1216 Research Article

IntroductionGrowth factors of the TGF� superfamily regulate cellproliferation, differentiation, migration and apoptosis and arecrucial for the development and maintenance of many differenttissues (Cheifetz, 1999; Francis-West et al., 1999). Bonemorphogenetic proteins (BMPs) in this superfamily inducebone marrow mesenchymal stem cell differentiation intoosteoblasts and enhance periodontal regeneration in surgicallycreated defects. BMP ligands, as well as other ligands of theTGF� superfamily, initiate cellular signaling by binding totype I and type II receptors, both of which are serine/threoninekinase receptors, thereby inducing the formation of a hetero-oligomeric receptor complex. The type II receptor thenphosphorylates and activates the type I receptor, whichsubsequently transiently associates with and phosphorylates asubclass of a unique family of intracellular signaling molecules– the receptor-regulated Smads or R-Smads (Godin et al., 1999;Ghosh-Choudhury et al., 1994). The R-Smad Smad1transduces signals from BMPs, whereas the R-Smads Smad2and Smad3 mediate signals from the TGF�/activin/nodalsubfamily members. The recruitment of R-Smads to thereceptor complex is the initial step of intracellular TGF� signaltransduction and plays a critical role in regulating signaltransduction by the TGF� receptor family.

TGF� signaling can be enhanced by a protein, Smad anchorfor receptor activation or SARA, which bindsunphosphorylated Smad2 and Smad3 and recruits them tomembranes (Tsukazaki et al., 1998). SARA contains a FYVEdomain adjacent to its Smad-binding domain. The double zinc-finger motifs in FYVE domains have been shown to bind

phosphatidylinositol 3-phosphate and anchor FYVE-containing proteins to cytoplasmic or endosomal membranes.This enables proteins containing the FYVE domain to recruitother molecules to membranes for endocytosis orrelocalization (Gaullier et al., 1998; Kutateladze and Overduin,2001). The FYVE domain in SARA is required for recruitmentof Smad2/3 to the appropriate subcellular locations thatposition it adjacent to the activated T�RI (Tsukazaki et al.,1998; Wu et al., 2000). Activation of TGF� signaling causesphosphorylation of Smad2, subsequent dissociation fromSARA, and nuclear translocation with the common Smad,Smad4 (Massague, 1998; Heldin et al., 1997).

Recent data suggest that SARA can also act as a negativeregulator of TGF� signaling. In Drosophila, the catalyticsubunit of protein phosphatase 1 (PP1c) has been shown tobind SARA and negatively regulate signaling through Dpp, aBMP homologue (Bennett and Alphey, 2002). Expression of amutant SARA with a mutation in the PP1c-binding domain(F678A) resulted in hyperphosphorylation of the type Ireceptor and elevated expression of dpp target genes (Bennettand Alphey, 2002). In our previous studies, we also found thatSARA facilitates dephosphorylation of T�RI in cell culturethrough a Smad7-mediated negative feedback loop. Theinhibitory Smad7 recruits a protein phosphatase holoenzyme,GADD34-PP1c, to inhibit TGF�-induced cell cycle arrest andconfer cell resistance to TGF� in response to UV irradiation.SARA facilitates the formation of GADD34-PP1c bypresenting PP1c to this holoenzyme, and thus acts as acomponent of Smad7-mediated negative feedback loop of thissignaling pathway (Shi et al., 2004). SARA is therefore a

Signaling through receptors of the transforming growthfactor � (TGF�) superfamily is mediated by cytoplasmicSmad proteins. It has been demonstrated that Smad anchorfor receptor activation (SARA) facilitates TGF� andactivin/nodal signaling by recruiting and presentingSmad2/3 to the receptor complex. SARA does not bindSmad1 and hence does not enhance bone morphogeneticprotein (BMP) signaling. Here we report for the first timethat the endosome-associated FYVE-domain proteinendofin acts as a Smad anchor for receptor activation inBMP signaling. We demonstrate that endofin binds Smad1preferentially and enhances Smad1 phosphorylation andnuclear localization upon BMP stimulation. Silencing ofendofin by RNAi resulted in a reduction in BMP-dependent

Smad1 phosphorylation. Moreover, disruption of themembrane-anchoring FYVE motif by point mutation ledto a reduction of BMP-responsive gene expression in cellculture and Xenopus ectodermal explants. Furthermore,we demonstrate that endofin contains a protein-phosphatase-binding motif, which functions to negativelymodulate BMP signals through receptordephosphorylation. Taken together, our results suggest thatendofin plays an important role in both positive andnegative feedback regulation of the BMP signalingpathway.

Key words: BMP, Smad, SARA, Protein phosphatase, Osteoblast

Summary

Endofin acts as a Smad anchor for receptor activationin BMP signalingWeibin Shi1, Chenbei Chang2, Shuyi Nie2, Shutao Xie1, Mei Wan1 and Xu Cao1,*1Department of Pathology and 2Department of Cell Biology, University of Alabama at Birmingham, Birmingham, AL 35294, USA*Author for correspondence (e-mail: cao@path.uab.edu)

Accepted 10 January 2007Journal of Cell Science 120, 1216-1224 Published by The Company of Biologists 2007doi:10.1242/jcs.03400

Jour

nal o

f Cel

l Sci

ence

JCS ePress online publication date 13 March 2007

1217Endofin in BMP signaling

central regulator of TGF� signaling and can act to eitherpromote or inhibit TGF� signals.

The enhancement of TGF�/activin signaling by anotherFYVE-domain protein, Hgs/Hrs, further indicates that otherSARA-like molecules may participate in the regulation ofTGF�/activin signaling (Miura et al., 2000). By analogy, itwould be anticipated that BMP signal transduction might alsoemploy SARA-like regulators to help bring Smad1 to thereceptor complexes; however, up to now, no SARA-basedmechanism had been identified for the BMP pathway. In thisstudy, we present evidence indicating an important role for theendosome-associated FYVE-domain protein endofin in BMPsignaling (Seet and Hong, 2001; Seet et al., 2004; Seet andHong, 2005). We show that endofin binds preferentially toSmad1 and regulates Smad1 signaling by modulation of itsphosphorylation and nuclear translocation upon BMPactivation. Endofin also has a functional PP1c-binding domain,which recruits PP1c and facilitates dephosphorylation of thetype I BMP receptor. Our results identify for the first time aBMP-specific Smad anchor as a receptor activation moleculethat plays a role in transduction and negative feedbackregulation of BMP signals.

ResultsEndofin interacts with Smad1 and PP1cEndofin has been shown to localize to early endosomes, as doesSARA; but it does not bind to Smad2 or influence TGF�signals (Seet and Hong, 2001). The relatively high divergenceof the Smad-binding domain suggested that endofin might bindR-Smads other than the TGF�-specific Smad-2, such as theBMP-specific Smad1. To test this possibility directly, weperformed a co-immunoprecipitation assay using the lysate ofCOS cells transfected with endofin and Smad1 or Smad2. Wefound that Smad1, but not Smad2, was co-precipitated withendofin (Fig. 1A), suggesting that endofin preferentially bindsto Smad1. To further confirm the interaction in situ, weperformed a protein-fragment complementation assay (PCA)using yellow fluorescent protein (YFP) as a marker. Theprinciple of the PCA strategy is that complementary fragments(F1 and F2) of a reporter protein (such as YFP) will fold intoan active form only when brought together in close proximityby fusion to two proteins that interact with each other (Remyet al., 2004; Li et al., 2006). In our experiments, we fusedSmad1 and endofin to N-terminal (a.a. 1-158, YFP1) and C-terminal (a.a. 159-239, YFP2) fragments of YFP, respectively.When expressed alone, neither YFP1-Smad1 nor endofin-YFP2 produced fluorescent signals in the cells. Co-expressionof the vectors encoding YFP1 and YFP2 did not generatefluorescent signals either (Fig. 1B). When the fusion proteinswere co-transfected into the COS cells, however, we observedstrong yellow fluorescence (Fig. 1B). The intensity of thefluorescence signal is similar to that observed when Smad1 andits interacting protein Hoxa1 were co-transfected in a similarway in parallel (Fig. 1B). These results indicated that Smad1and endofin interact in transfected cells. To confirm thisinteraction, we also examined the localization of tagged Smad1and endofin in transfected cells by fluorescenceimmunocytochemistry. Endofin and Smad1 co-localized in thecytosol in a punctuate pattern (Fig. 1C), suggesting that the twoproteins interact at intracellular vesicles in the cells. Toeliminate the possibility of artifacts generated by the

overexpression system, we also analyzed endogenous proteinsby using antibodies against Endofin and Smad1. As shown inthe merged panels in Fig. 1C, the endogenous proteins werealso co-localized in the cytosol of the cells. All our data areconsistent with the idea that endofin and Smad1 associate witheach other in the cell.

In addition to the Smad-binding motif, all known SARAhomologs also contain a canonical PP1c-binding motif(K/R)xVxF or (K/R)VxF (Bennett and Alphey, 2002) and thismotif is involved in negative feedback regulation of Dppsignaling in Drosophila. We therefore examined whetherendofin can also associate with PP1c. Co-immunoprecipitationanalyses indicated that endofin indeed interacts with PP1c.Interestingly, our results further showed that overexpression ofPP1c enhanced the interaction of endofin with Smad1 (Fig.1D). As PP1c can down-regulate BMP signaling resulting inhypophosphorylation of BMP signal components, ourobservation implies that endofin may preferentially bind tounphosphorylated Smad1. To verify this hypothesis, weassayed for endogenous interaction of endofin with Smad1,phosphorylated Smad1 (pSmad1) and PP1c in the presence orabsence of BMP2. We observed that BMP2 stimulated bindingof endofin to PP1c, and the phosphorylated form of Smad1 hasvery low affinity for endofin (Fig. 1E). To examine thespecificity of the interaction between endofin and Smad1, theinteraction of SARA and Smad1 was also detected. BMP2 doesnot stimulate the binding of SARA with either Smad1 andphosphorylated Smad1. Our results confirm that endofin, butnot SARA, binds preferentially to the inactive form of Smad1.

To further determine whether the putative Smad- and PP1-binding domains (SBD and PBD respectively) mediate theinteraction of endofin with Smad1 and PP1c, we performed co-immunoprecipitation experiments with mutant endofin. Threemutants were generated that contain point mutations in FYVEand PBD domains or deletion of the SBD motif (Fig. 2A). Wefound that disruption of the SBD or mutation of FYVE domainresulted in a diminishment of the Smad1-endofin interaction(Fig. 2B), while mutation of the PBD almost abolished theinteraction between PP1c and endofin (Fig. 2C). The resultsdemonstrate that SBD and PBD mediate the binding of endofinto Smad1 and PP1c respectively.

Endofin facilitates dephosphorylation of the BMP type Ireceptor by PP1Expression of a PP1c-binding mutant of SARA has beenshown to result in hyperphosphorylation of the type I receptorand elevation of expression of a target of TGF-� signaling(Bennett and Alphey, 2002). To assess whether endofin has asimilar activity, we performed an in vivo phosphorylationassay. As shown in Fig. 3A, overexpression of endofin reducedphosphorylation of the type I BMP receptor ALK3 in responseto BMP2, whereas the mutant endofin with impaired PBDenhanced ALK3 phosphorylation. Phosphatase inhibitor 1, aninhibitor for PP1c, also increased ALK3 phosphorylation. Theresults suggest that endofin can regulate ALK3 activationthrough its interacting phosphatase PP1c.

In our previous study of SARA, we found thatdephosphorylation of the type I TGF� receptor ALK5 ismediated by a protein phosphatase regulatory subunitGADD34, which binds to ALK5 through a bridgingmechanism involving an inhibitory Smad, Smad7 (Shi et al.,

Jour

nal o

f Cel

l Sci

ence

1218

2004). To see whether a similar mechanism applies to endofin,we tested the interaction of the BMP type I receptors ALK3and ALK6 with GADD34, as well as the potential requirementfor inhibitory Smads Smad6 and Smad7. We found that bothALK3 and ALK6 readily interact with GADD34 without anyauxiliary factors (Fig. 3B), and sequential co-

Journal of Cell Science 120 (7)

immunoprecipitation further clarified that although ALK3 andALK6 bind inhibitory Smads, they do not form a ternarycomplex with GADD34 and Smad6 or Smad7 (Fig. 3C). Ourdata indicate that unlike ALK5, ALK3 and ALK6 directlyassociate with the regulatory subunit of protein phosphatase 1in the absence of inhibitory Smads.

Fig. 1. Endofin interacts with Smad1. (A) Co-immunoprecipitation (co-IP) of endofin with Smads. Flag-tagged Smad1 or Smad2 was co-transfected with HA-tagged endofin into COS1 cells. The lysate was subjected to co-IP with anti-HA antibody and probed with anti-Flagantibody, or co-IP with anti-Flag and probed with anti-HA. (B) Confirmation of the interaction between Smad1 and endofin by protein-fragment complementation assays (PCA). The complementary fragments (YFP1 and YFP2) of YFP were fused with Smad1 and endofin,respectively, to create YFP1-Smad1 (a.a. 1-158 of YFP) and endofin-YFP2 (a.a. 159-239 of YFP), and then transfected into COS1 cells. Theinteraction between endofin and Smad1 is indicated here as yellow fluorescence. Transfection of pcDNA3, co-transfection of pcDNA3-YFP1and pcDNA3-YFP2 and co-transfection of YFP1-Smad1 and Hoxa1-YFP2 serve as negative and positive controls, respectively (Li et al., 2006).Bar, 10 �m. (C) Colocalization of Smad1 and endofin. Upper panel: C2C12 cells were transfected with HA-Endofin and Flag-Smad1. Aftertransfection for 36 hours, cells were processed for immunofluorescence using antibodies against HA (rabbit polyclonal) and Flag (mousemonoclonal). Lower panel: Endogenous protein co-localization. Naive C2C12 cells were subjected to immunofluorescent microscopy usingrabbit polyclonal anti-endofin and mouse monoclonal anti-Smad1. Nuclei were counterstained with Hoechst 33342. Representative images areshown. (D) Endofin interacts with both Smad1 and PP1c. COS1 cells were transfected with Flag-Smad1, HA-tagged endofin and Myc-taggedPP1c. Cells were treated with rhBMP2 (200 ng/ml) for 2 hours before lysis. The lysate was subject to co-immunoprecipitation with antibodyagainst HA, the presence of Smad1 and PP1c in the precipitate was detected with anti-Flag and anti-Myc antibodies, respectively. (E) Co-IP ofendogenous endofin, Smad1 and PP1c. C2C12 cells were cultured until subconfluent, with or without treatment of rhBMP2 (200 ng/ml, 2hours), cells were lysed, and the lysate was subjected to immunoprecipitation with antibody against either endofin or SARA. The presence ofSmad1 and PP1c in the precipitate was detected by anti-Smad1 and anti-PP1c antibodies. Bar, 5 �m.

Jour

nal o

f Cel

l Sci

ence

1219Endofin in BMP signaling

Endofin modulates Smad1 phosphorylation and nucleartranslocationTo examine the functional significance of the endofin-Smad1interaction, we next performed a loss-of-function study, usingsmall interfering RNA. Knockdown of endofin using specificsiRNA resulted in a reduction in the level of phosphorylationof Smad1 (Fig. 4A), which corroborated our hypothesis thatendofin acts as a SARA-like protein in BMP signaling and isrequired for Smad1 phosphorylation. Notably, endofin appearsto utilize a mechanism similar to that employed by SARA inwhich the PP1c-binding motif acts to negatively regulatephosphorylation of R-Smad.

As phosphorylation of Smad1 enables its translocation intothe nucleus, we next tested whether endofin also regulates thenuclear translocation of Smad1. Staining of phosphorylatedSmad1 in C2C12 cells expressing endofin confirmed thatendofin enhanced Smad1 phosphorylation and nucleartranslocation upon simulation of BMP2. Overexpression ofSARA, however, did not have a significant effect on Smad1activation and localization (Fig. 4B).

Endofin regulates expression of BMP downstreamgenes in C2C12 cells and mineralization in humanMSCsTo further examine the effect of endofin on BMP signaltransduction, we generated stable cell lines derived fromC2C12 and human MSCs infected with retroviruses thatexpress endofin, endofin mutants or the control greenfluorescent protein (GFP). The influence of stable endofinexpression on BMP-induced downstream gene transcription

was monitored in these cell lines using a BMP-responsivereporter construct (9xSBE-luc). As shown in Fig. 5A, wild-type endofin slightly enhanced the expression of the BMP-responsive luciferase gene. The FYVE(C753S) mutant, whichleads to cytosolic mislocalization of endofin and possibly itsassociated proteins including Smad1, decreased the reportergene transcription. As expected, deletion of the Smad-bindingdomain (Endofin�SBD) also inhibited BMP-induced geneexpression. By contrast, mutation of the PP1c-binding domainPBD(F872A) enhanced transcription of the reportertranscription; a finding that is consistent with our observationthat this mutant increased ALK3 phosphorylation (Fig. 3A).

BMPs were originally purified from bone matrix and havebeen shown to regulate bone development and induceosteoblast differentiation. We therefore examined the effects ofexpression of mutant endofin on BMP-stimulated expressionof the osteoblast differentiation marker, alkaline phosphatase(ALP), as well as osteoblast-mediated mineralization. C2C12cells with stable overexpression of the PBD(F872A) mutantshowed higher ALP activity than cells infected with wild-typeendofin (Fig. 5B,C). Similarly, human MSCs with stableoverexpression of the PBD(F872A) mutant showed enhancedmineralization (Fig. 5D).

Regulation of BMP-dependent mesodermal induction byendofin in Xenopus embryosBMPs play diverse roles during early Xenopus development,especially in embryonic dorsoventral patterning. Whenoverexpressed, BMPs can induce ventral types of mesoderm inXenopus ectodermal explants (animal caps). To determine

Fig. 2. Identification of the endofin interaction domain inSmad1 and PP1c. (A) Schematic representation of wild-typeand mutant endofin. Three mutants were generated to disruptthese conserved domains: FYVE domain (C753S), Smad-binding domain (deletion of amino acids 814-860) and PP1c-binding domain (F872A). (B) Deletion of Smad-bindingdomain in endofin diminishes its interaction with Smad1.COS1 cells were transfected with Flag-Smad1, HA-taggedendofin and its mutants. Immunoblots in the two lowerpanels show the protein expression levels of the COS1 celllysates. The upper panel shows anti-HA immunoprecipitatesprobed with anti-Flag antibody. (C) Co-IP of PP1c withendofin and the PBD(F872A) mutant. HA-tagged endofinand its mutant with disrupted PP1c-binding domain were co-

transfected with PP1c into COS1 cells. The lysate was subject to immunoprecipitation with anti-HA. Immunoblots indicating the proteinexpression levels of the lysates of COS1 cells are shown in the two lower panels. The upper panel shows anti-HA immunoprecipitates probedwith anti-Flag antibody.

Jour

nal o

f Cel

l Sci

ence

1220

whether endofin can modulate BMP signaling in vivo, weexamined the effect of endofin overexpression on BMP-dependent endogenous mesodermal marker expression. In thisexperiment, we injected BMP4 RNA with or without RNAsencoding endofin or its mutants in animal poles of two-cellstage embryos and dissected animal caps from injectedembryos at blastula stages 8.5-9. The caps were incubated togastrula stages before total RNA was extracted for use in areverse transcription (RT)-PCR assay of marker geneexpression. As shown in Fig. 6, BMP2 induced the expressionof the early mesodermal markers Xbra, Xhox3 and Xwnt8.Expression of wild-type endofin did not affect the level ofinduction of these genes, possibly because of the presence ofa high level of endogenous Xenopus endofin-like molecule.Expression of the FYVE(C735S) mutant led to a dramaticinhibition of Xbra, Xhox3 and Xwnt8, whereas expression ofthe endofin �SBD mutant resulted in a modest reduction in themarker gene expression (Fig. 6). By contrast, expression of thePBD(F872A) mutant enhanced the expression of Xbra, Xhox3and Xwnt8: a result further supporting the notion thatrecruitment of PP1c by endofin plays an important role innegative feedback regulation of BMP signaling in vivo.

DiscussionThe FYVE domain protein, SARA, was identified as a keymolecule that anchors Smad2/3 and promotes receptor

Journal of Cell Science 120 (7)

activation of the TGF� signaling pathway (Tsukazaki et al.,1998; Bennett and Alphey, 2002; Wu et al., 2000). Thesimilarities among members of the TGF� superfamilyprompted us to search for evidence of a SARA-like moleculeassociated with BMP signaling. GenBank search for sequenceswith homology to functional regions of SARA identified aFYVE domain protein, KIAA0305, which had a C-terminalregion that was closely related to the C-terminus of SARA.KIAA0305, also known as endofin (Seet and Hong, 2001), wasfirst sequenced by investigators at the Kazusa Institute inJapan, but its function remained unknown. The conserveddomain structures between SARA and endofin suggest thatendofin may function like SARA as a membrane anchorprotein for R-Smads; however, the Smad-binding domain ishighly divergent between the two proteins. Accordingly, it hasbeen shown that although endofin localizes to earlyendosomes, it does not bind to Smad2 nor influence TGF�signals (Seet and Hong, 2001). We therefore investigate thepossibility that endofin acts like SARA but binds different R-Smads. Here, we report that using co-immunoprecipitation,protein-fragment complementation assay andimmunolocalization approaches, we observed consistently thatendofin preferentially binds to Smad1, not Smad2, suggestingthat endofin may act as a Smad anchor for receptor activationin BMP signaling. This idea is consistent with our finding thatendofin, but not SARA, facilitated the BMP signalling, and

Fig. 3. The BMP type I receptor is the substrate of PP1.(A) C2C12 cells transfected with different combinations ofgenes were labeled with [32P]orthophosphate in thepresence or absence of BMP2 as indicated. ALK3-HA wasimmunoprecipitated from lysates of treated cells andseparated by 8.5% SDS-PAGE. Gels were dried andexposed to Biomax Mr film (Eastman Kodak). PBD(F872A) is a mutant with the disrupted PP1c-bindingdomain. Phosphatase inhibitor-1 is an inhibitor for PP1.(B) BMP type I receptors interact with PP1 regulatorysubunit GADD34. COS1 cells transfected with HA-taggedtype I receptors (ALK3 and ALK6) and Flag-taggedGADD34. Lysates were subjected to co-immunoprecipitation with anti-Flag and the presence of

ALK3 and ALK6 were probed with antibody against HA. (C) Unlike in the TGF� signaling pathway, GADD34 interacts with the BMP type Ireceptor without the bridging of inhibitory Smads. The GADD34-ALK3/6 complex was further confirmed by a sequentialimmunoprecipitation. COS1 cells were first co-transfected with ALK3/6-HA with or without Flag-Smad6/7 and GADD34. The lysates weresubjected to immunoprecipitation with HA antibody, and the resultant precipitates were eluted from protein-G-Sepharose beads by HA peptidecompetition and then subjected to a second immunoprecipitation with anti-Flag antibody. The final precipitates were immunoblotted withantibodies against all these components.

Jour

nal o

f Cel

l Sci

ence

1221Endofin in BMP signaling

knockdown of endofin with siRNA reduced the level of Smad1phosphorylation.

The specificity of the Smad anchor for receptor activationproteins may be determined by critical residues in the Smad-binding domains, which define preferential interactions withTGF�- or BMP-specific R-Smads (Wu et al., 2000). Thecrystal structure of the Smad2 MH2 domain bound by theSmad-binding domain of SARA has revealed that the MH2domain interacts with an extended proline-rich motifconsisting of a proline-rich coil, an � helix and a � strand inthe Smad-binding domain of SARA (Wu et al., 2000). Thesequence of this region differs in endofin and SARA, whichmay explain why the two proteins bind distinct R-Smads.Furthermore, it has been shown that an asparagine residue(N381) in Smad2 makes extensive contacts with the Smad-binding domain of SARA (Wu et al., 2000). Notably, in Smad1this residue is a serine. Substitution of N381 in Smad2 with

serine interferes with the binding of SARA to Smad2 and leadsto a failure of SARA to properly localize the mutant Smad2for activation. We also found that substitution of the serine atposition 379 in Smad1 with asparagine interfered with thebinding of endofin to Smad1 (our unpublished data). Thiswould suggest that a serine at this position is critical for contactof Smad1 with endofin.

The SARA-like proteins appear to be capable of bothenhancing and dampening the level of phosphorylation ofcertain phospho-proteins. Based on our results, it seems thatwhen ligand binding initiates the signaling events, the endofinrecruits the Smad effectors thereby enabling theirphosphorylation, subsequent nuclear translocation and furtherregulation of downstream gene expression (Tsukazaki et al.,1998; Bennett and Alphey, 2002; Shi et al., 2004). To balancethe stimulated signaling, PP1c is recruited by endofin fordephosphorylation of Smad1 and this enables the signaling

Fig. 4. Endofin modulates Smad1 phosphorylation and nuclear translocation. (A) Silencing of endofin expression by RNA interferenceinhibited Smad1 phosphorylation. C2C12 cells were transfected with Flag-tagged Smad1 and either vector, construct for GFP or endofinsiRNA. On the third day following 3-hour BMP2 treatment, Flag-tagged Smad1 was precipitated with anti-flag and its phosphorylation levelwas detected with anti-phospho-Smad1. (B) Endofin, but not SARA, regulates BMP-specific phosphorylation and intracellular translocation ofSmad1. C2C12 cells were transfected with either HA-SARA or HA-Endofin, then stimulated with BMP (100 ng/ml) for 3 hours. After fixationwith paraformaldehyde (PFA), cells were immunostained with either anti-HA (green) or anti-phospho-Smad1 (red). Nuclei were visualizedusing Hoechst 33342. Representative images are shown. Bar, 20 �m.

Jour

nal o

f Cel

l Sci

ence

1222 Journal of Cell Science 120 (7)

Fig. 5. Endofin regulates expression of BMP downstream genes and mineralization in human MSCs. (A) Transcriptional response assay. C2C12cells stably expressing GFP (control), or endofin or its mutants were transfected with BMP signaling reporter construct 9XSBE-luc. Transfectedcells were incubated in the presence or absence of BMP2 (200 ng/ml). Luciferase activity was normalized and plotted as the mean ± s.d. oftriplicates from a representative experiment. *P<0.05 compared with the second group. (B) Alkaline phosphatase activity assay. C2C12 cellsstably expressing GFP (control) or endofin or its mutant PBD(F872A) were cultured in 24-well plates treated with or without BMP2 andharvested at day 5 for alkaline phosphatase activity assay. Relative alkaline phosphatase activity was normalized and plotted as the mean ± s.d.of triplicates from a representative experiment. *P<0.05 compared with second group. (C) Alkaline phosphatase staining. C2C12 were culturedand treated as in alkaline phosphatase activity assay and harvested at day 3. After fixation with 4% paraformaldehyde, cells were subjected toalkaline phosphatase staining. Bar, 20 �m. (D) Overexpression of mutated endofin with disrupted PP1c-binding domain enhancedmineralization activity of human MSCs (von Kossa Assay). Human MSCs were cultured in Dulbecco’s Modified Eagle’s Medium: low glucose,1� penicillin-streptomycin, 10% fetal bovine serum (BioWhittaker, MSC serum), 10 mM �-glycerol phosphate, 50 �M Ascorbic acid 2-phosphate (AsAP) and 200 ng/ml BMP2. Mineralization assay was performed at day 24. After silver nitrate was added, a calcium deposit wasvisible as a black structure or focal dot. i, GFP; ii, GFP+BMP2; iii, Endofin+BMP2; iv, Endofin-PBD(F872A)+BMP2. Bar, 30 �m.

Fig. 6. Regulation of BMP-dependent mesodermal inductionby endofin in Xenopus embryos. Capped RNAs encodingendofin (wild type and mutants) and BMP4 were synthesizedin vitro and injected alone or in combination into both animalpoles of two-cell-stage embryos. The ectodermal explants(animal caps) from injected embryos were dissected atblastula stages (stage 8.5-9) and cultured until gastrula stages(stage 11) before total RNA was extracted. RT-PCR wasperformed using the primers for different mesodermal markergenes. EF1-� serves as a loading control. Although mutationsin the FYVE and SBD domains reduced expression of BMP-responsive genes, the mutation in the PBD domain enhancedBMP-induced marker gene expression.

Jour

nal o

f Cel

l Sci

ence

1223Endofin in BMP signaling

back to a basal unstimulated state. Endofin preferentially bindsunphosphorylated R-Smads. Therefore, once most R-Smadsare phosphorylated and dislodged from endofin, PP1c willreadily bind to endofin (Shi et al., 2004). The availability of R-Smads and phosphorylated R-Smads or the ratio of the twodetermines whether endofin facilitates BMP signalingpositively or negatively. Thus, the immediate question is theidentity of the substrate of the PP1c once it is recruited toendofin. Smad1 is not likely to be the substrate because itdissociates from endofin once phosphorylated (Bennett andAlphey, 2002; Shi et al., 2004). However, as shown previously,GADD34, a PP1 regulatory subunit (He et al., 1996; Novoa etal., 2001), interacts with Smad7 to direct the PP1 holoenzymeto the type I receptor of TGF�, ALK5 for its dephosphorylation(Shi et al., 2004). Here we demonstrated a Smad7-independentinteraction between GADD34 and the type I receptor of BMP.Endofin, acting as a Smad anchor for receptor activation,delivers PP1c to GADD34 to form a holoenzyme thatdephosphorylates BMP type I receptor. This dual function alsoexplains why overexpression of wild-type endofin and SARAhas only a modest affect on phosphorylation of Smads andosteoblast differentiation (Fig. 4B, Figs 5 and 6) (Bennett andAlphey, 2002; Shi et al., 2004). These dual regulatorymechanisms ensure the proper level of signaling activity.However, once the dephosphorylation activity is blunted bymutation, the positive regulatory effect of the Smad anchor forreceptor activation becomes evident immediately (Figs 5 and6). Endofin and its mutants affected the expression of theosteoblast differentiation marker alkaline phosphatase,osteoblast-mediated mineralization as well as transcription ofendogenous BMP-responsive genes in vivo in Xenopus uponBMP stimulation (Figs 5 and 6), which suggests that endofinutilizes membrane anchoring (FYVE domain) and PP1cdephosphorylation (PBD domain) to modulate the levels ofBMP signaling during in vitro osteoblast differentiation and invivo Xenopus development.

TGF� signaling is modulated mainly through regulation ofthe phosphorylation status of key components of the signalingpathway (Massague, 1998; Derynck, 1994). Continuousreceptor activity is required to maintain localization of activeSmads in the nucleus and for TGF�-induced transcription(Inman et al., 2002). This implies that depletion of activeSmads in the nucleus is a prerequisite for dampening of thesignaling. It has been reported that both Smads and receptorscan be degraded in the cytoplasm through proteosomedegradation (Datto and Wang, 2005; Kuratomi et al., 2005; DiGuglielmo et al., 2003). However, accumulating evidencesuggests that R-Smads are recycled and that R-Smads areshuttled continuously between the nucleus and cytoplasm (Xuet al., 2002; Inman et al., 2002). After phosphorylation by atype I receptor and translocation into the nucleus, the R-Smadsreappear in the cytoplasm unphosphorylated by adephosphorylation mechanism. Recently, such a phosphatasewas finally identified (Knockaert et al., 2006; Lin et al., 2006).In addition to the direct dephosphorylation mechanism of R-Smads, we identified a BMP-specific inhibitory mechanismby which endofin recruits PP1 complex to dephosphorylatetype I receptors. As a result, like SARA, endofin indirectlyregulates the R-Smads phosphorylation level positively andnegatively in controlling R-Smads recycling, and maintainsbalanced BMP signaling. Thus, upon BMP stimulation and

Smad1 phosphorylation, endofin releases the phosphorylatedSmad1 for its translocation to the nucleus. At the same time,there is a signal-dependent increase in binding of PP1c toendofin for negative feedback inhibition of the BMP signals.Endofin-regulated dephosphorylation occurs in a similarfashion to that of SARA, but with a distinct mechanism:inhibitory Smad7 is not involved, since type I receptors ofBMPs directly interact with GADD34 in the absence ofinhibitory Smads (Fig. 3). Furthermore, unlike directdephosphorylation of R-Smads by their phosphatase(Knockaert et al., 2006; Lin et al., 2006), SARA- and/orendofin-mediated dephosphorylation is at the receptor level,upstream of R-Smads. It is unclear whether endofininteraction with Smad1 and/or PP1c is also modulated byother proteins, and whether this constitutes another level ofcontrol for BMP signals in a cell-type-specific manner.

Materials and MethodsAntibodies and reagentsFor endogenous co-immunoprecipitation, with the assistance of Cytomol Corp.(Mountain View, CA) we developed a rabbit anti-endofin polyclonal antibody raisedagainst a peptide (amino acids 41-59, CSVSSELASSQRTSLLPKD) in the N-terminus of human endofin. All other antibodies were obtained from commercialsources: monoclonal anti-Flag M2 and anti-�-actin (Sigma-Aldrich), anti-HA(Babco), anti-phospho-Smad1 (Ser463/465) rabbit polyclonal IgG (Upstate), anti-phospho-Smad2 (Ser465/467) rabbit polyclonal IgG (Biosource), mousemonoclonal anti-Smad1 and monoclonal anti-PP1 (Santa Cruz). Texas-Red-conjugated and FITC-conjugated (donkey anti-mouse, donkey anti-rabbit)secondary antibodies were from Jackson Immunoresearch Laboratories (WestGrove, PA).

cDNA constructs and retroviral vectorsThe human cDNA of endofin (KIAA0305) was kindly provided by the Kazusa DNAResearch Institute and cloned into the pcDNA3 vector. For FYP fragment fusionexpression vectors, Smad1 and endofin were subcloned respectively at the 3� and5� end of the N-terminal fragment FYP1 (a.a. 1-158) and C-terminal fragment FYP2(a.a. 159-239) of YFP (Clontech, Palo Alto, CA) into the pcDNA3.1 vector. Themutations in the FYVE domain and PP1c-binding domain were generated by site-directed mutagenesis using the Quik-Change® Site-Directed Mutagenesis Kit(Stratagene, La Jolla, CA). The deletion mutant endofin�SBD was generated byPCR. We used the murine stem cell virus (MSCV) vector, pMSCVneo (Clontech),to generate cell lines expressing endofin or its mutants. Endofin and its mutantswere subcloned into pMSCVneo, and the viruses were prepared by transfection ofthese retroviral constructs into a packaging cell line, 293GPG, using Lipofectamine.Human mesenchymal stromal cells (MSCs) were infected with these viruses, andstable expression of endofin or its mutants was achieved through neomycinresistance gene selection. pMSCVneo-EGFP was used as a transfection andinfection control.

Immunoprecipitation and immunoblottingCells transfected by Lipofectamine (Gibco-BRL) were lysed withradioimmunoprecipitation buffer (50 mM Tris-HCl pH 8, 150 mM NaCl, 1%Nonidet-P-40, 0.5% sodium deoxycholate, 0.1% SDS) containing proteaseinhibitors (10 �g/ml aprotinin, 10 �g/ml leupeptin, 1.0 mM PMSF) and phosphataseinhibitors (10 mM sodium orthovanadate, 50 nM inhibitor-1 and 50 mM sodium �-glycerophosphate). Lysates were immunoprecipitated by incubation with theappropriate antibodies, followed by adsorption to protein-G-Sepharose.Immunoprecipitates were separated by SDS-PAGE, blotted onto a PVDF membrane(Bio-Rad Laboratories), and visualized by enhanced chemiluminescence (ECL Kit;Amersham Biosciences).

Small interfering RNA (siRNA) constructs and transfectionTo generate the construct for silencing endogenous endofin expression, a 21-nucleotide oligo (oligo 1) corresponding to nucleotides 390-410 of the mouseendofin coding region was first inserted into ApaI-HindIII-digested pBS/U6 (Sui etal., 2002). The inverted motif that contains the six-nucleotide spacer and five Ts(oligo 2) was then subcloned into the EcoRI-HindIII sites of the intermediateplasmid to generate BS–U6–si-Endofin for endofin silencing. The sequence of oligo1 is 5�-GGTAA CTTAG TGCAT GCCAC A-3� (forward) and 5�-AGCTT GTGGCATGCA CTAAG TTACC-3� (reverse). The sequence of Oligo 2 is 5�-AGCTTGTGGC ATGCA CTAAG TTACC CTTTT TG-3� (forward) and 5�-AATTCAAAAA GGGTA ACTTA GTGCA TGCCA CA-3� (reverse). For the controlconstruct for silencing of green fluorescence protein (GFP) expression, pBS-U6-

Jour

nal o

f Cel

l Sci

ence

1224 Journal of Cell Science 120 (7)

siGFP, a 22-nucleotide oligo (oligo 1) corresponding to nucleotides 106-127 of theGFP coding region was used as described previously (Sui et al., 2002). A BLASTsearch of the NCBI database ensured specific targeting of the cognate mRNA.C2C12 myoblasts were transfected using Lipofectamine (Invitrogen).

Detection of in vivo phosphorylationThirty-six hours after transfection with different combinations of genes, cells werewashed twice with phosphate-free DMEM containing 2% dialyzed fetal calf serum,incubated in the same medium for 4 hours and then treated with rhBMP2 (Sigma-Aldrich) for an additional 2 hours at 37°C. The cells were washed again with thesame medium and incubated with complete DMEM/2% FBS for a further 2 hours.The cells were then washed with ice-cold PBS and lysed withradioimmunoprecipitation assay buffer. Flag-Smad1 and HA-ALK3 wasimmunoprecipitated with anti-Flag and anti-HA, respectively, as described above.The resultant precipitates were separated by 8.5% SDS-PAGE, blotted onto PVDFmembranes and visualized using the ECLPlus western blotting detection system(Amersham Biosciences). For metabolic labeling, the cells were labeled with 1mCi/ml [32P]orthophosphate (PerkinElmer) for an additional 2 hours at 37°C in theabsence or presence of BMP2 (200 �g/ml), then subjected to the proceduresdescribed above. Finally, the gels were dried and exposed to Biomax Mr or MS film(Eastman Kodak Co.). For confirmation of equal loading, the dried gels wererehydrated with transfer buffer after autoradiographic analysis, reblotted onto aPVDF membrane, and the transfected HA-ALK3 visualized using the ECLPluswestern blotting detection system (Amersham Biosciences).

Subcellular localizationImmunolocalization was performed as described previously (Shi et al., 2004).Briefly, cells were fixed with 4% paraformaldehyde (PFA), permeabilized with0.1% Triton X-100, and the lysates incubated with primary antibody incubation,followed by incubation with chromophore-conjugated secondary antibody. Digitalpictures were taken using an Olympus, IX TRINOC camera fitted to an Olympus,IX70 Inverted Research Microscope (Olympus) with objective lenses of HoffmanModulation Contrast®, HMC 10 LWD PL FL, 0.3NA �/1, (OPTICS Inc.) at roomtemperature, and processed using MagnaFire® SP imaging software (Optronics). AZeiss TCs SP2 system was used for confocal imaging.

Alkaline phosphatase (ALP) and von Kossa assaysCells were cultured in osteogenic induction medium (100 nM ascorbic acid, 10 mMglycerophosphate) for osteogenic differentiation. ALP activity and staining weredetermined using the ALP activity assay kit (APF-1KT) and ALP staining kit (86c-1KT) from Sigma according to the manuals provided with the kits. For von Kossastaining, human MSCS cells were cultured in osteoinduction medium with orwithout BMP2 (200 ng/ml). Medium was changed every third or fourth day and themineralization assay was performed at day 24.

Xenopus animal cap assayCapped RNAs encoding endofin (wild-type and mutants) and BMP4 weresynthesized in vitro using the mMessage Machine kit (Ambion). The RNAs werethen injected alone or in combination into both animal poles of two-cell stageembryos. The ectodermal explants (animal caps) from injected embryos weredissected at blastula stages (stage 8.5-9) and incubated until they reached the gastrulastage (stage 11) before total RNA was extracted. Reverse transcription-PCR wasperformed using the primers for different marker genes as described previously(Chang et al., 1997; Chang and Hemmati-Brivanlou, 1999). First-strand cDNAs weresynthesized from 1 �g total RNAs with oligo(dT) priming using Multiscript ReverseTranscriptase (Promega) at 37°C for 1 hour. 1.5 �Ci [�-32P]dATP was added to eachPCR sample, and the PCR products were resolved by electrophoresis on a 5%polyacrylamide gel. As a negative control, PCR was performed with whole embryoRNA that had not been reverse-transcribed to check for DNA contamination.

ReferencesBennett, D. and Alphey, L. (2002). PP1 binds Sara and negatively regulates Dpp

signaling in Drosophila melanogaster. Nat. Genet. 31, 419-423.Chang, C. and Hemmati-Brivanlou, A. (1999). Xenopus GDF6, a new antagonist of

noggin and a partner of BMPs. Development 126, 3347-3357.Chang, C., Wilson, P. A., Mathews, L. S. and Hemmati-Brivanlou, A. (1997). A

Xenopus type I activin receptor mediates mesodermal but not neural specificationduring embryogenesis. Development 124, 827-837.

Cheifetz, S. (1999). BMP receptors in limb and tooth formation. Crit. Rev. Oral Biol.Med. 10, 182-198.

Datto, M. and Wang, X. F. (2005). Ubiquitin-mediated degradation a mechanism forfine-tuning TGF-� signaling. Cell 121, 2-4.

Derynck, R. (1994). TGF-�-receptor-mediated signaling. Trends Biochem. Sci. 19, 548-553.

Di Guglielmo, G. M., Le Roy, C., Goodfellow, A. F. and Wrana, J. L. (2003). Distinctendocytic pathways regulate TGF-� receptor signalling and turnover. Nat. Cell Biol.5, 410-421.

Francis-West, P. H., Parish, J., Lee, K. and Archer, C. W. (1999). BMP/GDF-signallinginteractions during synovial joint development. Cell Tissue Res. 296, 111-119.

Gaullier, J. M., Simonsen, A., D’Arrigo, A., Bremnes, B., Stenmark, H. and Aasland,R. (1998). FYVE fingers bind PtdIns(3)P. Nature 394, 432-433.

Ghosh-Choudhury, N., Harris, M. A., Feng, J. Q., Mundy, G. R. and Harris, S. E.(1994). Expression of the BMP 2 gene during bone cell differentiation. Crit. Rev.Eukaryot. Gene Expr. 4, 345-355.

Godin, R. E., Robertson, E. J. and Dudley, A. T. (1999). Role of BMP family membersduring kidney development. Int. J. Dev. Biol. 43, 405-411.

He, B., Chou, J., Liebermann, D. A., Hoffman, B. and Roizman, B. (1996). Thecarboxyl terminus of the murine MyD116 gene substitutes for the correspondingdomain of the gamma(1)34.5 gene of herpes simplex virus to preclude the prematureshutoff of total protein synthesis in infected human cells. J. Virol. 70, 84-90.

Heldin, C. H., Miyazono, K. and ten Dijke, P. (1997). TGF-� signalling from cellmembrane to nucleus through SMAD proteins. Nature 390, 465-471.

Inman, G. J., Nicolas, F. J. and Hill, C. S. (2002). Nucleocytoplasmic shuttling of Smads2, 3, and 4 permits sensing of TGF-� receptor activity. Mol. Cell 10, 283-294.

Knockaert, M., Sapkota, G., Alarcon, C., Massague, J. and Brivanlou, A. H. (2006).Unique players in the BMP pathway: small C-terminal domain phosphatasesdephosphorylate Smad1 to attenuate BMP signaling. Proc. Natl. Acad. Sci. USA 103,11940-11945.

Kuratomi, G., Komuro, A., Goto, K., Shinozaki, M., Miyazawa, K., Miyazono, K.and Imamura, T. (2005). NEDD4-2 (neural precursor cell expressed, developmentallydown-regulated 4-2) negatively regulates TGF-�(transforming growth factor-�)signalling by inducing ubiquitin-mediated degradation of Smad2 and TGF-� type Ireceptor. Biochem. J. 386, 461-470.

Kutateladze, T. and Overduin, M. (2001). Structural mechanism of endosome dockingby the FYVE domain. Science 291, 1793-1796.

Li, X., Nie, S., Chang, C., Qiu, T. and Cao, X. (2006). Smads oppose Hox transcriptionalactivities. Exp. Cell Res. 312, 854-864.

Lin, X., Duan, X., Liang, Y. Y., Su, Y., Wrighton, K. H., Long, J., Hu, M., Davis, C.M., Wang, J., Brunicardi, F. C. et al. (2006). PPM1A functions as a Smadphosphatase to terminate TGF� signaling. Cell 125, 915-928.

Massague, J. (1998). TGF-� signal transduction. Annu. Rev. Biochem. 67, 753-791.Miura, S., Takeshita, T., Asao, H., Kimura, Y., Murata, K., Sasaki, Y., Hanai, J. I.,

Beppu, H., Tsukazaki, T., Wrana, J. L. et al. (2000). Hgs (Hrs), a FYVE domainprotein, is involved in Smad signaling through cooperation with SARA. Mol. Cell. Biol.20, 9346-9355.

Novoa, I., Zeng, H., Harding, H. P. and Ron, D. (2001). Feedback inhibition of theunfolded protein response by GADD34-mediated dephosphorylation of eIF2�. J. CellBiol. 153, 1011-1022.

Remy, I., Montmarquette, A. and Michnick, S. W. (2004). PKB/Akt modulates TGF-� signalling through a direct interaction with Smad3. Nat. Cell Biol. 6, 358-365.

Seet, L. F. and Hong, W. (2001). Endofin, an endosomal FYVE domain protein. J. Biol.Chem. 276, 42445-42454.

Seet, L. F. and Hong, W. (2005). Endofin recruits clathrin to early endosomes via TOM1.J. Cell Sci. 118, 575-587.

Seet, L. F., Liu, N., Hanson, B. J. and Hong, W. (2004). Endofin recruits TOM1 toendosomes. J. Biol. Chem. 279, 4670-4679.

Shi, W., Sun, C., He, B., Xiong, W., Shi, X., Yao, D. and Cao, X. (2004). GADD34-PP1c recruited by Smad7 dephosphorylates TGF� type I receptor. J. Cell Biol. 164,291-300.

Sui, G., Soohoo, C., Affar, E. B., Gay, F., Shi, Y., Forrester, W. C. and Shi, Y. (2002).A DNA vector-based RNAi technology to suppress gene expression in mammaliancells. Proc. Natl. Acad. Sci. USA 99, 5515-5520.

Tsukazaki, T., Chiang, T. A., Davison, A. F., Attisano, L. and Wrana, J. L. (1998).SARA, a FYVE domain protein that recruits Smad2 to the TGF� receptor. Cell 95,779-791.

Wu, G., Chen, Y. G., Ozdamar, B., Gyuricza, C. A., Chong, P. A., Wrana, J. L.,Massague, J. and Shi, Y. (2000). Structural basis of Smad2 recognition by the Smadanchor for receptor activation. Science 287, 92-97.

Xu, L., Kang, Y., Col, S. and Massague, J. (2002). Smad2 nucleocytoplasmic shuttlingby nucleoporins CAN/Nup214 and Nup153 feeds TGF� signaling complexes in thecytoplasm and nucleus. Mol. Cell 10, 271-282.

Jour

nal o

f Cel

l Sci

ence