Differential regulation of gene expression in the digit forming area of the mouse limb bud by SHH and gremlin 1/FGF-mediated epithelial-mesenchymal signalling

In vertebrate embryos, the expression of morphoregulatory genes is highly dynamic and their expression levels and spatial distributions change during progression of embryo- and organogenesis. In particular, the temporally and spatially coordinated expression of members of the four Hox gene clusters regulates various embryonic patterning processes, including limb bud morphogenesis. The so-called co-linear expression of 5′Hoxd and 5′Hoxa genes is crucial to correct limb skeletal patterning, as alterations of their expression kinetics causes dysmorphic phenotypes(reviewed by Kmita and Duboule,2003; Zeller and Deschamps,2002). In particular, the expression of 5′Hoxd genes can be divided in two phases (reviewed by Deschamps, 2004): their posteriorly nested early expression domains in the limb bud mesenchyme (phase I) are established prior to activation of morphogenetic signalling by sonic hedgehog (SHH) (Chiang et al.,2001; Kraus et al.,2001). In fact, Hoxd and Hoxa genes have been implicated in Shh activation in the posterior limb bud mesenchyme(Kmita et al., 2005; Zakany et al., 2004). However,subsequent upregulation and distoanterior expansion of 5′Hoxd gene expression depends on SHH signalling and results in establishment of their presumptive digit expression domains in the distal limb bud mesenchyme (late domains/phase II). Spitz et al. (Spitz et al., 2003) have identified the large cis-regulatory landscape that regulates their expression in the digit-forming area, while the relevant transacting signals and factors regulating their dynamic expression remained largely unknown. Furthermore, extensive loss- and gain-of-function analysis in the mouse has established that their expression in the presumptive digit domains is indeed essential for patterning of the distal limb skeleton and specification of digit identities(Dollé et al., 1993; Zakany et al., 1997).

Shh is expressed by the polarizing region (or ZPA)(Riddle et al., 1993) and mainly specifies identities along the anteroposterior limb bud axis. In limb buds lacking Shh [e.g. mouse(Chiang et al., 1996)] distal development is disrupted and posterior identities are lost such that only one zeugopodal element and one digit form. Furthermore, anterior grafts of SHH-producing cells induce mirror-image digit duplications, in agreement with the proposal that the distal limb buds is patterned by long-range morphogenetic signalling (Riddle et al.,1993; Yang et al.,1997). However, genetic marking of Shh-expressing cells and their descendants has revealed that the ulna and digits 3 to 5 derive largely from descendants of cells that previously expressed Shh. In addition, the study by Harfe et al. (Harfe et al., 2004) revealed that an expansion-based temporal gradient of exposure to SHH probably specifies digits 3 to 5. In particular, the posterior mesenchymal cells that express Shh for the longest time period were shown to give rise to digit 5. Interestingly, this and an earlier study (Lewis et al., 2001) led to the conclusion that only specification of digit 2 would require long-range SHH signalling. Taken together, anteroposterior identities in the limb bud mesenchyme seem to be largely specified by a kinetic memory that integrates response to both autocrine and paracrine SHH signalling(Harfe et al., 2004). In addition, the cellular responsiveness to SHH signalling is modulated locally as the cells exposed to the highest SHH levels reduce their sensitivity to SHH over time (Ahn and Joyner,2004). All these studies emphasize the importance of identifying the molecular circuits that regulate the temporal and spatial kinetics of gene expression during progression of vertebrate limb bud development (reviewed by Zeller, 2004).

During limb bud morphogenesis, upregulation and maintenance of Shhexpression depends on the secreted BMP antagonist gremlin 1 (GREM1), which promotes epithelial-mesenchymal (EM) feedback signalling by the apical ectodermal ridge [AER, expressing several FGF genes(Sun et al., 2002)]. GREM1 acts in the posteriordistal limb bud mesenchyme downstream of the initial,GLI-mediated cellular response to SHH signalling(Zuniga et al., 1999). Experimental and genetic evidence indicates that the SHH/GREM1/FGF feedback loop upregulates and maintains the expression of Shh, Grem1 and 5′Hoxd genes in the posterodistal mesenchyme and of FGF genes in the AER(Haramis et al., 1995; Laufer et al., 1994; Niswander et al., 1994). Analysis of mouse embryos lacking Grem1 showed that GREM1-mediated BMP antagonism in the mesenchyme is essential to induce and/or upregulate the expression of FGF and BMP genes in the overlying AER(Khokha et al., 2003; Michos et al., 2004). As a consequence, the upregulation and maintenance of SHH signalling are disrupted,which is indicative of the failure to establish EM feedback signalling(Haramis et al., 1995; Michos et al., 2004). These molecular alterations result in the characteristic ld phenotype,which includes loss of posterior digit identities and fusion (ulna with radius) or loss (fibula) of the posterior zeugopodal element. By contrast,single and compound mutant mouse embryos lacking FGF genes in the AER of their limb buds did not display phenotypes, as would have been expected from disrupting EM feedback signalling (see Lewandoski et al., 2000; Sun et al., 2000; Sun et al., 2002). These studies left some doubts with respect to the requirements of FGF signalling from the AER for regulation of gene expression in the distal limb bud mesenchyme and specification of digit identities.

In the present study, we analyse the interactions of SHH, GREM1 and FGFs in the distal limb bud mesenchyme by combining analysis loss-of-function mutations in the mouse with manipulation of mouse limb buds in culture. First,we establish that SHH-dependent transcriptional upregulation of antagonists and signals such as Grem1, Bmp2 and Jag1 are controlled by localised and differential mesenchymal responsiveness to SHH signalling. The BMP antagonist Grem1 is an early transcriptional target of SHH signalling in the posterior limb bud mesenchyme, while the Notch ligand Jag1 is identified as a relatively late SHH target in the posterodistal mesenchyme. Grafts of SHH-producing cells into Shh^-/- mouse limb buds reveals that the spatially restricted competence to express a particular target signal is an inherent,SHH-independent property of the mesenchyme. Second, grafts of FGF producing cells into Grem1 deficient limb buds restore the expression of Shh and Hoxd13. In addition, blocking FGF signal transduction with SU5402 in wild-type limb buds results alters gene expression in a similar manner as is observed in Grem1^-/- limb buds. These results establish that GREM1-mediated BMP antagonism acts via FGF signalling to propagate gene expression in the distal limb bud mesenchyme. Third, SHH grafts are unable to restore Jag1 and Hoxd13expression in the distal limb bud mesenchyme of Grem1-deficient embryos. These results indicate that GREM1 is part of a timing mechanism that regulates expression kinetics in response to SHH signalling. Finally, to analyse the temporal requirement of SHH signalling, we block SHH signal transduction from specific time points onwards by treating mouse limb buds with cyclopamine. Rather unexpectedly, these studies reveal the differential and limited dependence of the expression of particular genes on SHH signalling. Upregulation of Grem1 expression requires SHH signalling,while its anterior expansion in the distal mesenchyme appears to be SHH independent. Jag1 expression depends on SHH only transiently during transiently and its expression in the presumptive digit area is largely SHH independent. Rather unexpectedly, only the establishment but not maintenance of the Hoxd11 expression domain in the distal limb bud mesenchyme requires SHH signalling. By contrast, establishment of the presumptive digit expression domain of Hoxd13 requires SHH signal transduction for much longer than Hoxd11. These studies show that the differential temporal dependence of 5′Hoxd genes on SHH signalling correlates well with the reverse co-linear establishment of their late expression domains (phase II)(reviewed by Deschamps,2004).

Heterozygous mice were intercrossed to obtain homozygous embryos for analysis. Noon of the day of vaginal plug detection was considered as day 0.5(E0.5). Wild-type and mutant embryos were age matched according to their somite numbers (variation of ±1 somite). Shh^-/-embryos were genotyped by PCR as described(St-Jacques et al., 1998). Mice homozygous for the Grem1-ld^In2 mutation(Grem1^In2) were intercrossed to generate embryos lacking Grem1 expression specifically in limb buds(Zuniga et al., 2004).

Whole-mount in situ hybridization using digoxigenin-labelled antisense riboprobes was performed as described by Haramis et al.(Haramis et al., 1995). Three or more independent embryos were analysed per stage and genotype, and yielded comparable results.

Mouse forelimb buds were grafted and cultured as described(Michos et al., 2004; Zuniga et al., 1999) with the following modifications. Trunks with forelimb buds attached were isolated from wild-type, Grem1^ln2/In2 and Shh^-/-mutant embryos from embryonic day E10.25 after counting their somites (32-34 somites). After isolation, spherical aggregates of cell beads expressing the desired signalling molecule were grafted in forelimb buds. SHH signalling was blocked by supplementing the culture medium with 10 μM cyclopamine (final,dissolved in ethanol) and FGF signal transduction was blocked with 10 μM SU5402 (final, dissolved in DMSO). Controls were treated with 0.16% ethanol(final) or 0.03% DMSO (final), which is equal to the solvent content in the respective experimental samples. Trunks were cultured between 3 and 32 hours in serum-free, high-glucose DMEM medium (GIBCO-Invitrogen), supplemented with penicillin/streptomycin, L-glutamine, non-essential amino acids, sodium pyruvate, D-glucose, L-ascorbic acid, lactic acid, d-biotin, vitamin B12 and PABA in 6.5% CO₂ at 37°C. When culturing for 32 hours, the medium was exchanged after 18-20 hours. Following culturing, samples were rinsed in PBS and fixed overnight with 4% PFA at 4°C. Each result shown is representative of minimally three independent embryos per genotype and type of manipulation, the analysis of which yielded identical results (in many studies significantly more embryos per result were analysed).

QT6 fibroblast cells expressing Shh, Grem1, Fgf4 or Fgf9(full-length coding sequences cloned into the pRc/CMV vector, Invitrogen)under control of the CMV promoter were generated by standard calcium phosphate transfection. About 24 hours after transfection, cells were plated at high density on bacterial plates, which results in formation of spherical cell aggregates. The following day, cell aggregates were treated with mitomycin-C to completely block their proliferation(Zuniga et al., 1999) and washed extensively before grafting into forelimb buds.

SHH signal transduction is required for positive transcriptional regulation of a variety of mesenchymal signals in the limb bud mesenchyme (data not shown). We had previously identified the BMP antagonist GREM1 as a signal activated prior to SHH, but its continued expression in the distal limb bud mesenchyme becomes rapidly dependent on SHH signalling(Zuniga et al., 1999). Like Grem1, the expression of signals such as Bmp2 and Jag1 is dependent on SHH signalling(Laufer et al., 1994; McGlinn et al., 2005).

To gain insight into the temporal and spatial kinetics by which SHH regulates the expression of mesenchymal signals, SHH-producing cell aggregates were grafted into the forelimb bud mesenchyme of Shh-deficient and wild-type embryos (E10.25, 31-33 somites, Fig. 1). In response to SHH, an anterior ectopic ring of Gli1 expression is induced within 6 hours in wild-type forelimb buds (blue arrowheads, Fig. 1A). These results establish that the initial transcriptional response to SHH signalling is rapid and comparable with endogenous Gli1 expression levels (open arrowheads, Fig. 1A). Localized and strong ectopic Grem1 expression is detected within 9 hours in the mesenchyme distal to the graft (blue arrowhead, Fig. 1B). Ectopic Jag1expression is also detected within 9 hours distally to the graft (blue arrowhead, Fig. 1C), but levels are much lower than the endogenous transcripts (big open arrowheads, Fig. 1C). After 15 hours the levels of ectopic Jag1 expression in the distal-anterior limb bud mesenchyme have increased significantly (blue arrowhead, Fig. 1D) and continue to rise(Fig. 4A and data not shown). These results reveal that mesenchymal cells respond to SHH signalling by differential and localized transcriptional upregulation of secondary signals. Indeed, Grem1 is expressed locally by the dorsal and ventral mesenchyme, while Jag1 (like 5′Hoxd genes) is expressed throughout the distal limb bud mesenchyme (see Fig. S1 in the supplementary material).

To analyse the responsiveness of nascent mesenchyme, SHH-expressing cells were grafted into the posterior limb bud mesenchyme of Shh-deficient limb buds. As a fraction of Shh^-/- mouse embryos die prematurely, a series of pilot experiments established that culturing mutant limb buds for 15 hours allows reliable and reproducible assessment of gene expression levels (Fig. 1E-H;data not shown). As in wild-type limb buds(Fig. 1A), the general response to SHH signalling is revealed by upregulation of Gli1 expression around the graft (Fig. 1E). Despite this widespread initial response, the transcriptional upregulation of the BMP antagonist Grem1 and the Notch ligand Jag1 is always restricted to cells located distal to the SHH graft (blue arrowheads, Fig. 1F,G). By contrast,expression of Bmp2 is mostly upregulated in cells located proximally to the graft (blue arrowhead, Fig. 1H) and the AER (broken arrowhead, Fig. 1H). Restoration of SHH signalling in Shh^-/- limb buds shows that the differential and spatial restricted competence to activate these SHH target signals is maintained in Shh^-/- limb buds.

Our previous genetic analysis indicated that Grem1 is required to establish EM feedback signalling and to propagate Shh expression in the distal mesenchyme (Michos et al.,2004; Zuniga et al.,1999). Likewise, the expression of Jag1 in distal limb bud core mesenchyme depends on GREM1-mediated EM feedback signalling (see Fig. S1 in the supplementary material). However, the analysis of Htuhomozygous mouse embryos [carrying a point mutation in the Jag1 gene(Kiernan et al., 2001)] shows that JAG1 itself is does not regulate the expression of 5′Hoxd and other morphoregulatory genes in the developing limb bud. Rather, this Notch ligand seems to regulate aspects of cell proliferation in the limb bud mesenchyme(L.P. and R.Z., unpublished). For the purpose of the present study, Jag1 is used as an additional marker expressed within the digit-forming territory of the distal limb bud mesenchyme (see Fig. S1 in the supplementary material).

To gain insight into the temporal and spatial kinetics by which GREM1-mediated EM feedback signalling regulates gene expression in the distal limb bud mesenchyme, GREM1 and FGF producing cell aggregates were grafted into the posterior limb bud mesenchyme of Gre1^In2/In2 embryos(E10.25, 31-33 somites). These experiments establish that Shhexpression in the posterior mesenchyme is restored (red arrowhead, Fig. 2A) and Fgf4expression in the AER activated within 9 hours (red arrowhead, Fig. 2B). In limb buds lacking Grem1, grafts of FGF4 and FGF9 expressing cells restore Shhexpression with identical kinetics (Fig. 2C; data not shown). This initial restoration of EM feedback signalling (Fig. 2A,B) is followed by upregulation of Jag1 expression in the distal mesenchyme within 15 hours (between graft and AER, Fig. 2D). The expression of 5′Hoxd genes in the distal mesenchyme is restored within 15 hours(Hoxd13, red arrowhead, Fig. 2E; Hoxd11 and Hoxd12: data not shown). Furthermore, grafts of FGF4 producing cells into Grem1 deficient limb buds restore Jag1 and Hoxd13 expression with identical kinetics (Fig. 2F,G).

As FGFs are able to propagate mesenchymal gene expression in the absence of Grem1 (Fig. 2), these results provide good evidence that FGF signalling is required downstream of GREM1. This is an important finding as the requirement of FGF signalling for patterning of the distal limb bud mesenchyme has been a matter of some debate(see Introduction). To investigate the requirement of FGF signal transduction further, limb buds (E10.5, 34-36 somites) were treated with the inhibitor SU5402 in culture (Montero et al.,2001). Culturing wild-type limb buds in the presence of 10 μM SU5402 blocks FGF signal transduction (Fig. 3A-D). Such treatment interferes with transcriptional upregulation of Jag1 and partly inhibits anterior expansion of its expression(Fig. 3A,B). Similarly,upregulation and anterior expansion of Hoxd13 expression, i.e. establishment of its late, presumptive digit expression domain is disrupted by treatment with SU5402 (Fig. 3C,D). By contrast, Grem1 expression is not significantly altered by blocking FGF signal transduction(Zuniga et al., 2004). In summary, treatment of limb buds with SU5402 results in similar alterations of the Jag1 and Hoxd13 expression domains, as observed in Grem1-deficient limb buds at E11.0 (42 somites, Fig. 3E,F; compare 3E with 3A,B and 3F with 3C,D). These results further support the proposal that GREM1 acts up-stream of and via FGF signalling during limb bud morphogenesis.

To further dissect the functional relevance of GREM1-mediated EM feedback signalling with respect to SHH-dependent regulation of gene expression,wild-type and Grem1^In2/In2 limb buds (E10.25, 31-33 somites) received anterior grafts of SHH-producing cells(Fig. 4) to clearly discriminate induced (blue arrowheads) from endogenous gene expression (open arrowheads). In wild-type limb buds, ectopic SHH signalling induces significant anterior expansion and upregulation of Jag1 and Hoxd13 gene expression in the distal mesenchyme within 15 to 22 hours(Fig. 4A,C; see also Fig. 1D and data not shown). By contrast, no or only little transcriptional upregulation of ectopic Jag1 and Hoxd13 transcripts is observed in the distal mesenchyme of Grem1-deficient limb buds after 22 hours (blue arrowheads, Fig. 4B,D; left panels). In fact, the ectopic Jag1 expression levels in Grem1-deficient limb buds after 22 hours are reproducibly lower than the ones in wild-type limb buds after 9 hours (compare Fig. 1C with Fig. 4B). Only after 32 hours,the levels of ectopic anterior Jag1 and Hoxd13 transcripts(blue arrowheads, Fig. 4B,D;right panels) become comparable with ones in wild-type limb buds (blue arrowheads, Fig. 4A,C; right panels). This indicates an at least 12 hours delay in efficient transcriptional response to SHH signalling in Grem1-deficient limb buds. This delay is probably due to the fact that grafts of SHH-producing cells into Grem1-deficient limb buds do not restore FGF gene expression by the mutant AER and thereby fail to restore EM feedback signalling (Michos et al.,2004; Zuniga et al.,1999). These results establish that GREM1-mediated EM feedback signalling regulates aspects of the temporal kinetics of the transcriptional response to SHH signalling.

To analyse the requirement of SHH signalling during EM feedback signalling and establishment of the presumptive digit expression domains of 5′Hoxd genes, SHH signal transduction was blocked at specific time points by culturing limb buds in the presence of the inhibitor cyclopamine(Chen et al., 2002). This approach was chosen as the genes of interest are either not expressed or downregulated in limb buds of Shh-deficient embryos much prior to establishing GREM1-mediated feedback signalling (Jag1, 5′Hoxd genes) (Chiang et al., 2001; Kraus et al., 2001). The presence of 10 μM cyclopamine in the culture medium significantly reduces Gli1 expression within 9 hours(Fig. 5A), indicative of blocking SHH signal transduction (E10.5, 34-36 somites). After 15 hours of cyclopamine treatment, Gli1 is no longer detectable by whole-mount in situ hybridization, while cellular apoptosis is not yet significantly increased (Fig. 5A; data not shown). Therefore, embryonic trunks were cultured for 15 hours in the presence of cyclopamine prior to analysis (Fig. 5B,C and Figs 6,7). Cyclopamine treatment induces loss of Fgf4 expression from the AER,indicating that Fgf4 requires sustained SHH signalling(Fig. 5B). By contrast, the expression of Fgf8 is only slightly reduced in comparison with wild-type controls (Fig. 5C).

During limb bud morphogenesis, Grem1 expression is activated independently of SHH signalling in the posterior limb bud and its expression is upregulated and expands progressively from posterior to anterior within the distal limb bud mesenchyme under control of SHH signalling(Zuniga et al., 1999). Cyclopamine treatment from E10.25 (31-33 somites) onwards does not alter the anterior limit of the Grem1 expression domain(Figs 6A,B; broken lines indicate the approximate anterior limits), while expression is lost from the distal mesenchyme (brackets in Cyc panel in Fig. 6A,B). This failure to upregulate Grem1 expression in the distal-most mesenchyme is the likely cause of the loss of Fgf4 expression from the overlying AER(Fig. 5B). Furthermore,concurrent inhibition of both SHH and FGF signal transduction neither alters the anterior boundary of the Grem1 expression domain nor decreases expression levels further (data not shown). In light of this rather unexpected maintenance of the anterior expression boundary, Grem1 expression was re-assessed in Shh deficient limb buds(Fig. 6C). As previously reported, Grem1 expression is rapidly downregulated in Shh-deficient limb buds (Fig. 6C, compare panel Shh^-/- with wild type)(Zuniga et al., 1999). Although overall expression levels are low, Grem1 transcripts are reproducibly detected in the distoanterior mesenchyme of Shh-deficient limb buds at E9.75(Fig. 6C, anterior limit indicated by broken line). By E10.5 (34-36 somites), Grem1 transcript levels are further reduced, but expression remains throughout in the distalmost mesenchyme in Shh-deficient limb buds(Fig. 6C). These results show that anterior expansion of the Grem1 expression domain occurs in the absence of SHH, while SHH signalling is required to upregulate Grem1expression in the distal limb bud mesenchyme.

In contrast to Grem1, transcriptional activation of Jag1requires SHH, as it is not expressed in Shh^-/- limb buds(Fig. 1G). Indeed, the initial upregulation of Jag1 expression is affected by blocking SHH signal transduction from E10.25 onwards (Fig. 6D). However, subsequent propagation of Jag1 expression does not depend on SHH signal transduction, as cyclopamine treatment from E10.5 onwards neither interferes with transcriptional upregulation nor with anterior expansion of the Jag1 expression domain in the distal limb bud mesenchyme (Fig. 6E,F).

Finally, the effects of cyclopamine treatment on the establishment of the presumptive digit expression domains of 5′Hoxd genes were assessed(Fig. 7). In contrast to Grem1 and Jag1, SHH signalling is continuously required for establishment of the late, presumptive digit expression domains of 5′Hoxd genes. Cyclopamine-treatment from E10.25 onwards interferes with transcriptional upregulation and anterior expansion of the late expression domains of 5′ most Hoxd genes (Fig. 7A,D; data not shown). By E10.5, the late Hoxd11 domain has been established and maintenance of its spatial distribution no longer requires SHH, while expression levels are lower in cyclopamine-treated animals than in wild-type controls (Fig. 7B). Blocking SHH signal transduction from E10.75 onwards (37-39 somites) no longer affects Hoxd11 expression, despite the fact that its presumptive digit expression domain continues to enlarge(Fig. 7C, compare T₀panel with wild type and Cyc panels). These results reveal that Hoxd11 expression depends on SHH until about E10.5, while subsequent maintenance and propagation of its presumptive digit expression domain occurs independently of SHH signal transduction. The expression of Hoxd12depends slightly longer on SHH as treatment at both E10.25(Fig. 7D) and E10.5 efficiently blocks its transcriptional upregulation and establishment of the late expression domain (compare Fig. 7E with 7B). Only around E10.75, is the establishment of the late Hoxd12 expression domain largely independent of SHH signalling. By contrast, Hoxd13 expression, the presumptive digit domain of which is established last and extends most anterior(Dollé et al., 1993),requires SHH signal transduction for longer (i.e. beyond E10.75, Fig. 7G,H). The results shown in Fig. 7 reveal the graded dependence of 5′Hoxd genes on SHH signal transduction. This differential temporal dependence correlates well with the kinetics by which their presumptive digit expression domains are established. In particular, the expression of the 5′ most Hoxd13 gene(Fig. 7G,H) depends significantly longer on SHH signal transduction than do the ones of the more 3′ located Hoxd12 and Hoxd11 genes(Fig. 7A-F), in agreement with reverse co-linear establishment of their presumptive digit expression domains.

We have analysed the temporal and spatial requirements of the SHH, GREM1 and FGF feedback signalling interactions for upregulation and propagation of their own expression, and for establishment of the presumptive digit expression domains of 5′Hoxd genes and Jag1. In contrast to the primary mesenchymal response to SHH signalling, secondary signals are activated and/or upregulated in a spatially restricted fashion. Our study shows that this spatially restricted competence is an inherent property of the limb bud mesenchyme, which is retained in Shh-deficient limb buds. This spatially restricted competence may be established by the mutual antagonistic interaction of GLI3 with HAND2, which pre-patterns the mouse limb bud mesenchyme prior to SHH signalling (te Welscher et al., 2002a). GLI3-mediated restriction of Hand2 expression to the posterior limb bud mesenchyme seems to regulate the activation of Shh, Grem1 and, possibly, 5′Hoxd genes (te Welscher et al.,2002a; Charite et al.,2000; Yelon et al.,2000; Zuniga et al.,1999; Zuniga and Zeller,1999). In addition, the transcriptional regulators Hoxd13(Chen et al., 2004), Tbx3(Rallis et al., 2005) and Twist1 (Firulli et al., 2005)interact with GLI and Hand2, respectively. In addition, genetic studies have implicated retinoic acid, 5′Hoxa and 5′Hoxd genes in activation of Shh expression in the posterior limb bud mesenchyme(Kmita et al., 2005; Mic et al., 2004; Niederreither et al., 2002). Subsequently, SHH signalling is required to upregulate and propagate mesenchymal gene expression in the distal mesenchyme as the expression of many genes is rapidly downregulated and/or lost in Shh-deficient limb buds(Chiang et al., 2001; Kraus et al., 2001; Litingtung et al., 2002; te Welscher et al., 2002b; Zuniga et al., 1999). Our study establishes that SHH is required for transcriptional upregulation but not distoanterior expansion of Grem1 expression during limb bud patterning. Therefore, the SHH-independent anterior expansion of Grem1 expression in early limb buds could be regulated by the pre-patterning mechanism acting upstream of SHH signalling (see before).

In Grem1-deficient limb buds, the propagation of Shhexpression is disrupted and there is a significant temporal delay in establishing the 5′Hoxd digit expression domains(Haramis et al., 1995; Michos et al., 2004). This delay cannot be rescued by grafts of SHH-expressing cells into the posterior limb bud mesenchyme (this study) and probably results in mesenchymal cells not receiving their positional identities at the correct time. This achronism provides a plausible explanation for the observed loss of posterior identities in Grem1-deficient limbs (Michos et al., 2004). In contrast to SHH, grafts of FGF-producing cells upregulate mesenchymal gene expression in Grem1-deficient limb buds with temporal kinetics comparable with GREM1-producing cells. Furthermore,inhibition of FGF signal transduction in wild-type limb buds phenocopies aspects of the molecular alterations observed in Grem1-deficient limb buds. These results provide good evidence that GREM1-mediated BMP antagonism regulates the mesenchymal response to SHH signalling indirectly via FGF signalling from the AER. Hence, our studies support an essential role of GREM1/FGF-mediated EM feedback signalling in regulation of the temporal kinetics of gene expression and patterning. As GREM1 is required to induce Fgf4, Fgf9 and Fgf17 expression in the posterior AER and to upregulate Fgf8 expression(Michos et al., 2004) (L.P. and R.Z., unpublished), the overall strength of FGF signalling by the AER may be most relevant to FGF-mediated EM feedback signalling. Indeed,transgene-mediated overexpression of Fgf4 in the AER of mouse limb buds lacking Fgf8 completely restores their development, which indicates that FGF4 functionally replace FGF8 in the mutant AER(Lu et al., 2006).

The importance of the temporal control of gene expression is emphasized by the fact that an expansion-based temporal gradient of SHH and modulation of SHH responsiveness over time control patterning of the anteroposterior limb axis (Ahn and Joyner, 2004; Harfe et al., 2004). In the present study, we uncover the temporal requirement of SHH signal transduction for establishment of the late, presumptive digit expression domains of 5′Hoxd genes. Their co-linear expression is reversed in the distal limb bud mesenchyme, such that expression of the 5′ most gene, Hoxd13 extends most anterior and is essential for digit patterning(Dollé et al., 1993). Genetic manipulation of the Hoxd complex in mice has revealed the existence of a large global control region (GCR) that regulates the establishment of their late expression domains in the limb bud mesenchyme(Spitz et al., 2003). A distant digit enhancer is located far 5′ to the Hoxd gene complex and strongest enhances the expression of Hoxd13, while the expression of the more 3′ located Hoxd12 and Hoxd11 genes is enhanced progressively less (Kmita et al.,2002). Our study reveals that establishment but not maintenance of the late 5′Hoxd expression domains requires SHH signalling. In good correlation with their differential regulation by the digit enhancer, Hoxd13 requires SHH signal transduction for significantly longer than the more 3′ located Hoxd12 and Hoxd11 genes. Therefore, their differential dependence on SHH signalling could be mediated by interaction of SHH targets such as the GLI transcriptional regulators with the digit enhancer (Kmita et al.,2002; Spitz et al.,2003). Our study shows that 5′Hoxd genes are rendered SHH-independent in a reverse co-linear fashion as their presumptive digit expression domains are being established. This progressive stabilization and SHH independent expression of 5′Hoxd genes may constitute a part of, or at least mark the kinetic memory that mesenchymal cells may acquire as a consequence of their overall exposure to SHH signalling(Harfe et al., 2004).

Our analysis, together with previous studies allows division of limb bud patterning in three distinct phases. During the initial setup phase, the AER-FGF and ZPA-SHH signalling centres and differential mesenchymal responsiveness are established under the influence of the GLI3R-HAND2 pre-patterning mechanism (Fig. 8, phase I) (e.g. te Welscher et al., 2002a). During this initial phase, the differential responsiveness to SHH signalling is probably setup in the nascent mesenchyme(this study) and Grem1 expression is activated in the posterior mesenchyme (Zuniga et al.,1999). In addition, the early posteriorly nested expression domains of 5′Hoxd genes are established (e.g. Zuniga and Zeller, 1999) and participate in activation of Shh expression(Kmita et al., 2005). The second, dynamic phase is initiated by concurrent establishment of SHH and GREM1/FGF-mediated EM feedback signalling, which coordinates temporal progression with anterior expansion of gene expression in the distal mesenchyme (Fig. 8, phase II;yellow arrow). The present study leads us to conclude that SHH constitutes the main `engine' for this dynamic phase, while the temporal kinetics are regulated in concert with GREM1/FGF EM feedback signalling. During this dynamic phase, the presumptive digit expression domains of Jag1 and the 5′Hoxd genes are progressively established and rendered independent of SHH signalling. Mesenchymal cells probably acquire their kinetic memory of exposure to SHH signal transduction (Harfe et al., 2004) and the expanding population of Shhdescendents displaces the posterior limit of the Grem1 expression domain towards anterior (Scherz et al.,2004). This widening gap between SHH-producing and Grem1-expressing cells eventually terminates SHH/GREM1/FGF-mediated feedback signalling and limb bud patterning(Fig. 8, phase III). The present study reveals the temporal and spatial kinetics by which mesenchymal SHH signalling and GREM1-mediated BMP antagonism function together with FGF signalling from the AER to pattern the distal limb bud mesenchyme. It will be of particular interest to identify the BMP ligands that are antagonized by GREM1 in the limb bud and participate in regulation of the gene expression kinetics. Finally, it will be important to gain more insight into how these highly dynamic morphoregulatory interactions are established(Zuniga et al., 1999; te Welscher, 2002a) and terminated at the appropriate developmental time points(Scherz et al., 2004).

The authors are grateful to C. Lehmann and C. Torres de los Reyes for technical assistance, to A. Roulier for artwork, and to C. Müller-Thompson for help in preparing the manuscript. We thank A. Gossler and F. Guillemot for providing probes for in situ hybridization. We are grateful to N. Matt for experimental suggestions and to anonymous reviewers and our group members for critical input into the manuscript. This study was supported by the Swiss National Science Foundation (R.Z.), both cantons Basel(A.Z., R.Z.), the Dutch NWO (R.Z.), KNAW (A.Z.) and the Stichting Catharine van Tussenbroek (L.P.).