Fgf9 signaling regulates small intestinal elongation and mesenchymal development

The mammalian small intestine undergoes dramatic developmental changes during the last third of embryogenesis. In mice, this period begins at embryonic day (E) 13.5, when the small intestine is lined on its inner surfaced by a pseudo-stratified, uniformly proliferative endoderm(Kaufman, 1992). From E15.5 to E18.5, a proximal-to-distal wave of cytodifferentiation converts the epithelium to a simple columnar histology, organized into nascent villi with intervillus regions (Schmidt et al.,1988; Calvert and Pothier,1990). A less appreciated developmental change that also occurs during this time period is a remarkable cephalo-caudal lengthening of the organ. Early in embryonic development, the lengthening of the small intestine occurs in proportion to total body length. However, in the latter third of embryogenesis, the rate of intestinal lengthening far outpaces that of the body, becoming exponential in both humans and rodents(Sbarbati, 1979; Weaver et al., 1991). As postnatal development commences, the rate of intestinal lengthening slows with a resumption of a linear relationship between the two.

In humans, the issue of intestinal lengthening becomes clinically important in individuals with short bowel syndrome(Ladefoged et al., 1996). This condition most often occurs when a section of intestine is surgically removed postnatally for a variety of conditions in which a portion of the small intestine is no longer viable (Nightingale and Lennard-Jones, 1993). Following resection, an adaptive response occurs in which the diameter of the small intestine increases in response to loss of surface area. In mouse models of this process, crypts and villi lengthen owing to increased epithelial progenitor proliferation and decreased epithelial turnover (Erwin et al., 2006; Dekaney et al.,2007; Wang et al.,2007). However, overall small intestinal length is never regenerated to any appreciable degree. Because the adaptive response is often inadequate to restore sufficient surface area for absorption, individuals with short bowel syndrome are often treated with total parenteral nutrition, which has a 3-year survival rate of 65-80%(Howard et al., 1995). Thus,understanding the exponential lengthening of the small intestine in the mouse during late-stage embryonic development has potential clinical implications.

However, little is known regarding the developmental pathways that regulate small intestinal lengthening during this crucial period. Most well-studied major developmental pathways play important roles in the proper patterning of the crypt-villus axis or the development of the outer muscle wall during embryogenesis. The formation of villus structures during the late prenatal period is dependent on an intact Pdgf pathway(Karlsson et al., 2000). The canonical Wnt pathway is essential for the regulation of epithelial stem cell homeostasis (Korinek et al.,1998). The Bmp arm of the Tgf-β superfamily regulates crypt census, and thus indirectly impacts stem cell number in the small intestine(Haramis et al., 2004). The ultimate cell fate decisions of committed stem cell daughters are regulated at least in part by the Notch signaling pathway(van Es et al., 2005; Fre et al., 2005; Vooijs et al., 2007). Finally,the outer muscular layer of the intestine depends upon hedgehog signals for its proper formation (Ramalho-Santos et al., 2000).

More recent studies have begun to elucidate how these disparate developmental pathways interact during intestinal embryogenesis. For example,Wnt and Notch perform synergistic roles in the maintenance of proliferation and homeostasis of the epithelial stem cell compartment(van Es et al., 2005). Wnt and Indian hedgehog play antagonistic roles in the determination of epithelial cell differentiation in the colon (van den Brink et al., 2004). Finally, the Pten/Akt family interacts with Wnt/β-catenin signaling to regulate epithelial homeostasis(He et al., 2004; He et al., 2007). Thus, the interaction of multiple developmental pathways plays a crucial role in several aspects of small intestinal development.

Fibroblast growth factors (Fgfs) comprise a large family of polypeptide growth factors that are found in organisms ranging from C. elegans to humans (Ornitz and Itoh,2008). Specifically, Fgf9 plays a key role in embryonic development of the lung, heart, cecum and testes(Colvin et al., 2001a; Colvin et al., 2001b; Lavine et al., 2005; White et al., 2006; Zhang et al., 2006). However,Fgf9 expression is not limited to these organs. We and others have previously shown that this growth factor is also expressed throughout the length of the intestine during late stage embryogenesis(Zhang et al., 2006; Sala et al., 2006). A careful study of the role of Fgf9 in the development of the intestine outside of the cecum has not been undertaken.

Here, by using mice that lack mesenchymal Fgf9 signaling (either through loss of the Fgf9 ligand or its mesenchymal receptors), we show that Fgf9 signals are required for longitudinal growth of the small intestine during late-stage embryogenesis. Fgf9 signaling to the intestinal mesenchyme regulates proliferation of fibroblasts, which in turn appears to drive gut lengthening. Additionally, we found that Fgf9 inhibits the fibroblast to myofibroblast transition in the small intestinal mesenchyme during embryogenesis. Interestingly, this process is regulated by an indirect cellular mechanism that controls the Tgfβ signaling pathway in fibroblasts.

All animal experiments were performed in accordance with approved protocols from the Washington University School of Medicine Animal Studies Committee. Mice used in this study were housed in microisolator cages, in a specified pathogen-free barrier facility, under a strict 12 hour light cycle. Mice were fed a standard irradiated chow diet (PicoLab Rodent Chow 20, Purina Mills) ad libitum. Fgf9^-/-, Fgfr1^fl/+,Fgfr2^fl/+ and Dermo1-Cre were previously generated(Colvin et al., 2001a; Trokovic et al., 2003; White et al., 2006; Yu et al., 2003) and all are on a C57Bl/6 background.

For embryo collections, time-mated females were sacrificed at E14.5-E18.5,1 hour after an intraperitoneal injection of bromodeoxyuridine (BrdU, 120μg/gm, Sigma). The GI tract was isolated from embryos, pinned on black wax,fixed in 2% paraformaldehyde for 4 hours and processed as previously described(Stappenbeck and Gordon,2000). Tissue samples were embedded in paraffin and 5 μm serial sections were prepared. Sections were then stained with Hematoxylin and Eosin. Adjacent, unstained sections were used for immunohistochemical studies (see below).

Paraffin sections were processed for immunohistochemistry using an antibody to BrdU (BD Biosciences). Sections were counterstained with Hematoxylin. Proliferation was measured in three cellular compartments in the proximal and distal small intestines as illustrated in Fig. 2 by scoring the number of BrdU-labeled nuclei to the number of total nuclei in each compartment.

Slide preparations for immunohistochemistry varied depending on primary antibody used. For antibodies directed against α-SMA (clone 1a4;Sigma-Aldrich; 1:2000 dilution), prolyl 4-hydroxylase (clone ER-TR7; BMA Biomeidcal; 1:100), desmin (clone ZC18; Zymed; 1:100), I-Fabp, Fgfr1 (Santa Cruz,; 1:50), Fgfr2 (Santa Cruz; 1:100), Fst (Santa Cruz; 1:10), Fstl1 (Santa Cruz; 1:10), Pecam1 (rat anti-mouse CD31; clone MEC13.3, BD Pharmingen;1:100), p-Smad1/5/8 (Cell Signaling; 1:100) and p-Smad2/3 (Cell Signaling;1:100), embryonic small intestines were embedded in paraffin and sectioned at 5 μm prior to staining. Sections were then deparaffinized in three washes of xylene, and rehydrated in isopropanol. Antigen retrieval was performed by varying methods depending on primary antibody used. For I-Fabp, Fst, Fstl1,p-Smad2/3 and p-Smad1/5/8, slides were heated at 95°C for 20 minutes in 10 mM sodium citrate buffer (pH=6.0). Slides were then washed in double distilled H₂O for 5 minutes prior to washing in blocking buffer (composed of 1% BSA, 0.1% Triton in PBS) for 20 minutes. For primary antibody to Pecam1,antigen retrieval was carried out by a 10-minute incubation with a 1%chymotrypsin solution, followed by a 10 minute wash in double distilled H₂O prior to application of blocking buffer. Antibodies toα-SMA, prolyl 4-hydroxylase, Fgfr2 and desmin required no antigen retrieval.

For antibodies to p-Erk1/2 (Cell Signaling; 1:100), CD29 (BD Biosciences;1:100), CD44 (BD; 1:100), CD45 (BD; 1:100), CD54 (BD; 1:100), CD105 (BD;1:100) and CD106 (BD; 1:100), embryonic small intestines were embedded in OCT compound (Sakura) and immediately frozen in liquified Cytocool (Richard-Allan Scientific). Frozen sections of 5 μm were then fixed for 5 minutes in 4%paraformaldehyde, rinsed in PBS and then immersed in blocking buffer for 20 minutes prior to incubation with primary antibody. Primary antibodies were incubated at 4°C overnight, followed by secondary antibody incubation for 1 hour at 24°C using a 1:500 dilution. Slides were then counterstained with bis-benzimide and coversliped with a 1:1 glycerol/PBS solution. Sections were viewed with a Zeiss Axiovert 200 with Axiocam MRM camera and Apotome optical sectioning slider.

A single wild-type E18.5 small intestine was minced with a razor blade followed by treatment with collagenase (Gibco) in Dulbecco's modified eagle medium (DMEM) containing 10 mM HEPES buffer for 1 hour at 37°C. After passage through a 70 μm filter, the cell suspension was centrifuged for 10 minutes at 400 g. The cell pellet was then suspended in DMEM containing 10% FBS and plated on standard plastic tissue culture plates. After incubation for 2 hours at 37°C, the growth media containing nonadherent cells was removed and discarded. Adherent cells were replenished with fresh media and passaged once weekly.

To differentiate MSCs into adipocytes, isolated cultured embryonic small intestinal MSCs were treated with 10^-8 M dexamethasone and 5μg/ml insulin for 21 days. Cells were then stained with Oil Red O to verify presence of lipid stores. To differentiate MSCs into osteocytes, MSCs were treated with 10^-8 M dexamethasone, 5 μg/ml ascorbic acid 2-phosphate, 10 mM β-glycerophosphate for 21 days. Cells were then stained with Alizarin Red to verify presence of Ca²⁺ deposits in the extracellular matrix.

Cultured cells were treated for 5 minutes with a trypsin/EDTA solution. Cells were pelleted and fixed in 2% formaldehyde in PBS for 10 minutes at 37°C. Cells were then chilled at 4°C for 1 minute, and pelleted by centrifugation at 250 g for 5 minutes. Cells were then permeabilized by 30 minute incubation in 90% methanol at 4°C. Cells were then washed twice with PBS, and then re-suspended in 1% BSA in PBS, containing the primary antibody for 1 hour at 24°C. The cells were then washed twice in PBS before incubation in fluorescently-conjugated secondary antibody(diluted 1:500 in 1% BSA in PBS for 1 hour). Cells were then rinsed in PBS,before finally suspending in 1% BSA in PBS for flow cyotometric analysis. All analysis was performed on a FACSCalibur flow cytometer and analyzed with FlowJo software.

MSCs were serum starved for 2 days in DMEM containing 0.5% FCS. Following starvation, 10 ng/ml Fgf9 (Peprotech) was added to the growth media. Following treatment, cells were lysed in RIPA buffer for 5 minutes on ice and then centrifuged at 14,000 g for 10 minutes. Protein content of the supernatant was quantified using a Bio-Rad D_c colorimetric protein assay (Bio-Rad). Protein (10 μg) was then reduced and loaded into 10%Bis-Tris gels and electrophoresed. Samples were then transferred onto PVDF membranes and blocked in a solution containing 4% BSA (gels and membranes from Invitrogen), 0.1% Tween in TBS for 1 hour at 24°C. Blots were then incubated in primary antibody overnight, washed in TBST and then probed with HRP-conjugated secondary antibody (Bio-Rad) for 1 hour at 24°C before development with the SuperSignal West Pico chemiluminescent kit (Pierce).

Cells were serum starved and treated with Fgf9 as above. Total cellular RNA was collected and purified using a Qiagen RNeasy mini kit. cDNAs were then synthesized using Superscript III reverse transcriptase (Invitrogen) and random primers according to the manufacturer's protocol. qRT-PCR that was performed in triplicate for each sample using SYBR-green master mix(Invitrogen). The following primers were used: 18S(5′-CATTCGAACGTCTGCCCTATC and 5′-CCTGTGCCTTCCTTGGA), Fst(5′-TACTGTGTGACCTGTAATCGG-3′ and 5′-TGATACACTTTCCCTCATAGGC-3′) and Fstl1(5′-CACGGCGAGGAGGAACCTA-3′ and 5′-TCTTGCCATTACTGCCACACA-3′).

A rat myofibroblast line (Mic216)(Plateroti et al., 1998) was co-cultured with MSCs using a transwell system (Corning). MSCs(5.0×10⁴) were plated on the upper chamber, which was separated from the lower chamber by a membrane containing 0.4 μm pores. The lower chamber was seeded with 1.0×10⁴ myofibroblasts. Experiments were performed was carried out in DMEM containing 0.5% FCS. When confluent (∼3 days), myofibroblasts were collected for FACS analysis as described above. Fst was obtained from R&D Systems for treatment of Mic216.

Ten thousand cells were separately procured from the mesenchyme and epithelium of control and DCR1R2 E18.5 embryonic small intestines as previously described (Pull et al.,2005). DNA was isolated by incubating the captured cells with proteinase K in TE buffer at 42°C for 10 hours. DNA was amplified by 40 cycles of PCR using primers as previously described(Yu et al., 2003).

We examined the gastrointestinal tract of wild-type, Fgf9 ^+/- and Fgf9^-/- E18.5 embryos. The length of the small intestine of Fgf9^-/- embryos was significantly shorter than the length of controls at E14.5, E16.5 and E18.5 but not at E12.5 [(Fig. 1A,B)defined as wild-type and Fgf9^+/- littermates]. In this and all other parameters examined in this manuscript, wild-type and Fgf9^+/- embryos were indistinguishable, indicating no significant gene dosage effect occurred. Thus, wild-type and Fgf9^+/- were taken together as controls. No statistically significant differences were noted when comparing the crown-rump lengths of control and Fgf9^-/- embryos at E18.5 (data not shown). In addition, the colon length was not significantly different when comparing control and Fgf9^-/- embryos (average colon length=1.4±0.2 cm for Fgf9^-/- and=1.5±0.3 cm for control). The ratio of small intestinal/colon length was 3.0 in Fgf9^-/- and 4.3 in controls (35% reduction). Fgf9 expression was previously shown to be detectable in the epithelium starting at E12.5 (Zhang et al., 2006). Fgf9 expression persists in a similar distribution during late stage embryogenesis (see Fig. S1A in the supplementary material).

Interestingly, the defect in small intestinal length in the Fgf9^-/- embryos was not accompanied by an alteration in its caliber. We measured caliber at the same relative position along the length of each individual small intestine(Fig. 1C). Thus, the Fgf9^-/- small intestine was not a miniaturized version of controls, but instead was growth altered in only one dimension (length). In addition, more severe developmental defects such as small intestinal atresia,duplication or malrotation were never observed as have been previously noted as frequent malformations in other gene knockouts such as Shh^-/- and Ihh^-/-(Ramalho-Santos et al.,2000).

Histological analysis revealed that Fgf9^-/- small intestines were phenotypically similar to littermate controls. Specifically, Fgf9^-/- small intestines contained no obvious abnormalities of villus morphology such as bifurcation and fusion, or ectopic crypt placement/polyposis (Fig. 1D). The only morphological abnormality that we observed in all E18.5 Fgf9^-/- small intestines was premature crypt-like structures in the proximal small intestine(Fig. 1D). Normally, crypt formation begins after birth and coincides with the appearance of Paneth cells in this structure. However, the premature crypt-like structures in the Fgf9^-/- mice did not express markers for Paneth cells such as lysozyme (data not shown).

Analysis of the intestinal epithelium provides a sensitive readout for interaction of Fgf9 with several major developmental pathways. Proliferative epithelial progenitors (as indicated by the presence of BrdU-positive cells)were present only in the intervillus epithelium of control and Fgf9^-/- mice (Fig. 1E), indicating that the balance of Wnt and Bmp signaling was not significantly perturbed. The proper allocation of secretory and absorptive lineages is controlled by Notch signaling(van Es et al., 2005). The small intestinal epithelium of Fgf9^-/- embryos contained similar proportions of goblet cells, enteroendocrine cells and enterocytes as those seen in their control littermates at E18.5(Fig. 1F), indicating no severe defects were present in this pathway.

However, we found that Fgf9^-/- small intestines showed a subtle defect in enterocytic differentiation at E16.5 and E18.5. These cells expressed lower amounts of intestinal fatty acid binding protein (I-fabp)(Cohn et al., 1992) in the proximal small intestine, as well as lower levels of ileal lipid-binding protein (Ilbp) in the distal small intestine(Fig. 1G; data not shown for E16.5). However, the enterocytes displayed no obvious alterations in polarity,apical junctions or size in either location of the small intestine.

Cell proliferation in the small intestine was quantified by BrdU labeling. We analyzed the three longitudinal layers of this organ (the intervillus epithelium, its underlying mesenchyme, and the muscularis propria) in control and Fgf9^-/- embryos as one or more of these layers could drive intestinal elongation (Fig. 2A). We determined the proliferative index (PI) for each compartment as defined by the number of nuclei incorporating BrdU divided by the total number of nuclei in that compartment. We performed this analysis in both the proximal and distal small intestine at three developmental time points (E14.5, E16.5 and E18.5). The average PI for the intervillus epithelium and for the mucularis propria was not significantly different when comparing control and Fgf9^-/- embryos at any location or timepoint(Fig. 2B,C shows data for the proximal small intestine at all timepoints). However, the PI for the mesenchyme was significantly reduced (P<0.01) in Fgf9^-/- embryos when compared with their littermate controls in both the proximal and distal small intestine at all time points between E14.5 and E18.5 (Fig. 2C). To examine the effects of Fgf9 signaling loss on cell death,TUNEL stained sections of control and Fgf9^-/-small intestines were examined. We observed no significant differences in apoptosis in any cellular compartment at any time point (E14.5-E18.5) during late-stage embryogenesis (data not shown).

The embryonic intestinal mesenchyme contains multiple cell types that may increase proliferation in response to Fgf9. The vast majority of the mesenchyme in E18.5 control small intestines was composed of prolyl 4-hydroxylase-positive, α smooth muscle actin (α-SMA)-negative fibroblasts (Powell et al.,2005). Two other smaller populations included Pecam1-positive endothelial cells and CD45-positive, F4/80-positive monocyte-derived cells. As expected, the monocyte-derived cells did not proliferate to any appreciable levels in the mesenchyme of these embryos (data not shown). In the absence of Fgf9, fibroblasts exhibited a significant decrease in proliferation(P<0.01; Fig. 2D),whereas endothelial cells maintained a proliferative rate similar to controls. These findings suggested that mesenchymal fibroblast proliferation contributed to small intestinal elongation during embryogenesis.

Fgfr1c and Fgfr2c show functional redundancy in the development of several mouse organ systems (Lavine et al.,2005; White et al.,2006; Yu and Ornitz,2008). To examine non-epithelial targets for Fgf9 and potential redundancy among Fgfrs, we conditionally inactivated floxed alleles of both Fgfr1 and Fgfr2 (Pirvola et al., 2002; Herbert et al.,2003; Trokovic et al.,2003; White et al.,2006) using the Dermo1-Cre allele(Yu et al., 2003). To define the targeted lineages in the small intestine, Dermo1-Cremice were mated with ROSA26 reporter mice(Soriano, 1999). At E14.5, the mesenchyme and muscularis propria expressed Dermo1-Cre, as indicated by positive X-Gal staining in these regions(Fig. 3A). All mesenchymal lineages, including fibroblasts, endothelial cells and macrophages contained detectable β-galactosidase whereas all epithelial cells were negative(Fig. 3A,B).

Dermo1-Cre; Fgfr1^+/-: Fgfr2^+/- mice were mated with mice containing two floxed alleles of both Fgfr1 and Fgfr2 to generate mice with the genotype Dermo1-Cre; Fgfr1^loxP/-;Fgfr2^loxP/- (DCR1R2). This breeding scheme also produced Dermo1-Cre Fgfr1^loxP/-Fgfr2^LoxP/+ (DCR1), Dermo1-Cre;Fgfr1^loxP/+; Fgfr2^loxP/- (DCR2) and control mice that lack Dermo1-Cre. We confirmed by immunofluorescence using Fgfr1 and Fgfr2 antisera that DCR1R2 mice contained neither Fgfr1 nor Fgfr2 expression in mesenchymal cells of the small intestine whereas the levels of expression in the epithelium were not perceptibly changed (data not shown). In addition, we performed PCR of laser capture microdissected epithelium and mesenchyme from control and DCR1R2 mice. We could only detect PCR products that indicted the presence of a recombined allele for Fgfr1 and Fgfr2 in the mesenchyme of DCR1R2 mice(Fig. 3C).

Late stage DCR1R2 embryos contained shortened small intestines when compared with littermate controls (P<0.01; Fig. 3D), whereas the crown-rump length was not significantly altered (data not shown). The gross morphology, histological analysis and epithelial differentiation showed defects similar to those described in the Fgf9^-/- mouse(data not shown). DCR1 and DCR2 embryos both showed milder but significant reductions in gut length (P<0.01; Fig. 3D). The effects of loss of Fgfr1 and Fgfr2 were additive as the small intestinal lengths of DCR1R2 mice were significantly different than both DCR1 and DCR2 (P<0.01; Fig. 3D). Proliferation data for the DCR1R2 mice showed that these embryos displayed a similar phenotype to Fgf9^-/-embryos. Only the mesenchyme contained a significant (P<0.01)decrease in proliferation (Fig. 3E). These data suggest that Fgf9 acts primarily through both Fgfr1 and Fgfr2 in the mesenchyme of the developing small intestine to drive mesenchymal proliferation and small intestinal lengthening.

As the small intestine transitions to the postnatal period, the majority of the mesenchymal fibroblasts are replaced by myofibroblasts, which are characterized by elevated expression of α-SMA and their lack of detectable desmin expression (Powell et al., 2005). To examine mesenchymal differentiation, sections of the small intestine of control and Fgf9^-/- embryos were stained for α-SMA and desmin (Fig. 4A-F; see Fig. S1B,C in the supplementary material). Interestingly, at E14.5, E16.5 and E18.5, Fgf9^-/- small intestines displayed a profound increase in α-SMA expression in the mesenchyme when compared with the same region in their littermate controls. These α-SMA-positive cells did not express detectable levels of desmin,consistent with myofibroblast differentiation (see Fig. S1B,C in the supplementary material).

Tgfβ signals are well known to drive fibroblast to myofibroblast transitions in other model systems, such as wound repair in skin(Powell et al., 2005). Smad2 and Smad3 serve as intermediate molecules in the Tgfβ signaling cascade that are phosphorylated in response to active binding of Tgfβ and activin ligands to their receptors, while Bmp signaling phosphorylates and activates Smad1, Smad5 and Smad8 (Waite et al., 2003). To assess the level of Tgfβsignaling in intestinal mesenchyme of control and Fgf9^-/-mice, we examined phospho (p)-Smad2/3 expression using antibodies specific for the active phosphorylated forms. At E14.5, E16.5 and E18.5, both Fgf9^-/- and control embryos showed high level expression of p-Smad2/3 in the nuclei of villus epithelial cells(Fig. 4G-L). At all three of these time points, Fgf9^-/- small intestines showed robust p-Smad2/3 staining in nearly all mesenchymal cells, whereas control mice showed only scattered p-Smad2/3-positive cells. By comparison, absence of Fgf9 did not similarly alter the levels of p-Smad1/5/8 in the mesenchyme (see Fig. S1D,E in the supplementary material). The staining pattern of p-Smad1/5/8 in control embryos corresponded to previously shown patterns in adult small intestines (Haramis et al.,2004).

Fgf9 signaling in the small intestine ultimately targets proliferation,differentiation and Tgf-β signaling of mesenchymal fibroblasts. A commonly used method for examining Fgf signaling activity is to evaluate the phosphorylation state of Erk1/2 (Thisse and Thisse, 2005; Que et al.,2007). If the mechanism of Fgf9 action was exclusively directed towards fibroblasts, then we would predict that p-Erk1/2 stained sections should label these cells. However, stained sections of E18.5 small intestines from control mice showed that only a small subset of mesenchymal cells showed detectable p-Erk1/2 staining (Fig. 5A, denoted as high p-Erk1/2). We found that these high p-Erk1/2 cells were located closer to the epithelium than the muscle wall and were typically clustered at the base of villi. Significantly fewer of these high p-Erk1/2 mesenchymal cells were detected in Fgf9^-/- small intestines (Fig. 5B,C), as well as in DCR1R2 mice (see Fig. 2A in the supplementary material). In control small intestines, the high p-Erk1/2 mesenchymal cells did not express well-characterized markers for either of the less abundant mesenchymal lineages [CD45-positive macrophages and Pecam1-positive endothelial cells(Fig. 5D,E)]. In addition, the p-Erk1/2 high cells were neither BrdU positive(Fig. 5F) nor prolyl 4-hydroxylase positive (data not shown) in control small intestines,suggesting these cells were distinct from the proliferative fibroblasts in the mesenchyme.

The enrichment of the Fgf9 responsive, high p-Erk1/2 mesenchymal cells suggested a mechanism of action whereby Fgf9 indirectly inhibits Tgfβsignaling in mesenchymal fibroblasts. This model would suggest that high p-Erk1/2 cells function as a signaling intermediary that inhibits Tgfβsignaling. Known secreted inhibitors of Tgfβ signaling include follistatin (Fst), a homologue, follistatin-like 1 (Fstl1), gremlin and noggin(Smith, 1999). We screened for mRNA expression of these candidates in the whole small intestines of wild-type mice by RT-PCR. We detected expression of Fst and Fstl (data not shown). Both of these genes were previously identified using expression profiles as preferentially expressed in the mesenchyme (versus the epithelium) of E15.5 small intestines (Li et al.,2007). We then evaluated expression of Fst and Fstl1 in control embryonic small intestines by immunohistochemistry. At E14.5 we found scattered mesenchymal Fst- and Fstl1-positive cells (see, for example, Fig. S2B in the supplementary material). At later times in embryonic development,we observed clusters of both Fst-and Fstl 1-positive mesenchymal cells near the base of villi of control (Fig. 5G,H) but not Fgf9^-/- mice. Double-label immunofluorescence for p-Erk1/2 with either Fst or Fstl1 showed that the subset of stromal cells that expressed high levels of pErk1/2 showed colocalization with both of these proteins(Fig. 5G,H). In addition, Fgfr2 colocalized with both p-Erk and Fst, indicating that epithelial Fgf9 signals through mesenchymal receptors in cells that contain high levels of Fst and Fstl1 (see Fig. S2C,D in the supplementary material).

One mesenchymal cell type that cannot, as yet, be identified on tissue sections with a single marker is the mesenchymal stem cell (MSC). Though these cells were originally considered to function as the source of many different types of mesenchymal lineages, recent work has suggested that they may play a role in signaling, particularly in injury repair(Gnecchi et al., 2006). Most mouse tissues from either adults (da Silva Meirelles et al., 2006; Wang et al., 2006a) or embryos (Lee and Tarantal, 2006) contain MSCs that can be readily isolated by well-established protocols (da Silva Meirelles et al., 2006). Therefore, we isolated MSCs from late stage wild-type embryonic small intestines using these same protocols.

The resulting embryonic small intestinal MSC (iMSC) lines displayed characteristics of previously described MSCs from other tissues. The iMSCs could be maintained in culture for at least 40 passages. The cell-surface marker expression matched previous studies of MSCs from other tissues (iMSCs were positive for CD29, CD44, CD54, CD105 and CD106 and negative for CD45 and Pecam1; Fig. 6A)(da Silva Meirelles et al.,2006; Wang et al.,2006a). Last, the iMSCs differentiated into either adipocytes or osteocytes when cultured under inductive conditions (see Fig. S3A,B in the supplementary material) (da Silva Meirelles et al., 2006).

The iMSCs also displayed similar characteristics as the high p-Erk/Fst/Fstl1 small intestinal mesenchymal cells. When cultured alone in 10%serum, the iMSCs expressed much higher relative amounts of mRNA encoding Fst(ranging from 10-128 fold higher than mRNAs from epithelial, macrophage and myofibroblast cell lines) and Fstl1 (ranging from 47-3500 fold higher than the same set of cell lines; Fig. 6B). Importantly, the iMSCs responded in vitro to Fgf9 stimulation in a time- and dose-dependent manner by increasing p-Erk1/2 expression (see Fig. S3C in the supplementary material; data not shown). We found that Fst was co-expressed with p-Erk1/2 in the iMSCs in vitro (see Fig. S3D in the supplementary material). Furthermore, exogenous Fgf9 increased expression of both Fst and Fstl1 in these iMSCs (Fig. 6C).

To test the hypothesis that the iMSCs could affect myofibroblast differentiation, we co-cultured them with a line of well characterized intestinal myofibroblasts (Mic216)(Plateroti et al., 1998) using a transwell system. We found that the presence of the iMCSs resulted in a de-differentiation of myofibroblasts towards fibroblasts, based on decreased expression of α-SMA in the Mic216 cells(Fig. 6D,E). This finding was partially recapitulated by the incubation of Mic216 cells with Fst(Fig. 6E). Mic216 cells did not significantly alter their αSMA expression in response to Fgf9(Fig. 6E). Last, we performed double-label immunofluorescence on sections of wild-type small intestines with p-Erk1/2 and markers of iMSCs. We found that iMSC markers colocalized with the high p-Erk1/2 mesenchymal cells (Fig,6F,G; data not shown). iMSCs are also targeted by Dermo1-Cre, as shown by the colocalization of β-galactosidase with markers of this cell type (e.g. Fig. 6H). Taken together, these data support the hypothesis that the p-Erk/Fst/Fstl1-expressing small intestinal stromal cells are iMSCs.

In this report, we demonstrated that Fgf9 is essential for small intestinal development during late stage embryogenesis. Loss-of-function models for both the Fgf9 ligand and its mesenchymal receptors showed that epithelial Fgf9 signals to the underlying mesenchyme to regulate elongation of the small intestine. A key cellular target of Fgf9 was ultimately mesenchymal fibroblasts. In the absence of Fgf9 signaling, these fibroblasts showed diminished proliferation, premature differentiation into myofibroblasts and increased Tgf-β signaling, a pathway well known to drive such cellular transitions (Powell et al.,2005). Surprisingly, the response to Fgf9 signaling in the small intestinal mesenchyme did not occur uniformly. Instead the response to Fgf9 signaling occurred robustly in a select subset of mesenchymal cells that also expressed high levels of Tgfβ inhibitors, Fst and Fstl1. We found that these cells most closely resembled iMSCs and that these cells appeared to act as a cellular intermediary in the signaling of Fgf9 to intestinal mesenchymal fibroblasts (Fig. 7).

An important feature of Fgf9 signaling in the small intestine is that it occurs across an epithelial-mesenchymal boundary to regulate elongation of this organ during late-stage embryogenesis. This type of crosstalk is a common mechanism of action for fibroblast growth factors. For example, epithelial to mesenchymal Fgf9 signaling, with reciprocal mesenchymal to epithelial Fgf10 signals, is known to be essential for formation of a cecum(Burns et al., 2004; Zhang et al., 2006). In the lung, epithelial-to-mesenchymal and mesothelial-to-mesenchymal Fgf9 and reciprocal Fgf10 signaling is required for airway branching(Bellusci et al., 1997; Colvin et al., 2001a; White et al., 2006). Finally,it has recently been shown that epithelial-to-mesenchymal Fgf9 is necessary for myocardial growth and proper coronary vascular development(Lavine et al., 2005). The function of Fgf9 in the small intestine is somewhat distinct from these other organs, as no branching of a targeted structure (epithelium or blood vessel)occurs. Thus far, we have not uncovered a functional reciprocal Fgf partner that is produced in the intestinal mesenchyme that signals back to the epithelium. Further studies will be ongoing to determine if this occurs.

Our report suggests that a single molecule (Fgf9) targets the mesenchyme to affect the elongation of the entire organ during development. Importantly,communication between the three functional layers of the intestine is absolutely required, as proper functioning of the organ as a whole requires each layer to be properly oriented to each other. Thus, extensive communication across epithelial-mesenchymal boundaries is necessary for proper intestinal development and adult homeostasis. An unanticipated interaction of Fgf9 was with Tgfβ but not the Bmp signaling pathways. Previous studies have demonstrated that development of the mouse small intestine uses mesenchymal-to-epithelial signaling to control epithelial proliferation and differentiation during development. Knockout mice for the transcription factors Nkx2.3 (Pabst et al.,1999) and Foxl1 (Kaestner et al., 1997) each show diminished expression of Bmps in the intestinal mesenchyme. In turn, loss of Bmp signaling has several effects,including increased epithelial proliferation and dysmorphic villi. In addition, Wang et al. (Wang et al.,2006b) have shown that expression of epimorphin in mesenchymal myofibroblasts is required for proper epithelial maintenance in adult animals,at least in part by mesenchymal modulation of the Bmp pathway. The relationship between Fgf9 signaling and Nkx2.3, Foxl1 and epimorphin is currently unknown but they appear to be distinct pathways as Fgf9 does not appear to affect Bmps based on the pattern of expression of p-Smad1/5/8 and on the absence of epithelial/villus defects in Fgf9^-/- mice. The antagonistic relationship of Fgf9 with Tgfβ that we observed in the developing mouse small intestine is the opposite of the synergistic interaction that has been extensively documented for Fgfs with Tgfβ in other systems, most notably the early development of Xenopus and zebrafish during gastrulation (e.g. Cornell and Kimelman, 1994; Mathieu et al., 2004). Antagonistic interactions of Fgf and Tgfβsignaling have been observed much later in development such as during melanocyte formation and maturation(Stocker et al., 1991).

Interestingly, we show that an important downstream function of Fgf9 is the inhibition of Tgfβ signaling that we hypothesize occurs through an indirect mechanism using iMSCs as cellular intermediaries. It is known that Erk, a downstream target of the Fgf receptor, and other MAP kinases, have the ability to directly modulate the activity of the Tgfβ superfamily through the targeting of Smad1 molecules for degradation by phosphorylation and later ubiquination (Kretzschmar et al.,1997). Interestingly, this ability of Erk to target Smad for degradation has been shown to be a property of Smads associated with the Bmp signaling pathway (Smad1, Smad5 and Smad8), and not the Tgfβ and activin specific Smad2 and Smad3. However, this is still a possible mechanism as both Smad2 and Smad3 have the ability to be ubiquitinated and degraded following prolonged activity (Lo and Massagué, 1999). Here, we propose an indirect mechanism that involves Fgf9 regulation of secreted inhibitors of Tgfβ. The overall effect of Fgf9 on the small intestinal mesenchyme may involve other mechanisms whereby Fgf9 targets all fibroblasts and additionally inhibits Tgfβ signaling in a cell autonomous manner. Such mechanisms need not be mutually exclusive. Indeed, considering the importance of Tgfβsignaling in GI tract development and the multiple levels through which Tgfβ is known to be regulated, such multifaceted control is very likely.

MSCs are most probably present in all adult and fetal organs and share the ability to differentiate into several mesenchymal cell types ex vivo(McTaggert and Atkinson,2007). However, their exact role in development is still unclear. Fgfs are known to modulate the behavior of MSCs, serving to increase proliferation and suppress differentiation of bone marrow-derived MSCs in vitro (Tsutsumi et al., 2001; Farré et al., 2007). We propose that iMSCs play a role in development by regulating the differentiation of adjacent mesenchymal cells (fibroblasts) through expression of secreted Tgfβ inhibitors. The location of the iMSCs near the base of villi may be important for this function as the range of signaling for Tgfβ is very limited (Reilly and Melton, 1996).

This potential function of MSCs is in stark contrast to the traditional role of stem cells, which is to divide only occasionally in order to produce daughter cells that, through repeated divisions via transit amplification,produce committed lineages of mature differentiated cells. However, growing evidence suggests that MSCs have the ability to function outside of this paradigm. For example, it has been recently shown that MSCs have the potential to play a role in disease by functioning as a nidus of paracrine signaling(Gnecchi et al., 2006; Karnoub et al., 2007). Here,we show that this idea of MSCs as paracrine signaling centers can be extended to developmental biology, and that these cells play an important role in gut development by functioning as cellular transceivers. In this role, iMSCs preferentially receive an active Fgf9 signal, and in turn communicate to the surrounding mesenchyme by secreting Tgfβ superfamily inhibitors, thereby preventing premature mesenchymal differentiation. In this manner, they create a permissive environment through which mesenchymal expansion occurs, which we propose contributes to the lengthening of the intestine.

It has been reported that the mesenchymal response to Fgf in other organ systems, particularly the trachea, occurs in a scattered population similar to what we observe in this study (Que et al.,2007). Based on our observations in the intestine, we would speculate that these scattered mesenchymal cells represent mesenchymal stem cells residing in these organs. Thus, the concept of Fgf preferentially signaling to MSC populations might extend to other organ systems. Understanding the regulation of the signaling function of MSCs in response to Fgfs may be informative in our understanding of their normal physiological roles in development and disease.

We thank Dr Deborah Rubin for her helpful comments on the manuscript and Sarah Brown for technical assistance. This work was supported by the Pew Foundation and DK071619 (T.S.S.) and a grant from the March of Dimes Foundation (D.M.O.). M.J.G. was supported by an MSRT training grant T32DK0713034.