Evidence for a dynamic spatiotemporal fate map and early fate restrictions of premigratory avian neural crest

A major question in development is when and how do apparently homogeneous progenitors segregate into distinct fates. The neural crest (NC) is a group of progenitors that disperses in the embryo and differentiates into many derivatives (Kalcheim and Burstyn-Cohen, 2005; Le Douarin and Kalcheim, 1999; Morales et al., 2005). Initial NC specification was suggested to occur during gastrulation (Basch et al., 2004; Basch et al., 2006). Then, progenitors sequentially emigrate from the neural tube (NT) following an epithelial-mesenchymal transition (EMT) (Burstyn-Cohen and Kalcheim, 2002; Burstyn-Cohen et al., 2004; Coles et al., 2007; Groysman and Kalcheim, 2008; Nieto, 2001; Sela-Donenfeld and Kalcheim, 1999; Sela-Donenfeld and Kalcheim, 2000; Sela-Donenfeld and Kalcheim, 2002; Shoval et al., 2007).

Vital dye labeling revealed that NC cells migrate in a stereotypical manner, leading to a general ventral to dorsal order of derivative colonization. In the trunk of avians and zebrafish, a ventralward migration leads first to colonization of the sympathetic ganglia (SG) followed by colonization of the dorsal root ganglia (DRG); the last cells to emigrate are melanoblasts that advance subectodermally (Erickson et al., 1992; Raible and Eisen, 1994; Raible et al., 1992; Serbedzija et al., 1989; Weston and Butler, 1966). Similarly, in mouse and Xenopus, the temporal segregation between ventrally migrating neural progenitors is kept, yet both ventral and lateral movements occur simultaneously (Collazo et al., 1993; Serbedzija et al., 1990).

How is this stereotypical order of cell homing related to fate acquisition? One extreme possibility would postulate that although emigrating cells are multipotent, they migrate precisely because migratory pathways open sequentially to allow cell progression. Alternatively, early emigrating cells are already fate-specified and therefore may be differentially marked regarding the time of delamination and/or the migratory paths to follow (Anderson, 2000; Harris and Erickson, 2007; Le Douarin and Kalcheim, 1999). Lineage analysis suggests the coexistence of multipotent along with fate-restricted NC progenitors (Bronner-Fraser and Fraser, 1989; Bronner-Fraser and Fraser, 1988; Collazo et al., 1993; Erickson and Goins, 1995; Frank and Sanes, 1991; Fraser and Bronner-Fraser, 1991; Greenwood et al., 1999; Lo et al., 2005; Perez et al., 1999; Raible and Eisen, 1994; Reedy et al., 1998; Wilson et al., 2004; Zirlinger et al., 2002). The relative proportion of either type of progenitors in the dorsal NT remains, however, unclear and is likely to vary in a stage and axial level-dependent manner. To approach these issues, it is imperative to understand whether this ordered mode of target colonization is related to events taking place in the dorsal NT. For instance, in the trunk of avian embryos, NC cells continuously delaminate over a period of two days yet the mechanism of replenishment of the premigratory pool remains elusive; this knowledge is important for elucidating the relationship between cell emigration and fate, for understanding the dynamic histogenesis of the dorsal NT and the mechanisms leading to segregation of the NC from other CNS lineages.

To begin addressing these questions, we lineage traced small cell populations in the avian dorsal NT at flank regions. Labeled cells delaminated without leaving residual progeny in the neuroepithelium, which excludes an asymmetric mode of cell emigration. This finding was complemented by the demonstration of a time-dependent cell relocation towards the dorsalmost region followed by emigration. This dynamic behavior resulted in a corresponding narrowing of the expression domains of FoxD3, Sox9 and Snail2. Consistently, the final localization and fate of prospective NC cells could be predicted from their relative position within the NT at a given stage, as well as from their time of emigration. This raised the question of whether fate acquisition depends upon the migratory environment or whether it is established prior to EMT. To address this issue, prospective neural progenitors were diverted into the melanocytic pathway, where they were still able to adopt neural traits. Hence, initial restriction to a neural identity is independent of the migratory environment and takes place prior to cell emigration. Altogether, these results demonstrate the existence of a dynamic spatiotemporal fate map of NC derivatives within the premigratory domain of the NT, and provide for the first time a link between the topography of premigratory progenitors, cell delamination and cell fate.

Chick (Gallus gallus) or quail (Coturnix coturnix japonica) eggs were from commercial sources. Experiments were restricted to the flank (somites 20-25) or the prospective flank in embryos younger than 20 somites.

CM-DiI (Molecular Probes, #C-7001) was iontophoretically applied (Cinnamon et al., 2001) either to the dorsal midline of closed NTs or to the apical surface of semi-open NTs. A slit was cut along the dorsal midline and then the right hemi-tube was flattened onto the somites. CM-DiI was locally delivered to the apical side and embryos were photographed immediately to assess for initial localization of the labeled cells. The relative positions of the dorsal and ventral limits of labeling were measured and expressed as a percentage of total dorsoventral extent of the NT. The accuracy of the time=0 measurements was validated in a parallel control series in which the localization of labeled cells in the live semi-open NTs was compared with their localization following immediate fixation and cross-sectioning. In 16 such embryos, the mean deviation of whole-mount measurements was 4.16±1.2% and 3.75±1.07% in dorsal and ventral directions in comparison with the mean values monitored in sections (n=16 cases). These highly comparable values validate the use of semi-open NT preparations.

pCAGGS-GFP, pCAGG-qEdnRB2 (Pla et al., 2005) or the Cdk4-binding 15 amino-acid domain of MyoD fused to nuclear localization signal (NLS) (Burstyn-Cohen and Kalcheim, 2002) were microinjected into the NT lumen. For hemi-tube electroporations, tungsten electrodes were placed on either side of the embryos. For focal electroporation of dorsal NT cells, an L-shaped electrode was placed underneath the blastoderm with a pointed electrode dorsal to the desired region. A square wave electroporator (BTX, San Diego, CA, USA) was used to deliver one pulse of current at 20 volts for 10 mseconds.

In addition, pCAGGS-GFP was microinjected using a micropipette with a tip diameter of about 8 μM. The epithelium was pierced once through the ectoderm and a minimal volume of DNA mixed with Fast Green was pressure injected (Ben-Yair and Kalcheim, 2005). Although developed for single cell transfections, the application of a higher DNA concentration (5 mg/ml) resulted in the labeling of 1-3 cells/embryo (n=7, data not shown).

Neural primordia of 18 somite stage (ss) intact quail embryos or of chick embryos unilaterally electroporated with pCAGGS-GFP or pCAGG-qEdnRB2/GFP were excised from the prospective flank, as described (Teillet et al., 1987). Donor tubes were grafted into the gap left after removal of the NT of older chick hosts (aged 38-40ss).

Antibodies against HNK1, GFP, MC/1 (Mochii et al., 1988), Sox9 (Morais da Silva et al., 1996), Islet-1, TH, Melem, P0 and QCPN (Hybridoma Bank) were used as described (Burstyn-Cohen and Kalcheim, 2002). Probes were as follows: Bmp4 (Sela-Donenfeld and Kalcheim, 1999), FoxD3 (Dottori et al., 2001; Kos et al., 2001), Sox9 (Cheung and Briscoe, 2003), Wnt1 (Burstyn-Cohen et al., 2004), Snail2 (Nieto et al., 1994), Cash1 (Jasoni et al., 1994), trkC (from G. Dechant, Innsbruck Medical University, Austria), and ckit (BBSRC clone ID chEST583a5).

The relative dorsoventral localization of DiI-labeled cells was measured as the range between the dorsal and ventral limits of each labeling event monitored at t=0. Results are expressed as a function of the distance from the dorsal midline (0%) and the ventral midline (100%) of the NT. The location of labeled NC derivatives at 33-35 hours after each labeling event was plotted as a function of this initial dorsoventral localization. Data were classified into six groups representing the various localizations of labeled progeny: ventral root cells (VR), DRG+VR, DRG, M+DRG, M, M+DRG+VR. Data were then subjected to statistical evaluation by Discriminant Analysis, in which the means of the ventral and dorsal variables reflecting the extent of labeling at t=0 were compared between the six groups using one-way ANOVA. There was a statistically significant difference between the groups (P<0.001). The distribution of NC derivatives as a function of the time of dorsal NT labeling was analyzed using non-parametric Kruskal-Wallis ANOVA. The Mann-Whitney non-parametric test was then applied to assess differences between sequential pairs of derivatives, using the Bonferroni correction of the significance level for multiple pairwise comparisons (P≤0.0125).

NC cells colonize their derivatives in a general ventral to dorsal order. To map the derivative generated as a function of age, we performed hemi-NT electroporations with pCAGGS-GFP and recorded the localization of labeled progeny at embryonic day (E) 4.5. To avoid axial level variabilities, we focused on the 23rd somite pair. Labeling between 15 and 25ss yielded by E4.5 GFP+/HNK1+ cells in SG, along ventral roots (VRs, Schwann cell precursors), in DRG and subectodermally (melanocytes; Fig. 1A,G; n=10). When electroporations were carried out at 27ss, labeled cells localized along VRs, in DRG and as melanocytes but were no longer detected in SG (Fig. 1B,G; n=10). When performed at 30ss, cells along VRs were no longer detected (Fig. 1C,G; see also Fig. 5B,F). Instead, only GFP+ fibers were observed from this stage onwards (Fig. 1C,G; data not shown). Upon transfection at 35-40ss, only melanocytes were apparent, and at 44ss, electroporations did not produce NC derivatives (Fig. 1D-F,G; n=10).

Furthermore, in each location, GFP+ cells expressed characteristic traits. In SG, they co-expressed Cash1 and TH; along VRs, cells co-expressed P0. Many GFP+ progenitors that homed to DRG differentiated into islet1+, Brn3+ or trkC+ neuroblasts at E4.5 (see Fig. S1A-L, arrowheads, in the supplementary material). Others, located to its mediodorsal part, remained unlabeled at this stage (see Fig. S1G-L, arrows, in the supplementary material). Finally, subectodermal GFP+ cells expressed melanocytic markers such as MC/1, but none of the precedent neural markers (see Fig. S1M-P in the supplementary material; data not shown). Labeled cells within ganglia can still differentiate into neurons or glia (Zirlinger et al., 2002), via Notch signaling (Wakamatsu et al., 2000), yet we did not follow labeled cells until differentiation was complete. These results confirm and extend previous reports invoking a ventral to dorsal order of colonization/formation of trunk NC derivatives as a function of time of delamination, and set the framework for exploring mechanisms underlying cell delamination in relation to this patterned population of homing sites.

To address the mechanism of replenishment of the premigratory pool of presumptive NC, we examined two alternative hypotheses. Upon cell division, premigratory NC could generate dissimilar progeny, one delaminating cell and one cell that remains in the NT. In such a case, the remaining cells would continue cycling to give rise again to one delaminating and one resident founder-like cell, respectively, until depletion of the entire NC pool (Fig. 2A). Alternatively, premigratory NC could generate daughter cells that both delaminate. Upon cell emigration from the dorsal midline area, progenitors localized at a slightly more ventral position would be directed dorsalwards prior to delamination (Fig. 2B). To discriminate between these possibilities, we performed discrete labelings of the dorsal NT opposite young epithelial somites and tracked cellular dynamics. If the asymmetric mode of cell delamination were correct, we would predict that labeled progeny would be detected in both NC derivatives and the dorsal NT. Conversely, no residual labeled progeny in the NT would be expected under the second mechanism. Iontophoretic injections of CM-DiI were directed to the dorsal midline of the NT. In controls fixed shortly thereafter, labeling in a restricted dorsal area (comprising five cells on either side of the midline) was detected in 31/42 embryos (74%; Fig. 2C). In nine out of 42 cases, only the ectoderm was labeled; in one out of 42 embryos, emigrating cells were detected at fixation; and in one out of 42 cases, labeling was seen beyond five cell diameters from the dorsal midline. In this experiment, DiI-injected embryos were re-incubated for 16 hours, when labeled cells were migrating ventralwards or in nascent ganglia. All labeled cells left the NT in 91% of cases (n=34, Fig. 2D). In addition, focal electroporations of pCAGGS-GFP were performed and labeled progenitors followed over time. Six hours after electroporation (control series), GFP+ cells were in the dorsal midline region in 13/16 cases (Fig. 2E). In two out of 16 cases, cells were detected outside of the NT and in the remaining case, labeling was too ventral. When fixed at 28 hours after electroporation, 88% of embryos (n=34) exhibited labeled cells in DRG and other characteristic NC homing sites, with no residual progeny in the NT (Fig. 2F). The few cases in which residual progeny were detected were likely to originate either from broader or initial ventral labelings, in line with results observed in a minority of short-term controls (see above). Consistently, when injections of pCAGGS-GFP attained only 1-3 dorsal NT cells, full emigration was seen in all cases over the period of NC production (see Fig. 5). Next, we followed the dynamics of cell delamination in individual embryos after discrete electroporation of the dorsal NT. Fig. 2G-I illustrates a typical experiment showing the distribution of several GFP+ cells along the dorsal NT 6 hours post-transfection (Fig. 2G) that had completely delaminated by 24 hours post-transfection (Fig. 2H); this was further confirmed in transverse sections at 48 hours post-transfection in which labeled cells contributed to neural derivatives (Fig. 2I).

These data rule out an asymmetric mode of NC emigration. Instead they suggest a progressive ventral to dorsal cell relocation followed by delamination until depletion of the NC pool.

To directly examine the possibility of ventral to dorsal relocation prior to cell exit, focal injections of pCAGGS-GFP were directed to NT sites ventral to the dorsal midline opposite epithelial somites in 25ss embryos (n=8). A dorsal view 6 hours later revealed labeled cells as pseudostratified progenitors spanning the apico-basal extent of the epithelium (Fig. 3A,B,B′). At this time, the dorsal midline was opened and labeled hemi-tubes were flattened over the somites to enable a follow up of cell relocation over time. This confirmed that labeling was distant from the dorsal midline/ectoderm at 6 hours post-transfection (28% from the dorsal midline, Fig. 3C), yet gradually approached the ectoderm by 24 hours (20%, Fig. 3D). By 48 hours post-transfection, cells reached the dorsal midline and also generated HNK1+ NC cells localized in dorsal NC derivatives (Fig. 3A,E,E′,F-I, n=8/8). Hence, progenitors located at relatively ventral sites progressively relocate dorsalwards prior to emigration. This relocation is likely to involve not only prospective NC cells but also, at least, future roof plate (RP) cells that will constitute the permanent dorsal spinal cord midline after completion of NC production and emigration.

The rostrocaudal expression of Snail2 is gradually restricted to caudal regions of the axis and shuts off from axial levels from which NC emigration is still underway (Sela-Donenfeld and Kalcheim, 1999). Likewise, expression of FoxD3 is transient (Kos et al., 2001). Here, we report that in whole embryos Snail2, FoxD3 and Sox9 mRNAs persist until about 35ss, but in older embryos transcripts were no longer detected (see Fig. S2A-H in the supplementary material; data not shown). In the flank, this stage corresponds to the end of ventral NC migration and the onset of lateral migration that generates melanocytes (Figs 1, 5).

Next, we sought to determine whether the ventrodorsal relocation of progenitors is paralleled by a corresponding narrowing of the pattern of gene expression. To this end, transverse sections were analyzed. Expression of Snail2, FoxD3 and Sox9 was relatively broad opposite flank epithelial somites (26ss), became significantly narrower opposite young dissociated somites (32ss) and had completely disappeared from the dorsal NT at the level of mature dissociated somites (37ss; Fig. 4A-I). A similar pattern was observed for Sox9 immunoreactivity (Fig. 4J-L). Hence, peripheral neural progenitors transcribe at least a combination of Snail2, FoxD3 and Sox9, and prospective melanoblasts become negative for these genes by the time of emigration.

The domains of Bmp4 and Wnt1 expression were also progressively reduced; however, in contrast to those of the above mentioned genes, they remained in the dorsal NT and later in the RP after NC emigration had ended (see Fig. S2I-O in the supplementary material). Taken together with our time-lapse analysis, these observations suggest that the definitive RP is born ventral to the prospective NC and reaches its final localization after NC cells have completed emigration.

Next, we directly examined whether the observed restriction of gene transcription domains results from the actual emigration of NC cells, as is inferred from the lineage tracing. To this end, NC emigration was prevented by overexpression of the 15 amino acid (aa) C-terminal domain of MyoD, which specifically inhibits the transition from G1 to the S phase of the cell cycle, a prerequisite for NC EMT in the trunk (Burstyn-Cohen and Kalcheim, 2002). Transfection of control GFP revealed 16 hours later the emigration of GFP+/FoxD3+ co-expressing progenitors, and the premigratory domain of FoxD3 was, as expected, bilaterally symmetrical and reduced (Fig. 4M-O, n=10). By contrast, in 15aa-treated hemi-NTs, the FoxD3 domain remained broader when compared with either the contralateral side or control GFP-transfected sides, and NC delamination was correspondingly inhibited (Fig. 4P-R, n=10). A wider expression domain was also observed for Sox9 (n=6, data not shown). Taken together, our results show that the progressive restriction of gene transcription domains reflects the actual exit of the expressing NC cells. This further supports the dynamic mode of premigratory NC behavior.

The preceding results suggest that the dorsal midline region is a `sink' through which NC precursors pass progressively on their way to leave the neural primordium. This suggests that the relative dorsoventral localization of progenitors in the NT at a given stage might reflect their final homing pattern and, hence, be associated with their fate. To test this hypothesis, we mapped distinct dorsoventral sites within the dorsal 30% of the NT of embryos aged 27-28ss opposite epithelial segments. At this stage, most cells destined to the SG had already emigrated from the NT (Fig. 1). CM-DiI was microinjected and the dorsoventral localization of labeled cells was monitored shortly thereafter. The final localization of labeled progeny was then assessed at 33-35 hours post-injection by serial transverse section analysis (n=32, 3-8 cases/group). Fig. S3 in the supplementary material shows that the more dorsal the labeling in the NT, the more ventral did the labeled progeny home, and vice versa. Furthermore, whereas discrete injections localized to single derivatives (see Fig. S3A-E in the supplementary material), broader labelings were found in two or even three sequential targets (see Fig. S3A in the supplementary material). A strong and significant correlation between the dorsoventral range of labeling within the NT and the final localization of NC derivatives was found, and the progeny contained within single derivatives was spatially separated from each other within the premigratory domain (see Fig. S3A in the supplementary material; P<0.001, discriminant analysis). Thus, there is an inverse spatial relationship between the dorsoventral localization of prospective NC cells in the neural primordium and their final localization in peripheral targets.

Given the dynamic behavior of NC cells and the reciprocal relationship between the localization of progenitors in the NT and their homing sites, we reasoned that the final NC targets are linked to the precise time of delamination. To test this, pCAGGS-GFP was injected into the dorsal NT midline and 1-3 cells were labeled; the accuracy of injections was assessed shortly after the onset of GFP fluorescence (Fig. 5A) (Ben-Yair and Kalcheim, 2005; Ben-Yair and Kalcheim, 2008). Labelings were performed at somitic level 23 in progressive stages, as defined in Fig. 1, and embryos were fixed at E4.5. Data from closely staged embryos were pooled and the results of 50 such injections (at least four cases/age range) are summarized (Fig. 5B). When transfections were performed in 15-25ss embryos, 72% of the progeny was detected in SG, with only 12 and 16% in VRs and DRG, respectively. By 28-29ss, we observed no colonization of SG and, instead, in 22% of cases labeling was along VRs, and in 78% it was in DRG. Injections between 30-31ss produced cells only in DRG, the colonization of which spanned a relatively long period. DRG colonization was over by 35ss when we detected melanocytes. The contribution to RP was apparent from 37ss onward and embryos older than 40ss revealed the presence of labeled progeny only in the RP, consistent with this stage marking the end of NC production (see Fig. 1). Hence, the final localization of NC-derived cells can be faithfully predicted from the time they delaminate from the dorsal NT (P<0.0001, non-parametric Kruskal-Wallis ANOVA). Further analysis of sequential pairs also revealed a significant temporal separation between most consecutive derivatives (SG/VR, P≤0.002; VR/DRG, not significant; DRG/melanocytes, P<0.0001; melanocytes/RP, P≤0.02, Mann-Whitney test; Fig. 5C). These results, together with the data shown above, suggest that a dynamic spatial and temporal fate map of NC derivatives exists within the approximate dorsal 25% of the neural primordium at a stage that precedes the onset of cell emigration.

The pooled results presented in Fig. 5B,C suggest a clear temporal bias among emigrating cells to generate discrete rather than multiple derivatives. Observation of individual embryos further revealed that in 47 out of 50 cases examined (94%), labeled derivatives populated a single anlage, either SG (8-16 cells/48 hours, n=7), VR (18-36 cells/48 hours, n=3), DRG (3-13 cells/40 hours, n=17), melanocytes (1-9 cells/28 hours, n=9) or RP (1-5 cells/24 hours, n=11; Fig. 5D-H). Notably, cases with labeled melanocytes or RP cells were also found to be mutually exclusive, suggesting that the two lineages are already segregated by this stage (Fig. 5G,H). This is even more significant given that injections were not designed to label single cells albeit the descendants were clearly derivatives of a very small group of initially transfected progenitors. In a minority of cases, in which two derivatives were observed, they were sequentially related along the order of derivatives normally observed (i.e. SG and VR, but not SG and DRG); this suggests that labeling in these cases was broader than expected.

Taken together, our results show the existence of a dynamic spatial and temporal map of NC derivatives in the premigratory domain of the NT. Furthermore, they raise the intriguing question of whether this map reflects an early fate restriction of premigratory NC progenitors.

The restriction of NC descendants to single targets raises the question of whether fate restrictions already occur in the dorsal NT prior to emigration, or whether they are imposed by the environment following cell exit. To directly test these hypotheses, we altered the migratory pathways of selected progenitors and asked whether they change their fates according to the new environment (i.e. still multipotent) or, alternatively, retain their original fates even in an ectopic environment (i.e. already restricted/specified). To this end, we electroporated Endothelin Receptor B2 (EdnRB2), a gene that directs lateral migration of melanoblasts (Harris et al., 2008). EdnRB2 transcription begins after melanoblasts have emigrated but it is not expressed within the NT or in neural derivatives (data not shown) (see also Pla et al., 2005). EdnRB2 was misexpressed in the early NT and fixation followed a day later to trace prospective `neural' progenitors, the only ones that exited the NT by this stage. As expected, by E3, labeled cells were diverted into the melanoblast-specific dorsolateral pathway between dermomyotome and ectoderm, instead of migrating normally along the ventral pathway (Fig. 6, compare B-D with A and E). Even if the transfected cells were exposed to the lateral `melanocytic' environment, they failed to adopt melanocyte phenotypes, as monitored by a lack of expression of MC/1, Melem or ckit (Fig. 6B-D,D′), which are otherwise expressed by melanocytes a day later (Fig. 6F-H). Instead, they adopted phenotypical markers characteristic of `neural fates', as defined by the expression of Cash1, Islet1, FoxD3, TrkC or neurofilament NF-M (Fig. 6N-R), and retained those markers at least until E4 (see Fig. S4 in the supplementary material; data not shown) when melanocyte differentiation is underway in the lateral environment. In control GFP-treated embryos, migration at equivalent stages was only through the ventral pathway, where cells expressed neural markers, as expected; these markers were never observed in the dorsolateral pathway under normal conditions at either early or late stages (Fig. 6I-M; see Fig. S1 in the supplementary material).

To further challenge the notion that some fate restrictions are made prior to cell migration and therefore are not influenced by the environment of the migratory routes, we exposed young neural tubes to an old `melanocytic' environment in which the ventralward migration of neural progenitors is completed. Neural primordia of 18ss donor chicks were electroporated with EdnrB2/GFP. In parallel, the NTs of old hosts aged 37-40 somites were removed and the gap was filled with young grafts (see Fig. S5A-E in the supplementary material). Forty-eight hours later, many EdnrB2/GFP-expressing progenitors, comprising both early-diverted `neural' precursors and normal melanocytes, colonized the dermis. Despite being within a bonafide melanocytic environment, a subset of EdnrB2-misexpressing cells, most likely the original neural population, kept the neural marker FoxD3 (Fig. 7E) which is never observed in migrating melanoblasts (Fig. 6K, Fig. 7A, n=5); cells also upregulated neural traits, such NF-M, Cash1 or Islet-1 (Fig. 7F-H, n=at least five chimeras/phenotype). By contrast, control GFP+ donor cells (from either chick NTs electroporated with pCAGGs-GFP or QCPN+ cells from intact quail NTs) that localized to dermis never expressed neural markers (Fig. 7A-D), yet they exhibited melanocyte traits (not shown). Notably, control cells also homed to neural derivatives even in the old environment. In these locations, both at 24 hours and 48 hours post-grafting, control labeled cells formed dorsal miniganglia (Fig. 7B) and/or integrated into the host DRG, where they co-expressed neural markers (Fig. 7A-D; see also Fig. S5F,G in the supplementary material).

Altogether, we show that the restriction/specification to neural fates is independent of the migratory environment and is likely to occur prior to cell emigration from the NT.

Our data suggest that the ventral to dorsal order of colonization of trunk NC derivatives can be explained by ordered progenitor delamination. Continuous cell exit is compensated by a corresponding ventral to dorsal relocation of progenitors towards the dorsal midline area of the NT, which therefore acts as a transition zone for the progressive influx and departure of cells (Fig. 8). We propose that initial NC delamination generates a dynamic progression of epithelial progenitors towards the midline. Consequently, a progressive narrowing of the pre-migratory NC domain occurs until disappearance and concomitant replacement by the definitive RP (Fig. 8).

The RP is considered as the dorsal midline center that develops upon NT closure (Chizhikov and Millen, 2004; Chizhikov and Millen, 2005). Our results show, however, that the prospective RP originates in a ventral stripe within the dorsal NT that continuously expresses Wnt1 but that only initially shares a common lineage with NC, and is brought to its characteristic dorsal position only after NC emigration. In addition, prior to melanoblast emigration, RP cells and melanoblasts are already lineally segregated. Hence, the dorsal NT is a dynamic area during NC ontogeny; despite progressive NC replacement, the dorsal NT continuously expresses members of the BMP, Wnt and Lmx families, whose functions change with development (Burstyn-Cohen et al., 2004; Chesnutt et al., 2004; Chizhikov and Millen, 2005; Sela-Donenfeld and Kalcheim, 1999). In light of our data further discriminating the genuine RP from the transiting NC, we propose that the term `roof plate' should be used to reflect only this definitive subset of dorsal NT cells left after the complete exit of NC progenitors.

Upon labeling the dorsal NT, GFP+ cells exited the NT leaving no residual progenitors behind. Thus, it is unlikely that cell delamination occurs via an asymmetric mechanism. These data are generally consistent with the findings of a study in which NC cells were followed by time-lapse imaging in slice cultures (Ahlstrom and Erickson, 2009). Although in these experiments there were cases of cell divisions in which one daughter remained in the NT, the duration of the imaging sessions (2-4 hours) might not have been long enough to track full emigration of the entire progeny. Along this line, we found that it takes about 5 hours for cells to leave the NT at the level of dissociating somites in vivo (Burstyn-Cohen and Kalcheim, 2002).

Previous studies highlighted the need for downregulating FoxD3 to enable the upregulation of MITF and melanogenesis (Thomas and Erickson, 2009). Because similar to the downregulation of FoxD3, Snail2 and Sox9 (at least) are also lost from the dorsal NT prior to melanoblast emigration, it is likely that the latter two, along with FoxD3, form part of a network that influences neural versus melanocyte development. The differential expression of these three genes to prospective neural lineages but not melanoblasts highlights early molecular differences between the above fates that are already apparent in the premigratory domain.

Altogether, these results not only emphasize for the first time the cellular mechanism of NC emigration, they also show that, at least shortly before the onset of NC delamination, NC progenitors generate a limited number of cell types and, thus, might not even undergo self-renewal. Consistently, we found that most of the cells labeled at a precise stage and axial level generated progeny that colonized single derivatives, raising the possibility that, at the stages concerned, NC cells become fate-specified prior to emigration.

When lysinated-rhodamine dextran was injected into single dorsal NT progenitors, 60% and 73% of the resulting clones contained labeled progeny in multiple NC sites, not always arranged according the normal order of colonization, and others were composed of both NT and NC, or only NT cells (Bronner-Fraser and Fraser, 1989; Bronner-Fraser and Fraser, 1988). These results were interpreted to favor multipotency of the premigratory NC. A difference between our study and those described above resides in the stages and levels analyzed. In our study, we systematically focused on the flank shortly before the onset of cell emigration; the high percentage of cells in single derivatives (94%) is thus unlikely to reflect a stochastic distribution within a multipotent population. In the analyses by Bronner-Fraser and Fraser, embryos ranged considerably in stages, between 10 and 17 HH or 11 and 18 HH, and in those a wide range of somitic levels (8-28 or 12-25, respectively) were labeled. Given this variability, many of the injections must have labeled primitive neuroectodermal progenitors, in which the segregation between NT and NC, or between the various NC fates, might not have yet occurred. In such cases, the single cells marked might have extensively proliferated within the neuroepithelium prior to emigration, yielding a combination of fate-restricted cells by the time they began exiting the NT, which therefore colonized multiple sites. According to this notion, we assume that the single type of progeny that was also observed in the above studies reflects injections performed at later stages.

The restriction of NC descendants to single targets raises the question of whether fate specification occurs in the dorsal NT or whether it is imposed on multipotent cells by the environment following cell exit. The observation that early NC cells, which would normally yield neural derivatives in ganglia, retained their identity even if forced to migrate ectopically sustain the former notion. Nevertheless, our results do not rule out the existence of multipotent progenitors or of plasticity of restricted progenitors, as evidenced in back-grafting experiments (reviewed by Le Douarin and Kalcheim, 1999). In the back-grafting experiments, however, chimeras were analyzed only late in development; restricted/specified progenitors might not have survived in ectopic locations until these stages.

Growing evidence supports the notion of fate restriction or at least fate bias prior to cell emigration in vivo (Anderson, 2000; Dorsky et al., 1998; Dorsky et al., 2000; Harris and Erickson, 2007; Le Douarin, 1986; Le Lievre et al., 1980; Raible and Eisen, 1994), and even in vitro (Baroffio et al., 1990; Ziller et al., 1987). Apart from differences in early gene expression patterns exhibited by NC subsets (Ernsberger et al., 2005; Luo et al., 2003; Robertson and Mason, 1995), it was found that the expression of neurogenins is associated with a fourfold increase in chance to develop a sensory rather than autonomic fate (Zirlinger et al., 2002), and that forced expression of neurogenins biases NC cells to become sensory neurons (Perez et al., 1999). Luo et al. demonstrated that TrkC+ progenitors generated neural derivatives only, whereas c-kit+ cells produced only melanocytes (Luo et al., 2003). An elegant lineage analysis in explants revealed a very rapid fate restriction of the emigrating NC cells (Henion and Weston, 1997). Furthermore, data suggested that melanocytes become specified before or shortly after delamination (Harris and Erickson, 2007; Thomas and Erickson, 2009). The knowledge gathered in the present study provides a basis for investigating further the timing of cell decisions during the segregation of NC-derived lineages, as well as the relative significance of cell intrinsic versus environmental cues in this process.

We thank all members of our group for discussions, Tallie Bdolach for assistance with statistical analysis, and J. Yisraeli and A. Klar for critical reading of the manuscript. This work was supported by grants from the DFG (SFB 488), the ICRF and the Israel Science Foundation to C.K.