Refinement of gene expression patterns in the early Xenopusembryo

One of the most remarkable aspects of embryonic development is its precision. For example, differentiating tissue invariably consists of a coherent mass of like cells with a sharp border separating them from other cell types (Gurdon, 1988). And at a more quantitative level, measurements show that the lengths of limbs on the left- and right-hand sides of the body are typically within 99% of each other (Summerbell and Wolpert,1973). In the early Xenopus embryo, the first example of developmental precision concerns the specification of ectoderm, mesoderm and endoderm, where the proportions of cells allocated to each germ layer are remarkably constant from individual to individual and where the embryo is able to regulate these proportions following experimental intervention(Cooke, 1989).

This developmental precision requires that the spatial and temporal expression patterns of genes involved in germ layer specification are themselves tightly regulated. How does this regulation occur? Cell transplantation experiments suggest that it is a gradual process. Cells of the vegetal region become committed to an endodermal fate through the action of the localised maternally encoded transcription factor VegT(Clements et al., 1999; Xanthos et al., 2001; Yasuo and Lemaire, 1999; Zhang et al., 1998). However,although these cells are specified to form endoderm from early in development,they are not completely committed to this fate until the early gastrula stage(Heasman et al., 1984; Wylie et al., 1987). Similarly, transplantation experiments reveal that animal cells also become committed to an ectodermal fate only gradually(Domingo and Keller, 2000; Snape et al., 1987), and indeed they are capable of changing their state of specification in response to mesoderm-inducing factors until gastrula stages(Heasman, 1997; Kimelman and Griffin, 2000). And most remarkably, equatorial cells, which become specified as mesoderm in response to signals derived from the vegetal hemisphere of the embryo, do not become irrevocably committed to their fate until the end of gastrulation(Godsave and Slack, 1991; Kato and Gurdon, 1993).

Many genes have been identified that are expressed in these different regions of the embryo. Sox17α, for example, is expressed in the prospective endoderm, the vegetal mass, at the early gastrula stage(Hudson et al., 1997), and Goosecoid is expressed in prospective anterior mesendoderm(Cho et al., 1991). In this paper, we use in situ hybridization and single-cell RT-PCR in an attempt to understand when the expression patterns of such genes become restricted. Our results at the early gastrula stage reveal a large number of cells expressing the `wrong' gene in the `wrong' place. For example, in situ hybridization and single cell RT-PCR reveal that in the early gastrula Goosecoid is expressed in significant numbers of ventral, Xwnt8-expressing cells,as well as in dorsal tissue, and Sox17α expression can be found in cells of the marginal zone as well as vegetal cells. In contrast, at the late gastrula stage very few of these `rogue' cells can be detected. Indeed,by late gastrula stages, the gene expression profiles of cells within the same region of the embryo have refined significantly, and the profiles of cells become more characteristic of their germ layer membership. Thus, these results reflect at the level of gene expression the embryological observation that cells of the early Xenopus gastrula become progressively and asynchronously committed to a specific germ layer.

In addition, we show that when ectodermal cells are exposed to the mesendoderm-inducing factor Activin, single cells respond by activating markers of both endoderm and mesoderm in the same cell, as well as markers of ventral and dorsal mesoderm. They thus recapitulate the gene combinations seen in the marginal zone of an early gastrula embryo.

Xenopus laevis embryos were maintained in 10% Normal Amphibian Medium (Slack, 1984), and staged according to Nieuwkoop and Faber(Nieuwkoop and Faber, 1967). All culture solutions and agarose dishes for dissection were freshly made and solutions were filter sterilized. This eliminated bacterial or fungicidal DNA/RNA contamination that could otherwise be amplified in the RT-PCR procedure.

Explants were dissociated in calcium-magnesium free medium (CMFM)(Sive et al., 2000). Single cells were picked in 0.5 μl of CMFM using a P2 pipette, transferred into 4.75 μl lysis mix [1×PCR buffer (Roche), 1.5 mM MgCl₂,0.005% Igepal, 5 mM DTT, 0.05 mM dNTPs, 0.2 ng/μl anchor primer(TATAGAATTCGCGGCCGCTCGCGAT₂₄; PAGE purified), 0.3 U/μl Prime RNase Inhibitor (Eppendorf) and 0.4 U/μl RNase Inhibitor (Roche)] in thin-walled PCR tubes or 96-well plates, and then subjected to RT-PCR described below. Medium (0.5 μl) was taken as a cell-free control for each sample.

Cells were dissociated from whole explants and picked for RT-PCR within 15 minutes to minimize the effects of dissociation on gene expression. Cells from ectodermal explants were taken from the inner layer only, as the outer layer is more resistant to dissociation. Explants contained between ∼200 cells(endodermal explant, stage 10; endodermal cells are larger than marginal or animal cells) and ∼1000 cells (marginal zone explants, stage 12). For the experiments described in Fig. 6, ectodermal explants were isolated at stage 10 and incubated in 20 U/ml Activin protein (Cooke et al.,1987) for up to 4 hours. Explants collected at different time points were dissociated in CMFM and handled as described above.

In situ hybridization was carried out on bisected embryos as described(Sive et al., 2000).

PCR tubes containing lysed cells were heated to 65°C for 1 minute then 0.25 μl of RT mix [133 U/μl Superscript II (Invitrogen), 1.7 U/μl Prime RNase Inhibitor (Eppendorf), 2.2 U/μl RNase Inhibitor (Roche), 1.1μg/μl T4 gene 32 product (Roche)] was added to each tube. The RT reaction was incubated at 37°C for 90 minutes, 50°C for 50 minutes,then heat inactivated at 65°C for 10 minutes. Next, 5 μl of tailing mix[1×PCR buffer (Roche), 1.5 mM MgCl₂, 3 mM dATP, 0.75 U/μl rTdT (Invitrogen), 0.05 U/μl RNaseH (Roche)] was added to each tube and the reaction incubated at 37°C for 20 minutes, then heat inactivated at 65°C for 10 minutes. Carrying out reverse transcription and terminal transferase reactions in PCR buffer increased the efficiency of these reactions compared to using standard RT and TdT buffers. PCR mix (100 μl)[1×Taq buffer (Takara), 0.25 mM dNTPs, 20 ng/μl anchor primer, 0.05 U/μl EX Taq Polymerase (Takara)] was then added to each tube and incubated as follows: 95°C, 2 minutes; 37°C, 5 minutes; 72°C, 20 minutes;then 40 cycles of 95°C 30 seconds; 67°C, 1 minute; 72°C, 6 minutes plus 6 seconds extension for each cycle; then 72°C, 10 minutes. cDNA (5μl) from each reaction was run on a 2% agarose gel to check for amplification and integrity. cDNA (5 μl) from each reaction was then Southern blotted and/or dot blotted, and probed for each marker using standard protocols (Saitou et al.,2002; Sambrook et al.,1989). Samples showed the same gene expression profile regardless of whether the cDNA was Southern or dot blotted, and most samples were therefore dot blotted for experimental convenience.

For the split cell lysate assay, 36 single cells were picked from four regions of the embryo and the cell lysate was split equally between 2 tubes (A and B samples; Fig. 2). Each sample was then subjected to RT-PCR, dot blotted and probed for up to 8 markers. For spiking experiments, single cell lysates were spiked with 2×10^–6pg, 2×10^–5pg,1×10^–4pg, 2×10^–4pg,1×10^–3pg, 2×10^–3pg and 1×10^–2pg of polyadenylated eGFP mRNA in 1 μl of water and subjected to single cell RT-PCR as described. These correspond to∼1, 10, 50, 100, 500, 1000 and 5000 transcripts, respectively. Ten transcripts are reliably detected by this method. Fig. 4 shows a dot blot where samples from two separate RT-PCR reactions were analyzed at the same time.

For each marker gene, 300-450 bp of DNA sequence positioned no more than 500 bp from the poly-A tail was PCR amplified from gastrula stage cDNA, and used as a template to make probe (PCR primer sequences available on request). Probes, incorporating ³²PdCTP, were made using Prime-It RmT Random Primer Labelling Kit (Stratagene) according to the manufacturer's instructions. Samples showing a signal above background were scored as positive, and absence of signal was scored as negative. Signal intensity was not taken into account, and nor were signals normalized for levels of ODC or EF1α, which are frequently used as loading controls in experiments studying gene expression in Xenopus. This is because if more than 100 transcripts (as assayed by spiking experiments) of an mRNA are present in the reaction, the signal no longer falls within the linear range (Fig. 4).

Total RNA used as a template for real-time RT-PCR in the LightCycler instrument (Roche) was prepared from five pooled ectodermal explants as described (Trindade et al.,2003). Primers and PCR conditions for Xbra, Mix.1 and Sox17α were as described(Kofron et al., 1999; Xanthos et al., 2001). XK70A primers were: 5′ CGACCACCAGTCTTTGGAGTATAAG (forward) and 5′ TCGGATGCGTTATCCCTAAGG (reverse). PCR conditions for XK70Awere as follows: melting temperature, 95°C; annealing temperature/time,58°C/10 seconds; extension temperature/time, 72°C/10 seconds;acquisition temperature/time, 83°C/3 seconds.

Goosecoid (Gsc) is expressed in the organiser at early gastrula stages (Cho et al.,1991). However, whole-mount in situ hybridization analysis of bisected Xenopus laevis embryos at early gastrula stages (stage 10-10.5) reveals that Gsc-expressing cells, which are surrounded by non-expressing cells, can be detected further from this dorsoanterior mesendodermal domain, in the ventral marginal zone and vegetal mass(Fig. 1). Expression in these cells is not as strong as in the canonical domain and often appears nuclear and peri-nuclear. These `rogue' cells were seen both on the cut surface of the embryo and on the outer surface of the embryo in 42% of bisected embryos sampled at the early gastrula stage (n=56).

Analysis of other markers that are expressed in the early embryo in a restricted pattern revealed a similar phenomenon of individual `rogue' cells or of expression of a marker straying into the non-canonical domain. Xwnt8, for example, is a marker of ventral and lateral mesoderm(Christian et al., 1991). However bisected embryos hybridized with an anti-Xwnt8 probe showed individual dorsal and vegetal cells expressing Xwnt8 at stage 10-10.5(24% vegetal expression, 3% dorsal expression; n=33; Fig. 1). These cells were seen on both the cut surface and outer surface of the embryo. Expression of Sox17α, which is expressed strongly in the vegetal mass, was also seen to stray into the marginal zone at stage 10-10.5 (40%; n=35; Fig. 1) with no clear boundary of expression. When embryos are bisected horizontally these cells are adjacent to Sox17α non-expressing marginal zone cells.

These observations were confirmed by assaying expression of these genes in single cells by RT-PCR (see Materials and methods; Fig. 2). In addition, we looked at a range of other markers expressed in the early gastrula stage embryo to determine whether these markers were also expressed outside their canonical domains and whether the `rogue' cells we observe also express markers characteristic of the domain they are in. Markers were chosen to represent the three germ layers. In the Xenopus early gastrula, the three germ layers broadly correspond to regions along the animal-vegetal axis: the ectoderm arises from animal cells, mesoderm from the equatorial region, and endoderm from vegetal cells. XK70A, a type I cytoskeletal keratin(Krasner et al., 1988), was chosen as a marker of ectoderm. Xbra was chosen as a pan-mesodermal marker (Smith et al., 1991). Mix.1 is another marker of endoderm(Hudson et al., 1997; Rosa, 1989), although it is reported to have some expression in the vegetal marginal zone at the early gastrula stage (Lemaire et al.,1998). Derrière (Der), a marker of both mesoderm and endoderm at the early gastrula stage, was also assayed(Sun et al., 1999). EF1α and ODC were chosen as ubiquitous markers(Bassez et al., 1990; Krieg et al., 1989).

Explants of stage 10 embryos were taken from the dorsal marginal zone, the dorsal blastopore lip, the ventral marginal zone and the vegetal mass, and the animal pole region. Fig. 3shows that as expected, individual dorsal marginal zone cells and cells taken from the dorsal lip expressed pan mesodermal markers such as Xbra and Der (Fig. 3), in addition to Gsc. One cell out of 69 dorsal cells expressed Xwnt8, a ventral and lateral mesodermal marker, and some dorsal marginal zone cells (3/46, 6%) expressed Sox17α, consistent with the in situ hybridization results. In fact, 7/69 dorsal cells expressed markers of all three germ layers: XK70A, Xbra and Sox17α. No expression of these markers was detected in the cell-free control samples.

We note that dorsal marginal zone cells express Gsc in only 7/46(15%) of cells sampled (Fig. 3). This is because dorsal marginal zone explants were cut from above the forming dorsal lip, and so also included cells above the organizer region that do not express Gsc(Fig. 1, Fig. 2C). However, 22/23 (96%)cells taken from immediately above the dorsal lip do express Gsc, as would be expected (Fig. 3). As suggested by the in situ hybridization results at this stage Gsc is also expressed in some ventral (5/46; 11%) and vegetal cells (17/44; 39%; Fig. 3).

In the ventral marginal zone 52% (24/46) of cells sampled expressed Xwnt8 (Fig. 3). Significantly, the `rogue' cells that express Gsc on the ventral side also express Xwnt8, confirming that these are indeed ventral cells with unorthodox Gsc expression. There were also cells in the ventral marginal zone (2/46; 4%) that expressed markers characteristic of each germ layer: XK70A, Xbra and Sox17α.

In the animal explants, 42/47 (89%) of cells sampled expressed the ectodermal marker XK70A. Of these, however, nine (19%) also expressed another marker, including Xbra, Xwnt8, Der or, in one case, even Sox17α, none of which would be predicted to be expressed in animal cells. No expression of these markers was seen in the cell-free control sample.

A large number of cells in the vegetal mass express Sox17α, Mix.1 or Der (39/44; 89%), as expected. However, many also express mesodermal markers, such as Gsc, Xbra and Xwnt8(32/44 cells; 73%), consistent with the in situ hybridization results. At this stage, fewer individual vegetal cells also appear to express `ubiquitous'genes such as ODC or EF1α than do individual cells from other regions of the embryo. For example, vegetal cells express both ODC and EF1α in only 24/44 (55%) cells compared with 46/46 (100%) and 44/47 (97%) in dorsal and animal cells, respectively(Fig. 3). Part of this may be due to inefficiencies in the RT-PCR reactions, or to some cells having low levels of transcripts that are not detectable by this procedure (see below). Indeed, we and others have previously observed that in comparison to other regions of the embryo, whole vegetal mass explants express lower amounts of ODC and EF1α overall as a percentage of total RNA levels than other regions of the embryo(Darken et al., 2002; Saka and Smith, 2004).

Fig. 3 suggests individual cells in the early gastrula have highly heterogeneous gene expression profiles. However, some of this variability in gene expression, particularly where a cell does not express a marker it might be expected to, may be due to the inefficiencies in the reverse transcription, terminal transferase and PCR amplification reactions during RT-PCR(Neves et al., 2004). In order to assess the reproducibility of our assay, we picked 36 single cells and split the cell lysate between two tubes (A and B samples; Fig. 4; Materials and methods). These samples were then subjected to RT-PCR and dot blotted, and then analyzed for expression of up to eight markers. In the majority of cases amplification was remarkably consistent between each split sample(Fig. 4B). However, in 14/160 hybridized A+B samples (8.75%), asymmetric marker expression was seen, where one sample hybridized to the probe and the other sample showed no signal (see ventral cell 1, which expresses Xbra in the B sample but not in the A sample; Fig. 4). Although we are unable to measure this, it is also possible that in a proportion of cases a marker was present but was not amplified in either sample. For example, out of 25 split lysates of single cells taken from the marginal zone, three failed to express Xbra in either sample(Fig. 4 and data not shown).

Some of this variability may be due to a cell containing a small number of transcripts, close to the threshold number that can be detected, particularly as the cell lysate and thus the number of transcripts in each reaction is halved in the experiment above. In order to test the number of transcripts we can detect with this method, single cell lysates were spiked with between∼1 and 5000 transcripts of polyadenylated eGFP mRNA and subjected to RT-PCR and dot blot as normal (Fig. 4C). In two separate experiments, samples containing ∼10 transcripts produced a robust signal, whereas in one experiment the∼1-transcript sample also produced a weak signal above background; no signal above background was detected in the control cell lysate that contained no eGFP transcripts. This is consistent with previous studies which detected over ∼10-25 transcripts reliably, and in some experiments could detect as few as ∼1 (Chiang and Melton,2003; Tietjen et al.,2003). These results indicate that our protocol can detect as few as 10 copies of a transcript in a single cell lysate.

Single cell transplant experiments suggest that during gastrulation, cells become progressively more committed to a particular germ layer(Godsave and Slack, 1991; Heasman et al., 1984; Kato and Gurdon, 1993; Snape et al., 1987; Wylie et al., 1987). To determine whether this commitment is reflected in the gene expression profiles of cells from later embryos, explants were cut from the vegetal region, animal region, dorsal marginal zone and ventral marginal zone of a late gastrula stage 11.5 embryo (Fig. 5A). Dorsal marginal zone cells were taken from the anterior region corresponding to those cells that involuted through the dorsal lip first at early gastrula stages. Cells were then dissociated from these explants and picked for RT-PCR.

The results show that at this stage, in contrast to the early gastrula stage, single cells are more likely to express markers that are characteristic of the appropriate germ layer, and distinct populations of cells are observed. Fig. 5B shows that at stage 11.5 there are two populations of animal cells. One population (20/35 cells;57%) expresses only the ectodermal marker XK70A. The other (12/35 cells; 37%) expresses both XK70A and Gsc. Although these may represent a population of `rogue' cells that was not seen in the animal pole region at early gastrula stages, the presence of Gsc in animal cells may also correspond to a neurectodermal domain of Gsc that is normally seen by in situ hybridization at stage 13(Fig. 5C, arrow), but that is detected by the more sensitive RT-PCR assay at stage 11.5. In contrast to early gastrula stage ectodermal cells, where 19% of the cells sampled (9/47)expressed a non-ectodermal marker (such as Xwnt8, Xbra or Der), only 2/36 cells (6%) sampled at the late gastrula stage expressed detectable levels of such genes (Der or Sox17α).

Endodermal cells also show a more uniform gene expression profile at stage 11.5 than at stage 10 (Fig. 5B). At stage 11.5, only 8/34 of endodermal cells sampled (24%)also express markers of mesoderm and ectoderm, compared with 32/44 (73%) of cells at stage 10.

Dorsal marginal zone cells and ventral marginal zone cells show more variability than animal or vegetal cells at stage 11.5, but even so their gene expression profiles are more representative of their germ layer and dorsal-ventral position than cells taken from the dorsal or ventral marginal zone at stage 10 (Fig. 5B). For instance, none of the ventral marginal zone cells that were assayed (0/35; 0%)express Gsc. Fig. 5Balso shows fewer ventral marginal zone cells (7/35; 20%) and dorsal marginal zone cells (3/35; 9%) expressing endodermal markers at stage 11.5 than at stage 10 when 36/45 (80%) ventral cells and 28/69 (41%) dorsal cells expressed an endodermal marker. The results are consistent with the observation that early in gastrulation the domains of expression of Xbra and Mix.1 overlap, but become separate as gastrulation proceeds(Lemaire et al., 1998). It is also reminiscent of experiments in zebrafish that show at the onset of gastrulation expression of Ntl (zebrafish Brachyury)overlaps with that of the endodermal marker Gata5, but that later in gastrulation the expression domains of these two markers becomes distinct(Rodaway et al., 1999).

It is possible that at these later stages cells express transcripts at a higher level compared with early gastrula stages, and so representative markers are more likely to be reliably amplified and detected in more cells. However, regions also lose expression of some markers (e.g. Sox17α is lost from marginal zone cells). We also did not observe `rogue' cells by in situ hybridization at late gastrula stages, even when the time of colour reaction was lengthened (not shown).

Our results suggest that at late gastrula cells in a particular region of the embryo express genes more characteristic of their position when compared with the early gastrula.

In response to secreted signals from the vegetal region, cells of the marginal zone in the Xenopus embryo form mesoderm, and in turn express secreted signals that act to maintain and pattern mesoderm. As such,cells of the marginal zone and vegetal region in the early embryo are exposed to a wide variety of signals, including members of the TGFβ, Wnt and FGF families (Heasman, 1997). Blastomeres in the same region, particularly the marginal zone, may therefore experience different concentrations and combinations of these factors. TGFβs in particular induce different cell fates at different concentrations. For instance, higher concentrations of the TGFβ family members Activin, Der, Xnr1 and Xnr2 are required to induce endoderm in ectodermal explants than are required to induce mesoderm(Clements et al., 1999; Henry et al., 1996; Sun et al., 1999). Similarly,when mesodermal markers are assayed, anterior and dorsal markers such as Gsc or muscle actin are activated at higher concentrations than ventral and posterior markers such as Xbra(Agius et al., 2000; Green et al., 1992; Gurdon et al., 1994; Jones et al., 1995; Sun et al., 1999; Wilson and Melton, 1994).

The results presented in Fig. 3 suggest that at early gastrula stages, cells of the marginal zone and vegetal region comprise a population with individual cells expressing genes characteristic of both mesoderm and endoderm. By the late gastrula stage, however, this co-expression is less marked, and individual cells tend to express a combination of genes characteristic of either mesoderm or endoderm. We therefore asked whether we could recapitulate this early response in ectodermal cells induced to form mesendoderm by exposure to a single growth factor, Activin.

Ectodermal explants from the animal pole regions of embryos at early gastrula stage 10 were incubated in the presence or absence of 20 U/ml Activin, which induces both mesoderm and endoderm (Green et al., 1990; Rosa, 1989). One explant was dissociated immediately, before the addition of Activin (0 hours), then every hour one cultured explant was dissociated and single cells were picked for RT-PCR (Fig. 6A). Intact explants remaining at 4 hours were assayed for expression of XK70A, Xbra,Mix.1 and Sox17α by quantitative real-time RT-PCR(Fig. 6B). These intact caps expressed high levels of XK70A and Xbra following Activin treatment, while induction of Mix.1 was less marked, and no significant activation of Sox17α was observed(Fig. 6B). How is this induction reflected at the single cell level?

The results indicate that ectodermal cells exposed to Activin form a population of cells expressing mesodermal and endodermal markers in a manner resembling that seen in the whole embryo in the marginal zone(Fig. 6C). Untreated cells, for the most part, maintain their expression of XK70A throughout the culture period and just one cell at 1 hour and one at 2 hour activated expression of Der as well as XK70A(Fig. 6C), in manner resembling ectodermal cells in an intact embryo (Fig. 3).

Little response to Activin is observed 1 hour after treatment(Fig. 6C), but at 2 hours and beyond, mesendodermal markers are activated, with many cells are expressing Xbra (7/11 cells at 4 hours; 64%) and Mix.1 (4/11 cells;36%), reflecting the levels of expression of these genes observed in intact,induced ectodermal explants (Fig. 6B). Only 1/11 cells expressed Sox17α at 4 hours,again consistent with the low level of induction observed in whole caps by real-time RT-PCR (Fig. 6B). Thus, animal cap cells exposed to Activin respond in a similar way to marginal zone cells with Mix.1 and occasionally Sox17αexpressed in the same cells as mesodermal markers. These results also reflect those seen in cell culture, where studies have shown that when exposed to a growth factor cells do not respond uniformly, but rather individual cells activate a subset of characteristic markers (e.g. Ko, 1992; Levsky et al., 2002).

The levels of RNA seen in intact caps measured by conventional PCR methods reflects the number of cells expressing those transcripts by single-cell RT-PCR, also suggesting that embryonic cells respond to the addition of a growth factor in a stochastic manner, similar to cultured cells. However, it is also possible that some transcripts are expressed at levels too low to detect by single cell RT-PCR but are nonetheless expressed in all cells.

Strikingly, almost all Activin-treated animal cells maintain their expression of XK70A throughout the 4 hour culture period, even though by 2 hours and thereafter 30/36 cells also express mesodermal or endodermal markers (Fig. 6C). The maintenance of XK70A expression is consistent with the high level of expression of XK70A detectable in intact Activin-treated animal caps(Fig. 6B). Although it is possible that XK70A transcripts have a half-life in excess of 2 hours, these observations suggest that Activin can activate the expression of mesendodermal genes without significantly downregulating ectodermal markers.

Fig. 6 also shows that Xbra and Gsc are expressed in the same cells at 2 and 3 hours. This confirms previous results indicating Activin initially induces Gsc and Xbra in the same cells, but that after 5 hours Gsc and Xbra become expressed in separate domains, perhaps through secondary cell interactions (Papin and Smith, 2000; Wilson and Melton, 1994). It also reflects the situation in vivo where Gsc and Xbra are expressed initially in an overlapping dorsal mesodermal domain, before becoming separate as gastrulation proceeds(Artinger et al., 1997).

In this study, we show that at the level of gene expression, many cells in the early gastrula Xenopus embryo express markers of more than one germ layer, and that `rogue' cells expressing markers associated with other regions of the embryo are often observed. This was first inferred from in situ hybridization experiments in which Gsc expression was observed in some ventral cells of the embryo, Xwnt8 was observed in vegetal and dorsal cells, and Sox17α was observed to be discontinuous at the boundary of its expression in marginal zone cells(Fig. 1). Single cell RT-PCR experiments confirmed that cells might frequently express a transcript even though they are not within the canonical expression domain of the gene in question (Fig. 3). During later stages of gastrulation, however, gene expression in individual cells from a particular region of the embryo becomes more characteristic of their position,and `rogue' cells are observed less frequently(Fig. 5).

It is unlikely that the inappropriate gene expression we observe by single-cell RT-PCR is due to contamination with cells from other areas of the embryo because the `rogue' cells also express markers characteristic of their original site in the embryo. For example, Gsc-expressing cells on the ventral side of the embryo also express Xwnt8; and Sox17α, Der or Xbra-expressing cells in the ectoderm also express XK70A (Fig. 3).

It is possible that some part of the heterogeneity we observe here is due to the single-cell RT-PCR method, which slightly underestimates the number of cells expressing a particular transcript, because just under 9% of split reaction samples fail to amplify a transcript uniformly(Fig. 4). Nevertheless heterogeneous expression of genes within a population of embryonic precursor cells has been observed in other systems. For example, different combinations of Dscam splice variants are expressed by individual photoreceptor cells during Drosophila development(Neves et al., 2004). In differentiating myofibres, individual nuclei within the syncytium express different genes, with only some expressing muscle-specific markers(Newlands et al., 1998).

Members of the TGFβ family, including Activin, Xnr2 and Derrière, are required for mesoderm induction and for maintenance of mesodermal and endodermal fate in the early Xenopus embryo (reviewed by Yasuo and Lemaire, 2001). Animal cap experiments suggest that high concentrations of TGFβ proteins induce anterior and dorsal mesendodermal markers (such as Gsc), while lower concentrations induce ventral and posterior mesodermal markers (such as Xbra). In addition to these TGFβ family members, cells of the blastula and gastrula are exposed to many other secreted factors such as FGFs and Wnts (reviewed in Heasman,1997). Our single-cell RT-PCR results suggest that the initial outcome of this induction in the embryo is the expression in cells of the marginal zone and vegetal mass of anterodorsal and posteroventral mesodermal markers, or mesodermal and endodermal markers, in the same cell. To test whether gene activation is similar when cells are exposed to a single growth factor, we incubated animal cells in Activin, an inducer of mesendodermal cell types. We observe that these cells too respond by activating anterior dorsal mesendoderm and ventral posterior mesodermal markers in the same cell(Fig. 6).

The level of induction of different markers assayed in whole explants by real-time RT-PCR reflects the number of individual cells that express each marker, as assayed by single cell RT-PCR. This suggests that cells in the animal pole do not all respond equally by activating expression of the same genes. Although this observation may in part be due to some variability in the single cell RT-PCR method, this conclusion is consistent with experiments in other systems that indicate that gene expression in a field of cells exposed to a given concentration of inducer is heterogeneous (see above). It contrasts, however, with work showing that dispersed Xenopusectodermal cells exposed to Activin and cultured for 2 hours express mesodermal markers uniformly when analyzed by in situ hybridization(Gurdon et al., 1999). However, our experimental procedure differs in several respects from this study (Gurdon et al., 1999),which may account for our different observations. In our study, intact explants were continuously exposed to Activin throughout the culture period,whereas Gurdon and colleagues exposed dispersed cells to a 10-minute pulse of Activin before washing. In addition, our cells were cultured in agarose dishes, whereas in the other study dispersed cells were transferred onto fibronectin-coated slides and cultured.

Our finding that members of a group of cells refine their patterns of gene expression to conform to their position within the embryo as gastrulation proceeds reflects previous observations that cells in the Xenopusembryo become committed to a specific germ layer gradually, and asynchronously, during blastula and gastrula stages(Domingo and Keller, 2000; Godsave and Slack, 1991; Heasman et al., 1984; Kato and Gurdon, 1993; Snape et al., 1987; Wylie et al., 1987). Thus, at the start of gastrulation, when commitment to ectoderm, endoderm and particularly mesoderm is not complete, cells frequently express markers of two or more germ layers (Fig. 3),perhaps explaining why they are not yet fully committed. At later stages when more cells become committed to a particular germ layer, gene expression profiles of cells within the same region of the embryo are more uniform and representative of their germ layer (Fig. 4).

The asynchronous nature of commitment to germ layer fate seen in single cell transplant studies (Heasman et al.,1984; Wylie et al.,1987) may be a consequence of the stochastic nature of gene expression (reviewed by Fiering et al.,2000). For example, commitment could arise through the co-expression of a particular combination of lineage-specific factors. If lineage-specific factors are activated at different concentrations of a signal or in response to combinations of signals, and if receptors are also expressed stochastically, commitment will appear asynchronous. The observation that cells in a population show heterogeneous gene expression has led to the idea that transcription in eukaryotic cells occurs in a digital fashion, with gene expression being regulated at the level of the probability that the mRNA will be produced by a cell rather than by the rate of transcription (reviewed by Fiering et al., 2000; Hume, 2000). The gene expression profile of a cell will thus be a function of the probability that a pulse of transcription will occur and the stability of the mRNA in question,leading to stochastic gene expression and the detection of `rogue cells'.

What is the mechanism by which gene expression is refined as the embryo develops? One possibility is that cells expressing inappropriate markers die by apoptosis if they find themselves in the wrong part of the embryo, and indeed limited apoptosis does occur in the Xenopus embryo after stage 10.5 (Hensey and Gautier,1998). Another mechanism by which a cell population may refine its patterns of gene expression is through a local feedback loop. One example of this involves Xbra and eFGF, which are both expressed in the marginal zone during gastrulation and which act in a positive feedback loop to maintain each other's expression (Casey et al.,1998; Isaacs et al.,1994). Thus, if Xbra were erroneously expressed in an ectodermal cell, its expression would not be maintained as eFGF is not present in the animal cap at early gastrula stages(Isaacs et al., 1992). Interaction between groups of cells, the community effect, is important for muscle precursor cells to maintain muscle-specific gene expression(Gurdon et al., 1993a; Gurdon et al., 1993b; Standley et al., 2001). Similar interactions are also likely to be important in establishing boundaries between germ layers during gastrulation. Mixer, for example, has been implicated in forming the boundary between mesoderm and endoderm by limiting expression of mesoderm-inducing signals during gastrulation(Kofron et al., 2004). An interesting future challenge will be to identify other factor(s) involved in restriction of germ layer boundaries and to understand how these interact to refine gene expression and cell fate commitment.

This work is supported by the Wellcome Trust. We thank Hiro Matsunami for generously communicating the single cell RT-PCR protocol before publication,Rick Livesey for help and advice, and our colleagues for discussion through the course of this work.