Dusp6 (Mkp3) is a negative feedback regulator of FGF-stimulated ERK signaling during mouse development

Fibroblast growth factors (FGFs) comprise a family of small, highly basic proteins that typically are secreted and bind to and activate high-affinity FGF receptors (FGFRs), a family of single-pass transmembrane proteins with intracellular tyrosine kinase activity. FGF signals are transduced to cellular targets by several distinct intracellular signaling pathways. Cells respond to FGF signals in a context-dependent manner; in different instances they may be stimulated to divide, to differentiate, to migrate or to survive. Loss-of-function genetic studies are beginning to reveal the ways in which these FGF-dependent cell behaviors are deployed during the development of many different organs (Itoh and Ornitz,2004; Thisse and Thisse,2005). The responses to FGF signals are dosedependent. Just as a reduction in FGF signals can lead to developmental abnormalities, so too does an increase in FGF signaling. Indeed, some of the most frequently observed human mutations are activating mutations in Fgfr genes that cause a variety of dominant skeletal disorders(Cohen, 2004; Wilkie, 2005).

Many FGF signals travel and are amplified through the extracellular-signal-regulated kinase (ERK) mitogen-activated protein kinase(MAPK) pathway. FGF binding stimulates autophosphorylation of the FGFR, which recruits signal adapters and leads to sequential phosphorylation and activation of cytoplasmic protein kinases, ultimately resulting in the diphosphorylation of ERK on threonine and tyrosine residues. This activated form of ERK has substrates in all cellular compartments, including the nucleus, where it phosphorylates and activates transcription factors,effecting changes in gene expression(Powers et al., 2000; Chen et al., 2001; Tsang and Dawid, 2004; Eswarakumar et al., 2005). Significantly, Corson and colleagues(Corson et al., 2003) showed that most of the diphosphorylated ERK (dpERK, an indicator of ERK pathway activity) in early mouse embryos [embryonic day (E) 6.5-10.5] is dependent on FGFR activity. This suggests that FGFs, as opposed to other signals that also act through receptor tyrosine kinases, are the major input into the ERK pathway at these stages.

Signaling through MAPK pathways can be attenuated at several levels and one class of dual-specificity phosphatases, the MAPK phosphatases (MKPs) inhibit MAPK signaling by dephosphorylating activated MAPKs. For example, in Saccharomyces cerevisiae, mating pheromone-induced signaling through the MAPK Fus3p induces the MKP Msg5p, which feeds back to turn off the signal by inactivating Fus3p (Zhan et al.,1997). Similarly, during Drosophila embryogenesis,signals required for dorsal closure activate the DJNK (basket) MAPK pathway,leading to transcriptional induction of the MKP puckered, which feeds back to inactivate DJNK and dampen the signal(Martin-Blanco et al., 1998). In addition, Gomez and colleagues(Gómez et al., 2005)found that Drosophila MKP3 functions as a negative feedback regulator of epidermal growth factor receptor-stimulated ERK signaling during wing vein development.

Mammalian genomes contain at least 11 DUSP genes encoding the MKPs(Alonso et al., 2004), several of which have been analyzed biochemically. Some MKPs are relatively non-specific toward different dpMAPKs in vitro. Other MKPs, however, show substrate specificity, including the structurally related cytosolic MKPs -3(DUSP6), -X (DUSP7) and -4 (DUSP9), which inactivate dpERK in preference to other activated MAPKs (Camps et al.,2000; Keyse, 2000; Theodosiou and Ashworth,2002). Dusp6 (also known as Mkp3 or Pyst1) transcripts can, in some circumstances, be induced in cultured cells by growth factors that stimulate the ERK pathway(Mourey et al., 1996; Muda et al., 1996a), but clues to physiologic inducers have come from embryonic expression analyses. We and others noticed that Dusp6/Mkp3 is expressed during embryonic development of several vertebrate species in a pattern that corresponds with areas of active FGF signaling, suggesting that it could be a conserved transcriptional target of FGF signals(Dickinson et al., 2002a; Klock and Herrmann, 2002; Eblaghie et al., 2003; Kawakami et al., 2003; Tsang et al., 2004; Gómez et al., 2005)(C.L. and S.L.M., unpublished). Indeed, ectopic FGF signals activate Dusp6/Mkp3 transcription in chick, zebrafish and frog embryos, as well as in explanted mouse neural tube cultures; ectopic Dusp6/Mkp3expression reduces dpERK levels, and local siRNA, global morpholino-mediated knockdown, or dominant-negative experiments suggest roles for Dusp6/Mkp3 in chick limb development, axial patterning of zebrafish embryos and anterior development of frog embryos, respectively(Eblaghie et al., 2003; Kawakami et al., 2003; Tsang et al., 2004; Echevarria et al., 2005; Gómez et al., 2005; Smith et al., 2005). Taken together, these results show that Dusp6/Mkp3 can be activated by FGF signaling and suggest a negative feedback role for DUSP6/MKP3 in FGF/ERK signaling, but genetic loss-of-function data are lacking.

To address the hypothesis that mouse DUSP6/MKP3 plays a role in FGF-stimulated ERK signaling analogous to the MKPs that play clearly established negative feedback roles in regulating invertebrate MAPK signaling pathways, we studied Dusp6/Mkp3 expression in FGFR-deficient mouse embryos and generated and analyzed a targeted loss-of-function Dusp6/Mkp3 allele. Our data show that in mouse embryos, Dusp6/Mkp3 transcription depends on FGF signaling and that dpERK is a physiologic DUSP6/MKP3 substrate. Furthermore, we find that loss of Dusp6/Mkp3 leads to dominant, incompletely penetrant and variably expressive phenotypes that have features, including skeletal dwarfism, coronal craniosynostosis and hearing loss, in common with dominant gain-of-function mutations in human and mouse FGFRs.

Dusp6 is the official symbol for the mouse gene encoding MKP3(MGI:1914853), and this nomenclature is used from here on.

Whole mouse embryos isolated from timed pregnancies were hybridized with digoxigenin-labeled anti-sense RNA probes, which were detected according to Henrique et al. (Henrique et al.,1995). Stained embryos were cryoprotected and sectioned at 14μm as described (Wright and Mansour,2003). The Dusp6 probe was complementary to nucleotides 1-413 of the Dusp6 reference sequence (5′ UTR; GenBank Accession number NM_02628). The Erm probe was complementary to nucleotides 1309-1766 of the Etv5 reference sequence (GenBank Accession number NM_023794). The Fgfr1 probe was complementary to nucleotides 1-459 of the Fgfr1 cDNA (5′ UTR; GenBank Accession number BC010200). The Fgf8, Fgf10, Fgfr2b and Fgfr2cisoform-specific and Fgfr4 probes were generated as described(Wright et al., 2003; Wright and Mansour, 2003; Ladher et al., 2005). Whole mouse embryos were stained with antibodies directed against diphosphorylated ERK1/2 exactly as described by Corson et al.(Corson et al., 2003).

A Dusp6-containing lambda genomic phage was isolated from a strain 129SV/J library (Stratagene). Standard techniques were used to isolate a 6.9 kb fragment containing the entire gene and transform it into a LacZknock-in gene targeting vector (see Fig. 3A). An nlsLacZSV40pA cassette was excised from a gene-trap vector(Yang et al., 1997), excluding the splice acceptor sequence, and inserted into the Dusp6 MscI site in exon 1, fusing the first 56 codons of the DUSP6 open reading frame (ORF)in-frame with the βgal ORF. A self-excising Neo^rexpression cassette (ACN) (Bunting et al.,1999) was placed immediately downstream of the LacZ gene for positive selection of targeted cell lines. In anticipation of the small possibility of spurious transcription initiation from inside the LacZcassette, followed by translation initiation from an internal Dusp6AUG and/or the remote possibility of translation initiating downstream of the DUSP6/βGAL ORF on an SV40pA read-through transcript, we added a linker containing a stop codon in the DUSP6 frame into the BstBI site in exon 3. Any hypothetical DUSP6 fragment produced from such transcripts would be missing portions of both the amino- and carboxy-terminal domains that are essential for DUSP6 activity (Camps et al.,1998; Zhou et al.,2001). The targeting construct was flanked by two different thymidine kinase expression cassettes for negative selection. The targeting DNA was linearized and electroporated into R1-45 embryonic stem (ES) cells. After selection in 380 μg/ml G-418 by weight (Invitrogen) and 2 mmol/l ganciclovir (Sigma), drug-resistant colonies were cloned and expanded. DNA isolated from each colony was screened by Southern blot hybridization using NdeI digestion and a 3′ flanking probe. DNAs showing the 7.0 kb targeted band were further screened by PCR with primers 376(5′-GGTATCAGCCGCTCTGTCAC-3′) and 331(5′-GGACACGGTTGTCACAAGG-3′) for the presence of the stop codon in exon 3. The wild-type allele was 187 bp and the mutant (stop-containing)allele was 198 bp. DNAs from targeted cell lines that carried the stop codon were further analyzed by Southern blotting using a variety of enzymes and probes. Of 192 drug-resistant cell lines tested, 12 had a targeted insertion in the Dusp6 locus and of these, eight also carried the stop codon.

All work with mice complied with protocols approved by the University of Utah Institutional Animal Care and Use Committee. Several correctly targeted cell lines were aggregated with C57Bl/6 morulae, cultured overnight and implanted into pseudopregnant females(Khillan and Bao, 1997). Three of these cell lines gave rise to chimeric males that transmitted the targeted allele (less the self-excising Neo cassette) to offspring. In subsequent crosses, the mutant and wild-type alleles were distinguished in tail or yolk sac DNA using either of two 3-primer PCR mixes. The mix containing primers 344 (5′-CTGTGTCGCTTTCCCTAACC-3′), 309(5′-ACGCTGCTGCTGTTGC-3′) and 315(5′-GACCGACTCTCCACCAGTGT-3′) produced a wild-type band of 499 bp and a mutant band of 363 bp. The mix containing primers 344 (as above), 357(5′-CCAGGGTTTTCCCAGTCA-3′) and 430(5′-CTAGCTCCCCTAAGCGCAAT-3′) produced a wild-type band of 661 bp and a mutant band of 522 bp. No difference between intercross offspring produced from different targeted cell lines was immediately apparent; so one line was selected for all further analysis and backcrossing.

Mice carrying the Fgfr1α (Fgfr1^tm1Cxd,MGI:2153353) and Fgfr2ΔIgIII(Fgfr2^tm1Cxd, MGI:2153790) hypomorphic alleles were generously provided by Dr Chuxia Deng and genotyped as described(Xu et al., 1998; Xu et al., 1999).

Total RNA was isolated from E11.5 embryos or adult brains, mRNA was purified and 3 μg of each genotype was analyzed by northern blot hybridization according to standard protocols(Yang et al., 2001). The 448 bp 3′ UTR probe was generated by PCR amplification of mouse genomic DNA with primers 311 (5′-ACCCCTTGAGACACTGTAAGC-3′) and 329(5′-GGGTATAGTGGAGCCAAAGAGA-3′). The 5′ UTR probe was the same as that used for in situ hybridization. The LacZ probe was generated by PCR amplification of plasmid DNA with primers 24(5′-GGGTTGTTACTCGCTCACA-3′) and 25(5′-AAAGCGAGTGGCAACATGG-3′).

To visualize the skeleton, animals were asphyxiated in CO₂,skinned, fixed in 95% ethanol, defatted in acetone and then stained with Alcian Blue 8GX and Alizarin Red S and cleared as described(Mansour et al., 1993). MicroCT scans of pups sacrificed at P10 were performed as described(Keller et al., 2004). Cranial vault measurements were made using reconstructed mid-sagittal sections. For bone histology, the proximal tibial metaphyses were stained en bloc with Villanueva bone stain, dehydrated in graded concentrations of alcohol,defatted in acetone and embedded in methyl methacrylate monomer. Longitudinal frontal sections of the tibia were cut at 4 μm. One set of sections was stained with 5% silver nitrate and counterstained with 0.5% basic fuchsin,another set of sections was stained with 0.1% Toluidine Blue O(Jee et al., 1997).

Mice were anesthetized using 0.02 ml/g Avertin. Auditory brainstem response(ABR) thresholds for click (47 μs), 8 kHz, 16 kHz and 32 kHz tone pip stimuli (3000 μs, Exact Blackman envelope) were determined according to Zheng et al. (Zheng et al.,1999), using high frequency transducers controlled and analyzed by SmartEP software (Intelligent Hearing Systems).

It has been noted previously that in mouse, chick and zebrafish embryos, Dusp6 (also known as Mkp3) is expressed in cells adjacent to those expressing Fgfs, particularly Fgf8, suggesting that transcription of Dusp6 is regulated by FGF signals(Dickinson et al., 2002a; Klock and Herrmann, 2002; Eblaghie et al., 2003; Kawakami et al., 2003; Tsang et al., 2004). Areas of the embryo potentially competent to respond to FGF signals can be recognized by their expression of FGF receptor genes (Fgfrs). If Dusp6is a transcriptional target of FGF signaling, then it should be coexpressed with Fgfrs. We found that between E9.5 and 11.5, Dusp6expression was highly correlated with several areas of Fgfrexpression (Fig. 1). For example, Fgfr1 and Dusp6 were expressed in the branchial arches and limb buds at E9.5 (Fig. 1A,B), Fgfr4 and Dusp6 were expressed in the somites at E10.5 (Fig. 1D,E)and Fgfr2b and Dusp6 were expressed in the branchial arches and otic vesicle at E11.5 (Fig. 1G,H).

One of the intracellular pathways by which activated FGF receptors signal to the nucleus, is the ERK pathway, activation of which can be visualized by immunostaining for dpERK. Indeed, most expression domains of dpERK in E7.5-10.5 mouse embryos depend on FGF signaling(Corson et al., 2003). If Dusp6 is a transcriptional target of FGF signaling through the ERK pathway, then it should be coexpressed with dpERK. We found that there was a striking correlation between the dpERK and Dusp6 expression domains between E9.5 and 11.5 (Fig. 1). For example, dpERK and Dusp6 were found in the limb buds, branchial arches and midhindbrain boundary at E9.5(Fig. 1B,C), in the somites,limb buds and branchial arches at E10.5(Fig. 1E,F) and in the limb buds, branchial arches, otic vesicles and olfactory pits at E11.5(Fig. 1H,I).

Sections taken through the developing limb buds(Fig. 1J-O) and branchial arches (Fig. 1P-S) at E10-10.5 illustrate potential FGF signaling pathways in more detail. In the limb, the genes encoding the ligands FGF8 (Fig. 1J) and FGF10 (Fig. 1M) were expressed in the apical ectodermal ridge (aer) and mesenchyme (mes), respectively. FGF8 and FGF10 could signal to their preferred receptors, mesenchymal FGFR2c (Fig. 1K) and ectodermal FGFR2b (Fig. 1N), respectively. Dusp6 and dpERK were primarily mesenchymal (Fig. 1L,O),suggesting that they may participate in the mesenchymal response to FGF signals. Similar FGF signaling pathways were present in the branchial arches. FGF8 and FGF10 were ectodermal and mesodermal, respectively (data not shown),whereas genes encoding their receptors, FGFR2c(Fig. 1Q) and FGFR2b(Fig. 1P), were expressed in the mesoderm and ectoderm, respectively. As in the limb bud, branchial arch Dusp6 and dpERK were primarily mesenchymal(Fig. 1R,S). Taken together,these data show that Dusp6 expression correlates with Fgfrexpression domains and with sites of FGF-activated ERK signaling in the mesenchyme.

Ectopic activation of FGF signaling in zebrafish and chick embryos or mouse neural tube explants is sufficient to induce ectopic Dusp6 expression(Eblaghie et al., 2003; Kawakami et al., 2003; Tsang et al., 2004; Echevarria et al., 2005; Smith et al., 2005). Furthermore, genetic or physical ablation of FGF4 and FGF8 signals from the aer in mouse, chick and zebrafish embryos(Eblaghie et al., 2003; Kawakami et al., 2003),application of the FGF receptor inhibitor SU5402 to developing chick somites or mouse neural tube explants (Smith et al., 2005; Vieira and Martinez, 2005), or expression of dominant-negative FGF receptors in zebrafish embryos (Tsang et al.,2004) reduces endogenous Dusp6 expression in the target tissue. To determine whether FGF signaling is required globally for Dusp6 transcription in mouse embryos, we assayed Dusp6expression by in situ hybridization of whole mouse embryos homozygous for hypomorphic mutations in either Fgfr1 or Fgfr2(Xu et al., 1998; Xu et al., 1999). Dusp6 transcripts were severely reduced in both E9.5 Fgfr1α (Fig. 2A,B) and E8.5 Fgfr2ΔIgIII(Fig. 2C-H) homozygous embryos relative to similarly staged control embryos stained under identical conditions. These data show that Dusp6 expression requires signaling through FGFR1 or FGFR2.

If Dusp6 is a transcriptional target of FGF-activated signaling through the ERK pathway and its biochemical function is to inactivate dpERK,then ablation of Dusp6 should lead to an increase in FGF-activated ERK signaling. Indeed, Dusp6 morpholinoinjected zebrafish embryos exhibit axial patterning phenotypes similar to those induced by ectopic Fgf8 expression (Tsang et al.,2004). To determine the role of mouse Dusp6, its function was disrupted using standard gene targeting techniques to insert an inframe LacZ cassette into the first coding exon and a stop codon in the third exon (Fig. 3A). Correctly targeted ES cell lines were identified by Southern blot hybridization analysis of NdeI-digested DNA using 3′ flanking and internal Cre probes (Fig. 3B). Additional digestions and probes confirmed the expected structure of the targeted allele (data not shown). Correctly targeted cell lines were further screened for the presence of the stop codon by PCR using primers flanking the newly introduced stop codon (Fig. 3C). During germline transmission of the correctly targeted, stop codoncontaining allele, the Neo gene was automatically deleted and this was confirmed by Southern hybridization analysis (data not shown). The LacZ knock-in allele was designated Dusp6^L. A 3-primer PCR assay was developed to distinguish the wild type (+) and mutant(-) alleles (Fig. 3D).

To determine the efficacy of the disruption strategy, we hybridized a northern blot of E11.5 mRNAs with a Dusp6 exon 3 3′ UTR probe(Fig. 3E). The expected ∼3 kb Dusp6 transcript and an uncharacterized, weakly expressed ∼4 kb transcript were evident in wild-type and heterozygous samples, but were absent from homozygous mutant mRNA, showing that the targeting strategy effectively disrupted production of Dusp6 mRNA. A novel, lowabundance transcript of ∼7.5 kb was apparent in heterozygous and mutant samples. This is the expected size for mutant transcripts initiated at the Dusp6 promoter and terminated at the Dusp6, rather than the SV40 polyadenylation signal, followed by removal of the Dusp6 introns(see Fig. 4G,I for further hybridization data supporting this interpretation). Even if translation of this ∼7.5 kb fusion mRNA could initiate at a Dusp6 AUG codon downstream of the DUSP6/βgal ORF, a functional DUSP6 protein could not be produced from the mutant allele, as portions of both the amino- and carboxy-terminal regulatory domains required for DUSP6 activity would be absent (Camps et al., 1998; Zhou et al., 2001). Thus, the targeted allele is likely to be null.

One objective of the LacZ knock-in strategy was to enable a simple reporter assay for Dusp6 expression. Surprisingly, we found that heterozygous embryos exhibited little or no βgal activity, whereas homozygous E10.5-16.5 embryos produced readily detectable βgal in a pattern that largely, though not perfectly, mimicked that of transcripts detected by whole-mount in situ hybridization in embryos of both genotypes by using either Dusp6 5′ UTR or LacZ probes (data not shown, but see Fig. 4G for northern data). This apparently post-transcriptional phenomenon is under investigation.

To assess the role of Dusp6 in embryonic development, the F1 Dusp6^+/L offspring of germline chimeras were intercrossed and embryos were collected and genotyped between E8.5 and 17.5. Of 1160 intercross offspring, 244 were wild type, 459 were heterozygous, 226 were homozygous mutant and 11 genotypes could not be determined. These results are consistent with a normal Mendelian distribution of wild-type and mutant alleles. No obvious abnormalities were seen in embryos of any genotype,indicating that Dusp6 does not have a unique and visibly evident function during embryonic mouse development.

DUSP6 has a high degree of substrate specificity in vitro,dephosphorylating dpERK in preference to other MAP kinases(Groom et al., 1996; Muda et al., 1996b). To determine whether ERK signaling was perturbed in homozygous mutants, we stained whole embryos at E9.5 and 10.5 with antibodies directed against dpERK in parallel with wild-type littermate control embryos. In four out of eight homozygotes analyzed, the extent of dpERK staining was increased, particularly in the limb (Fig. 4A-C). In the remaining cases, no differences in dpERK staining between wild-type and homozygous mutant embryos could be discerned (data not shown). Thus, as predicted, DUSP6 dephosphorylates dpERK in vivo.

One of the responses to activation of the ERK pathway is an increase in Erm (Etv5 - Mouse Genome Informatics) transcripts(Raible and Brand, 2001; Roehl and Nüsslein-Volhard,2001; Firnberg and Neubüser, 2002; Liu et al., 2003). To determine whether Erm expression was affected in Dusp6 mutants, we subjected E10.5 embryos (nine of each genotype) to RNA in situ hybridization analysis with an Erm probe. After an identical staining reaction, stopped when Erm transcripts were just becoming evident in wild-type embryos(Fig. 4D), three out of nine heterozygous embryos had greater Erm expression(Fig. 4E) and three out of nine homozygotes had even more Erm signal(Fig. 4F). No differences from wild-type controls were apparent in the remaining heterozygous or homozygous embryos. These results confirm that inactivation of Dusp6 can increase signaling through the ERK pathway. Furthermore, the intermediate phenotype seen in heterozygotes suggests that signaling through the ERK pathway to induce Erm is sensitive to the level of Dusp6.

If DUSP6 has a negative feedback role in ERK signaling, transcription from its promoter should be increased in Dusp6 mutants. Northern hybridization analysis of E11.5 mRNA samples using a Dusp6 5′UTR probe located upstream of the LacZ insertion showed a small increase in Dusp6-initiated transcripts present in homozygous mutant mRNA relative to wild-type and heterozygous samples(Fig. 4G; note levels of the Gapdh control hybridization). A second, independent comparison between genotypes showed equivalent levels of Dusp6 5′UTR-containing transcripts (data not shown), presumably because multiple embryos of the same genotype were pooled to prepare the mRNA samples and the phenotype was variable. More strikingly, a similar hybridization analysis of mRNA prepared from individual adult brain samples showed a notable increase in Dusp6 5′ UTR-containing transcripts in homozygous mutant samples relative to wild-type and heterozygous samples (n=2 out of 4, Fig. 4H). Also as expected, in both embryonic and adult sources of mRNA, the LacZ probe detected transcripts from the mutant allele of ∼5 kb and ∼7.5 kb, reflecting production of Dusp6/LacZ fusion transcripts with polyadenylation probably occurring at the SV40 or Dusp6 polyadenylation sites,respectively (Fig. 4G-I). This result shows that Dusp6 transcription is subject to negative autoregulation.

To determine whether Dusp6 is required for any aspect of postnatal development, Dusp6^+/L offspring of chimeric males and C57Bl/6 females were produced and offspring of these F1 Dusp6^+/L intercrosses were genotyped at weaning (∼4 weeks postnatal). The numbers of wild-type and heterozygous F1 progeny of the chimeras did not deviate significantly from those expected(Table 1). By contrast, of 580 F1 intercross offspring genotyped, 168 were wild type, 287 were heterozygotes and only 125 were homozygotes (Table 1), all of which appeared normal. This genotypic distribution deviated slightly, but significantly, from the Mendelian expectation(P=0.04) and suggested reduced viability of homozygotes and possibly even heterozygotes. Increasing the C57Bl/6 contribution to the genetic background by crossing F1 heterozygotes to C57Bl/6 animals caused a significant reduction in the observed numbers of heterozygous offspring relative to wild type. Of 540 F1 × C57Bl/6 offspring, 309 were wild type and only 231 were heterozygous (P=0.00078, Table 1). The genotypic distribution of the F2 Dusp6^+/L intercross offspring was even more significantly distorted than that of the F1 intercross. Among 505 total offspring, 142 were wild type, 269 were heterozygous and 94 were homozygous mutant (P=0.0036, Table 1). The very mixed genetic background of this intercross may explain why the heterozygous lethality seen in the backcross was not obvious. Taken together, these breeding results suggest that the Dusp6mutation is associated with dominant, but incompletely penetrant, postnatal lethality that is exacerbated on the C57Bl/6 background.

Observations of F1 intercross litters before weaning indicated that most litters had pups that appeared smaller than their littermates after about P5. Therefore, we collected and genotyped intercross pups between P5 and P16. Pups were scored as small if their weight was less than 75% of the littermate average. An example of a small P10 heterozygote and its normally sized wild-type littermate is shown in Fig. 5A. Of 115 intercross offspring (predominantly from the F1 intercross, but including some litters from the F2 intercross), 31 were wild type (three small), 60 were heterozygous (four small), 23 were homozygous mutant (ten small) and one could not be genotyped because it was cannibalized after initial observation. This genotypic distribution was not significantly different from the Mendelian expectation, but about 40% of the P5-16 homozygotes were small. Small homozygous or heterozygous animals were not routinely observed prenatally or at weaning, suggesting that pre-weaning lethality of the subset of small heterozygous or homozygous individuals can account for the reduced number of these animals genotyped at weaning.

As small Dusp6 mutants did not appear to have any physical impediments to feeding, such as malocclusion or cleft palate (data not shown),we examined their long bones for disturbances of the growth plate. The proximal tibia of a P14 wild-type animal had typical histology, with well-ordered chondrocytes proceeding through proliferative, hypertrophic and ossification stages (Fig. 5B,D). By contrast, the proximal tibia of a small P14 Dusp6^-/- pup had severely reduced hypertrophic and ossification zones (Fig. 5C). In addition, chondrocytes in the mutant proliferating zone were disorganized;they did not form typical long straight columns of cells(Fig. 5E). This phenotype was observed in all three small mutants examined and is very similar to that of mice bearing activating mutations in FGFR3(Brodie and Deng, 2003).

Mildly activating mutations in FGFR1 and FGFR2 typically cause craniosynostoses (premature fusions of the cranial sutures) in humans and mice, and hand and foot malformations in humans. In addition, the P250R substitution in FGFR3 causes `non-syndromic' craniosynostosis in humans(Zhou et al., 2000; Brodie and Deng, 2003; Chen et al., 2003; Wang et al., 2005; Wilkie, 2005). Therefore, we examined small Dusp6^+/L and Dusp6^L/Lanimals for additional phenotypes associated with activation of FGF signaling. Staining the carcasses of small P5-15 animals to reveal cartilage and bone showed coronal craniosynostosis in small heterozygous and homozygous individuals (Fig. 5G,I), but not in wild-type (Fig. 5F,H) or normally sized Dusp6^+/L or Dusp6^L/Lpups (data not shown). Furthermore, microCT scanning of small P10 Dusp6^+/L and Dusp6^L/L skeletons showed that the skull vault height:length and height:width ratios were significantly larger than those of the corresponding wild-type control ratios, which is consistent with mild craniosynostosis(Table 2).

By contrast, no obvious hand or foot malformations were apparent (data not shown), but the ossification sequence of the phalanges was slightly delayed in small animals. For example, at P0, the primary ossification center of the middle phalanx of the fifth digit of the hand was clearly visible in a wild-type pup but had not yet ossified in its small heterozygous littermate(Fig. 5J,K). In addition, all of the middle phalangeal ossification centers for digits 2-4 could be seen in wild-type feet but were not yet ossified in a small heterozygous littermate(Fig. 5L,M). By P5,ossification in small heterozygotes or homozygotes resembled that of wild-type littermates (data not shown). Thus, reduction or loss of Dusp6 caused at least two phenotypes, skeletal dwarfism and craniosynostosis, which are also consequences of activated FGF signaling, but this mutation was not associated with significant distal limb malformations.

Clinical descriptions of patients with dominant, activating FGFR mutations often include reports of sensorineural and/or conductive deafness(Gorlin, 2004). Therefore, we measured ABR thresholds at about 6 weeks of age in normally sized animals of all three Dusp6 genotypes, but no significant differences between genotypes were observed (data not shown). In addition, five small intercross offspring survived to 21-28 days and could be evaluated for an ABR. Four of the five small animals had ABR thresholds that were significantly elevated in one or both ears relative to wild-type littermates and other similarly aged control mice (Table 3). Examination of inner ear histology from a small heterozygote (I607) with bilaterally increased ABR thresholds did not reveal any obvious abnormalities relative to a normal hearing control littermate, whereas in another small bilaterally affected heterozygote (I1034), the opening of the middle ear cavity, to which the tympanic membrane attaches, was distorted(Fig. 6A,B). That animal had ossicles that appeared normal (data not shown). Similar results were obtained for the left (affected) ear of the unilaterally affected heterozygote, I1008. Another unilaterally affected homozygote (I919) died unattended before any tissues could be recovered for analysis.

Further examination of the small homozygous or heterozygous Dusp6mutant skeletons revealed that, in comparison with wild-type littermates, they had variable abnormalities of the middle ear bones and otic capsule. For example, the only small pup (a heterozygote) observed and collected at P0 lacked the left incus (Fig. 6C,D). A small P5 homozygote lacked cartilage staining in the developing right otic capsule surrounding the semicircular canals(Fig. 6E,F). The right ossicles of the same P5 homozygous mutant ear were missing tissue and/or stained weakly for cartilage (Fig. 6G,H). Finally, there was a large dorsolateral hole in the left otic capsule of a small P10 homozygote (Fig. 6I,J) and the malleus from that ear was misshapen. These data suggest that the hearing loss found in the small surviving Dusp6-deficient pups was likely to be conductive and could be caused by developmental delays and/or malformations of the middle ear components.

To address the hypothesis that DUSP6 functions in vivo as a negative feedback regulator of MAPK signaling, we used gene targeting to disrupt mouse Dusp6. Homozygous mutant embryos showed variably penetrant increased levels of dpERK, the proposed DUSP6 substrate, in the limb bud, and globally increased levels of Erm, a transcriptional target of the ERK pathway,as well as increased transcripts initiated from the Dusp6 promoter itself. Taken together, these molecular data are consistent with a pathway in which signals that activate ERK lead to increased transcription of Dusp6. The resulting DUSP6 protein then feeds back to dampen the signal by inactivating ERK.

One signal that initiates the negative feedback loop is likely to be FGF,because we found that Dusp6 transcription depends on signaling through FGFR1 or FGFR2. Furthermore, we found that abrogation of Dusp6 function in mice leads to variably penetrant and expressive postnatal phenotypes that are similar to some of the features of ectopic FGF ligand expression or dominant activating mutations in FGFRs, which are major inputs into the ERK pathway (Powers et al., 2000; Tsang and Dawid,2004; Eswarakumar et al.,2005). Beginning around P5, affected Dusp6 mutants were small, exhibited coronal craniosynostosis and middle ear and otic capsule malformations, and affected individuals that survived past P21 frequently had uni- or bilateral hearing loss.

Skeletal dwarfism manifesting in the early postnatal period is also characteristic to various extents of mice carrying Fgf2- or Fgf9-expressing transgenes (reviewed by Ornitz and Marie, 2002), and of knock-in mice carrying the Apert syndrome equivalent mutation (FGFR2S252W)(Chen et al., 2003; Wang et al., 2005), and all of the characterized FGFR3 syndromic gain-of-function mutations(Brodie and Deng, 2003). In addition, osteoglophonic dysplasia, which can be caused by any of several activating mutations in FGFR1(White et al., 2005), is characterized by short stature. In general, the more strongly activating receptor mutations lead to the strongest dwarfing phenotypes(Ornitz and Marie, 2002; Chen and Deng, 2005). This is consistent with the idea that FGF signaling limits endochondral bone growth and that inactivating mutations in negative regulators of FGF signaling, such as Spred2 (Bundschu et al.,2005) or Dusp6, lead to increases in FGF signaling and decreased bone growth. The bone histology of small Dusp6 mutants(reduced hypertrophic and ossification zones, disorganized proliferation zone)is more similar to that of mice with Fgfr3 gain-of-function mutations(e.g. Li et al., 1999) than to the Apert knock-in models, which in one case showed a slightly reduced proliferation zone (Chen et al.,2003) and in the other subtle irregularity of the hypertrophic zone (Wang et al., 2005). This suggests that DUSP6 is more likely to regulate signaling downstream of FGFR3 than of FGFR2 in the growth plate. It would be interesting to make similar comparisons of growth plate histology when a mouse model of osteoglophonic dysplasia is produced.

The craniosynostosis seen in Dusp6 mutants is also likely to be FGF-mediated, as similar phenotypes are seen consequent to ectopic Fgf2 expression or retrovirally mediated increases in Fgf3and Fgf4 expression. In addition, many of the models of FGF receptor activation have more severe craniosynostoses, in some instances involving the coronal as well as the interfrontal and sagittal sutures (reviewed by Ornitz and Marie, 2002; Chen and Deng, 2005), than do affected Dusp6 mutants, in which only the coronal suture is affected. Suture formation is regulated by opposing FGF signaling pathways that control the balance of cellular proliferation (through FGFR2) and differentiation(through FGFR1) (Iseki et al.,1999). Thus, it is conceivable that Dusp6 has differential effects on the two pathways, perhaps regulating signaling through FGFR1 rather than through FGFR2 in the developing calvarium.

No information on middle ear or otic capsule morphology or auditory status is available for any of the mouse Fgf or Fgfrgain-of-function mutations, but the common findings of hearing loss and otopathology in Apert, Pfeiffer, Crouzon and Muenke syndrome patients(Gorlin, 2004) and the mouse Dusp6 mutant phenotype suggest that these pathologies may yet be found in the mouse models as well. Apert and Pfeiffer syndromes are characterized by limb malformations (syndactyly and broad first digits,respectively) (Muenke and Wilkie,2000; Wilkie,2005), but no such malformations were apparent in the small Dusp6 mutants. This may not be surprising, as none of the mouse models for these FGFR1 and FGFR2 syndromes have limb findings, potentially reflecting slight differences in the regulation of FGF signaling between mice and humans.

Taken together, the Dusp6 mutant phenotypes we observed do not precisely mimic any particular mouse Fgfr gain-of-function mutation,suggesting the possibility that Dusp6 is downstream of more than one FGFR, but that it does not serve as a negative feedback regulator of all FGF signaling events. Our finding that hypomorphic mutations in either Fgfr1 or Fgfr2 reduce, but do not entirely eliminate, Dusp6 expression at early embryonic stages further supports this idea. Genetic interaction studies between the Dusp6 mutant allele and various Fgf or Fgfr mutations could be used to address this hypothesis and learn which specific FGF signaling events are subject to regulation by DUSP6. Indeed, our preliminary studies suggest that loss of Dusp6 exacerbates the small size and lethality of the Fgfr1^P250R allele. Whether this effect is a result of changes to craniofacial or limb skeletal elements or both is currently under investigation.

The variable penetrance of the embryonic molecular and postnatal morphologic phenotypes resulting from Dusp6 loss, and the absence of discernible morphologic changes in the mutant embryos, suggest that there could be redundant ERK phosphatases that compensate to some extent for Dusp6. These may or may not be part of a negative feedback loop regulating ERK. DUSP7 and DUSP9 are both relatively ERK-specific in vitro(Dowd et al., 1998; Dickinson et al., 2002b), and their transcripts have some areas of overlap with Dusp6, particularly in the developing limb buds and branchial arches (data not shown)(Dickinson et al., 2002b). The Dusp7 mutant phenotype has not yet been reported, but mutation of the X-linked Dusp9 gene leads to failure of placental development and consequent lethality. Tetraploid rescued embryos develop normally, however(Christie et al., 2005). Thus,conditional Dusp9 (and Dusp7) alleles will have to be generated in order to assess potential redundancy with Dusp6.

Finally, the similarity between the Dusp6 mutant phenotypes and those of humans with activating FGFR mutations suggests that mutations in DUSP6 or other negative regulators of FGF signaling, such as SPRY genes, SPRED genes or SEF (IL17RD - Human Gene Nomenclature Database), or indeed other DUSP genes, are good candidates for molecularly unexplained cases of FGFR-like syndromes. Furthermore, the increasing lethality of the Dusp6 mutation as the allele was backcrossed to C57Bl/6 shows that there are likely to be genes that interact with Dusp6 and suggests that genetic variation among negative regulators of FGF signaling is a potential source of modifiers of the variable expressivity of human FGFR mutations.

We are very grateful to Drs Mario Capecchi for providing an aliquot of R1-45 ES cells and the TK1+TK2 vector, Chuxia Deng for providing mice carrying the Fgfr1α and Fgfr2ΔIgIII alleles,Charles Keller for carrying out the microCT scans, Webster Jee for assistance with the bone histology and Lisa Urness for critiquing the manuscript. Xiaofen Wang provided excellent technical assistance with mouse production; Albert Noyes generously provided the dpERK data for Fig. 1C; and Samantha Covington generated the plasmid bearing the Fgfr1 5′ UTR fragment. This work was supported by grants from the NIH/NIDCD: R01DC02043 and R01DC04185,from the Deafness Research Foundation, and from the Children's Health Research Center. D.A.S. was a Primary Children's Medical Foundation Scholar.