Clonal and molecular analysis of the prospective anterior neural boundary in the mouse embryo

During gastrulation, pluripotent epiblast cells that do not ingress through the primitive streak (PS) go on to form the neuroectoderm, the surface ectoderm and the amnion ectoderm (Snow, 1977; Tam, 1989; Lawson et al., 1991; Quinlan et al., 1995). The anterior neuroectoderm is a uniform sheet of neuroepithelial cells that gives rise to the forebrain: by the end of somitogenesis, this complex structure comprises the telencephalon and the eyes dorsally and ventrally, the hypothalamus and the diencephalon.

In the mouse, embryological and molecular evidence indicates that, from embryonic day (E) 7.0, signals from the anterior mesendoderm initiate ectoderm regionalisation, well before molecular and morphological signs of neural plate formation are detected (Ang and Rossant, 1993; Ang et al., 1994; Pevny et al., 1998; Wood and Episkopou, 1999; Pfister et al., 2007). Otx2, Six3 and Hesx1, which encode transcription factors essential to forebrain development, are expressed in specific domains within the anterior ectoderm of the gastrulating embryo (Rhinn et al., 1999; Martinez-Barbera et al., 2000; Martinez-Barbera et al., 2001; Simeone and Acampora, 2001; Lagutin et al., 2003). At the beginning of neurulation (E8.5), further regionalisation of the anterior neural plate involves cell-specific responses to extrinsic signals produced by axial midline tissues and non-neural ectoderm (Rubenstein and Shimamura, 1998; Wilson and Houart, 2004). FGF signalling in the anterior neural ridge (ANR), at the junction between anterior neural and non-neural ectoderm, regulates forebrain development and patterning (Shimamura and Rubenstein, 1997; Shanmugalingam et al., 2000; Paek et al., 2009). Various anterior defects, including absence or reduction of the telencephalic vesicles, eyes, olfactory placodes and frontonasal structures, and truncation of the structures rostral to the zona limitans intrathalamica (ZLI), have been consistently associated with a reduction of Fgf8 expression in the ANR (Dattani et al., 1998; Meyers et al., 1998; Acampora et al., 2000; Crossley et al., 2001; Kobayashi et al., 2002; Suda et al., 2001; Tian et al., 2002; Zoltewicz et al., 2009; Vieira et al., 2010).

Genetic studies and earlier investigations of cell fate in the anterior ectoderm, relying on orthotopic grafts (Beddington, 1981; Tam, 1989), provided important insights into the early steps of forebrain development. Yet, we still have a limited understanding of how anterior ectoderm develops into neural and non-neural derivatives and, in particular, how the ANR signalling centre forms (Eagleson et al., 1995; Houart et al., 1998). Lineage analysis of single cells (Lawson et al., 1991) is required to define whether, at late gastrulation stage, the anterior ectoderm is composed of distinct region-specific progenitors, a mixed population of progenitors with distinct fates, or a single multipotent cell population. Single-cell resolution is also needed to understand whether the ANR is a separate lineage or whether it arises in response to spatial cues as the neural plate matures. Finally, clonal analysis is essential to investigate whether a given gene marks a prospective tissue and functions as a putative lineage determinant.

We traced the descendants of single cells to describe the transformation of late gastrulation stage anterior midline ectoderm into the surface ectoderm and forebrain at early somite stages. Evidence for segregation of neural and non-neural progenitors and significant regionalisation of prospective fate was found. Importantly, our results delineate a rostral limit to forebrain contribution at a position that is significantly more distal than previously described. We provide a comprehensive spatiotemporal atlas of anterior gene expression at stages crucial for anterior ectoderm regionalisation and neural plate formation. Its comparison with the fate map showed that several transcription factors are restricted to subdomains that correlate with regions of prospective fate, but that these factors do not define a specific lineage. A 3D representation of clone spatial distribution at the early somite stage showed that the anterior neural border is not fully defined at late gastrulation stage, nor are early regional subdivisions of the presumptive forebrain territory established. In addition, our results show that ANR progenitors are dispersed over a broad area of the anterior ectoderm where neural precursors reside. ANR and forebrain descendants often share a common progenitor.

Mouse embryos were obtained from [C57BL/6J × CBA] F1 matings. The stage at injection ranged from late streak (LS) to late streak/early bud (LSEB). At the LS stage, the anterior end of the streak has reached the distal tip and the amniochorionic fold is nearly complete. At the LSEB stage, the node starts to be obvious, the amniochorionic fold component parts have just fused and the allantoic bud is just visible. Experiments were performed in accordance with European and French Agricultural Ministry guidelines for the care and use of laboratory animals (Council directives 2889 and 86/609/EEC).

The procedures for embryo culture, iontophoretic injection into single cells and labelled cell identification were as previously described (Beddington and Lawson, 1990; Lawson et al., 1991; Perea-Gomez et al., 2001) with the following modifications. Impaled ectoderm cells showing a stable drop in potential of 10 mV or more were injected with a mixture of 8.7% horseradish peroxidase (HRP; ∼1000 U/mg, Boehringer/Roche) and 1.3% lysinated tetramethylrhodamine dextran (LRDX; 10³M_r, Molecular Probes) in 0.05 M KCl, using 500 msecond per second pulses of 1-3 nA depolarising current for 15-20 seconds. Only one ectoderm cell per embryo was injected, although occasionally two cells were labelled owing to the passage of the dye between mitotic pairs still connected by a cytoplasmic bridge (Lawson et al., 1991; Gardner and Cockcroft, 1998). An additional injection for 5 seconds was made into an extra-embryonic visceral endoderm (VE) cell near the embryonic/extra-embryonic junction and in the same focal plane as the injected ectoderm cell. The radial position of the resulting compact VE clone relative to the midline multiplied by the value for 25% of the circumference was used to define, retrospectively, the circumferential position (CX) of the injected ectoderm cell. The circumference was calculated using the mean diameter of the ectoderm layer at the level of the embryonic/extra-embryonic junction from longitudinal and frontal views. HRP-labelled cells were identified in cultured embryos after staining, fixation in 4% paraformaldehyde, ethanol dehydration and clearing in BABB (1:2 benzoyl alcohol:benzyl benzoate), before embedding in epoxy resin (Agar 100, Agar Scientific). Serial sections cut at 6 μm were recorded using DIC optics; images of those with HRP-labelled cells were stacked manually to count the cells and assess their position.

The procedures for the preparation and reconstruction of 1 μm frontal serial sections of the head of the cultured embryo used to plot the clones in 3D can be found at http://www.emouseatlas.org/emap/ema/home.html (EMA Anatomy Atlas/EMA Protocols and TS12/EMA:145). The Woolz image processing system (Piper and Rutovitz, 1985) was used. Anatomical and clonal cell domains were defined within the reconstruction using a combination of non-interactive image processing tools (grey level thresholding and morphological operations) and MAPaint (Baldock, 2004) for interactive voxel labelling and for identifying the sectioning plane in experimental embryos.

Triangulated mesh surface models were created corresponding to all the defined domains, and a simple affine scaling transformation was used to achieve an isotropic 1 μm voxel size. The mesh was cropped, smoothed and remeshed to produce a simply connected open surface (as required for parameterisation) using a combination of Woolz and MeshLab (Cignoni et al., 2008). A mesh-based transform was established between the cropped ectoderm surface mesh and its cropping plane. The mesh node displacements were computed using least squares conformal mapping with boundary nodes fixed to the cropping plane. The transformation was defined using least squares conformal parameterisation (Levy et al., 2002). The mesh transform defined a bi-directional mapping between the ectoderm surface and the planar flat map. Barycentric interpolation within mesh elements allowed the transformation of arbitrary positions between the 3D surface and 2D flat map. The parameterisation resulted in large variations in scale within the flat map, which made the distribution of clone cells hard to interpret. Fixed size circular markers were placed at their location to overcome this problem. A full description of the procedure will be published elsewhere. The clone cells, the ectoderm and other anatomical domain surfaces were visualised using VTK (Schroeder et al., 2007).

Clonal growth is assumed to be exponential; then N=N ^bt₀ or lnN=bt+lnN₀, where N is the number of labelled cells at time t (hours after injection), N₀ is the number of progenitors (1, or 2 when siblings were initially labelled). Clone doubling time (cdt) is given by ln2/b=ln2×t/lnN–lnN₀.

The number of population doublings (n) in the targeted region is estimated by 2ⁿ=ΣN/ΣN₀, or n=lnΣN–lnΣN₀/ln2, where ΣN is the total number of labelled cells in all embryos with label after culture and ΣN₀ is the total number of labelled progenitors in these embryos.

WISH and histology were performed as described (Perea-Gomez et al., 2004), using antisense probes from published sources (see Results). Embryos were cryosectioned at 20 μm.

In order to trace cell lineage shortly before neural plate formation, single ectoderm cells of LS and LSEB stage embryos were microinjected by iontophoresis with a mixture of HRP and LRDX. The region targeted in the embryonic ectoderm covered the anterior midline, from the attachment of the amnion to the embryonic ectoderm, proximally to half of the distance to the node along the anterior midline, distally (distance PD; Fig. 1A). Laterally, the targeted region extended over 25% of the ectoderm circumference spanning the midline. The fluorescent label was used to immediately visualise the position of the injected cell (Fig. 1B; see Materials and methods) and HRP to identify its descendants after 1 day of culture (mean 24.7 hours, range 20.2-28.0 hours). Embryo development was coordinated normally, as judged by heart and neural morphogenesis and somite number [3-11 somite pairs, mean 5.60±1.85 (mean±s.d.)]. Of 93 single-cell-injected embryos, 58 (62%) had HRP-labelled descendants after culture. Of these, 47 were labelled in ectoderm, eight in mesoderm and three in endoderm. The clones in the last two categories were identified as the descendants of mesoderm and endoderm cells accidentally impaled instead of ectoderm and were not considered further. Informative histological sections were obtained from 41 of the embryos with ectoderm clones. Their analysis forms the basis of this study.

The frequency distribution of clone size (Fig. 1C) showed peaks at four, eight and 16 cells, indicating some synchronisation of the cell cycle (Lawson et al., 1991; Tzouanacou et al., 2009). The clones in seven embryos were derived from two labelled siblings (see Materials and methods). These clones fell into two groups in the distribution: the four largest clones, and three clones with no more than eight labelled cells. For the latter group, the clone doubling time (cdt) was calculated on the basis of one surviving sibling (see Materials and methods). The total number of labelled cells (509) from 45 progenitors (37 singletons and four doublets) implies an 11.4-fold increase in the size of the targeted population, or 3.5 population doublings. The cdt showed a skewed distribution with median 7.65 hours and 95% confidence limits 6.91 and 8.17 hours (Fig. 1D). Clones with a very long or very short cdt were examined for spatial clustering that could indicate a lengthening of the cell cycle accompanying early differentiation on the one hand, or a local area of very rapid proliferation on the other. Progenitors of both groups were scattered throughout the targeted area (Fig. 1E), indicating that the anterior ectoderm comprises a uniformly proliferating population.

Anatomical boundaries, cell shape and cell arrangement were used to score descendant cells as belonging to the neural primordium or to one of three distinguishable regions of the non-neural ectoderm: the ectoderm in contact with the overlying anterior prosencephalon neuroectoderm layer, without intervening mesoderm (VEAP for ‘ventral ectoderm of the anterior prosencephalon’); the buccal ectoderm (oral or stomodeum epithelium); and the surface ectoderm (Fig. 2A,B). The buccal ectoderm, which includes the oral plate, is a cuboidal epithelium that is continuous with the VEAP and extends to the squamous surface ectoderm covering the heart. Examples of labelled cells in neural or non-neural ectoderm derivatives are shown in Fig. 2C-G.

A pie chart spatial representation (Fig. 3) shows the original positions of clone progenitors and the tissue contributions of their respective descendants. We were able to identify a proximal region that did not contribute to the forebrain. The most proximal forebrain progenitor was 77 μm from the embryonic/extra-embryonic junction (or 0.43 when normalised to PD length) and established the limit for forebrain contribution (LFBC) (Fig. 3). By contrast, contribution to non-neural ectoderm was found throughout the targeted area. However, progenitors further than 106 μm from the embryonic/extra-embryonic junction (0.56 normalised) contributed to the VEAP exclusively or in combination with neuroectoderm, but not to buccal or surface ectoderm. A limit for surface ectoderm contribution (LSEC) was thus defined (Fig. 3). We therefore distinguished three zones along the anterior midline: a proximal zone (PROX), a distal zone (DIST) and, in between, an intermediate zone (INT) delineated by the LFBC and LSEC (Fig. 3). With the exception of two labelled cells in the hindbrain, all descendants of the PROX zone clone progenitors colonised the surface ectoderm and buccal ectoderm (Table 1, Fig. 3). By contrast, 88% of the descendants of the DIST zone clone progenitors populated the neuroectoderm, with a majority in the forebrain (Table 1, Fig. 3); the remaining 12% were in the VEAP (Table 1). The INT zone was characterised by progenitors that generated a wide variety of fate combinations, diversely associating all neural and non-neural fates described for the PROX and DIST zone progenitors (Table 1, Fig. 3). These results show that, at late gastrulation stages, the anterior ectoderm exhibits a significant regionalisation of prospective fate and a marked segregation of neural and non-neural progenitors.

Clones yielding exclusive colonisation of neuroectoderm derivatives (15/41) originated from the INT and DIST zones. By contrast, labelled progenitor cells showing descendants exclusively in non-neural derivatives (16/41) were found over the entire anterior ectoderm. For these progenitors, a clear correlation was observed between their original location and the specific non-neural structure to which they contributed (Fig. 3). Nevertheless, a number of progenitors contributed to both neural and non-neural structures (10/41). Among these progenitors, two types could be distinguished: mixed fate progenitors confined to the INT zone (5/10) that generated an array of neural and non-neural fate combinations; and mixed fate progenitors from the DIST zone (4/11) that consistently generated descendants contributing to the forebrain and the VEAP. Together, these findings demonstrate that neural and non-neural fates are regionalised but not clonally separated. Some descendants in the VEAP, which is continuous with the buccal and surface ectoderm at E8.5, are clonally related to neural precursors.

We investigated possible links between the regionalisation of prospective cell fate and the expression patterns of anterior-specific genes. We examined the spatial and temporal evolution of both lineage-specific markers and regional markers: AP-2.2 (Tfap2c – Mouse Genome Informatics) and Dlx5 [non-neural ectoderm (Chazaud et al., 1996; Yang et al., 1998)], Sox1 [neuroectoderm (Pevny et al., 1998)], Sox2 [neuroectoderm and surface ectoderm precursors (Avilion et al., 2003)], Hesx1 and Six3 [prospective forebrain (Thomas and Beddington, 1996; Yang and Klingensmith, 2006)], Otx2 [prospective forebrain and midbrain (Ang et al., 1994)] and Irx3 [posterior forebrain, midbrain and hindbrain (Kobayashi et al., 2002; Braun et al., 2003)].

Analysis of expression patterns from LS to late headfold (LHF) stage (Downs and Davies, 1993) revealed several unreported features (Fig. 4). Six3 and Sox1 were expressed at LSEB stage, earlier than previously reported. Both transcripts were found at the same axial level, just rostral to the node (Fig. 4D,F, compare with Gsc and Foxa2 in supplementary material Fig. S1), but the Sox1 domain was broader, extending laterally, and no Sox1 transcript was detected in the midline. Strikingly, Six3 ectodermal expression started at a very distal position compared with Hesx1 expression at the same stage (Fig. 4C,D). Later, Six3 expression spread rostrally and slightly laterally and, by early headfold (EHF) stage, colocalised with Hesx1 expression. Irx3 transcripts were first detected at LS stage, lateral to the Six3 domain. Later, the two expression domains overlapped although Irx3 expression appeared weaker where Six3 was expressed. AP-2.2 transcription was consistently weaker than that of Dlx5 and clearly absent from ectoderm cells of the anterior midline (Fig. 4A,B). This analysis highlights dynamic changes in exclusive and overlapping domains of ectodermal gene expression during early embryonic development.

The proximal and distal expression boundaries of the Dlx5, Hesx1, Sox1 and Sox2 genes were defined along the anterior midline to establish how they relate to cell fate. These are classical markers of anterior, non-neural and neural identities. Only LS to late bud (LB) stage embryos were analysed as developing headfolds prevented measurements. Actual values of the expression boundaries were plotted against embryo size (Fig. 5). At all stages examined, the rostralmost gene was Dlx5 with, slightly caudal to it, Hesx1. The Sox2 proximal boundary coincided early on with the Hesx1 proximal boundary. By contrast, when Sox1 was first detected, its proximal boundary of expression abutted the Hesx1 distal boundary (Figs 4, 5). Normalised values indicate that Sox1 expression expands anteriorly as the embryo grows (data not shown). Sox1 and Dlx5 expression domains were mutually exclusive at all stages examined. Two phases could be distinguished: an early phase from LS to EB (PD=160-200 μm), when the Hesx1 and Dlx5 expression domains overlap rostrally, is followed, from EB to LB (PD=200-300 μm), by the onset of Sox1 expression, the domain of which progressively overlaps that of Hesx1 caudally. At all stages examined, a significant area of the anterior ectoderm positive for Hesx1 remained negative for both the non-neural marker Dlx5 and the neural marker Sox1.

We compared the positions of the LFBC and LSEC defined in the fate map with gene expression boundaries at equivalent stages (Fig. 5). The PROX zone that contributes to both surface and buccal ectoderm lay within a domain in which Dlx5 is expressed alone or is co-expressed with Hesx1. The mixed progenitors confined to the INT zone were located just distal to the Dlx5 domain and, like the progenitors of the DIST zone contributing to forebrain and VEAP, were in a region of ectoderm expressing Hesx1 exclusively. Together, these results demonstrate that, at E7.5, markers of forebrain, neuroectoderm and surface ectoderm do not define lineages. However, their collective expression defines ectoderm subdomains that correlate with areas showing distinct prospective fates.

The anterior ectoderm undergoes substantial growth and complex morphogenetic curvature during early neurulation. In order to gain insight into this process we analysed the clone spatial distribution using a high-resolution 3D reconstruction of the head of a 5- to 6-somite stage embryo, cultured from the LSEB stage (supplementary material Fig. S2). Cells occupying positions that matched those of the labelled cells in sections were painted in the reconstruction, enabling a representation of the position and general orientation of the clones. Anatomical domains were also painted in the reconstruction. The ectoderm layer was then extracted as a 3D surface with the associated painted cells and anatomy (Fig. 6A-E). The clone spatial distribution could be viewed from any angle (supplementary material Movie 1) and plotted clones checked against the record of clones in the intact embryo (Fig. 7). In order to compare with published data on other species, the 3D surface was transformed into a flat map (Fig. 8A; supplementary material Fig. S3 and Movie 1).

In practice, only 24 embryos had sufficient congruency for matching into the reconstruction. Most of these embryos had 5-8 somites and had been injected at a significantly more advanced stage (9 at LS, 15 at LSEB) than the remaining embryos (15 at LS, 2 at LSEB; χ_c²=10.14, 1 degree of freedom, P<0.005). Neural clones from the latter group consistently showed extension into midbrain and even into anterior hindbrain (‘not-plotted’ clones), indicating that caudal extension from the prospective forebrain region starts to decrease sometime between the LSEB and early somite stage. Other not-plotted clones were included in the description when their distribution could be grouped with plotted clones that had similarly situated progenitors.

The following description of clone expansion is based on the three zones that have been distinguished by their clonal composition. During gastrulation, clones from anterior epiblast are non-coherent and anisotropic, typically expanding towards the PS, i.e. aligned approximately parallel to the embryonic/extra-embryonic junction (Lawson et al., 1991). The behaviour of clones generated in the PROX zone shortly before neural plate formation followed this pattern (Fig. 8A,B): clones in buccal and surface ectoderm were either elongated laterally away from the midline (clones 3, 4, 6 and not-plotted 9), or had colonised the surface ectoderm at the level of the midbrain and anterior hindbrain, leaving no descendants in buccal ectoderm (clones 2, 5, 7). Clone 1 was an exception: initially labelled very close to the embryonic/extra-embryonic junction, it extended parallel to the midline as squamous ectoderm over the heart, ending in buccal ectoderm.

Clones generated in the INT zone showed a mixture of behaviours (Fig. 8A,C). Two clones (12, 20) spread laterally away from the midline in the VEAP. Four clones (14, 15, 16 and the caudal group of cells in 17) colonised some buccal ectoderm, but mainly surface ectoderm dorsal and generally rostral to the region occupied by clones 5 and 7 from the PROX zone. Three clones diverged from the expansion pattern typical of epiblast clones. They were confined, away from the midline, in the junctional region between forebrain, VEAP, buccal and surface ectoderm, oriented from surface ectoderm and buccal ectoderm through the most caudal part of the VEAP and rostrally to neural (clone 11, not-plotted 18), or from VEAP slightly caudally through surface ectoderm to neural (rostral group of clone 17).

All plotted clones generated in the DIST zone contributed to the neuroectoderm, within which three types of behaviour could be distinguished (Fig. 8A,D). First, clones that spanned the midline (26, 40 and not-plotted 33) did so only anterior to the foregut, rostral to the nascent optic pit, and were presumably prospective telencephalon. Second, clones that were localised in the forebrain on the side on which they originated spread from the midline rostrally and laterally (clone 35), laterally and caudally (24, 28) or primarily caudally for the most lateral progenitors (22, 36), plus a small clone in the midbrain (31). Third, clones contributing to both neuroectoderm and VEAP (34, 39) were located in the VEAP further away from the midline than clones 12 and 20 from the INT zone and spread into the neuroectoderm near the junctional region between forebrain, VEAP, buccal and surface ectoderm, i.e. into a similar position as clone 11 from the INT zone. Of the two not-plotted clones contributing to the VEAP from the DIST zone, clone 29 was similar to clone 34 and clone 32 spread more widely throughout the left VEAP, extending into neural on the right side, and so showed elements of different behavioural categories. Importantly, whether a clone was localised in prospective telencephalon or diencephalon was not predictable from the proximodistal position of its progenitor. Therefore, we concluded that cells of the rostral neural boundary are not spatially organised at the end of gastrulation and the prospective forebrain territory is not regionally subdivided at this stage.

We identified putative NCCs (Fig. 2F) in two plotted clones (17, 22) from the INT zone and one not-plotted clone (37) from the DIST zone (Fig. 3), outside the presumptive AP-2.2 expression domain (Fig. 4). Thus, at E7.5, the anterior limit of AP-2.2 does not define the rostral limit of NCC production (Osumi-Yamashita et al., 1994). In support of previous ultrastructural (Nichols, 1981) and DiI-labelling (Serbedzija et al., 1992) studies of cranial NCCs in the mouse embryo, we found emigrating NCCs at the level of the forebrain (clones 17, 22), located caudal to the optic pit (Fig. 6F) and in the mesencephalon (clone 37).

To characterise the molecular identity of the tissues generated by anterior ectoderm cells at E8.5 and to determine which clones had descendants in the ANR, we compared the expression profiles of Six3, Hesx1, Foxg1, Dlx5 and Fgf8 (Oliver et al., 1995; Xuan et al., 1995; Thomas and Beddington, 1996; Yang et al., 1998; Hatini et al., 1999). Longitudinal and frontal histological sections were obtained from 3- to 7-somite stage embryos ex utero stained by WISH (Fig. 9A-P; data not shown). In the mouse embryo, the ANR is located at the junction between the most rostral part of the neural plate and the non-neural ectoderm; it expresses Fgf8 from the 4-somite stage onward (Shimamura and Rubenstein, 1997), so it comprises both forebrain and VEAP cells at the anatomical anterior ridge (Fig. 9M-O′). The overlap between the Dlx5 and Fgf8 expression domains corresponds to the non-neural, i.e. VEAP, part of the ANR (Fig. 9J-L′).

Information about the Fgf8 expression domain was used to locate the presumed ANR in sections of the HRP-labelled embryos. Thirteen clones had descendants in the ANR. The progenitors were scattered throughout the INT and DIST zones (Table 2, Fig. 2G, Fig 9Q). Four clones had labelled cells only in the ANR: two clones generated in the INT zone (12, 18) and two in the DIST zone (34, 39). About half of the progenitors contributing to the ANR also had descendants spreading widely in the forebrain (six out of 13). Interestingly, three of those originating in the DIST zone (clones 26, 27, 40) had descendants in the ANR and the forebrain only (Fig. 9Q). Consistent with the observation that the INT zone is characterised by progenitors with a variety of fate combinations, four out of six clones with descendants contributing to the ANR (11, 13, 20, 23) also showed colonisation of neuroectoderm, buccal ectoderm and surface ectoderm. Together, these findings demonstrate that prospective ANR cells arise from progenitors that are distributed over a broad area of the anterior ectoderm and interspersed with precursors of other tissues. Furthermore, these findings reveal that, clonally, prospective ANR cells are more closely related to forebrain than to non-neural tissues.

In agreement with previous fate map studies, we found that anterior ectodermal cells differentiate into neuroectoderm, epidermis and intermediate cell types characteristic of the border between the neural plate and non-neural ectoderm. However, our findings diverge with regard to other tissue contributions. A previous report showed that a small fraction of the descendants of anterior ectoderm grafts colonised cranial mesenchyme and heart mesoderm (Tam, 1989). A retrospective analysis of genetically labelled single cells showed that common neuromesodermal progenitors persist long after the segregation of endoderm and surface ectoderm lineages, which occurs during gastrulation (Tzouanacou et al., 2009; Petit and Nicolas, 2009). We found no contribution to mesoderm, indicating that, if present, neuromesodermal progenitors are rare within the anterior part of the embryo at E7.5. This strongly suggests that lineage segregation occurs earlier anteriorly than for caudally and laterally located progenitors. Nevertheless, restricted cell fate is not a demonstration that these cells have lost the potential to differentiate into mesoderm (Chan and Tam, 1986).

Previous studies using orthotopic grafts showed that, although non-neural and neural precursors are intermingled, a rostrocaudal organisation of forebrain, midbrain and hindbrain neuroectoderm precursors could be distinguished at E7.5 (Beddington, 1981; Beddington, 1982; Tam, 1989). The higher resolution provided by clonal analysis revealed a tendency for the anterior midline cells to be restricted in their prospective fate and showed that their spatial segregation is also well underway. Proximal ectodermal progenitors expressing Dlx5 contributed to surface and buccal ectoderm, whereas progenitors located more distally and expressing Hesx1 predominantly contributed to the neuroectoderm and the VEAP. Together, our findings establish that neural and non-neural lineages are not yet clonally separated at E7.5. Importantly, our revised fate map delineates a rostral limit, beyond which no forebrain contribution is found, at a position that is more distal, i.e. farther from the embryonic/extra-embryonic junction than that previously reported.

Three-dimensional reconstruction allowed us to visualise cell dispersion and the mode of clonal growth in the anterior cephalic region. The behaviour of clones generated in the proximal ectoderm shortly before neural plate formation followed the lateral and caudal expansion expected from epiblast cells at earlier stages. By contrast, for more distally located ectoderm cells, in particular for forebrain and VEAP progenitors, different combinations of cell behaviours were observed. No correlation between the anteroposterior position of these progenitors and the subsequent dispersion patterns of their descendants was observed. These findings lead us to conclude that the anteriormost border of the neural plate is not spatially defined at the end of gastrulation, nor is there early regional subdivision of the prospective forebrain territory. Therefore, the organisation of the mouse embryo appears to differ from that of the zebrafish (Mathis and Nicolas, 2006). Late in gastrulation, discernible domains in the zebrafish dorsal blastoderm can be assigned to major subdivisions of the forebrain (Woo and Fraser, 1995), which resembles the neural fate map of Xenopus (Eagleson and Harris, 1990) and chick (Couly and Le Douarin, 1985; Cobos et al., 2001).

In line with molecular evidence of anterior ectoderm regionalisation at the end of gastrulation, patterns of cell behaviour demonstrate significant regionalisation in prospective fate. However, overlapping regions of transcription factor expression often persist until LHF stage. Together, these findings demonstrate that molecular regionalisation and, in particular, classical surface ectoderm and forebrain markers, do not strictly define lineages at late gastrulation stage. In the mouse, the establishment of strict clonal boundaries appears to be a later event that could arise by progressive restriction of cell dispersion, rather than by a process of lineage specification, resulting from distinct combination of prepattern factors (Fraser et al., 1990; Inoue et al., 2000; Mathis and Nicolas, 2002; Puelles et al., 2005; Veitia and Salazar-Ciudad, 2007).

Our results reveal the existence of a distinct population of precursor cells that generates a wide range of neural and non-neural fate combinations. Strikingly, these mixed fate progenitors are confined to a small area of the anterior midline, the INT zone, delineated by the LFBC and the LSEC. It is interesting to compare these findings with data obtained in other vertebrate models. FGF, WNT and BMP signals cooperate in the specification and positioning of the border cells between neural and non-neural ectoderm (Streit and Stern, 1999; Litsiou et al., 2005; Bailey et al., 2006; Patthey et al., 2009). Cell-labelling studies have established that, at early somite stages, neural plate border cells give rise to sensory placodes at the rostral level and NCCs at caudal levels of the neuraxis (Whitlock and Westerfield, 2000; Bhattacharyya et al., 2004; Jones and Trainor, 2005). Dlx5 is a highly conserved transcription factor which, in the chick embryonic ectoderm, promotes the formation of border cells that subsequently strongly express Msx1, Bmp4 and Six4 (McLarren et al., 2003). Dlx5 mouse mutants have late defects in structures derived from border cells, such as the olfactory and otic placodes (Depew et al., 1999; Acampora et al., 1999). Cells in the INT zone could, at an earlier stage (before the formation of the neural plate), be the equivalent of the border cells described in the chick embryo at early somite stages. These mixed fate precursors in the mouse, however, do not express Dlx5 but do express Hesx1. Furthermore, their descendants are not restricted to the prospective placodal region but are widespread throughout the non-neural and neural structures. Alternatively, these progenitors may lie at the prospective boundary between neural and non-neural ectoderm. They might coincide with the region of decay of morphogenetic gradients responsible for the specification of the two neighbouring territories. Often, regions of overlap, where specific precursors intermix, persist at the border of two distinct domains (Woo and Fraser, 1995; Veitia and Salazar-Ciudad, 2007). The intermediate region depicted here is different: it is composed of mixed fate progenitors. The early phase of anterior head formation in the mouse requires considerable growth and complex morphogenesis and therefore a putative pool of mixed fate progenitors could be crucial in providing both neural and non-neural derivatives.

An important finding is that the non-neural VEAP, which is in direct contact with the anteriormost neuroectoderm but also continuous with buccal and surface ectoderm, has a closer relationship, both spatially and clonally, with the forebrain than with neighbouring, non-neural ectoderm.

The ANR is a well-known signalling centre involved in the specification and regionalisation of the anterior prosencephalon (Houart et al., 2002; Lagutin et al., 2003; Paek et al., 2009). The definition of the ANR varies between vertebrate models, resulting in some inconsistencies in fate maps obtained by grafting or DiI-labelling methods. Fgf8 expression is often used to define the location of the ANR (Shimamura and Rubenstein, 1997; Vieira et al., 2010), but the onset and extent of Fgf8 expression vary depending on the species and the developmental stage considered. Here, we define the ANR, on the basis of Fgf8 expression, as the region encompassing the rostralmost part of the forebrain neuroectodem and the underlying VEAP at the 5- to 7-somite stage. In the chick embryo, the ANR has been described as differentiating from the margin of the neural plate rostral to the anterior limit for NCC formation. A common observation in all studies at early somite stages is that the ANR generates the following ectoderm derivatives: the ventral cephalic epithelium, the olfactory placodes and Rathke's pouch, i.e. the primordium of the anterior pituitary (Couly and Le Douarin, 1985; Couly and Le Douarin, 1987; Eagleson and Harris, 1990; Osumi-Yamashita et al., 1994; Eagleson et al., 1995; Cobos et al., 2001). The Hesx1-expressing non-neural ectoderm that contacts the floor of the ventral diencephalon in E8.5 mouse embryos evaginates to form Rathke's pouch at E9.5 (Hermesz et al., 1996; Thomas and Beddington, 1996; Rizzoti and Lovell-Badge, 2005; Gaston-Massuet et al., 2008). Although the HRP-labelled embryos are too young to show any morphological evidence of Rathke's pouch, it is likely, given the results obtained in other animal models, that some precursors of the VEAP are founders of the anterior pituitary primordium. This hypothesis is strongly supported by the genetic fate map of Hesx1-expressing cells in the normal mouse embryo (Andoniadou et al., 2007).

Little is known about the origin of the ANR. A study of mid-gastrula stage zebrafish embryos revealed the existence of one row of signalling cells at the margin of the anterior neural plate before the formation of the ANR. These cells, called the anterior neural border (ANB), later contribute to ANR derivatives and telencephalon (Houart et al., 1998). Our data suggest that ANR precursors in the mouse are not confined to the border of the neural plate, as seems to be the case in zebrafish, but are dispersed over a broad area of the anterior ectoderm. The ANR arises both from the intermediate zone composed of mixed fate progenitors and from a more distal region where neural progenitors reside, sometimes sharing common progenitors.

This study, combining cell lineage and gene expression analyses, provides a basis for further investigations into the specification and early regionalisation of the anterior neural plate and for the interpretation of mutant phenotypes.

We thank D. Sabéran-Djoneidi, R. Baldock, C. Houart, D. Morello, A. Zwijzen, K. Pakdaman and D. Touati for advice and discussion; H. Morrison for the optical projection tomography work; L. Graham for producing the images for the reconstruction; and A. Perea-Gomez, C. Papanayotou and V. Wilson for critical reading of the manuscript.