Prentice-Hall Inc.

WILSON, H. V. (1891). The embryology of the sea bass. Bull. U.S.Fish Comm. 9, 209-277.

WILSON, P. A., OSTEJI, G. AND KELLER, R. (1989). Cell

rearrangement and segmentation in Xenopus: direct observationof cultured explants. Development 105, 155-166.

WOOD, A. AND TIMMERMANS, L. P. M. (1988). Teleost epiboly:reassessment of deep cell movement in the germ ring.Development 102, 575-585.

{Accepted 28 January 1990)