A Self-Regulatory System of Interlinked Signaling Feedback Loops Controls Mouse Limb Patterning

See allHide authors and affiliations

Science  20 Feb 2009:
Vol. 323, Issue 5917, pp. 1050-1053
DOI: 10.1126/science.1168755


Embryogenesis depends on self-regulatory interactions between spatially separated signaling centers, but few of these are well understood. Limb development is regulated by epithelial-mesenchymal (e-m) feedback loops between sonic hedgehog (SHH) and fibroblast growth factor (FGF) signaling involving the bone morphogenetic protein (BMP) antagonist Gremlin1 (GREM1). By combining mouse molecular genetics with mathematical modeling, we showed that BMP4 first initiates and SHH then propagates e-m feedback signaling through differential transcriptional regulation of Grem1 to control digit specification. This switch occurs by linking a fast BMP4/GREM1 module to the slower SHH/GREM1/FGF e-m feedback loop. This self-regulatory signaling network results in robust regulation of distal limb development that is able to compensate for variations by interconnectivity among the three signaling pathways.

Tissue morphogenesis depends on self-regulatory mechanisms that buffer genetic and environmental variations. With the exception of self-regulatory bone morphogenetic protein (BMP) signaling during gastrulation, the mechanisms endowing vertebrate development with robustness are largely unknown (1, 2). The vertebrate limb bud is a classical model to study organogenesis, and its development is driven by signaling interactions between two instructive centers (3, 4). The sonic hedgehog (SHH)/Gremlin1 (GREM1)/fibroblast growth factor (FGF) feedback loop (57) coordinates SHH signaling by the mesenchymal zone of polarizing activity (ZPA) with FGF signaling by the apical ectodermal ridge (AER) (4, 810). In Grem1-deficient mouse limb buds, this feedback loop is not established, which disrupts both signaling centers and distal development as revealed by fusion of ulna and radius and loss of digits (Fig. 1A). Previous studies (57) suggested that GREM1-mediated antagonism of mesenchymal BMPs is key to SHH-mediated specification of digits 2 to 5 and proliferative expansion of the digit territory (autopod) (8, 10). Three BMP ligands are expressed in limb buds (fig. S1) (11), and genetic studies revealed that mesenchymal BMP signaling inhibits anterior expansion of AER-Fgf expression and polydactyly (formation of additional digits) (12, 13) and that a minimal BMP threshold is required to initiate chondrogenesis of posterior digits (14, 15).

Fig. 1.

Genetic reduction of BMP4 preferentially restores Grem1-deficient forelimb buds. Alcian blue and alizarin red stained skeletal preparations of mouse forelimbs at E14.5. Digit identities are indicated by numbers 1 (thumb, anterior) to 5 (little finger, posterior). Fused digits 2 and 3 are indicated as “2/3,” digits with unclear anterior identity (digit 2 or 3) as “a,” and hypoplastic digits with an asterisk. Question marks indicate digits with unknown identity, a hallmark of Grem1-deficient limbs. r, radius; u, ulna; r/u, fused radius and ulna. (A) Wt, wild-type; Grem1Δ/Δ, Grem1 deficient; Grem1Δ/ΔBmp2Δ/+ and Grem1Δ/ΔBmp7Δ/+, Grem1Δ/Δ embryos heterozygous for Bmp2 or Bmp7, respectively. (B) Allelic series to reduce the Bmp4 gene dosage in a stepwise manner in Grem1Δ/Δ forelimbs using both the hypomorphic Bmp4hf and loss-of-function Bmp4Δ alleles. The activity of the Bmp4hf allele is reduced to slightly less than 50% (fig. S1). Digit identities were determined using morphological criteria in combination with carpal elements and Sox9 expression (fig. S3). Digits with restored identity are indicated in red. All panels are oriented anterior to the top and posterior to the bottom.

To identify the BMP ligand(s) antagonized by GREM1, we performed a genetic interaction screen in mouse embryos. Halving the Bmp2 or Bmp7 gene dosage only slightly improved limb development (Fig. 1A), whereas inactivation of one Bmp4 allele restored the zeugopod (ulna and radius) and the posterior-most digits (Fig. 1B) (n = 24/24). Complete inactivation of Bmp7 resulted in similar restoration (fig. S2), which indicates that GREM1 may reduce overall BMP activity. Cross-regulation among BMP ligands is unlikely because genetic lowering of one Bmp did not alter the expression of the others (figs. S1 and S2). Hindlimb development was also restored, but we only show forelimbs because the molecular alterations in Grem1Δ/Δ limb buds have been best characterized in forelimbs (57). In addition, the Prx1-Cre transgene was active in forelimb buds from an early stage.

Because genetic reduction of Bmp4 in Grem1Δ/Δ embryos was most potent, we used a hypomorphic floxed Bmp4 allele (Bmp4hf) (16) with de creased activity (fig. S1) to reduce the Bmp4 gene dosage in a stepwise manner. In this allelic series, forelimb development was progressively restored with proximal-to-distal and posterior-to-anterior sequence (Fig. 1B). In particular, low Bmp4 levels restored pentadactyly in Grem1Δ/ΔBmp4Δ/hf limb buds (Fig. 1B) (n = 10/14). About one-third of these limbs had five almost normal digits [Fig. 1B, right-most panel (n = 3/10) and fig. S3], whereas digits 2 and 3 remained proximally fused in all others [Fig. 1B (n = 7/10) and fig. S3]. This restoration is a likely consequence of rescuing cell survival and the distal 5′Hoxd expression domains, which are disrupted in Grem1-deficient limb buds (fig. S4) (57).

To monitor BMP activity, we evaluated the transcription of Msx2, a direct and early target of BMP signal transduction in limb buds (11). Msx2 expression was increased by a factor of about 2 in the mesenchyme of Grem1-deficient limb buds, whereas it was reduced to wild-type levels in Grem1Δ/ΔBmp4Δ/hf limb buds (Fig. 2A and fig. S5). Shh expression and AER-Fgf expression were drastically reduced in Grem1-deficient limb buds (57) but were partially restored in Grem1Δ/ΔBmp4Δ/hf limb buds (Fig. 2B and fig. S5). Thus, the aberrantly high BMP4 activity in Grem1Δ/Δ limb buds opposes distal limb development and specification of digit identities by interfering with transcriptional up-regulation of ZPA-SHH (810) and AER-FGF (4) signaling.

Fig. 2.

Restoration of signaling in Grem1Δ/ΔBmp4Δ/hf limb buds. (A) Detection of Msx2 transcripts by in situ hybridization in wild-type, Grem1Δ/Δ and Grem1Δ/ΔBmp4Δ/hf (G1Δ/ΔB4Δ/hf) forelimb buds (35 somites). Bottom panels show posterior views (dorsal, top; ventral, bottom) of the limb buds in the top panels. (B) Detection of Shh (top, 37 somites) and Fgf8 (bottom, 36 somites) transcripts.

To probe these interactions further, we inactivated one Shh allele in Grem1-deficient embryos with reduced Bmp4 gene dosage. In an otherwise wild-type context, inactivation of one Shh allele was completely compensated (Fig. 3A). In contrast, heterozygosity for Shh caused loss of an anterior digit in Grem1Δ/ΔBmp4Δ/hf limb buds, as a likely consequence of further reducing SHH signal transduction (Fig. 3B) (n = 12/12). Posterior digit identities were lost in ShhΔ/+Grem1Δ/ΔBmp4Δ/+ embryos in concert with even lower SHH and increased BMP signal transduction (Fig. 3C) (n = 6/8). These results suggest that BMP and SHH activities are opposing one another and establish that intact epithelial-mesenchymal (e-m) feedback signaling buffers heterozygosity for Shh (Fig. 3A), whereas its disruption causes sensitivity to the Shh gene dosage (Fig. 3, B and C).

Fig. 3.

Heterozygosity for Shh is no longer compensated in Grem1Δ/ΔBmp4Δ/hf andGrem1Δ/ΔBmp4Δ/+limb buds (ShhΔ/+G1Δ/ΔBmp4Δ/hf and ShhΔ/+G1Δ/ΔBmp4Δ/+). Left and middle panels, forelimb skeletons at E14.5; right panels, QPCR quantification of the Shh, Gli1, and Msx2 transcript levels; bars represent the average of nine limb bud pairs (34 to 38 somites) with SDs. (A) Heterozygosity for Shh (ShhΔ/+) does not affect digit specification or Gli1 and Msx2 transcription when e-m feedback signaling is intact. (B and C) Disruption of e-m feedback signaling renders digit specification and SHH and BMP signal transduction (assessed by Gli1 and Msx2 transcript levels) sensitive to the Shh gene dosage. The genotypes are indicated for Shh (S), Grem1 (G1), and Bmp4 (B4) as follows: G1Δ/ΔB4Δ/hf, Grem1Δ/ΔBmp4Δ/hf; SΔ/+G1Δ/ΔB4Δ/hf, ShhΔ/+Grem1Δ/ΔBmp4Δ/hf; G1Δ/ΔB4Δ/+, Grem1Δ/ΔBmp4Δ/+; SΔ/+G1Δ/ΔB4Δ/+, ShhΔ/+Grem1Δ/ΔBmp4Δ/+. Shh transcription is reduced about 30% when one Shh allele is inactivated (ShhΔ/+; SΔ/+G1Δ/ΔB4Δ/hf; SΔ/+G1Δ/ΔB4Δ/+). Error bars, mean ± SD. ***, P < 0.001; **, P < 0.01.

Similarly, reducing the Bmp4 gene dosage alone did not alter limb skeletal patterning (fig. S6). However, Grem1 expression was reduced in Bmp4Δ/hf limb buds (Fig. 4A, B), which buffered BMP signal transduction such that Msx2 was only slightly affected and Shh remained normal (fig. S6). It is known that BMPs up-regulate the expression of their antagonist Grem1 (17, 18), but the functional relevance of this self-regulatory interaction remained unclear. To determine the kinetics by which BMP4 and SHH coregulate Grem1 expression (5, 6, 18), carrier beads soaked with recombinant ligands were implanted into cultured mouse limb buds. The initial response to BMP4 was detected within 1 hour (fig. S7) and Grem1 was up-regulated within 2 hours (Fig. 4C and fig. S7), whereas SHH required about 6 hours (Fig. 4D). In turn, GREM1 required minimally 6 hours to up-regulate AER-Fgfs (fig. S7). Thus, the SHH/GREM1/FGF feedback loop (57) operates with a loop time of about 12 hours, whereas the BMP4/GREM1 feedback module is about 6 times as fast (Fig. 4E). The temporal dynamics of these dual-time feedback loops were simulated using an ordinary differential equation model (Fig. 4, F to H) (15, 19) to probe the underlying network properties. These simulations indicated that BMP4, which functions upstream of Grem1 and Shh (fig. S8), initiates Grem1 expression around embryonic day (E) 9.0 (Fig. 4F). This increase in GREM1 rapidly lowered BMP4 activity, which in turn enabled the rise of SHH, GREM1, and AER-FGF activities (i.e., establishment of SHH/GREM1/FGF feedback signaling) in combination with low, but persistent, BMP4 activity (Fig. 4F). In particular, these equations were able to simulate the restoration of SHH activity in compound mutant limb buds (fig. S9). These results point to a switch from BMP4-dependent initiation to SHH-dependent progression of morphogenetic signaling (Fig. 4F). Indeed, simulations without the positive regulation of Grem1 by BMP4 failed at initiation (Fig. 4G), whereas simulations without transcriptional input from SHH still enable up-regulation of morphogenetic signaling (Fig. 4H). This analysis identifies the BMP antagonist GREM1 as the critical node (20) linking the fast BMP4/GREM1 initiator module to the slower SHH/GREM1/FGF feedback loop.

Fig. 4.

GREM1-mediated interlinking of a fast GREM1/BMP4 module with the slower SHH/GREM1/FGF feedback loop. (A and B) Genetic lowering of Bmp4 is compensated by down-regulation of Grem1 expression at E10.75 (37 somites). (C and D) Differential transcriptional regulation of Grem1 by BMP4 (C) and SHH (D). (C) BMP4 induces Grem1 within 2 hours. (D) SHH induces Grem1 within 6 hours. (E) Scheme depicting how the GREM1 transcriptional node (shaded gray) interlinks the BMP4/GREM1 module and SHH/GREM1/FGF feedback loop. (F) Computer-aided simulations of the signaling activities in wild-type limb buds. The starting BMP4 activity was defined as high (Hi), whereas GREM1, SHH, and FGF are low (Lo). The initial increase of GREM1 is controlled by BMP4, whereas the second phase is controlled by SHH. Steady-state levels are reached by E10.0 and are characterized by high SHH/GREM1/FGF feedback loop and low BMP4/GREM1 module activities. (G) Disruption of BMP4-mediated up-regulation of Grem1. (H) Disruption of SHH-mediated up-regulation of Grem1.

We tested this predicted early requirement for BMP4 by Prx1-Cre mediated inactivation of the hypomorphic Bmp4hf allele in Bmp4Δ/hf embryos. This resulted in specific loss of forelimb mesenchymal Bmp4 expression by about E9.0 (Fig. 5A and fig. S10) (21) and severe truncation of the forelimb skeleton (Fig. 5B). This early loss of Bmp4 disrupted Grem1 (Fig. 5C), Shh activation (figs. S10 and S11), and formation of a functional AER as revealed by aberrant or lack of Fgf8 expression (Fig. 5D). The predicted transient requirement of BMP4 was studied by inactivation from specific time points onward using a tamoxifen-inducible Cre transgene (22). Inactivation from E8.75 onward disrupted AER formation (Fig. 5E), whereas inactivation from E9.25 no longer impaired AER formation and Shh expression (Fig. 5F and fig. S10). However, Fgf8 expression was expanded in these forelimb buds (Fig. 5F), indicating that the persistent low activity of the BMP4/GREM1 module is required to restrict AER length. This is relevant because anterior expansion of AER-FGF signaling causes digit polydactylies (12). Thus, BMP4 functions in the mesenchyme to initiate Grem1 expression and signals to the ectoderm to regulate formation and length of the AER. The mesenchymal BMP4 signal is likely transduced in the AER by BMP receptor 1A (BMPR1A) because early AER-specific inactivation results in similar disruption of AER formation, whereas later inactivation alters AER-Fgf signaling (23, 24). This early requirement of BMP4 (Fig. 5) may have been overlooked in previous studies because Bmp4 was not inactivated early enough (15). In the limb bud mesenchyme, the BMP4 signal is likely also transduced by BMPR1A because its mesenchyme-specific inactivation disrupts Grem1 up-regulation and distal limb development (17).

Fig. 5.

BMP4 is required during initiation of limb bud development. (A) Detection of Bmp4 transcripts in the mesenchyme (left) and AER (right) of Bmp4Δ/Δc forelimb buds (26 somites). (B) Skeletal phenotypes of Bmp4Δ/Δc forelimbs at E14.5 after inactivation of the Bmp4hf allele by the Prx1-Cre transgene. (Left) Truncation distal to scapula (n = 8/14). (Right) Loss of posterior elements (n = 6/14). h, humerus; sc, scapula; r, radius; asterisks indicate rudimentary digits. (C and D) Detection of mesenchymal Grem1 (28 somites, n = 6) and AER-Fgf8 (26 somites) transcripts at E9.75. Note the patchy Fgf8 expression (n = 8/10) or loss of Fgf8 expression (n = 2/10). (E) Tamoxifen (TM)–Cre mediated inactivation of Bmp4 by tamoxifen injection at about E8.75, i.e., before initiation of forelimb bud development. Detection of Fgf8 transcripts in forelimb buds at E10.25 (32 somites, n = 6). Control, TM-Cre heterozygous embryo. (F) Detection of Shh and Fgf8 transcripts in forelimb buds at E11.75 (50 somites) after tamoxifen injection at E9.25, i.e., after initiation of forelimb bud development (n = 4). Arrowhead points to the anterior expansion of Fgf8 expression in Bmp4Δ/Δc forelimb buds. Control: Bmp4Δc/+TM-Cre forelimb bud.

Our analysis provides evidence that linking the fast and self-regulatory BMP4/GREM1 initiator module to the slower SHH/GREM1/FGF feedback loop constitutes a crucial component of the limb patterning system. This system controls fail-safe specification of digit identities by coordinating the opposing SHH and BMP4 activities. Although genetic analysis in the mouse has not revealed major roles for BMP2 and BMP7 in the limb patterning system described here, they probably contribute to its robustness and might be more relevant for limb development in other vertebrate species (18, 25). Another fascinating aspect of the SHH/GREM1/FGF feedback loop concerns its self-terminating properties, because the expanding population of Shh descendants is refractory to Grem1 expression, which eventually disrupts e-m feedback signaling and autopod development (18, 26). This refractoriness is caused by activation of an inhibitory FGF/GREM1 feedback loop in Shh descendants (27). Simulations reveal that the limb signaling system progresses in a self-regulatory manner from BMP4-dependent initiation by SHH-dependent digit specification and growth (propagation phase) to FGF-induced termination (fig. S12). This occurs because of the differential impact of BMP4, SHH, and FGF signal transduction on Grem1 expression over time. Finally, our analysis reveals how variation can be compensated by so-called distributed robustness (28) due to interconnectivity between different signaling pathways and not simply by intrapathway compensation. During tetrapod evolution, fine-tuning the signaling interactions described here may have contributed to shaping and stabilizing the pentadactylous autopod (15).

Supporting Online Material

Materials and Methods

SOM Text

Figs. S1 to S13

Tables S1 to S4


References and Notes

View Abstract

Stay Connected to Science

Editor's Blog

Navigate This Article