Cell tracing reveals a dorsoventral lineage restriction plane in the mouse limb bud mesenchyme

Patterning of embryonic primordia involves tight regulation of cell behavior during tissue growth, morphogenesis, lineage diversification and differentiation. Understanding how complex organs develop therefore requires the deciphering of how a cell's behavior is defined by its lineage history,the external cues it senses and its position within the developing primordium. A key feature underlying organ patterning during development is the subdivision of developmental fields into independent units of growth and patterning known as compartments. Compartments were originally described in Drosophila imaginal discs as subdivisions (normally hemisections) of organ primordia occurring on an otherwise homogeneous epithelial cell field and whose coherence is maintained by cell lineage(Garcia-Bellido et al., 1973; Lawrence, 1973; Morata and Lawrence, 1975). Compartment boundaries are therefore unique lines at stereotyped positions in a developing organ, across which cell intermingling does not take place, even though it is allowed elsewhere in the primordium. Drosophilacompartment borders do not necessarily coincide with or anticipate any morphological or cell-type discontinuity even after overt differentiation(reviewed by Blair, 2003). Thus, Drosophila compartmental organization is a background subdivision of embryonic fields that serves to establish positional references in the primordium and is not necessarily related to anatomical boundaries in the organism.

Lineage restriction units resembling Drosophila compartments have also been described in vertebrates. Pioneering work in the chick, involving labeling of multiple neuroepithelial cells with a lipophilic dye, identified cell lineage restriction boundaries at the frontiers between rhombomeres(Fraser et al., 1990). Similarly, fate mapping of multiple cells identified a cell lineage restriction border at the dorsoventral (D-V) boundary of the chick limb ectoderm (Altabef et al.,1997). In the mouse, the use of inducible site-specific recombination (Brocard et al.,1997) has allowed the prospective lineage analysis of groups of cells defined by the expression of particular genes/enhancers(Kimmel et al., 2000). These approaches, collectively known as genetic inducible fate mapping (GIFM), have been used extensively in the analysis of nervous system development, providing insightful information on neural tube regionalization, brain morphogenesis,neural cell-type specification mechanisms and neural stem cell dynamics(reviewed by Joyner and Zervas,2006). The use of GIFM in the mouse has also identified several other lineage restriction borders in the vertebrate neuroepithelium, such as the midbrain-hindbrain boundary (MHB), the zona limitans intrathalamica (ZLI)and others (reviewed by Kiecker and Lumsden, 2005). Similarly, GIFM has revealed the lineage compartmentalization of the mouse limb ectoderm into dorsal, dorsal apical ectodermal ridge (AER) and ventral compartments(Kimmel et al., 2000).

A common characteristic of the lineage compartments described so far in insects and vertebrates is their occurrence in epithelial structures, so that each compartment is a two-dimensional epithelial area and the compartment border is a line between two adjacent compartments (reviewed by Blair, 2003; Kiecker and Lumsden, 2005; Vincent, 1998). Furthermore,most lineage restriction borders described in both vertebrates and insects are associated with signaling centers (Kiecker and Lumsden, 2005), which suggests that a major role of lineage compartments during embryonic development is signaling-center stabilization.

In contrast to Drosophila compartments, however, all lineage restrictions described so far in vertebrates coincide with, or anticipate,anatomical or cell-type discontinuities. The known restrictions in vertebrates may thus not be a background subdivision of embryonic fields, but might instead largely correlate with strategies to allocate cells fated to different anatomical structures. Systematic searches for cell lineage restrictions during vertebrate embryonic development, however, have been limited, and therefore our knowledge of the complete set of restrictions that occur is likely to be incomplete. Efforts in this direction have made use of the low-frequency spontaneous recombination of a mutant lacZ transgene(laacZ) (Bonnerot and Nicolas,1993). This transgene, driven either by ubiquitous or specific promoters, has been successfully used to determine fundamental parameters of tissue growth dynamics and regionalization in the developing heart(Meilhac et al., 2004; Meilhac et al., 2003),paraxial mesoderm (Nicolas et al.,1996) and CNS (Mathis and Nicolas, 2000; Mathis et al.,1999; Wilkie et al.,2004). This method is constrained, however, by the inability to control the timing and frequency of recombination events. More recently, GIFM at low recombination frequency has been successfully used in a random retrospective analysis of cell lineage at clonal resolution in the hair follicle (Legue and Nicolas,2005).

Here, we apply a novel strategy based exclusively on knock-in alleles of ubiquitous expression, which allows unrestricted clonal analysis of cell lineage from the two-cell stage to the adult mouse. Using this strategy, we have analyzed the topology of cell clone distribution in the developing vertebrate limb, with a particular focus on the mesenchyme. The lineage analysis we present demonstrates that there is no lineage compartmentalization at any position along the proximodistal (P-D) or anteroposterior (A-P) axes,indicating that patterning along these axes does not involve restrictions of cell dispersion at specific axial positions. By contrast, we have identified a D-V lineage restriction boundary in the limb mesenchyme. This is the first example from any organism of a lineage boundary restriction operating within a mesenchymal tissue. The resulting D-V lineage compartments are three-dimensional and the compartment border is a plane that is neither associated with any obvious signaling function nor coincident with any anatomical or cell-type discontinuity.

RERT and R26R or R26R-EYFP knock-in alleles were maintained in double homozygosity in a mixed genetic background, a condition that did not show any obvious adverse phenotype. To generate experimental litters, males of the double-homozygous stock were mated with CD1 outbred females, so that all embryos from these crosses were double-heterozygous for the inducer and reporter alleles. To induce recombination, pregnant females from timed matings were injected intraperitoneally with 4-hydroxy-tamoxifen(4OHTM) dissolved in corn oil at 0.5 mg/ml. lacZ staining [i.e. staining for β-galactosidase (β-gal) activity] was performed as previously described (Torres,1998). Embryos were post-fixed in 4% paraformaldehyde (PFA) and stored in 80% glycerol, 4% PFA.

Embryos were fixed for 30 minutes in 4% PFA, immersed in 30% sucrose until equilibrated, and frozen in OCT embedding compound. Sections (10 μm) were cut, air dried for at least 1 hour and washed twice for 5 minutes each in PBS containing 0.1% Tween 20 (PBT), once for 15 minutes in 0.1% Triton X-100 in PBT and a further twice for 5 minutes each in PBT. Sections were blocked in 5%donkey serum in PBT for 1 hour, followed by incubation for 1 hour at room temperature with goat anti-Lmx1b antibody (Santa Cruz Biotechnology, sc-21231)diluted 1:100 and anti-GFP antibody (Becton Dickinson, Living Colors, 632460)diluted 1:100 in 5% donkey serum in PBT. Sections were then washed three times for 5 minutes each in PBT and incubated with donkey anti-goat Cy3 and donkey anti-rabbit Alexa-488 secondary antibodies for 30 minutes, washed three times for 5 minutes each in PBT and mounted in Vectashield (Vector Laboratories).

Paraffin sections (8 μm) from whole-mount embryos stained for lacZ expression were blocked in 3% H₂O₂ in methanol for 25 minutes and washed in distilled water. Antigen retrieval was performed by a 20-minute incubation in a microwave oven at maximum power in tri-sodium citrate buffer pH 6. Sections were then washed in TBST (TBS plus 0.1% Tween-20) for 5 minutes, blocked in 10% goat serum in TBS (15 mM Tris-ClH, 4.5 mM Tris-Base, 150 mM NaCl, pH 7.6) for 30 minutes and subsequently incubated with anti-Lmx1b antibody diluted 1:20 (Abcam, ab-25504)and a biotinylated goat anti-rabbit secondary antibody. Sections were stained using the Vectastain ABC Peroxidase Staining Kit. A similar procedure was used for the simultaneous detection of β-gal and PECAM, except that it was performed on cryostat sections and the antigen retrieval step was omitted.

To calculate the probability of polyclonal specimens, independence of the recombination event was assumed. We therefore estimated the expected frequency of polyclonal specimens affecting adjacent regions by multiplying the observed frequency of positive cells in each region in isolation.

OPT was performed essentially as described(Sharpe et al., 2002), but with a reduction in the time of dehydration in methanol and clearing in BABB,because longer exposure causes the lacZ stain to fade.

Volume rendering and surface mapping were performed using the Visualization Tool Kit software(http://public.kitware.com/VTK/index.php)and the Edinburgh Mouse Atlas software for three-dimensional data processing.

To perform retrospective clonal analyses in mouse embryos, we used a genetic strategy based on the site-specific Cre recombinase(Sauer and Henderson, 1988). Labeled cells are produced by the activity of CreERT2, an inducible recombinase activated by tamoxifen (TM)(Feil et al., 1997). Ubiquitous CreERT2 expression is provided by a knock-in insertion of the CreERT2 cDNA into the 3′ UTR of the RNA polymerase II gene(Guerra et al., 2003), which yields viable homozygotes with no obvious phenotype. Cre-mediated recombination is monitored by expression from either the recombination-activatable R26R or R26R-EYFP knock-in alleles(Soriano, 1999; Srinivas et al., 2001).

To achieve lineage tracing at clonal resolution, we established the conditions for low frequency recombination, such that when a group of positive cells is detected, the probability of polyclonal origin remains low. We achieved this by titrating the dose of 4OHTM. We established the optimal induction dose as 2.5 μg 4OHTM/g of body weight, which yielded frequencies of lacZ-positive limbs of 1-20%.

Calculation of the frequency of polyclonal limbs, however, can only be deduced from the frequency of positive limbs if recombination events behave as independent events; that is, if the occurrence of a recombination event does not correlate with the occurrence of a second one in the surrounding cells. To determine this, we induced recombination in embryos carrying both the lacZ and the EYFP Rosa26 recombination reporters. After inducing recombination, we recorded the frequency of limbs positive for EYFP, lacZ, or both reporters. The frequencies we obtained were in complete agreement with the occurrence of independent recombination events(Table 1).

Another relevant parameter is whether recombination takes place homogeneously across the limb bud. To study this, we induced recombination at a high 4OHTM dose and analyzed the results 24 hours later, to minimize cell movements that might distort the original recombination map. A homogeneous distribution of positive cells was obtained across the three limb bud axes(see Fig. S1A in the supplementary material).

Background recombination in this system is consistently low, with only two out of 204 limbs showing spontaneous positive cells at E14.5. These two clones were composed of very few cells (see Fig. S1B in the supplementary material;data not shown), indicating a late induction event.

Finally, to establish the effective period of clone induction, we determined clone frequency at different times after 4OHTM injection. At the low 4OHTM dose used, we first detected lacZ-positive cells 12 hours after 4OHTM injection at E9.5, but few recombination events were detected at this stage (see Fig. S1C in the supplementary material; Table 2). Most induced clones are already detectable 24 hours after injection, with little increase in clone frequency after this (see Fig. S1D,E in the supplementary material; Table 2). To determine the minimum time of induction, we cultured E9.5 embryos in vitro in a high 4OHTM concentration (50 μM) and found that 6 hours were necessary to observe the first lacZ-positive cells (data not shown). We thus estimate that the majority of recombination events take place between 12 and 18 hours after 4OHTM injection.

To perform systematic retrospective clonal analysis in the developing mouse embryo, we established a general method for time- and frequency-controlled induction of random-labeled cells (see Materials and methods for details). The method relies on the use of ubiquitously expressed knock-in alleles and is suitable for non-invasive permanent cell labeling during embryonic development. As confirmation of the suitability of this approach for systematic lineage analysis in the developing mouse embryo, we were able to detect the previously characterized lineage relationships of the D-V compartments of the limb ectoderm (Fig. 1A-C) (Kimmel et al.,2000). In the current study, we have analyzed the topological distribution of cell clones in the limb mesenchyme.

Mice pregnant with embryos at stages E8.5, E9.5 and E10.5 were injected with 4-hydroxy-tamoxifen (TM), and the distribution of induced clones examined between E12.5 and E14.5. Analysis of positive cell distribution within the limb was performed both by direct whole-mount visualization of lacZstaining and by optical projection tomography (OPT) (n=37)(Sharpe et al., 2002). At all injection stages, considerable dispersion of positive cells along the P-D and A-P axes was observed, indicating a strong tendency for limb mesenchyme cells to disperse and intermingle with neighbors(Fig. 2A). Considerable cell dispersal was also observed along the D-V axis; however, in most specimens dispersal was restricted to one side of a plane dividing the limb into dorsal and ventral sectors, and this affected dorsal and ventral clones similarly(Fig. 2B-H and Table 3). OPT analysis confirmed that the distribution of cells at the ventral limits of dorsal clones (and vice versa) define D-V planes(Fig. 3).

Apart from restriction at this D-V boundary, pronounced cell dispersal was observed across the dorsal or ventral regions. This apparent freedom of movement, coupled with the consistent and remarkably straight boundary at the D-V mid-plane, shows that the lineage restriction cannot be the result of chance.

To determine whether the frontiers respected by dorsal and ventral clones are consistently located at a reproducible position along the D-V axis, we measured the relative extension of a collection of dorsal and ventral clones at specific positions along the P-D axis of digits 2, 3 and 4 at E13.5 and E14.5. The positions of the boundary measured for each independent dorsal or ventral clone were remarkably consistent, with the majority of boundaries formed at a position displaced dorsally from the midline in 5% of the total D-V extension (Fig. 2I). These results thus identify a single specific boundary at a precise position along the D-V limb axis, which is similarly respected by dorsal and ventral cells.

In a fraction of limbs, however, we detected positive cells, both at dorsal and at ventral positions, that did not respect this boundary(Table 3). A statistical analysis to determine whether these violations might be explained by the presence of multiple clones predicted that 4.0% of E8.5 injections and 4.4% of E9.5 injections would contain independent clones in both dorsal and ventral compartments. However, the observed proportion of limbs showing no D-V restriction was 15% for E8.5 injections and 14% for E9.5 injections. By contrast, at E10.5 we found that the proportion of unrestricted clones was slightly higher but not significantly different from that expected for double recombination events. These results indicate a strong but not complete D-V lineage restriction in early limb bud stages. Full D-V lineage restriction appears to take place only after ∼E10.

Limb mesenchyme is composed of resident cells derived from the lateral plate mesoderm and several other incoming cell populations. Our strategy labels not only resident limb mesenchymal cells, but also other cell lineages such as skeletal muscle, endothelium, smooth muscle, melanoblasts and cells of hematopoietic origin. The observed D-V restriction operates on derivatives of resident limb cells, as the vast majority of restricted clones colonized skeletal elements and other derivatives of resident mesenchyme (see below). However, the incoming cell populations that contribute to the limb could not be assessed from the lacZ whole-mount stainings alone. An exception to this limitation was the skeletal muscle clones, which could be identified by morphology, were found in a low proportion of ∼1% (3 out of 312) and were found restricted to either dorsal or ventral muscle masses. This observation reflects the independent ingression into the limb of muscle precursors for ventral and dorsal masses (for a review, see Christ and Brand-Saberi,2002).

We therefore performed an analysis of markers for other incoming cell populations on sections from limbs containing D-V-unrestricted lacZ-positive clones. We used PECAM (also known as Pecam1 - Mouse Genome Informatics) to track endothelial lineages and CD45 (also known as Ptprc) to track hematopoietic lineages in 12 specimens containing clones that violated the D-V boundary. In none of these specimens did we find colocalization of CD45 with clone cells; however, we found that in one specimen all clone cells colocalized with PECAM(Fig. 4). lacZ-positive cells of this clone had a characteristic appearance and distribution in the whole-mount specimen. This characteristic pattern could easily be identified in other unrestricted clones; we analyzed three of these clones for colocalization with PECAM in sections, and found PECAM-positive clone cells in all three cases (not shown). The ability to identify clones contributing to the vascular network in whole-mounts allowed us to estimate that clones of this class account for ∼8% of unrestricted clones (9 out of 109 screened). Unfortunately, we found no histological marker that would allow simultaneous detection of clone cells and melanoblasts, a cell population that colonizes the limb bud by E13.5 (Mackenzie et al., 1997). We were thus unable to determine the contribution of this population to unrestricted clones.

Dorsoventral patterning of the vertebrate limb involves the dorsalizing action of the secreted molecule Wnt7a, produced by the dorsal ectoderm(Parr and McMahon, 1995). Countering this, BMP signaling in the ventral ectoderm promotes ventralization by activating the transcription factor En1(Ahn et al., 2001; Logan et al., 1997; Pizette and Niswander, 2001). In the dorsal mesoderm, the transcription factor Lmx1b (Lmx1 in the chick)responds to Wnt7a signals and specifies the dorsal mesenchymal structures(Chen et al., 1998; Riddle et al., 1995; Vogel et al., 1995).

We noted that the dorsal mesenchymal limb compartment we identified resembles the expression domain of Lmx1b(Dreyer et al., 2004; Schweizer et al., 2004). In distal undifferentiated limb regions, the Lmx1b-positive dorsal domain is slightly smaller than the Lmx1b-negative ventral domain(Fig. 5A,C). At single-cell resolution, the Lmx1b expression domain border was not completely straight, showing some interdigitation between Lmx1b-positive and Lmx1b-negative cells (Fig. 5E,I). To determine the correlation between D-V compartmentalization and Lmx1b expression, we double-stained limbs for Lmx1b and clone cells, and found that the dorsal-most cells of ventral clones were located next to the ventral-most Lmx1b-expressing cells (n=5) (Fig. 5F-I). Cells of ventral clones did not express Lmx1b,even when surrounded by Lmx1b-expressing cells(Fig. 5I). Consistent with the results obtained for ventral clones, ventral-most cells of dorsal clones coincided with ventral-most Lmx1b-expressing cells (n=6)(Fig. 5D,E; Fig. 6E-J). The Lmx1bexpression domain thus coincides with the dorsal limb mesenchyme compartment,and Lmx1b expression is excluded from the ventral compartment.

In more-proximal limb regions, chondrogenic condensations take place at the boundary between dorsal and ventral cells, as defined by Lmx1bexpression. Within the early prechondrogenic condensation, Lmx1b was expressed in a dorsal domain, showing that chondrogenic cells are recruited from both dorsal and ventral compartments, and therefore that the condensations contain the boundary between dorsal and ventral cells(Fig. 5B; Fig. 6F). As condensations mature, Lmx1b expression was lost in the chondrogenic area but remained in the surrounding tissues, maintaining its relative position with respect to the condensation (Fig. 5A; Fig. 6G).

The D-V restriction of the clones affected both the distal undifferentiated limb tissues and the more-proximal differentiating regions. Whereas the D-V frontier was remarkably straight in distal limb regions, it became less uniform in differentiated regions of the limb, suggesting that the frontier between dorsal- and ventral-derived tissues is deformed as they terminally differentiate. In correlation with the Lmx1b expression pattern, the D-V compartment boundary appeared to run across the skeletal condensations,dividing them into dorsal and ventral domains(Fig. 5D-G; Fig. 6E-J). In the prechondrogenic condensations, the expression limit of the fading Lmx1b expression domain corresponded with the ventral-most location of dorsal clone cells (Fig. 6F,I). In the surrounding differentiating soft tissues the limits of clone expansion also correlated with the position of the frontier within the chondrogenic condensation (Fig. 5F,G; Fig. 6G,J).

In our analysis, clonal A-P expansion varied according to clone size. Early-induced large clones colonized up to nearly a half of the total A-P axis extension, whereas small clones expanded less widely(Fig. 7A-C, see also Fig. 3A and Fig. 8F). Strikingly, despite extensive mixing of clone cells with their neighbors, most clones displayed straight A-P boundaries running parallel to the main limb axis, so that the fraction of the A-P axis colonized by a given clone remained constant along most of the limb P-D axis. Although this growth pattern suggested possible A-P lineage restrictions, mapping the boundaries for different clones identified no specific frontiers respected by limb cells in their expansion across the A-P axis (not shown).

Clone cell distribution in the autopod showed additional interesting features. We observed two characteristic complementary classes of clones:inter-digital and mid-digital clones (Fig. 7C-F). Inter-digital clone cells were loosely distributed across the whole inter-digital area and the lateral regions of adjacent digits, but stopped abruptly at straight boundaries lateral to the midline of the two neighboring digits. Mid-digital clone cells, conversely, were densely packed and restricted to the central region of a single digit. Mixed clones that partly colonize both areas were also detected and, again, mapping the boundaries for different clones identified no specific respected frontiers. These results indicate there are no A-P compartment restriction boundaries and, in fact, that there is a very notable heterogeneity in cell behavior across the autopod A-P axis. Whereas cells in the inter-digital and lateral digit areas actively disperse and intermingle with neighbors, cells at the central digit regions retain the ordered growth and reduced mixing observed in other P-D limb segments.

Vertebrate limbs develop in a proximal-to-distal sequence, being patterned along the three axes by the coordinated action of three signaling centers. One of these centers, the AER, is located at the interface of dorsal and ventral ectoderm. The AER produces signals that maintain a population of distal undifferentiated mesenchymal cells in the underlying region. As the limb bud grows, mesenchymal cells exit the distal region and differentiate to generate limb structures in a proximal-to-distal sequence(Rowe and Fallon, 1982; Saunders, 1948).

We found that clone expansion along the P-D axis was considerable, with a large proportion of clones spanning more than one main limb segment (stylopod,zeugopod, autopod) (Fig. 8A-D). The results obtained were similar for injections at E8.5 and E9.5, and for large clones and small ones, even if composed of very few cells(Fig. 8A-C). Violations of P-D segment frontiers occurred both with clones affecting the skeletal elements and with clones affecting surrounding tissues(Fig. 8D). In view of the frequency of lacZ positives for each segment, we estimated the probability of polyclonal origin in cases of multi-segment lacZpositives. For this analysis we only considered clones affecting skeletal elements, as these constitute the defining elements for limb P-D specification. We also pooled the data from fore- and hindlimbs, as we observed similar ranges of violation frequencies. In all cases, the observed frequencies of clones violating the frontiers between adjacent P-D segments were incompatible with an independent polyclonal origin(P<10^-4) (Table 4), indicating that single-cell descendants can indeed contribute to more than one limb segment. When a clone was restricted to one segment, we never observed accumulation of labeled cells at the edges of any given P-D segment (Fig. 8E,F), as sometimes happens at the borders between lineage compartments. We conclude that the main P-D limb segments do not behave as lineage restriction domains at the stages examined. Furthermore, a high proportion of clones (up to 50%for injections at E8.5), besides contributing to differentiating limb regions,included cells in the undifferentiated distal mesenchyme(Table 4). This indicates that cells in this region have a strong tendency to produce descendants that are able both to contribute to the differentiating structures and to remain undifferentiated under the AER.

We have presented here the characterization of a general method for clonal analysis in the mouse. The system has the principal characteristics required for an informative clonal analysis: random homogeneous and ubiquitous clone induction, control of induction frequency and timing, low background clone induction, and the possibility of using two independent reporters. The method is validated by its successful detection of previously determined lineage compartments. Given the ubiquitous expression of both the reporter and the inducer transgenes, the system is valid for lineage tracing studies at any stage during embryonic development or adult life, and can be applied to the study of any tissue or organ.

We report the topological distribution of clones induced during limb development. Vertebrate limb development has been extensively used as a suitable model to understand patterning mechanisms (reviewed by Tickle, 2003). Detailed fate maps of the chick limb mesenchyme have been obtained recently by labeling groups of cells with lipophilic dyes(Altabef et al., 1997; Dudley et al., 2002; Saadi et al., 1993; Sato et al., 2007; Vargesson et al., 1997) or by viral infection (Dudley et al.,2002). In the mouse, GIFM of Shh-expressing and Shh signal-receiving cells has been used to determine the fate of ZPA cells(Harfe et al., 2004; Scherz et al., 2004) and of cells receiving Shh signal (Ahn and Joyner,2004). These approaches have provided important information about limb growth dynamics and have suggested important modifications to established models of limb A-P and P-D patterning; however, they were not done at clonal resolution.

We found that early-induced large clones derived from single cells colonize considerable A-P extensions by mixing extensively with neighboring cells. At the same time, these clones have remarkably straight A-P borders, so that their relative contribution to the total A-P limb extension is maintained at constant size and position along the P-D axis. These borders do not represent lineage restrictions, but rather indicate that limb cell precursors keep their relative positions along the A-P axis as they contribute progressively to the P-D axis. These results suggest that during an early phase, cells disperse vigorously across the limb primordium, so that descendants of single cells can colonize large sectors of the total initial limb bud. A second phase, likely to start when the limb bud starts to elongate, would be characterized by the cessation of cell mixing and therefore the maintenance of the relative A-P positions of cells in the undifferentiated distal mesenchyme. During autopod generation, ordered growth is maintained at the mid-digital region, while intense cell mixing behavior is regained in the inter-digital areas. Multiple cell-labeling experiments in the chick autopod also suggested extensive mixing in the inter-digital areas versus restricted cell movement in the mid-digital regions (Omi et al., 2000). As a result of this restricted movement, mid-digital areas constitute barriers for the dispersion of inter-digital cells, a phenomenon that might be relevant in the light of the recent model proposing that inter-digital regions are responsible for specifying digit identity(Dahn and Fallon, 2000).

Two alternative models have been proposed to explain P-D specification in the limb bud. One model proposes that an autonomous mechanism for progressive distalization is operative in the progress zone (PZ), which is defined as the region of undifferentiated mesenchymal cells under the influence of AER signals (Summerbell et al.,1973). The more recent early allocation model proposes instead that cells in the early limb bud are first simultaneously specified to the main limb P-D segments in the early limb bud and then expand, get irreversibly determined, and differentiate in a proximal-to-distal sequence(Dudley et al., 2002; Sun et al., 2002). The results of our clonal analysis indicate that there are no lineage restrictions at any position along the P-D limb axis at the stages examined. Lineage tracing,however, does not assess cell specification status and therefore does not exclude the existence of a pre-pattern of P-D regional specification in the early limb bud. The results obtained nonetheless argue against a strong formulation of the early allocation model, in which individual cells would be assigned to specific P-D fates early in limb bud development. Similar results have recently been reported in the chick(Pearse et al., 2007; Sato et al., 2007).

The most significant result from our analysis is the detection of a lineage boundary that subdivides the resident limb mesenchyme into dorsal and ventral compartments. The compartmentalization observed may extend to some incoming cell lineages that enter the limb from external sources, such as the skeletal muscle precursors, whose dorsal mass cells acquire Lmx1b expression as they enter the limb (Schweizer et al.,2004), but not to others, such as endothelial cells, which we found to be insensitive to the D-V border. A D-V lineage restriction has also been detected in parallel studies by retroviral clonal analysis in the chick(Pearse et al., 2007) and GIFM from the Lmx1b locus in the mouse (R. Johnson, personal communication).

This is the first example in metazoans of a lineage compartment operating in a mesenchymal structure. As such, these compartments are organized in three dimensions and their borders are not lines, as is the case in epithelial compartments, but rather a plane between mesenchymal cells. This finding suggests that compartmentalization strategies during embryonic development might be more widely used than previously thought, especially in vertebrates,in which patterning of three-dimensional mesenchymal structures is a frequent process.

The identification of lineage compartment borders in Drosophilawas achieved through the use of cell clones that have a growth advantage with respect to the cells of the background [wild-type cells versus Minutemutant cells] (Garcia-Bellido et al.,1973). Despite the reduced tendency of imaginal disc cells to disperse, this growth advantage allowed wild-type clones to colonize enough territory to delineate the compartment borders, demonstrating the robustness of the mechanisms restricting cell mixing between compartments. In the case of the vertebrate limb, the mesenchymal cells exhibit a strong tendency to disperse during organ growth, resulting in the occupation of large volumes by clones composed of a relatively small number of cells, such that the compartment border is effectively delineated by the majority of the clones in the absence of any growth advantage. It remains to be explored, however,whether the restriction mechanisms operating in the vertebrate limb mesenchyme are sufficiently robust to avoid cell mixing in the case of clones that do have a growth advantage.

One of the most notable characteristics of the vertebrate limb D-V compartmentalization is that it subdivides the primordium into parts that do not correspond to or anticipate any anatomical or cell-type interface. They might thus be different from compartments described so far in vertebrates,which do correlate with transient or definitive anatomical frontiers. The vertebrate limb D-V subdivision therefore appears to result from a background modular design of the developmental field, unrelated to the need to sort cells destined for different anatomical elements or that belong to diverse histotypes. In this sense, the compartments described here might be more similar to classical Drosophila compartments than to others previously described in vertebrates.

Compartments are independent patterning units that express specific selector genes essential for maintaining compartment-specific properties and for regulating pattern formation within their territory(Morata and Lawrence, 1975). We found that the territory corresponding to the limb dorsal compartment correlates with the expression domain of the transcription factor Lmx1b. Eliminating Lmx1b function in the mouse results in double-ventral limbs (Chen et al., 1998), and overexpression of the ortholog gene Lmx1 in chicken results in double-dorsal limbs (Riddle et al.,1995; Vogel et al.,1995), demonstrating this factor's essential role in dorsal specification. Qiu and co-workers (R. Johnson, personal communication) report that at least some of the cellular properties involved in maintaining the dorsal compartment are encoded by Lmx1b and that after an initial unstable period, Lmx1b expression is transmitted stably to daughter cells. This is supported by results with chicken explants and recombinant limbs, which suggest that after an initial phase of activation by Wnt7a, Lmx1 expression is stably maintained independently of external cues(Piedra et al., 2000; Riddle et al., 1995). However,expression of Lmx1b is not dorsally restricted in the early mouse limb bud (Loomis et al.,1998), a result which suggests that the mesenchymal D-V compartment might not be established in the early limb bud and might appear only after Lmx1b expression becomes restricted to dorsal regions. In fact, our results suggest that D-V restriction is incomplete in 4OHTM injections at or before E9.5. Previous studies in the chick failed to identify any D-V lineage restriction in limb mesenchyme before HH16(Altabef et al., 1997),indicating that compartmentalization in the chick may take place after this stage and after the initial Lmx1 activation at stage HH15+(Altabef and Tickle, 2002). Indeed, the initial D-V specification of limb mesoderm depends on external signals, first from paraxial mesoderm and lateral somatopleura, and later from ectoderm (Michaud et al.,1997).

Similar to classical compartments in Drosophila and others described later in vertebrates, the dorsal limb mesenchymal compartment therefore appears to constitute an independent unit of genetic regulation, and Lmx1b appears to be a major regulator of its properties.

Besides the Lmx1b expression transition, there was no other feature that we could associate with the mesenchymal D-V boundary, even at electron microscope resolution (data not shown). A barrier mechanism for the maintenance of the lineage restriction border is therefore unlikely. The compartment boundary might instead be maintained by a differential adhesion mechanism that restricts the strong tendency of mesenchymal limb cells to disperse and mix with neighbors. Such a mechanism would fit well with the observed irregular shape of the border between Lmx1b-positive and Lmx1b-negative cells.

One of the most widely accepted ideas about compartments is that lineage restriction serves to stabilize signaling centers at compartment borders,which constitute organizers for adjacent compartments (reviewed by Vincent, 1998). Classical insect compartment borders, as well as the AER and the MHB in vertebrates, are signaling centers with organizer properties. Compartments at both sides of these borders are independent developmental fields that respond to the organizer activity of the signaling center. Lineage boundaries with no organizer activity, such as rhombomeric limits, also express signaling molecules with essential patterning functions. By contrast, in the case of the D-V limb mesenchymal border, there is no evidence for any organizer or specific signaling activity, as none of the known signaling molecules is specifically expressed at the D-V interface. The subdivision of the developmental field into independent units of patterning and growth might thus be a more fundamental characteristic of lineage compartments than their association with signaling regions. In the case of mesenchymal compartments,in addition to defining independent patterning units, compartmentalization might be relevant to the complex morphogenetic movements that three-dimensional mesenchymal structures such as limb buds have to go through(for example, D-V flattening and bending). Exploring the existence of further compartments of this sort will determine how widely this mechanism is used to pattern mesenchymal fields during metazoan development.

We thank Mariano Barbacid, M^a Victoria Campuzano and Carmen Guerra for the RERT mice, Shankar Srinivas for the R26R-EYFPmice, Jesús García-Herrero for advice on the statistical calculations, Estefanía Escudero for mouse colony management, Cristina Herranz for excellent technical assistance and Cristina Clavería for advice on antibody stainings. We are grateful to Alfonso Martínez-Arias, Peter Lawrence, Jose Luis de la Pompa, Marian Ros and Juan José Sanz-Ezquerro for helpful advice on manuscript preparation. We thank Simon Bartlett for editorial corrections to the manuscript. C.G.A. was supported by a fellowship from the Ministerio de Educación y Ciencia. This work was supported by grants from the Spanish Ministerio de Educación y Ciencia (BFU2006-10978/BMC), the Human Frontiers Science Program (RGP0008/2004) and the European Commission (LSHG-CT-2003-5032259). CNIC is supported by the Spanish Ministry of Health and Consumer Affairs and the Pro-CNIC Foundation.