Fibronectin mediates mesendodermal cell fate decisions

Cells that contribute to the heart arise from distinct pools of mesodermal progenitors during embryonic development and are typically characterized by expression of specific transcription factors (Srivastava, 2006). Several lineage-tracing studies have demonstrated the origin and location of distinct populations of precardiac mesodermal cells (Buckingham et al., 2005). The specification, fate and expansion of cardiac mesodermal progenitors are controlled by diffusible morphogens, such as Wnt, Nodal, BMP and FGF (Murry and Keller, 2008; Mercola et al., 2011). Such non-cell-autonomous signals from early endoderm have long been implicated in the formation of cardiac progenitors (Schultheiss et al., 1995; Schultheiss and Lassar, 1997; Marvin et al., 2001; Ueno et al., 2007). Furthermore, the function of transcription factors, such as Gata4 and Sox17/18 in the endoderm, is required for appropriate cell-cell signaling (Liu et al., 2007; Holtzinger et al., 2010). However, the precise mechanisms by which endoderm-mesoderm communication occurs leading to the specification of precardiac mesoderm remain unclear.

The ability of embryonic stem (ES) cells to spontaneously differentiate into cardiomyocytes offers an opportunity to study non-cell-autonomous signals in vitro, as co-culturing ES cells with endoderm-like (End2) cells is known to enhance cardiac differentiation (Mummery et al., 2003; Rudy-Reil and Lough, 2004). Conversely, loss of the endodermal HMG-box transcription factor Sox17/18 in ES cell culture suppresses cardiomyogenesis non-cell-autonomously, but the suppression can be rescued by co-culturing ES cells with normal endodermal cells (Liu et al., 2007).

Here we investigated the mechanisms of endoderm-mediated differentiation of the cardiac lineage and found that endodermal cells promote precardiac mesoderm through signals containing a crucial short-range component involving the extracellular matrix protein fibronectin 1 (Fn1). Fn1 in turn activated β-catenin signaling through integrin-β1, and this signaling event biased cells, probably in conjunction with other secreted signals, toward the precardiac mesoderm fate.

Mouse ES (mES) cells were propagated in an undifferentiated state on gelatin-coated cell culture plastic (Nunc) in GMEM supplemented with 10% FBS, 0.1 mM non-essential amino acids, 2 mM GlutaMAX, 0.1 mM sodium pyruvate (Invitrogen), 0.1 mM 2-mercaptoethanol (Sigma-Aldrich) and 1500 U/ml leukemia inhibitory factor (LIF; Millipore). ES cells were passaged every 2-3 days with TrypLE Express (Invitrogen) with daily medium changes.

For co-culture experiments, ES cells and End2 cells were dissociated to single cells and plated on ultra-low attachment plastic surface (Corning) in IMDM/Ham's F12 (Cellgro) (3:1) supplemented with N2, B27, penicillin/streptomycin, 2 mM GlutaMAX, 0.05% BSA, 5 ng/ml l-ascorbic acid (Sigma-Aldrich) and α-monothioglycerol (MTG; Sigma-Aldrich) at a final density of 75,000 cells/ml.

For cardiac mesoderm induction, cells were propagated as described above (75,000 cells/ml) for 48 hours. Embryoid bodies (EBs) were collected and dissociated to single cells. Cells were replated on ultra-low attachment plastic surface as described above and induced for 40 hours with activin A (5 ng/ml), BMP4 (0.5 ng/ml) and VEGF (10 ng/ml) (R&D Systems).

For cardiac differentiation, GFP⁺ cells from co-cultures were plated on gelatin-coated cell culture plastic (Nunc) in StemPro34 (Invitrogen) supplemented with 10 ng/ml l-ascorbic acid, penicillin/streptomycin, 2 mM GlutaMAX, FGF (20 ng/ml), FGF10 (50 ng/ml) and VEGF (10 ng/ml) (all from R&D Systems). Unbiased differentiation was achieved by growing EBs in 20% FBS in GMEM supplemented with 2 mM GlutaMAX and 0.1 mM non-essential amino acids.

ES^Bry-GFP cells were a kind gift from G. Keller (McEwen Centre for Regenerative Medicine, Toronto, Canada). ES^Nkx2.5-GFP cells were generated from the mouse E14 ES cell line carrying RP11-88L12/NKX2-5-Emerald GFP as described (Hsiao et al., 2008). The ES^GFP cell line is a variant of the E14 line that expresses GFP under control of the CMV/β-actin (CAG) promoter. All mES cells used in this study were E14 derivatives. Neural stem cells (NSCs) were obtained from ATCC. End2 cells were a kind gift from C. Mummery (Leiden University Medical Centre, Leiden, The Netherlands).

For TOP/FOP-flash luciferase assays, transfected mES cells were treated with 20 ng/ml Wnt3A, 2 μM BIO and/or anti-CD29 (Itgb1) (BD Biosciences) for 24 hours before analysis. TOP/FOP-flash luciferase constructs were kindly provided by Randall Moon (University of Washington, WA, USA).

Visceral endoderm or extra-embryonic ectoderm (ExE) was isolated by microdissection of 20 mouse embryos at embryonic day (E) 6.5. Endoderm or ExE was dissociated into single cells and co-cultured with ES^Bry-GFP cells in a 1:10 ratio.

RNA was extracted with TRIzol (Invitrogen). Quantitative reverse transcription PCR (qRT-PCR) was performed using the Superscript III first-strand synthesis system (Invitrogen) followed by use of TaqMan probes on the ABI 7900HT real-time PCR system (Applied Biosystems) according to the manufacturer's protocols. Optimized primers from TaqMan gene expression arrays were used. Expression levels were normalized to that of Gapdh. All samples were run at least in triplicate. Real-time PCR data were normalized and standardized with SDS2.2 software (Applied Biosystems).

Cells were dissociated and percentages of GFP⁺ and cTnT⁺ cells were analyzed using a FACSCalibur flow cytometer (BD Biosciences) and FlowJo software (TreeStar). For cTnT expression, cells were incubated with anti-cTnT antibody (Neomarkers) followed by incubation with secondary antibody conjugated with Alexa Fluor 647 (Invitrogen). For FACS, cells were dissociated as described above, resuspended in PBS containing 0.1% FBS, 20 mM Hepes and 1 mM EDTA, and sorted on a FACSAria I or II cell sorter (BD Biosciences). FACS analyses were performed on days indicated in the figures.

For knockdown studies, ON-TARGET Smartpools of fn1 siRNA, collagen Ia1, collagen IVa1, collagen IVa2, Sparc siRNA, scrambled siRNA (control) (Dharmacon) or Block-iT Alexa Fluor Red (Invitrogen) were used at 75 nM. End2 cells were transfected using Lipofectamine RNAiMAX (Invitrogen) 24 hours before co-culture experiments. For antibody inhibition studies, anti-CD29 (anti-integrin-β1, BD Biosciences) or anti-Fn1 (Developmental Studies Hybridoma Bank) was added from the start of the experiment at the concentrations described.

E6.75 embryo cryosections were stained with anti-Fn1, anti-collagen I, anti-collagen IV or anti-Sparc (Abcam), followed by incubation with secondary antibodies conjugated with Alexa Fluor 488 or 546 (Invitrogen).

The ECM RT² Profiler PCR Array (PAMM-013) was obtained from SABiosciences. The following samples were used: (1) ES^CAG-GFP cells; (2) ES^CAG-GFP cells that had been differentiated with End2 cell co-culture for 2.5 days and sorted by FACS; (3) NSCs; (4) End2 cells from condition (2) that were negatively sorted by FACS to remove ES cells; and (5) End2 cells. RNA was extracted with TRIzol (Invitrogen). qRT-PCR was performed using the Superscript III first-strand synthesis system (Invitrogen). The ECM RT² Profiler PCR Array was run on the ABI 7900HT according to the manufacturer's instructions (Applied Biosystems). Data from conditions (4) and (5) were compared with data from conditions (1-3). Results were analyzed and visualized using software provided from the manufacturer with the arrays. ECM gene comparative expression data are given in supplementary material Table S1.

Cell lysate was resolved by SDS-PAGE and electroblotted onto PVDF membranes. The membranes were incubated with primary antibodies in 5% nonfat milk overnight at 4°C, and secondary antibodies for 1 hour at room temperature. Detection was by chemiluminescence (Amersham ECL, GE Healthcare Life Sciences).

Zebrafish (Danio rerio) care and breeding were carried out as described (Westerfield, 1995). For in vivo knockdown studies in zebrafish, previously described morpholino antisense oligonucleotides against the translational start sites of zebrafish fibronectin 1 (MO-fn1, 5′-TTTTTTCACAGGTGCGATTGAACAC-3′) (Trinh and Stainier, 2004) and zebrafish fibronectin 1b (MO-fn1b, 5′-TACTGACTCACGGGTCATTTTCACC-3′) (Jülich et al., 2005) and a control scrambled non-targeting morpholino were injected individually (6-8 ng) or in combination into one-cell stage zebrafish embryos.

Whole-mount in situ hybridization of stage-matched zebrafish embryos was carried out as described (Fish et al., 2008). Embryos were staged before the procedure as described (Kimmel et al., 1995). Zebrafish ntl and gsc expression vectors used as templates for digoxigenin-labeled RNA antisense probe synthesis were kindly provided by D. Stainier (Max Planck Institute for Heart and Lung Research, Bad Nauheim, Germany).

The two-tailed Student's t-test, type II, was used for data analyses. P<0.05 was considered significant.

Since endoderm has an inductive role in cardiogenesis (Schultheiss et al., 1995; Mummery et al., 2003; Rudy-Reil and Lough, 2004), we investigated the range and nature of this inductive signal. Given the proximity of endoderm and nascent mesoderm during early development in vivo (Nijmeijer et al., 2009), we hypothesized that the induction of precardiac mesoderm involves close contact with endoderm. To test this, mouse ES (mES) cells were differentiated by aggregation with or without End2 cells, or in End2 cell-conditioned medium. After 8 days of differentiation, 65% of embryoid bodies (EBs) co-cultured with End2 cells had beating foci, compared with 25% or 20% when co-cultured with mES cells alone or in conditioned medium, respectively (Fig. 1A). Correspondingly, the number of cardiomyocytes, marked by the sarcomeric protein cardiac troponin T (cTnT; Tnnt2 – Mouse Genome Informatics), was fourfold greater in EBs formed with End2 cells, as quantified by fluorescence-activated cell sorting (FACS) (Fig. 1B). Furthermore, expression of the cardiac transcription factor gene Nkx2.5 and the cardiac sarcomeric genes α-cardiac actin (Actc1) and myosin heavy chain 7 (Myh7) was significantly elevated upon co-aggregation with End2 cells (Fig. 1C), compared with mES cells cultured alone or in conditioned medium. Increasing the ratio of End2 cells to mES cells resulted in a dose-dependent increase in expression of Actc1 and of the cardiac transcription factor gene Hand2 (supplementary material Fig. S1A).

To test the relationship between cell-cell contact and mesoderm formation in cells exposed to the same medium, we mixed EBs formed from co-culture of mES cells stably expressing GFP (mES^CAG-GFP) and End2 cells, with EBs formed from unlabeled mES cells without End2 cells. After 8 days of differentiation in a mixed suspension culture, beating EBs were scored for fluorescence to determine the influence of End2 cells. The percentage of GFP⁺ EBs that were beating was fivefold greater than for EBs cultured without End2 cells (supplementary material Fig. S1B). These experiments suggest that an important component of the inductive effect of End2 cells is being mediated by a short-range signal. End2 cells did not proliferate in EB suspension culture (supplementary material Fig. S1C), and treatment of End2 cells with mitomycin C to arrest cell division before co-culture did not affect their cardiogenic potency (data not shown).

One of the earliest lineage markers of mesodermal precursors is brachyury (Bry), a T-box-containing transcription factor expressed in the primitive streak/early mesoderm (Wilkinson et al., 1990; Kispert and Herrmann, 1994; Inman and Downs, 2006). To determine whether End2 cells affect the emergence of Bry⁺ cells, we utilized an ES cell line with GFP knocked into the endogenous Bry locus (mES^Bry-GFP) (Fehling et al., 2003), and aggregated mES^Bry-GFP cells with End2 cells (1:1) to form EBs in suspension. EBs co-aggregated with End2 cells showed an increased percentage of Bry⁺ cells compared with control cells (Fig. 1D,E). In addition, the co-aggregation resulted in increased emergence of Bry⁺ cells in a dose-dependent manner, suggesting an inductive role of the End2 cells at day 3 (Fig. 1E-G). No induction was observed when mES^Bry-GFP cells were co-aggregated with ectodermal cells or other mES cells (Fig. 1H). The Bry-promoting effect appeared to involve an essential short-range component, as neither conditioned medium nor concentrated conditioned medium increased the percentage of Bry⁺ cells (supplementary material Fig. S1D). To determine whether endoderm had a similar effect, mES^Bry-GFP cells were co-aggregated at a 10:1 ratio with visceral endoderm or extra-embryonic ectoderm (ExE) cells isolated by manual dissection from E6.5 mouse embryos. Similar to End2 cells, the endodermal cells were able to induce Bry-GFP⁺ cells from mES^Bry-GFP cells, whereas ExE cells had no effect (Fig. 1I,J).

Bry transiently marks mesendoderm, and Bry⁺ cells can give rise to a variety of different mesendodermal lineages (Hansson et al., 2009). By qPCR, we found that End2 cell-induced Bry-GFP⁺ cells, sorted at day 4 of differentiation, expressed markedly higher levels of Mesp1, the earliest marker for cardiovascular progenitors (Fig. 2A), and of the well-established precardiac mesoderm markers Kdr and Pdgfra (Kitajima et al., 2000; Kattman et al., 2011), compared with Bry-GFP⁺ cells isolated from normal ES cell differentiation (Fig. 2B). By contrast, the endodermal HMG-box transcription factor Sox17 was expressed at significantly lower levels compared with non-induced Bry-GFP⁺ cells (Fig. 2C). To determine the fate of the End2 cell-induced Bry-GFP⁺ cells, FACS-isolated Bry-GFP⁺ cells at day 4 were replated and differentiated for 3 additional days in conditions that favor cardiac differentiation (Kattman et al., 2006). Under these conditions, End2 cell-induced Bry-GFP⁺ cells formed a spontaneously contracting sheet of cardiomyocytes (supplementary material Movie 1) with 50% of the cells staining positive for cTnT (Fig. 2D), compared with less than 5% in the control Bry-GFP⁺ cells, suggesting that the endoderm-induced signal favored precardiac mesoderm.

To determine if End2 cells further improve cardiac differentiation after formation of precardiac mesoderm, we examined whether End2 cells affect the progression of precardiac mesoderm to Nkx2.5⁺ cardiac progenitors. Using an established combination of activin A and BMP4 (Kattman et al., 2006; Yang et al., 2008), we induced precardiac mesoderm (Bry⁺ Mesp1⁺) at day 4 of differentiation in mES^Nkx2.5-GFP cells carrying a GFP reporter in the Nkx2.5 locus (Hsiao et al., 2008), followed by co-culture with End2 cells. Interestingly, co-culture with End2 cells at this later stage inhibited the progression of Bry⁺ Mesp1⁺ precardiac mesoderm cells into Nkx2.5⁺ progenitors and cardiomyocytes (Fig. 2E,F), suggesting that the inductive signal functions during a narrow window and does not act beyond precardiac mesoderm induction.

Our earlier experiments suggested that the mesoderm-promoting signal from endoderm might have an important short-range component. We hypothesized that, in addition to known diffusible proteins (Freund et al., 2008), this effect might be mediated via extracellular matrix (ECM) components secreted by the endoderm. To screen for ECM candidates responsible for the observed effect, we used ECM-specific gene expression arrays to compare End2 cells with mES cells or ectodermal cells. We identified nine ECM genes with transcripts that were highly enriched (>4-fold) in End2 cells (Fig. 3A). Of these genes, Fn1, Sparc, collagen Ia1 (Col1a1), collagen IVa1 (Col4a1) and collagen IVa2 (Col4a2) were temporally expressed around the egg cylinder and gastrulation stages in mice, according to published EST profiles.

We examined the expression of these five genes in E6.5 mouse embryos by immunofluorescent staining of the proteins, and found that they were spatially expressed in the endoderm, but not in the epiblast (Fig. 3B). To determine whether any of the candidate proteins were involved in promoting Bry⁺ mesoderm, we used siRNA to knock down each candidate in End2 cells, and then co-cultured these End2 cells with mES^Bry-GFP cells. Knockdown efficacies were confirmed by qPCR (supplementary material Fig. S2A). Knockdown of Fn1 reduced the formation of Bry-GFP⁺ cells by 50%, whereas knockdown of Sparc, Col1a1, Col4a1 or Col4a2 had no effect (Fig. 3C). Similarly, blocking Fn1 activity by the addition of antibodies against extracellular Fn1 significantly reduced the formation of Bry-GFP⁺ cells (supplementary material Fig. S2B). Likewise, we noted an End2 dose-dependent increase in Fn1 expression in our initial co-culture experiment (supplementary material Fig. S2C). Further analysis of EBs formed in the co-culture showed that Bry-GFP⁺ cells generally emerged adjacent to fibronectin matrices (supplementary material Fig. S2D). These results suggested that Fn1 might be a key short-range signal that promotes mesoderm formation.

Loss of Fn1 has been linked to cardiac as well as generalized mesodermal defects in vivo, but the phenotypes are thought to arise much later and be secondary to structural and migratory defects, such as failure to form cadherin junctions as well as mislocalized atypical protein kinase C (aPKC). To examine whether these structural defects could account for the decrease in Bry⁺ cells, we immunostained for cadherins and aPKC in End2 cells or in End2 cell-containing EBs treated with fn1 siRNA or control siRNA. The presence or absence of Fn1 did not alter the ability of End2 cells to form cadherin junctions or alter aPKC localization (supplementary material Fig. S2E-G) Furthermore, in EBs containing End2 cells with reduced Fn1 levels, we observed no increase in cell death, no decrease in mesoderm proliferation, nor any increase in expression of pluripotency genes (supplementary material Fig. S2H-J). To examine whether knockdown of Fn1 affected any other ECM components, we compared expression of various ECM components in fn1 siRNA-treated cells with control-treated cells using ECM-specific gene expression arrays. Fn1 was the only ECM gene that was significantly downregulated by more than twofold in End2 cells and EBs expressing fn1 siRNAs (supplementary material Fig. S2K), further confirming the specificity of our knockdown and its effects. These data suggest that Fn1 might play a direct signaling role during the emergence of mesoderm.

To investigate whether fibronectin also promotes the formation of Bry⁺ cells in vivo, we knocked down fibronectin in zebrafish embryos by injecting a morpholino-modified antisense oligonucleotide directed against fn1 and fn1b (MO-fn1/fn1b) (Jülich et al., 2005), as there are two fibronectin genes in zebrafish (Trinh and Stainier, 2004; Jülich et al., 2005). As described previously (Jülich et al., 2005), the fibronectin morphants had severe defects in somite boundary formation (data not shown), confirming the efficacy and specificity of the oligonucleotides. The resulting morphants showed no obvious developmental delays or perturbation in gastrulation. However, both the expression level and domain of the zebrafish Bry ortholog no tail (ntl) were markedly decreased at 50% epiboly, a stage equivalent to mid-gastrulation in mouse (Fig. 3D). Expression of goosecoid (gsc), another mesodermal gene, was similarly decreased at late gastrulation (90% epiboly) (Fig. 3E). At this stage, the decrease in the expression domain of ntl was less apparent, suggesting partial late recovery (supplementary material Fig. S2L).

Given that early mesoderm is derived from mesendodermal progenitors (Rodaway and Patient, 2001), we tested whether the loss of fibronectin also affected endodermal cells in vivo. We injected MO-fn1/fn1b into transgenic zebrafish embryos expressing GFP under control of the sox17 promoter to label endodermal cell populations (Sakaguchi et al., 2006). The knockdown of fn1/fn1b resulted in expansion of the sox17-GFP expression domain at mid-gastrulation (50% epiboly) (Fig. 3F). Consistently, the number of GFP⁺ cells was greatly increased in fn1/1b morphants (Fig. 3F,G). Double labeling of cells for ntl revealed the abnormal presence of sox17-GFP⁺ cells within the expected ntl expression domain (Fig. 3I). Six hours later, at the 9-somite stage [~12 hours postfertilization (hpf)], the expanded population of sox17-GFP⁺ cells in the morphants followed a stereotypical endodermal migration and merged at the midline; however, the overabundance of sox17-GFP⁺ cells was less dramatic at this point (Fig. 3F,H). In agreement, numerous endodermal genes were upregulated in mouse EBs formed in Fn1-deficient End2-ES cell co-culture (Fig. 3J; supplementary material Fig. S3A). These results might reflect an early expansion of endoderm at the expense of mesoderm and are consistent with fibronectin playing a role in normal mesodermal cell fate specification.

Loss of fibronectin results in cardiac defects in both mouse and zebrafish, presumably as a consequence of cell migration abnormalities (George et al., 1993; Trinh and Stainier, 2004). To determine whether fibronectin is also involved in an earlier process – the proper emergence of cardiac progenitors – we performed siRNA knockdown of Fn1 in End2 cells and in those co-cultured with mES^Nkx2.5-GFP cells. Depleting Fn1 in End2 cells significantly reduced the number of Nkx2.5-GFP⁺ cardiac progenitors by day 5 of differentiation (Fig. 3K).

We investigated whether the decrease in Nkx2.5-GFP⁺ progenitors was due to a general loss of mesoderm or to a specific loss of precardiac mesoderm. Bry-GFP⁺ cells were induced by End2 cells treated with control or fn1 siRNA, and the percentage of precardiac mesodermal cells, identified by co-expression of the surface markers Kdr and Pdgfrα (Kattman et al., 2011), was analyzed among the resulting Bry-GFP⁺ cell populations by FACS. The number of Kdr⁺ Pdgfrα⁺ cells was reduced by 60% (from 25% to 11% of cells) in the Bry-GFP⁺ cell population exposed to Fn1-deficient End2 cells (Fig. 3L). In agreement with this reduction, the Bry-GFP⁺ cell population isolated from Fn1-deficient End2 cells differentiated poorly into cardiomyocytes (cTnT⁺ cells) under differentiation conditions favoring the cardiac lineage (Fig. 3M,N; see Materials and methods). When sorted Bry-GFP⁺ cells were differentiated in a non-biased manner, we observed a decrease in several mesoderm derivatives, although the cardiac lineage was most adversely affected (supplementary material Fig. S3B).

These data show that decreased levels of endodermal fibronectin result in fewer mesodermal cells and in a disproportionally smaller fraction of precardiac mesoderm. Similarly, in fibronectin-deficient zebrafish embryos, we observed a reduction in cardiac progenitors, as indicated by a decrease in gata5⁺ cardiac primordia at the 10-somite stage, as well as a reduction of the myosin light chain 7 (myl7)⁺ cardiac domain at the 16-somite stage (Fig. 3O,P). These results suggest that fibronectin promotes the emergence of Bry⁺ mesoderm and biases toward the formation of precardiac mesoderm.

Integrins often play a crucial role in outside-in signaling mediated by ECM proteins (Ieda et al., 2009). Since integrin-β1 is known to mediate fibronectin signaling (Vuori and Ruoslahti, 1993), we disrupted the Fn1–integrin-β1 interaction by treating mES cells with integrin-β1-blocking antibodies (Ieda et al., 2009). This treatment efficiently inhibited the ability of End2 cells to induce Bry-GFP⁺ mesoderm, similar to that of anti-Fn1 antibodies (Fig. 4A). Correspondingly, End2 cells were less effective in increasing the number of Bry-GFP⁺ cells in mES cells expressing a dominant-negative form of integrin-β1 (supplementary material Fig. S3C), which affects cell signaling but not cell adhesion or cytoskeletal organization (Ross et al., 1998; Ieda et al., 2009). Combined knockdown of Fn1 and Col4a, the second most differentially expressed ECM protein in End2 cells, resulted in a further decrease in Bry-GFP⁺ cells (supplementary material Fig. S3D). Similarly, ES cell-specific knockdown of the key integrin signaling component integrin-linked kinase (Ilk) decreased the number of Bry-GFP⁺ cells in End2 co-culture (supplementary material Fig. S3E). Furthermore, the number of Bry-GFP⁺ cells was modestly increased when End2 cells were replaced with mouse embryonic fibroblasts (MEFs), which express moderate levels of fn1 (supplementary material Fig. S3F,G). These results suggest that direct integrin-β1 activation by fibronectin promotes the emergence of mesoderm, probably along with other secreted proteins.

During somitogenesis, integrin-β1 activates Wnt/β-catenin signaling by phosphorylating and inactivating Gsk3β, resulting in accumulation of the active form of the transcriptional mediator β-catenin (Rallis et al., 2010). We tested whether an analogous integrin-β1-dependent cascade could affect Wnt/β-catenin signaling in mesendodermal cells and thereby the emergence of Bry⁺ mesodermal cells. Indeed, Bry-GFP⁺ cells were induced in a Wnt dose-dependent manner in our system (supplementary material Fig. S3H). Introduction of integrin-β1-blocking antibodies decreased the levels of active β-catenin protein and β-catenin/TCF-dependent transcriptional activity (TOP-flash) (Veeman et al., 2003) in Wnt3a-treated differentiating mES cells (Fig. 4B,C). ES cells cultured on exogenous Fn1 also showed higher TOP-flash activity than those cultured on poly-D-lysine (supplementary material Fig. S3I). Similarly, FACS-isolated mES^CAG-GFP cells co-cultured with End2 cells showed decreased levels of active β-catenin in the presence of integrin-β1-blocking antibodies (Fig. 4C). Importantly, knockdown of Fn1 also decreased the End2 cell-responsive accumulation of active β-catenin in sorted mES cells (Fig. 4C), suggesting that Fn1/integrin-β1 signals to positively regulate active β-catenin levels.

Since activated integrin can negatively regulate Gsk3β activity through Ser9 phosphorylation during somitogenesis (Rallis et al., 2010), we investigated whether the Fn1/integrin-β1/β-catenin effect in the mesoderm involved this residue of Gsk3β. We found that Gsk3β co-immunoprecipitated with Ilk, a key signaling component of integrin activation (Fig. 4D). Immunoblotting with an antibody specific for phosphorylated Gsk3β revealed diminished Ser9p-Gsk3β upon non-cell-autonomous decrease in fibronectin (Fig. 4E). Consistent with the epistatic relationship of this model, addition of the Gsk3β inhibitor BIO at day 2 of differentiation to mES cells co-cultured with Fn1-deficient End2 cells rescued the Bry induction defects caused by Fn1 knockdown (Fig. 4F). Addition of soluble Wnt3a, which acts upstream of Gsk3β, increased the absolute number of Bry⁺ cells but could not rescue the relative defect in induction upon Fn1 knockdown (Fig. 4F). Importantly, in zebrafish we observed that the expression domain of ntl was restored in vivo when BIO was added to fibronectin morphants at 2.5 hpf, whereas BIO had no significant effect on the ntl expression domain in control fish (Fig. 4G).

Together, these results suggest that the endoderm positively regulates mesoderm formation through a fibronectin/integrin-β1/β-catenin signaling cascade.

In this study, we demonstrate a previously unrecognized function of endoderm in the development of mesoderm that involves the ECM component fibronectin. Although secreted signals are the predominant morphogens during early development, we demonstrate that ECM components can also play important roles in modulating cell fate decisions. We found that fibronectin signals through integrin-β1 and stimulates Wnt/β-catenin signaling to promote the mesodermal fate, with a specific bias toward precardiac mesoderm. Knockdown of fibronectin resulted in impaired mesoderm induction both in vitro and in vivo and could be rescued by activating β-catenin signaling intracellularly. These findings address a fundamental aspect of developmental biology regarding cell-cell interactions during early specification.

In vivo, fibronectin has a crucial role during early development (George et al., 1993; Jülich et al., 2005). The reported defects upon loss-of-function suggest a structural role for fibronectin. In zebrafish, for example, disruption of the fibronectin-integrin interaction ablates somite boundary formation (Jülich et al., 2005) and causes cardiac bifida with impaired epithelial organization (Trinh and Stainier, 2004). In addition to the structural role, we found that fibronectin is necessary to adequately specify the domains of mesoderm and endoderm during gastrulation, suggesting a morphogenic role of fibronectin.

It is worth noting that fibronectin deficiency reduced expression of the mesendodermal gene bry through mid-gastrulation, but the reduction was largely rescued by late gastrulation, probably through a compensatory mechanism. This is consistent with the reported presence of precardiac mesoderm in zebrafish fibronectin morphants, although mesoderm markers were not examined at earlier stages and quantitative measurement of cardiac progenitors was not reported (Trinh and Stainier, 2004). Similarly, earlier reports of fibronectin deletion in mice revealed apparently normal Bry expression at the time points examined; however, morphological distortions in mesoderm derivatives were seen as early as E8.0, and thinning of the myocardium was apparent by E9.5 (George et al., 1993; Georges-Labouesse et al., 1996). The persistently abnormal expression of the mesodermal gene gsc in our model suggests an incomplete rescue of mesoderm and is in agreement with the more subtle defects in mesoderm derivatives despite the presence of Bry⁺ cells.

Our findings may address an interesting conundrum in evolutionary developmental biology. Morphogenesis during development is thought to be governed by conserved three-dimensional signal gradients. However, events prior to organogenesis vary widely in both shape and motion from one species to another. Although the gastrula differ in their shape, the effect of the signal gradients that induce and specify mesoderm, such as Wnt, FGF, Nodal and BMP, is highly conserved (Kimelman et al., 1992; Mercola et al., 2011).

A unifying explanation of this phenomenon might be the existence of a non-diffusible, yet conserved, regulator of gastrulation that signals independently of the three-dimensional structure and signal gradient. It is interesting to note that, in spite of the different gastrulating motions and architectures across vertebrates, initiation of mesoderm always occurs adjacent to endoderm. The fibronectin matrix between endoderm and mesoderm is conserved from mammals to sea urchins (Spiegel et al., 1980). Also conserved is the manner in which gastrulating mesoderm migrates through the secreted fibronectin matrix. Such a high degree of conservation, combined with the novel signaling role uncovered in our study, raises the possibility that fibronectin might be a key mediator of the response of gastrulating cells to diffusible cues in a localized fashion.

An interesting question raised by this study is the specificity of fibronectin in its signaling role. Although integrin-α5β1 is thought to be the primary receptor of fibronectin, mice lacking integrin-α5 have less severe developmental defects than those lacking fibronectin (Yang et al., 1993). This is frequently explained by the ability of ECM components to activate multiple integrin dimers. Complete loss of integrin-β1 signaling results in lethality as early as E6.5 (Chen et al., 2006). The observation that zebrafish fibronectin morphants initiate gastrulation normally implies that integrins are at least partially activated despite fibronectin knockdown. Redundant mechanisms of integrin activation might explain the partial rescue of ntl expression in our model. However, the rescue remains incomplete, as evidenced by the persistence of endoderm expansion and other mesodermal defects, including the decrease in gsc expression and disproportional loss of cardiac progenitors in vitro and in vivo (Trinh and Stainier, 2004). The integrin signaling that is involved in normal gastrulation movement therefore appears to be distinct from that involved in the specification and determination of mesoderm. The source of this specificity warrants further study.

We demonstrated that the fibronectin-integrin interaction promotes mesoderm formation at least in part through Wnt signaling. The decrease in mesoderm upon disruption of fibronectin-integrin interaction was rescued by exogenous inhibition of Gsk3β activity. These findings are consistent with previous observations in somites, where integrin-β1 acts upstream of Wnt signaling and activates Wnt signaling through phosphorylation-dependent inactivation of Gsk3β (Rallis et al., 2010). Although modulation of Wnt signaling can explain the earliest observed defect in Bry induction, it is unlikely to account for all the observed downstream defects. Secondary insults that result from alterations in cell polarization, such as have been observed in earlier studies in zebrafish (Trinh and Stainier, 2004) and Ciona (Cooley et al., 2011), are most likely to be important for later specification events that contribute to the complex phenotype observed.

Wnt/β-catenin signaling has biphasic roles during cardiac development. Early activation of Wnt/β-catenin signaling directly regulates Bry expression (Yamaguchi et al., 1999) and is required for mesoderm formation (Lindsley et al., 2006), whereas Wnt signaling suppresses the initial induction of Nkx2.5⁺ cardiac progenitors before positively regulating further expansion of cardiac progenitors (Kwon et al., 2007; Qyang et al., 2007; Ueno et al., 2007; Kwon et al., 2009). This is consistent with the biphasic role of fibronectin observed in our study. Interestingly, Wnt/β-catenin signaling has been reported to regulate the expression of ECM components, including fibronectin (Gradl et al., 1999). This suggests a potential positive-feedback loop for activation and propagation of Wnt signaling. It will be interesting to determine whether such ECM-mediated spatial localization of the signal response is employed more widely in cell specification during development and adulthood.

We thank G. Keller for ES^Bry-GFP cells; C. Mummery for End2 cells; D. Stainier for transgenic Sox17-GFP zebrafish and use of the zebrafish facility; R. Ross for Tacβ1 constructs; K. Tomoda for mouse embryo sections; members of the D.S. and C.K. laboratories for thoughtful discussions; S. Elmes from the UCSF Flow Core for cell sorting; G. Howard and S. Ordway for editorial assistance; B. Taylor for assistance with manuscript and figure preparation; and B. Bruneau for critical reading of the manuscript.