Cell lineage analysis of the amphipod crustacean Parhyale hawaiensis reveals an early restriction of cell fates

In the early embryos of several animals, such as nematodes, ascidians and leeches, complete cell cleavages generate individual cells whose fates have been determined either by following cells in living animals or through the injection of various tracers (Sulston, 1981;Nishida, 1987;Weisblat et al., 1984). In these taxa, specific patterns of invariant cell lineage are found, and thus individual cells (blastomeres) of the early embryo contribute to distinct parts of the organism. Subsequent studies, especially in the nematode C. elegans, have revealed that these lineage patterns are generated through a wide variety of mechanisms, including the asymmetric distribution of cellular components and cell-cell signaling between groups of cells whose positions are highly reproducible. By contrast, most insects, includingDrosophila, display superficial cleavage and lack patterns of invariant cell lineages during early cleavage stages. The early development of this insect is generally characterized as one in which positional information plays the major role in cell fate decisions, although invariant cell lineages are found much later in Drosophila development, particularly within the nervous system.

Many crustaceans (and a small number of insects), however, do display total cleavage during early embryogenesis, but very few studies have been undertaken to determine the extent to which invariant cell lineages occur in these animals. In some cases, observations based on tracking cell morphologies in crustacean embryos have suggested the presence of invariant lineages(Grobben, 1879;Bigelow, 1902;Fuchs, 1914; Hertzler et al.,1992), and a single study making use of the injection of a tracer has demonstrated the origin of mesendoderm material from a single blastomere at the four-cell stage in the indirect developing shrimp Sicyonia(Hertzler et al., 1994).

In this study, we describe experiments designed to trace the cell lineage pattern in the amphipod crustacean, Parhyale hawaiensis. Previous authors have noted the unique arrangement of blastomeres found in this group of crustaceans, but had not established lineage data(Langenbeck, 1898;Weygoldt, 1958;Scholtz, 1990). In amphipods,the first and second cell divisions are slightly unequal, but the third division is highly unequal and thus generates a set of four macromeres and four micromeres. Through the injection of various lineage tracers into the blastomeres of the eight-cell stage embryo, we demonstrate that the fates of the macromeres and micromeres in Parhyale are restricted to individual germ layers. The ectoderm is generated by three macromeres, the visceral mesoderm by the fourth macromere, the somatic mesoderm is generated by two micromeres, and the endoderm and the germ cells are generated by the two other micromeres. With the notable exception of germline formation in some insects (Kahle, 1908), this is the first demonstration of distinct blastomere cell fates in an arthropod. There is, however, no obvious resemblance between the lineage patterns ofParhyale and those of the nematodes, spiralians and deuterostomes,the lineages of which are known. Thus, it would appear that this level of blastomere fate determination has evolved independently in this group of crustaceans. Finally, we believe that this crustacean has several properties,including ease of culturing, ready accessibility to all embryonic stages and relatively rapid generation time, that make it a useful system for detailed analyses of many aspects of crustacean development.

Parhyale hawaiensis is a direct-developing marine amphipod that is simple to raise and propagate year-round in the laboratory, and we are currently preparing a detailed description of its care and embryonic staging(W. B., M. G., A. P. and N. P., unpublished). Parhyale hawaiensisembryogenesis takes approximately 10.5 days at 26°C and below is a very brief description of the stages that are relevant to this study.

The first three cleavages take place within the first 8 hours of development and result in an eight-cell embryo composed of four macromeres and four micromeres (Fig. 1A,B). By 12 hours, there are roughly 100 cells distributed relatively evenly at the surface of the egg (Fig. 1C)and all the cells are approaching the same size (as the macromeres have divided more than the micromeres). At 18 hours many of the cells have condensed towards specific regions of the egg and the onset of gastrulation begins shortly after this, as some cells move to more internal position within the egg. At ∼3 days (Fig. 1D), a distinct germband with head lobes is visible. At this time,the ectodermal cells begin to arrange themselves into a precise pattern of rows and columns, and this organizational process proceeds in an anterior-to-posterior direction across the germband. The initial rows that are formed undergo a subsequent precise pattern of divisions to yield individual parasegments, and again these divisions progress anterior to posterior along the germband. At 4 days (Fig. 1E), the germband has lengthened considerably and is folded in its posterior region, and appendages are clearly visible in the anterior regions of the animal. At 6 days (Fig. 1F), all the appendages are visible and internal organs such as the gut can be seen forming. At 9 days(Fig. 1G) organogenesis appears nearly complete, and the embryo has the same morphology as a hatchling, which in turn is very similar in morphology to a full grown adult(Fig. 1H).

In our study, we injected the blastomeres of two-, four- and eight-cell stage embryos to track their lineage. TF-4 needles (World Precision Instruments) are pulled with a horizontal puller P-90 (Sutter) and filled with the appropriate labeled dextran or mRNA. To keep them in place during the injection, the embryos are placed on a slide next to small strip of 2% agar in 50% seawater. Once properly oriented, the embryos are injected with an IM 300 Microinjector (Narishige) on a Zeiss Axiovert microscope. The following tracers were used: rhodamine-conjugated dextran [2.0 μg/μl TRITC dextran(M_r 500,000), Sigma], fluorescein-conjugated dextran [2.0μg/μl FITC dextran M_r 150,000) Sigma],Biotin-conjugated dextran [Biotin dextran, 1 μg/μl (70,000M_r), Sigma], and capped mRNA (1 μg/μl) encoding either green fluorescent protein (GFP) or Discosoma Red fluorescent protein (DsRed. T1) (Bevis and Glick,2001). For double injections of fluorescent tracers, several embryos were injected with one tracer, the location of the tracer was confirmed by fluorescence microscopy, the embryos were placed back on a slide,oriented appropriately and injected for a second time (in a different cell)with a different tracer.

The embryos injected with Biotin dextran were fixed in one of two ways,either by formaldehyde fixation or by boiling. Fixation by formaldehyde is done for 15 minutes in 3.7% formaldehyde in PBS (pH 7.0) at room temperature. While in the fixation solution, a hole was poked in the egg and the two outer membranes were removed with tungsten needles. When fixed in this manner (and stained as described below), it was possible to further dissect the embryos and flatten them afterwards on a slide; the tissue and the yolk stayed white and the DAPI staining was bright and clear. However, because a hole must be made initially, on either the ventral or the dorsal side, the distribution of clones could be scored accurately only on side or the other side. Fixation by boiling was achieved by immersing the embryos for 10 seconds in 95°C PBS(pH 7.0) followed by transfer to ice-cold PBS. This method of fixation makes dissections easier, the two membranes can be easily removed without damaging the embryo, and the whole embryo can be scored from all sides. However, the boiling hardens the embryos so that they cannot be flattened later, it turns the yolk a yellow color, and the DAPI staining of boiled embryos is weaker and has more background than that of formaldehyde-fixed embryos. After either method of fixation, embryos were incubated with HRP-conjugated streptavidin(Molecular Probes) at a dilution of 1:1000 in PT (PBS + 0.01% Triton), washed in PBS, developed with 1 mg/ml DAB + 0.6 mg/ml NiCl + 0.01%H₂O₂ for 10 minutes, washed in PBS, stained with DAPI at 1 μg/μl in PBS, and cleared in 50% and 70% glycerol in 1×PBS. Pictures were taken with a Zeiss Axiophot using a Kontron 3012 (Jenoptik)digital camera. Data were assembled using Adobe Photoshop 6.0.

The embryos injected with fluorochrome-coupled dextrans or mRNA for the fluorescent proteins EGFP or DsRed.T1 can be scored live over the whole period of embryogenesis. Pictures were taken at approximately 12 hours, 18 hours(just before gastrulation), 3 days (germband) and 6 days (organogenesis). Pictures were taken with a Zeiss Axiophot using a Sony digital camera. Data were assembled using Adobe Photoshop 6.0.

FITC dextran has a higher background problem because of tissue autofluorescence than does TRITC dextran. mRNAs for the fluorescent proteins EGFP and DsRed.T1 were made from expression vectors that were made by cloning the GFP- and DsRed.T1-coding regions from pEGFP-1 (Clontech) and pDsRed.T1(Bevis and Glick, 2001) into the expression vector pSP (gift of Angus MacNicol) and capped transcripts generated using the SP6 Ambion mMessageMachine kit. Expression was detected by fluorescence 1.5 hours after injection of DsRed.T1 mRNA and 2-3 hours for EGFP mRNA. The GFP signal was relatively weak in our hands, although the DsRed.T1 signal was as strong as the signal from TRITC labeled dextrans.

Biotin dextran is a useful tracer because, after fixation and subsequent enzymatic development, there is a high signal/noise ratio and the preparations are permanent. The Biotin dextran method also has a spatial resolution at the single cell level and allows for simultaneous DAPI staining. The main drawback of the Biotin dextran method is that embryos must be fixed, and thus each injection yields data for only a single time point. Fluorescent tracers are useful as they allow for continuous in vivo observation of the clones. However, this method does not allow us to collect DAPI data simultaneously,and there is a loss of fluorescent signal upon fixation. TRITC dextran and FITC dextran labeling provide excellent spatial resolution until gastrulation;after that, this technique does not produce as good a spatial resolution as the Biotin dextran method (although this may be resolved by improved optical techniques). The fluorescent proteins GFP and DsRed.T1 show the same advantages and disadvantages as the fluorochromes. In some cases, we have used a 1:1 mix of Biotin dextran plus one of the fluorescent tracers to take advantage of the strengths of each technique.

During the first 8-9 hours of development, three complete cleavages of theParhyale embryo result in the formation of an eight-cell embryo(Fig. 1A). It is the third cleavage that is highly asymmetric and yields the eight-cell pattern of four macromeres and four micromeres. The individual macromeres and micromeres can be distinguished according to their size, morphology and contacts made with neighboring cells (Fig. 1A). We have named these cells in accordance with their eventual progeny(Fig. 1B, see below). The smallest macromere is named `Mv', the other three macromeres, when viewed from the dorsal side, are named clockwise `Er', `Ep' and `El'(Fig. 1B). The smallest micromere (which is the sister cell of the smallest macromere `Mv') is named`g', and this micromere has the most prominent nucleus of all the micromeres. The other three micromeres, when viewed from the dorsal side, are named clockwise (starting from `g') `mr', `en' and `ml'(Fig. 1B). As the next few paragraphs will demonstrate, the prospective fate of the blastomeres is distributed in the following way: `Mv' generates all of the visceral mesoderm.`El', `Ep' and `Er' generate three different portions of the ectoderm (roughly distributed anterior left, posterior and anterior right, respectively, in the later embryo). The cells of the `g' clone are the germ cells. `ml' and `mr'generate the left and right regions of the somatic mesoderm, respectively.`en' generates the endoderm.

In our initial experiments, we injected individual cells of the eight-cell embryo with either Biotin-dextran or DsRed.T1 mRNA as lineage tracers (see Materials and Methods). We then analyzed the distribution of clones in germband stages (3-4 days of development) and during organogenesis (6-7 days of development). Table 1summarizes the number of clones analyzed at various stages for each injected blastomere. The fate of `El', `Ep', `Er', `ml' and `mr' is easy to recognize at the germband stage, but the fate of `g', `en' and `Mv' only become clear at about 6 days of development when organ formation has begun. Having established the fate of these clones, we then went back to analyze the distribution,proliferation and migration of the clones during earlier stages (between the time of injection and the establishment of the germband at 2-3 days). Below,we begin with a description of the fate of the clones at the germband and organogenesis stages, and then describe the way in which these clones behave and move during earlier stages of development.

The progeny of `El', `Er' and `Ep' are strictly ectodermal; all their progeny are restricted to the ectoderm (and ectodermal derivatives such as the nervous system) and the entire ectoderm can be traced back to these three macromeres.

At ∼3 days, when the initial germband is well organized, the cells from`El' and `Er' make up, respectively, the anterior left and anterior right ectoderm of the germband, and arrange themselves into a very precise pattern of rows and columns of cells (Fig. 2A,C). The `Ep' clone forms the posterior ectoderm of the germband(Fig. 2E), the cells of this clone will also eventually organize into rows and columns, but do so later than the more anterior ectoderm (Fig. 2F). In addition, the cells of the `Ep' clone also produce the midline of the ectoderm extending all the way up to the gnathal region of the embryo, and thus generate the central midline that separates the `El' and `Er'clones (Fig. 2E,F) along much of the length of the embryo. Interestingly, distinct behaviors of midline cells have also been found at later stages of development in the amphipodOrchestia (Gerberding and Scholtz, 1999; Gerberding and Scholtz, 2001).

The `El', `Ep' and `Er' clones intermix but remain a monolayer. As the germband first begins to form, `El' and `Er' form clones with a relatively small cell size and high cell density positioned at the anterior part of the forming germband. By contrast, the `Ep' clone displays a relatively larger cell size and lower cell density and is spread out over a region of the posterior ventral side and most of the posterior dorsal side of the egg; as the germband continues to condense, the `Ep' clone proliferates and compacts to form the most posterior part of the germband (data not shown). Given that the germband is formed by the condensation of cells from the surface of the egg, it is not surprising that some mixing of the cells between the three clones does occur, but the mixing is restricted in a predictable way. In the head (anterior to the future gnathal region), there is no distinct midline and cells from `El' and `Er' mix extensively across the midline. In the region of the gnathal and thoracic segments, however, the `Ep' clone establishes a well-defined midline, and the `El' and `Er' clones maintain a strictly left-right distinction (Fig. 2A-D). The anteroposterior boundary between `El'+`Er' domain versus the `Ep' clone varies from embryo to embryo, but is generally somewhere within the posterior thorax or anterior abdomen, in a few cases the contribution of `Ep' can be surprisingly small(Fig. 2B,D,F). Possibly, this variability in the composition of the germband ectoderm from the three clones results from the variable degree of inequality in the first and second cleavages that determine the relative sizes of the different macromeres. This boundary is usually quite irregular (i.e. not defined by any specific row of cells), and in addition, there can be scattered cells from `El' and `Er' that end up incorporated in a random manner into the developing abdomen. In summary, `El', `Er' and `Ep' clones can be characterized as occupying anterior left, anterior right and posterior ectoderm, respectively, but with some expected patterns of mixing occurring.

The `ml' and `mr' clones were analyzed during germband formation and organogenesis and were found to generate the somatic mesoderm and produce no other cell type than somatic mesoderm.

The germband mesoderm is assembled out of two clones. At the germband stage, the clones originating from `ml' and `mr' are found immediately underneath the ectoderm (Fig. 3A). The cells originating from `ml' and `mr' are found on the left and right sides of the embryo respectively, and we never observed any violation of this left-right allocation. At day 3, each unilateral clone is subdivided into two populations, one consists of a randomly arranged anterior population of cells (non-teloblastic mesoderm) that will form the mesoderm of the head and heart, and the second is a stereotypically arranged set of posterior cells (teloblastic mesoderm) that will go on to form all the rest of the somatic mesoderm (Fig. 3A,B).

The anterior, non-teloblastic parts of the `ml' and `mr' clones form the mesoderm of head (and its associated appendages) and two distinct circular structures on either side of the head that are not associated with segments or appendages (Fig. 3B). During organogenesis, these circles disperse and the cells migrate dorsally(Fig. 3D), moving jointly with endoderm and visceral mesoderm (see below). Before hatching, labeled cells can also be found mediodorsally in the putative heart rudiment (data not shown).

The posterior region of the mesoderm is the teloblastic mesoderm. On each side of the embryo, four mesodermal stem cells called mesoteloblasts differentiate at the very posterior end of the clones(Fig. 3A). As these stem cell divide, they move posteriorly one segment at a time in the embryo, leaving behind a row of four progeny in each segment as they do so. This establishes a pattern of four mesodermal precursor cells per segment on each side of the embryo (Fig. 3B). By day 4.5,the mesoteloblasts have finished the generation of segmental mesoderm progeny. These segmental mesodermal progeny then begin to divide(Fig. 3C) and eventually form the muscle cells of the appendages and body wall(Fig. 3D,E).

The lineage of `g' is restricted to the germ line and there is no other source for the germ cells. During development, the `g' clone settles at the prospective gonads.

During early germband formation at day 3, the cells in the `g' clone lie in single cluster underneath the ectoderm at the level of the future mandibular segment (Fig. 4A). As development proceeds, the clone splits at the midline into two bilaterally symmetric populations of cells (Fig. 4B) and by day 4 reach a position lateral to the germband at the level of the future gnathal segments (Fig. 4C). The two clusters of cells derived from `g' keep migrating until they reach the dorsal median where the heart rudiment forms. By day 9,the cells of the `g' clone are found within the developing gonads in a dorsal position adjacent to the gut at the level of the fourth thoracic segment(Fig. 4D), and it is at this position that the ovaries and testes are centered in adult animals.

The fate of `en' and `Mv' clones is most obvious during organogenesis (6-7 days). At this stage, it is clear that `en' and `Mv' generate the gut, i.e. endoderm and visceral mesoderm. Endoderm and visceral mesoderm are in immediate proximity and therefore so close together that overlap can not be exclude in all cases. However, the variety of methods used all provide evidence that `en' generates the entire endoderm and `Mv' generates the entire visceral mesoderm, and the lineages are restricted to these tissues respectively.

During organogenesis, the endoderm and visceral mesoderm clones form the gut tube around a central yolk. The central yolk is formed early by a separation of the outer cytoplasm from the inner yolk. The redistribution of the cytoplasm can easily be visualized as all the three dextran-coupled tracers as well as the DsRed protein are found preferentially in the cytoplasm. From the one-cell to the eight-cell stage, cytoplasmic signal is found throughout the whole injected cell. Starting at the 16-cell stage, the cytoplasm becomes successively localized at the surface and excluded from the inner yolk. This observation, however, does not allow us to conclude whether the inner yolk is cellular or acellular. If the yolk is cellular, the yolk cells could either originate from cells that stay central from early on or from cells that settle within the yolk secondarily. Alternatively, an acellular yolk could be formed by having the cells divide without a corresponding nuclear division. Indeed, previous observations of dissociated cells from early embryos of gammarid amphipods suggested that the cytoplasm and the yolk become separated by tangential divisions of the cells without divisions of their nuclei, resulting in small outer cells with cytoplasm and nuclei, and a bigger inner yolk compartment that is anuclear(Rappaport, 1960).

At germband formation, the `en' clone comprises no more than eight cells(data not shown). The clone is the most dorsal clone and its cells are flat and spread out over the dorsal yolk. At the same time, the `Mv' clone comprises about two dozens of cells and is located between the ectoderm clones of `Er' and `El' and the dorsal `en' clone (data not shown). During formation of the midgut, the `en' clone extends ventrally and posteriorly from its initial dorsal anterior position (Fig. 5A-D), while the `Mv' clone extends dorsally and posteriorly from its initial anterior lateral position (Fig. 5E-G). During this movement, both clones extend together and the leading edge comprises cell of both clones. In double labels with FITC- and TRITC-labeled dextrans, however, it is clear that `en' cells are situated internal to the `Mv' cells (Fig. 5J-L). At the end of this extension process, `en' and `Mv' form the two layered sheath of midgut around a tube-shaped yolk mass(Fig. 5D,H). We also analyzed sections through the midgut of labeled 7-day-old embryos to confirm that the`en'-derived cells were internal to the `Mv'-derived cells (data not shown).

Having analyzed the cell fate of each of the four macromere and four micromere lineages at germband and later stages, we decided to investigate some earlier stages in order to understand more about the behavior of these lineages in the steps leading up to the formation of the initial germband. We focused our analyses on two stages. First, at 12 hours of development, at which time the cells are more or less uniformly distributed around the surface of the egg (Fig. 1C,Fig. 6B). Owing to the appearance of the embryo at this time, we have nicknamed this the `soccerball'stage. Second, at 18 hours of development, when gastrulation is just about to begin. At this stage, there is a rosette shaped cluster of cells that is easily visible in living embryos (Fig. 6C), and we have nicknamed this the `rosette' stage.

The events prior to germband formation are addressed by double fluorescent labels. Because there are no clear morphological landmarks for orienting the embryo at the soccerball stage, it is difficult to compare the position of individually labeled clones. In addition, the overall shape of the egg at this stage does not provide a reproducible way of orienting the embryo forming within. However, by injecting pairs of blastomeres with different tracers (one with FITC dextran and another with TRITC dextran, see also row four ofTable 1) we found that we could understand the relative orientation of all the clones. The `Mv'+`Ep' and`g'+`en' pairs are situated along the AP axis, and the AP axis is the line connecting them. The `El'+`Er' and `ml'+`mr' pairs are situated to the right and the left side of the AP axis, therefore the line connecting them is perpendicular to the AP axis.

Proliferation rates and relative area are different between clones and change over time within clones. During the 4 hours between the eight-cell stage and the 100-cell soccerball stage, the numbers of cells, in each clone,the area on the egg surface that they cover, and their relative locations are all changing simultaneously. At the soccerball stage, `El', `Ep' and `Er'comprise about 12 to 15 cells (Fig. 6E), `Mv' comprises about eight cells(Fig. 6B). The micromeres have undergone two to three divisions, and thus there are four to eight progeny of each micromere at the soccerball stage(Fig. 6H,L). At the eight-cell stage, the macromeres of course cover a greater area than the micromeres. By the soccerball stage, progeny of the three macromeres `El', `Ep' and `Er' have increased their area relative to the progeny of macromere `Mv'. The progeny of micromere `en' increase their area of coverage relative to the other micromere lineages (Fig. 6H).

Migration patterns are different between clones. Relative to each other,the clones move extensively up to the formation of the germband and beyond. The clone that is proliferating and moving the least seems to be the `en'clone; thus, we define the position of the `en' clone to be fixed, and describe the movements of the other cells relative to the `en' clone. After the eight-cell stage, the progeny of the three other micromeres, `mr', `g' and`ml', leave their dorsal and superficial positions next to `en' and migrate to ventral and internal positions. During the movements, they pass through the anterior tip of the egg. The `g' clone takes a medial path and crosses over the `Mv' clone (Fig. 6H,J). The`mr' and `ml' cells take lateral paths (right and left) and do not cross the progeny of `Mv' (Fig. 6L,M). The two macromeres `Er' and `El' expand from their ventromedial position towards the anterior tip of the egg (Fig. 6E,F). The `Ep' cells follow `Er' and `El' anteriorly, but remain posterior to them at all times (Fig. 6B,C). In terms of movements, the behavior of the `Mv' clone is simialr to the `en' clone in that its cells move very little and remain superficial until germband formation (Fig. 6B,C).

During the rosette stage, the somatic mesoderm ingresses laterally, the endoderm remains superficial. At the rosette stage, the cleavage mode changes to superficial cleavage and cells start to condense at two locations. One of these condensations has the shape of a rosette that consists of about 15 cells and is situated at the future dorsal side. The rosette can be further subdivided into a central, deeper ring of cells and an outer, more superficial ring of cells. The other condensation appears a bit later than the rosette and is located at the future ventral side of the embryo. It comprises more cells than does the rosette, but all these cells remain superficial. Double labels elucidate the differential contributions of the clones to these two condensations. The deeper cells of the rosette are the `Mv' progeny(Fig. 6C,J). Double labels of`El'+`Er' show that the ventral condensation initially comprises cells derived from `El' and `Er'. The cells joining it later and more posterior originate from `Ep' (data not shown). Double labels of `ml' and `mr' show that the progenitors for the somatic mesoderm ingress separately at the anterior-right and anterior-left edge of the ventral condensation a few hours after the rosette stage (data not shown). The `en' clone remains superficial throughout the soccerball stage and only ingresses during the germband stage.

Through the observation of living embryos and the injection of tracers into two- and four-cell embryos, we were also able to deduce the division pattern that leads to the eight-cell stage (data not shown). A summary of the results is shown in Fig. 7.

There are two arrangements of the blastomeres at the eight-cell stage that show mirror symmetry (Fig. 7E). In one case, the sister cells `Mv' and `g' are located to the left of the prospective AP axis, in the other case, they are located to the right. The two arrangements also affect the cell pedigree. If `Mv' and `g' are to the left,they share a common progenitor at the two-cell stage with `Er' and `mr'(Fig. 7F). If they are located to the right, they share a progenitor with `El' and `ml'(Fig. 7G). Note that this does not affect the architecture of the pedigree, in both cases, each of the two cells at the two-cell stage generates paired ectoderm and mesoderm progenitors and non-paired progenitors for either ectoderm and endoderm or mesoderm and germ cells. A similar mirror symmetry pattern during early cleavage that still yields identical embryos has also been reported for other crustaceans and for spiralians (Baldass, 1941;Luetjens, 1995).

Some progenitors for the germ layers are paired, some unpaired. The four clones derived from macromeres `Mv' and `Ep' and micromeres `g' and `en'demarcate the AP axis of the embryo and start as bilateral cell populations situated on the median axis. Conversely, the four clones derived from macromeres `El' and `Er' and micromeres `ml' and `mr' begin as unilateral cell populations on either the left or the right side of the embryo. The `ml' and`mr' clones maintain their perfect left/right allocation while the `El' and`Er' clones display some left/right mixing (see above).

The blastomeres at the eight-cell stage can be depicted as a fate map that predicts where the daughters of the blastomeres end up at the germband(Fig. 7A-D). At the eight-cell stage, the material for the germ layers is located along the AP axis in the following orientation: `Mv' most anterior; `g'; `Er'/`mr'/`ml'/`El' in the middle; `en'; `Ep' most posterior (Fig. 7A,C). At the germband stage (i.e. after the initial processes of proliferation, migration and mesoderm ingression), the material is reconfigured along the AP axis. The order then is endoderm `en', visceral mesoderm `Mv', anterior ectoderm `El'+`Er', germ cells `g' and somatic mesoderm `ml'+`mr' (underneath `El' + `Er'), and `Ep' derived ectoderm at the posterior (Fig. 7B,D). During organogenesis, there are further rearrangements so that the endoderm `en' and visceral mesoderm `Mv' have formed the midgut, which runs almost the entire length of the embryo.

We labeled specific blastomeres at the eight-cell stage in the crustaceanParhyale hawaiensis and analyzed the resulting clones at subsequent stages. In each of the eight-cell lineages, patterns of proliferation, changes in shape and migration are distinct and invariant. More surprising, each blastomere contributes to only one of the germ layers. Each of the eight cells after the third cleavage gives rise exclusively to either germline, ectoderm,mesoderm or endoderm. The germ layers are derived from either one cell, as in the case of the germ cells and the endoderm, or three cells, as in the case of the mesoderm and the ectoderm. In short, the main features of the cell lineage patterns in Parhyale are their simplicity and their exclusivity. This is the first time that arthropod blastomere cell lineages have been followed through germband formation up to hatching. Our results provide new material for the study of the evolution of arthropod development. The findings can also be used to analyze the independent evolution of cell lineages in the Bilateria and the extent to which they share common features. Finally, we believe that our results indicate that Parhyale may be a useful system for the study of many aspects of crustacean development and suggest that experimental manipulations may be possible in this organism that are not feasible in arthropods with superficial cleavage.

Previous studies of crustacean development had established early fate maps for several species (reviewed by Shiino,1957; Anderson,1973; Weygoldt,1994). With one exception, these fate maps are not the result of labeled lineage analysis, but instead are based on tracing cells of particular morphology during the first few division of the embryo (and usually in sectioned material). For example, in several species, germline cell are picked out because of the unique appearance of the cytoplasm and their relatively slow proliferation rate and the endoderm is picked out by its very internal position in the embryo. Furthermore, these lineage analyses do not follow the fate of the cells up to the time that the final body plan is established. While these fates maps are incomplete, and need to be tested by the injection of lineage tracers, they nevertheless help to illustrate the diversity seen in early crustacean development (Weygoldt,1979). The fate map and cell lineage pattern we have established here for Parhyale bears similarities to that in other crustaceans,but surprisingly not to those of closely related malacostracan taxa, but instead to those of more distantly related non-malacostracan taxa.

Crustaceans are generally divided into five major groups of largely unresolved evolutionary relation, Remipedia, Cephalocarida, Branchiopoda,Maxillopoda and Malacostraca, and it is the latter to which Parhyalebelongs. Our results allow us to establish a fate map for Parhyaleand compare it with descriptions of other crustacean fate maps at a similar stage. By definition, the position where the prospective mesoderm is internalized is the blastopore. In Parhyale the prospective mesoderm ingresses underneath the ectoderm in an arc and thus the blastopore is more like a lip (A. Price and N. Patel, unpublished), while in at least a few other crustaceans a simpler pore-shaped blastopore exists. In Parhyale, the prospective endoderm is situated anteriorly of the prospective mesoderm(Fig. 8A). The inner germ layers move beneath the ectoderm from an anterior blastopore. In the closest relatives of the amphipods, the peracaridan malacostracans, the blastopore is located posterior to the ectoderm and the endoderm is located posterior of the mesoderm (McMurrich, 1895;Manton, 1928). In more distantly related malacostracans that have total cleavage, the endoderm is located dorsal, not anterior, to the mesoderm(Fig. 8E)(Taube, 1909;Hertzler, 2002). However, in the case of the branchiopods, the situation is similar to Parhyale. The blastopore is posterior, but the endoderm material is situated in front of the mesoderm (Grobben, 1879;Kühn, 1913;Baldass, 1941;Weygoldt, 1994)(Fig. 8C,I). In the maxillopodans (cirripeds and copepods), the situation is like that inParhyale and the branchiopods(Fig. 8C,G)(Bigelow, 1902;Fuchs, 1914;Delsman, 1917;Shiino, 1957). The lack of similarity between the fate map of Parhyale and those of more closely related malacostracan crustaceans and its similarity to those of distantly related non-malacostracan crustaceans suggests that this may be an example of convergent evolution.

Some cell lineage data, again based usually on tracing cells by their morphology, is also available for several crustacean taxa with total cleavage. In some of the taxa with total cleavage, cleavage is equal in the sense that early blastomeres cannot be distinguished(Müller-Calé,1913; Benesch,1969). In others, cleavage is more or less unequal and early blastomeres are distinguishable and are described as progenitors for germ layers. In general, the putative progenitors for mesoderm, endoderm and germ cells are derived from few cells at an early stage. A comparison betweenParhyale and these other crustaceans reveals that different numbers of cell cycles can occur before lineages are restricted to specific germ layers and different numbers of cells are used to generate the germ layers. For reasons of space, only comparisons for the endoderm and germ line are discussed here.

In the malacostracan shrimp, Sicyonia, the endoderm originates jointly with the mesoderm and the germline from one of the four blastomeres after the third division, as the only previous lineage tracer injection experiments carried out in crustaceans shows(Hertzler et al., 1994). The fifth and sixth division at the 31- and 62-cell stage each generate an endoderm progenitor (Fig. 8F)(Hertzler, 2002). In the maxillopodan barnacles, the fourth division at the 16-cell stage generates single progenitors of endoderm and mesoderm, and the endoderm progenitor is considerably bigger than all other cells(Fig. 8C)(Bigelow, 1902;Delsman, 1917;Shiino, 1957;Anderson, 1969) In the branchiopodan waterfleas, the fourth division at the 16-cell stage sets up single endoderm and germline progenitors of average size(Fig. 8I) (Grobben, 1878;Kühn, 1913). In the maxillopodan copepod Cyclops, the same process happens at the fifth division at the 32-cell stage (Fig. 8G) (Fuchs, 1914). This five taxa comparison shows that the endoderm progenitor `en' ofParhyale is generated earlier than in other crustaceans. In addition,it is a micromere that is the sister of the ectoderm progenitor `Ep'(Fig. 8B); in other crustaceans, however, the endoderm progenitors are either average sized or macromeres and, in lineage terms, are most related to the progenitors of either the mesoderm or the germ line (Fig. 8D,F,H,J).

Similarly, a comparison between our results for Parhyale and those for other crustaceans shows that the germ line progenitor `g' inParhyale is generated earlier than in other crustaceans. InSicyonia, the germ cell lineage separates from the mesoderm at the seventh division at the 122-cell stage(Fig. 8F)(Hertzler, 2002). In the barnacles, no germline is detected at the 64-cell stage(Bigelow, 1902;Delsman, 1917;Shiino, 1957;Anderson, 1969). In the waterfleas and copepods, the germline is set up as the sister cell of the endoderm progenitor at the 16- and 32-cell stages, respectively(Fig. 8H,J) (Grobben, 1878;Kühn, 1913;Fuchs, 1914).

These comparisons of the relative timing of cell lineage restrictions can be extended outside the crustaceans as well because the determination of germ layers is an ancient process that dates back to the common ancestor of protostomes and deuterostomes. The insect Miastor, a midge, offers an example where a germline progenitor is separated from the rest of the egg at the syncytial eight-cell stage by the deployment of a membrane surrounding only the germline progenitor (Kahle,1908). Outside of the arthropods, early embryonic patterns of invariant cell lineage are found in species off nematodes, annelids and ascidians. C. elegans generates a gut progenitor at the third division and a germline progenitor at the fourth division(Fig. 8K)(Sulston et al., 1981). In basal spiralians, the progenitors for the endoderm and mesoderm are generated at the sixth cell division, in leeches and other clitellates they are generated much later (Fig. 8L);the germline in clitellates is found to descend jointly with the muscle mesoderm from mesoderm stem cells (Goto et al., 1999; Kang et al.,2002). In ascidians, the various germ layers are composed from the combination of many cell lineages of the 64-cell stage; a germline has not been detected by this stage (Nishida,1987). In conclusion, germ layer determination usually takes place between the third and sixth division and the generation of germ cells can occur much later than this. It is debatable whether there are features of germ layer determination that are homologous between arthropods, nematodes,spiralians and ascidians, but it is clear that the lineage patterns found inParhyale are particularly noteworthy because they occur much earlier than in other animals.

Cell lineage and cell fate are linked to various degrees in different developmental systems (Goldstein and Freeman, 1996; Moody,1999). The nematode C. elegans has an invariant cell lineage, and some aspects of cell fate are linked to cellular asymmetries set up during the pattern of cell division, but other experiments show that several cell fate decisions can be uncoupled from the cell division pattern(Schnabel, 1997). Within annelids, comparisons of cell division patterns that at first appear rather different do reveal the conservation of certain patterns, hinting that a certain series of divisions may be necessary to determine the different cell fates and are therefore conserved during evolution(Schneider et al., 1991;Dohle, 1999). In the frogXenopus, the arrangement of blastomeres varies, but if embryos are selected for a stereotypic arrangement of blastomeres, the resulting lineages are invariant in the sense that blastomere fate is predictable and restricted(Moody, 1987a;Moody, 1987b). However, the fact that there are also non-stereotypical arrangements that give rise to identical animals demonstrates a primacy of regional determinants over cell lineage (Moody, 1990). Although we find a stereotyped arrangement of micromeres and macromeres inParhyale, and an invariant lineage pattern with regards to the formation of different germ layers, we do not know how yet how tightly cell fate is tied to cell lineage in Parhyale. The isolation of blastomeres as done in shrimp (Kajishima, 1951;Hertzler et al., 1994;Wang and Clark, 1996), and cell ablation experiments will allow us to investigate this issue and assess the contributions of cell lineage and positional information duringParhyale development. In addition to these questions of cell fate determination during early embryogenesis, we believe that Parhyaleholds promise as a useful crustacean for the investigation of many developmental problems, particularly comparative questions aimed at understanding the evolution of pattern formation within the arthropods.

We thank Carsten Wolff, Gerhard Scholtz and Phil Hertzler for sharing results on Orchestia and Sicyonia development prior to publication, Mark Martindale and Eric Wieschaus for their suggestions on the use of Biotin dextran and heat fixation respectively, Angus MacNicol for the pSP plasmid, Brooke Bevis and Ben Glick for the DsRed.T1 plasmid, and Yu-Chiung Wang for his help in constructing the EGFP plasmid for mRNA synthesis. We are especially grateful to Oliver Coleman for identifying the species we work with as Parhyale hawaiensis. We also thank Alivia Price for many helpful discussions and contributions including comments on the manuscript, and all the members of the Patel laboratory involved in maintaining our Parhyale colony. We thank the anonymous reviewers for their suggestions. M. G. and N. H. P. are, respectively, a research associate and an investigator of Howard Hughes Medical Institute.