Two highly related regulatory subunits of PP2A exert opposite effects on TGF-β/Activin/Nodal signalling

TGF-β superfamily ligands signal through complexes of type I and type II receptors, both of which are serine/threonine kinases. Binding of a TGF-β superfamily ligand allows the type II receptor kinase to phosphorylate and activate the type I receptor kinase, which in turn phosphorylates receptor-regulated Smads or R-Smads. Phosphorylated R-Smads form complexes with Smad4, which accumulate in the nucleus and directly regulate target gene transcription (ten Dijke and Hill, 2004). The pathway consists of two distinct branches, which are defined by the type I receptors and the R-Smads that are activated. The type I receptors ALK4, ALK5 and ALK7 specifically phosphorylate Smad2 and Smad3, whereas the type I receptors ALK1, ALK2, ALK3 and ALK6 are specific for Smad1, Smad5 and Smad8(Schmierer and Hill, 2007). Broadly speaking, the Activin, Nodal and TGF-β subfamilies of ligands induce phosphorylation and activation of ALK4/5/7 and thus Smad2/3, whereas ligands of the BMP subfamily activate ALK1/2/3/6 and consequently Smad1/5/8. These two branches are frequently referred to as the TGF-β/Activin/Nodal branch and the BMP/GDF branch, respectively(Feng and Derynck, 2005).

Receptor levels are an obvious determinant of the responsiveness of a cell to TGF-β superfamily ligands and are extensively regulated. Irrespective of the absence or presence of a signal, endocytosis of receptors by a clathrin-dependent mechanism operates in parallel with a caveolin-dependent mechanism, the former recycling receptors to the membrane, the latter promoting receptor degradation through the proteasomal pathway(Di Guglielmo et al., 2003). A balance between the two is thought to determine the amount of receptors that are present at the plasma membrane and thus competent for signalling. Lysosomal degradation of ALK4 and ALK5 also occurs and is promoted by the protein Dapper2 in zebrafish (Zhang et al., 2004), and degradation of BMP receptors in Xenopusis promoted by a phosphatase, Dullard(Satow et al., 2006).

In addition to receptor levels, the phosphorylation status of both the receptors and the Smads is also tightly controlled. Thus, protein phosphatases have long been postulated to influence TGF-β superfamily signalling, but concrete roles have only recently started to emerge. PP1 is targeted to active receptors by signal-induced feedback to dephosphorylate and inactivate the type I receptor (Shi et al.,2004) in both mammalian cells and Drosophila(Bennett and Alphey, 2002). Downstream of the receptors, a number of different phosphatases have been implicated in the removal of activating phosphates from the R-Smads(Chen et al., 2006; Knockaert et al., 2006; Lin et al., 2006). Finally,PP2A, a regulatory subunit of which can be phosphorylated by ALK5, has been implicated in the TGF-β signalling pathway as a downstream effector(Griswold-Prenner et al.,1998; Petritsch et al.,2000).

Here, we report an entirely novel role for specific regulatory subunits of PP2A in modulating TGF-β/Activin/Nodal signalling at the receptor level. PP2A is a multimeric serine/threonine protein phosphatase consisting of a 36 kDa catalytic subunit (PP2A_C) and a 65 kDa scaffolding subunit(PR65 or A subunit) (Janssens et al.,2005). An additional regulatory subunit, of which four distinct classes exist (B, B′, B″ and B‴), associates with this dimer. The B family (PR55) comprises four highly homologous, mammalian genes(PPP2R2A, PPP2R2B, PPP2R2C and PPP2R2D) with PPP2R2A and PPP2R2D being widely expressed, and PPP2R2B and PPP2R2C expression being restricted to neural tissues (Janssens and Goris,2001; Strack et al.,1999). To date, only redundant functions of these family members have been described (Adams et al.,2005). Here, we show that PPP2R2A (referred to throughout as Bα) and PPP2R2D (referred to throughout as Bδ) have distinct and opposing roles in the regulation of TGF-β/Activin/Nodal signalling.

Full-length mouse Bα, Bδ, B′δ and XenopusBδ were PCR-amplified from EST clones, sequence checked and cloned into pEF-Flag and pFTX9 (Howell et al.,2002) for N-terminal Flag tagging. Anti Xenopus Bαand Bδ in situ hybridisation probes corresponding to nucleotides 1988-2176 and 9-194 of the full-length mRNA, respectively, were generated by PCR from Xenopus cDNA libraries and cloned into pCR2.1 (Invitrogen). Xbra and gsc probes(Howell et al., 2002),pCS2-GFP, pFTX4K-EGFPSmad2 and the plasmid expressing activated ALK4(Batut et al., 2007), the plasmid expressing HA-ALK4 (Jullien and Gurdon, 2005), pCMV5-HA-ALK5 and pCMV-HA ALK5^ca(Nicolás and Hill,2003) and EF-LacZ (Bardwell and Treisman, 1994) were as described. HA-Raf1 was a kind gift from Richard Marais. Recombinant full-length human Smad2 was purified from bacterial lysates as a GST fusion by affinity purification followed by thrombin cleavage. Concentration and quality of the recombinant Smad2 were checked by SDS-PAGE and Coomassie staining. In vitro translations using the TnT system (Promega) were performed according to manufacturer's instructions.

Extraction of total mRNA from Xenopus embryos, reverse transcription and q-PCR were performed as described(Batut et al., 2007). RT-PCR was performed as described (Levy and Hill,2005). Details of the sequences of the morpholino oligonucleotides(Gene Tools) and the oligonucleotides used for RT-PCR and q-PCR can be provided on request.

HeLa TK- and HaCaT cell lines expressing EGFP-Smad2 were generated and cultured as previously described (Nicolas et al., 2004; Schmierer and Hill, 2005). TGF-β (PeproTech) was used at 2 ng/ml,Bafilomycin A1 (Calbiochem) at 10 nM, MG132 and lactacystin (Sigma) at 25μM. Treatment of extracts with PNGase F was as described(Dorey and Hill, 2006). Cells were transfected with plasmids using Lipofectamine 2000 (Invitrogen) or Fugene HD (Roche), and with siRNAs using Dharmafect II (Dharmacon). Pools of four siRNA oligos (SMARTpools, Dharmacon) targeting Bα, Bδ,Dapper-2 or TβR-II were transfected at a final concentration of 75 nM. Experiments were performed 72 hours after the transfection. As controls, a SMARTpool of non-targeting siRNAs were used. Knockdowns using the individual oligonucleotides of a SMARTpool were performed as above at a final concentration of 75 nM. siRNA sequences can be provided on request. Knockdown efficiency was assessed by RT-PCR and/or immunoblotting.

As substrates, endogenous and EGFP-tagged phospho-Smad2 were immunoprecipitated from TGF-β-treated HaCaT EGFPSmad2 cells(Batut et al., 2007; Schmierer and Hill, 2005) with an anti-Smad2/3 antibody in lysis buffer [50 mM Tris-Cl (pH 7.5), 100 mM NaCl,10% glycerol, 0.1% NP40, 2 mM DTT, protease inhibitors). HA-Raf-1 was expressed in HeLa TK- cells and immunoprecipitated with anti-HA antibody. Flag-Bα, Flag-Bδ or Flag-B′δ was expressed in HeLa TK-cells. PP2A holocomplexes containing these subunits were purified by Flag pulldown and eluted with Flag peptide. Alternatively, complexes were purified from stable HEK T-Rex cell lines harbouring Flag-Bα or FLAG-Bδ in the tetracycline-inducible pcDNA5/TO construct(Adams et al., 2005) induced with 1 μg/ml tetracycline for 48 hours, and treated, or not, for 1 hour with TGF-β. Phosphatase activity in the eluates was routinely assessed by phosphate release from a synthetic phosphopeptide using a colorimetric assay(Upstate). Phospho-Smad2 or HA-Raf-1 bound to protein G beads were incubated with equal phosphatase activities for 1 hour at 37°C in lysis buffer and analysed for dephosphorylation by immunoblotting.

Active, endogenous ALK5 receptor kinase was immunoprecipitated from TGF-β treated HaCaT cells lysed in lysis buffer using anti-ALK5 antibodies that had been chemically crosslinked(Harlow and Lane, 1988) to protein A beads. Kinase activity was assessed by in vitro phosphorylation of 800 ng recombinant human Smad2 protein in the presence of 5 mM ATP, 2 mM MgCl₂ and 2 mM MnCl₂ at 37°C for 90 minutes. For IP kinase phosphatase assays, purified PP2A holocomplexes containing Flag-Bα, Flag-Bδ or Flag-B′δ were included in the reaction mixture.

Fertilisation, culture, staging, preparation of synthetic mRNAs and microinjection of Xenopus embryos were performed as described(Howell et al., 2002). Activin A (R&D Systems) was used at 20 ng/ml. Okadaic acid (Calbiochem) was used at 25 nM. In vitro transcribed mRNAs were injected into Xenopus at 500 pg for EGFP-Smad2, 250 pg for HA-ALK4, 100 pg for activated ALK4 and 200-500 pg for Bα and Bδ. GFP mRNA was used as a tracer at 50 pg. Total concentrations of morpholinos were 20 ng. In all cases, the injection volume was 4-5 nl. Animal caps were dissected at stage 8-9 and harvested at the indicated stages. Whole-mount in situ hybridisation and immunofluorescence were performed as described (Batut et al.,2007). Antisense probes for in situ hybridisation were labelled with digoxigenin-UTP (Roche).

The following fly strains were used: (1) w P{GAL4}A9(Brand and Perrimon, 1993); (2) w; P{UAS-tws.B}23 (II) (Bajpai et al., 2004); (3) w; +/CyO; P{UAS-dSmad2.Z}8D3 (III)(Zheng et al., 2003); (4) y[1] w[1118]; Sp; +/TSTL, CyO, TM6B, Tb[1]; (5) tws[60]/TM6b, Tb Hu (Uemura et al., 1993);and (6) w[67c23] P{w[+mC]=lacW}Smox[G0348]/FM7c(Peter et al., 2002).

Flies were reared on cornmeal-agar-dextrose. Loss-of-function interactions were tested at 25°C. smox/FM7a females were mated to a loss-of-function tws strain or to y[1] w[c67c23]. Progeny that were smox/Y; tws/+ had no difference in viability or visible phenotypes compared with control smox/Y. Gain-of-function interaction tests were performed at 22°C and 25°C in males, which had stronger tws overexpression phenotypes. Wings were mounted in Euparal and imaged digitally using a 4× objective. It has previously been reported that larger wings result from A9-GAL4-driven expression of P{UAS-Smox.A} (Marquez et al.,2001). The genotype used here produced only a subtle increase in the average wing dimensions compared with the control genotype. Control matings to produce w P{GAL4}A9/Y; P{UAS-tws.B}23/+ were performed at 23-24°C to permit recovery of mature wings.

Whole-cell extracts from Xenopus embryos and tissue culture cells,Western blotting and immunoprecipitation procedures were as described(Batut et al., 2007; Howell et al., 1999). Confocal microscopy was performed as described(Batut et al., 2007; Schmierer and Hill, 2005). The following commercial antibodies were used: anti-phospho-Smad2 (S465/S467),anti-phospho-Smad2 (S245/S250/S255), anti-phospho-Raf-1 (S259),anti-phospho-ERK (all Cell Signaling Technology); anti-Smad2/3 (BD Biosciences Pharmingen); anti-ALK5 (v-22, Santa Cruz Biotechnology); anti-TβRII,anti-pan B subunit and anti-PP2A catalytic subunit (Upstate);anti-β-Catenin, anti-Flag, anti-Flag-HRP and anti-Flag beads (Sigma);anti-HA, anti-HA-HRP and anti-GFP (Roche); anti-α-tubulin (YL1/2,Abcam).

We initially identified the PP2A Bδ subunit as a protein whose overexpression in early Xenopus embryos causes loss of anterior structures. Embryos overexpressing Bδ exhibited a delayed gastrulation,as judged by the time of blastopore closure, and showed greatly reduced anterior structures at the tailbud stage(Fig. 1A; see Table S1 in the supplementary material). By contrast, knockdown of Bδ using a specific morpholino resulted in embryos with a shortened axis compared with wild-type embryos and slightly larger anterior structures relative to the trunk(Fig. 1A; see Table S1 in the supplementary material). The morphant phenotype was identical when two distinct morpholino oligonucleotides were used (data not shown), and was rescued by co-expression of a mouse Bδ mRNA(Fig. 1B).

To understand how specific these effects were for this particular B subunit, we investigated the effects of manipulating levels of the highly homologous Bα, which we found also to be expressed in early Xenopus embryos in a similar pattern to Bδ (see Fig. S1 in the supplementary material). Surprisingly, morpholino knockdown of Bαresulted in a very different phenotype to the Bδ knockdown. The embryos exhibited a short anterior-posterior axis, but in this case anterior structures were much reduced (Fig. 1A; see Table S1 in the supplementary material). In fact, the phenotype was similar to that caused by overexpression of Bδ. The effect of Bα knockdown was rescued by overexpression of mouse Bα(Fig. 1B). Embryos overexpressing Bα were phenotypically normal(Fig. 1B), perhaps because Bα levels are not limiting in the early embryo. Consistent with the distinct phenotypes resulting from knockdown of Bα and Bδ,overexpression of Bα could not rescue Bδ morphant embryos and overexpression of Bδ could not rescue Bα morphant embryos (see Fig. S2 in the supplementary material). We thus conclude that these two regulatory B subunits have distinct functions in the early Xenopusembryo.

The phenotype of Bα morphant and Bδ-overexpressing embryos was similar to the phenotypes of various Nodal signalling mutants in fish(Schier, 2001), and also to phenotypes of fish and frog embryos in which Nodal signalling had been inhibited by the pharmacological type I receptor inhibitor SB-431542(Batut et al., 2007; Ho et al., 2006; Sun et al., 2006). Conversely,the phenotype of Bδ morphant embryos was consistent with increased Activin/Nodal and/or Wnt signalling (Tada et al., 2002; Whitman,2001). To investigate this in more detail, we examined the effect of manipulating the levels of Bα and Bδ on the expression of the Activin/Nodal target genes gsc and Xbra(Howell et al., 2002). Both in situ hybridisation and quantitative PCR revealed that either overexpression of Bδ or knockdown of Bα greatly inhibited the expression of gsc and Xbra in early gastrula embryos(Fig. 2). By contrast,knockdown of Bδ increased the expression of these genes(Fig. 2). We also analysedβ-catenin localisation as a readout for Wnt activity, but found no evidence for enhanced Wnt activity in Bδ morphant embryos or reduced Wnt activity in Bα morphant and Bδ-overexpressing embryos (see Fig. 5A). Thus, the observed phenotypes are most probably due to a modulation in the intensity of Nodal signalling, with knockdown of Bδ promoting Nodal signalling and knockdown of Bα or overexpression of Bδ inhibiting Nodal signalling.

To corroborate our data from whole embryos, we investigated the effects of Bα and Bδ on Activin-dependent animal cap elongation, which is a functional readout for Activin/Nodal activity(Smith, 1993). Animal cap explants cultured in buffer alone heal into balls of ciliated epidermis,whereas those treated with Activin elongate and differentiate to form mesoderm. Activin-induced elongation of animal cap explants was completely abrogated by overexpression of Bδ, but not of Bα(Fig. 3A). Conversely,morpholino knockdown of Bδ enhanced cap elongation, whereas knockdown of Bα completely abolished elongation(Fig. 3B). Importantly, the effects of Bα knockdown could be rescued by overexpression of an EGFP-tagged version of Smad2, which is the intracellular mediator of the Activin/Nodal pathway (Fig. 3C). As a control, overexpression of EGFP-Smad2 had no effect on Activin-induced animal cap elongation by itself(Fig. 3C).

Altogether, these results indicate that altering the expression levels of Bα or Bδ in the early embryo modulates the Activin/Nodal signalling pathway. Importantly, Bα and Bδ affect the strength of Activin/Nodal signalling in Xenopus in opposite directions, with Bα normally acting positively and Bδ acting negatively.

We next extended our analysis of the role of the B family of PP2A regulatory subunits in Activin/Nodal signalling by analysing the genetic interaction between twins and smox in Drosophila. Twins is the only Drosophila B family member(Mayer-Jaekel et al., 1993)and Smox is Drosophila Smad2(Henderson and Andrew, 1998),which transduces signals from the Activin type I receptor Baboon(Brummel et al., 1999; Parker et al., 2006; Serpe and O'Connor, 2006). Overexpression tests yielded a genetic interaction between smox and twins, whereas no trans-heterozygous interaction was evident with loss-of-function mutations (see Materials and methods). Overexpression of Twins alone throughout the developing wing primordium with the UAS-Gal4 binary expression system (Brand and Perrimon,1993) yielded very small wings with minimal venation in males,which had reduced survival (Fig. 4B); control wings that lacked the A9-Gal4 expression driver were wild type in phenotype (Fig. 4A). Overexpression of Smox alone yielded wings with abnormal venation (n>40, Fig. 4C). When the two transgenes were co-expressed, however, 24 out of 30 wings were completely suppressed to wild-type shape and venation; six wings had partially suppressed phenotypes (see Fig. S3 in the supplementary material). Co-expression of the Baboon A isoform with Twins also suppressed,but not as well as Smox (see Fig. S3 in the supplementary material). In summary, the effects of Twins overexpression on wing size and vein structure were strongly or completely suppressed by increased levels of Smox/dSmad2 in 93% of cases, suggesting that Twins antagonises Activin/Smad2 signalling in the wing primordium. In Drosophila therefore, the only B subunit Twins acts similarly to Bδ in Xenopus.

We next analysed in detail at what point in the TGF-β/Activin/Nodal pathway these phosphatase subunits acted. TGFβ/Nodal/Activin stimulation leads to C-terminal phosphorylation of Smad2 (and Smad3), which then accumulate in the nucleus (Massaguéet al., 2005). At early gastrula stages, endogenous Nodal signalling in Xenopus embryos is stronger dorsally(Lee et al., 2001) as seen by nuclear localisation of Smad2 on the dorsal, but not on the ventral side(Fig. 5A, compare parts a and c). Knockdown of Bα or overexpression of Bδ suppressed this Smad2 nuclear accumulation on the dorsal side, whereas knockdown of Bδpermitted Smad2 nuclear accumulation even on the ventral side(Fig. 5A, top row). The effects were specific to Nodal signalling as levels of nuclear β-catenin, which reflect active Wnt signalling, were not similarly affected(Fig. 5A, middle row). In animal cap explants expressing a constitutively active version of the Activin/Nodal type I receptor ALK4 to mimic Nodal signalling(Wieser et al., 1995),overexpression of Bδ prevented nuclear accumulation of EGFP-Smad2 (see Fig. S4A in the supplementary material). Taken together, these results indicate that Bα and Bδ act on Activin/Nodal signalling downstream of the ligands.

TGF-β/Activin/Nodal-induced nuclear accumulation of Smad2 is driven by Smad2 C-terminal phosphorylation(Massagué et al.,2005). In accordance with the observed changes in Smad2 nuclear accumulation, we found that overexpression of Bδ caused a decrease in Smad2 phosphorylation in whole embryos in response to endogenous Nodal signalling, whereas overexpression of Bα had no effect(Fig. 5B). The same was true in animal caps in response to exogenous Activin (see Fig. S5 in the supplementary material; data not shown). In knockdown experiments, Smad2 phosphorylation was increased by Bδ knockdown and decreased by Bα knockdown(Fig. 5C). The effect of Bδ knockdown could be mimicked by a 1-hour treatment of animal caps with the PP2A catalytic subunit inhibitor, okadaic acid(Fig. 5D), suggesting that the action of Bδ on TGF-β/Activin/Nodal signalling requires the phosphatase activity of PP2A (see also Discussion). Previous work has suggested that B family members have cell type-specific effects on the ERK MAPK pathway (Adams et al.,2005; Strack,2002; Van Kanegan et al.,2005). However in Xenopus embryos, knockdown of Bδor Bα or their overexpression had no effect on ERK phosphorylation in response to endogenous receptor tyrosine kinase signalling(Fig. 5B,C).

The observed effects on Smad2 phosphorylation and nuclear accumulation are not confined to Xenopus embryos but are conserved in mammalian systems. We used siRNAs to knock down Bα and Bδ in a HeLa cell line stably expressing EGFP-Smad2. As in Xenopus, knockdown of Bα decreased the nuclear accumulation of Smad2 in response to TGF-βin these cells (Fig. 5E) and inhibited Smad2 phosphorylation (Fig. 5F). By contrast, knockdown of Bδ enhanced Smad2 nuclear accumulation relative to control cells(Fig. 5E) and increased Smad2 phosphorylation (Fig. 5F). These effects were observable in a number of different human and mouse cell lines (data not shown) and were specific, as several different individual siRNA oligonucleotides which were proven to specifically knock down Bαor Bδ (see Fig. S6A,B in the supplementary material), elicited the same effect (see Fig. S6B,C in the supplementary material). Consistent with the decrease in TGF-β-induced Smad2 phosphorylation caused by Bαknockdown, TGF-β was also less effective at mediating growth arrest in these conditions (see Fig. S7 in the supplementary material). Overexpression of either Bα or Bδ had little effect on the level of phosphorylated Smad2 in tissue culture cells (data not shown), perhaps because their levels are not limiting or because free subunits that are not incorporated into holocomplexes, are unstable(Strack et al., 2002).

In conclusion, Bα and Bδ modulate the level of active phosphorylated Smad2 and hence its nuclear accumulation in both Xenopus embryos and tissue culture cells, with Bα normally promoting Smad2 phosphorylation and Bδ inhibiting Smad2 phosphorylation.

The simplest hypothesis to explain the effects on Smad2 phosphorylation was that Bδ acted directly on the C-terminal phosphates of activated Smad2 and that Bα negated this action. We therefore immunopurified heterotrimeric active PP2A complexes containing either Flag-tagged Bα or Bδ or a distinct regulatory subunit, B′δ(Janssens and Goris, 2001),together with the catalytic and scaffolding subunits(Fig. 6A-C; data not shown). These complexes were all active as they dephosphorylated a control phospho-peptide (Fig. 6C); the PP2A complexes containing either Bα or Bδ, but not complexes containing B′δ, efficiently dephosphorylated an immunopurified Raf substrate (Fig. 6D), as previously reported (Adams et al.,2005). However, we could not detect any dephosphorylation of an immunopurified C-terminally phosphorylated Smad2 substrate by any of the phosphatase complexes (Fig. 6D). This was also true for PP2A holocomplexes purified from TGF-β-induced cells (Fig. 6E,F). Furthermore, we could not detect significant dephosphorylation of a number of phosphorylated residues within the Smad2 linker region (Kretzschmar et al.,1999) (Fig. 6G). Thus, neither Bα nor Bδ seems to act directly on phosphorylated Smad2.

As Bα and Bδ do not act on Smad2 directly, but do affect Smad2 phosphorylation in vivo, we asked whether they could regulate the activity of the TGF-β receptor complex. The TGF-β receptor complex comprises two type II receptors (TβR-II) and two type I receptors (ALK5). ALK5 activity requires its phosphorylation by the constitutively active type II receptor,and thus we reasoned that dephosphorylation of ALK5 by a PP2A complex in vitro would reduce ALK5 activity. In an in vitro kinase assay (see Fig. 7A for the experimental scheme), immunopurified active receptor complexes from TGF-β-induced cells phosphorylated recombinant Smad2 at its C terminus(Fig. 7B, lane C). However,incubation of receptor and substrate with either phosphatase complex had no significant effect on the ability of receptors to phosphorylate Smad2(Fig. 7B, lanes Bα,Bδ, B′δ). Thus, in vitro, neither phosphatase subunit affected receptor kinase activity.

We therefore investigated whether Bα or Bδ affected receptor activity in vivo. We have recently shown that in Xenopus ALK4 receptor clustering provides a convenient readout of receptor activity in vivo(Batut et al., 2007) as it is induced in response to ligand and requires ALK4 kinase activity (see Fig. S8 in the supplementary material). In untreated animal caps, which exhibit very low levels of Activin/Nodal signalling, morpholino knockdown of Bδ or inhibition of PP2A catalytic activity by okadaic acid, was sufficient to induce ALK4 clustering (Fig. 7C, compare untreated animal caps). Thus, a reduction of Bδactivity lowers the threshold of ligand required for ALK4 signalling,indicating that Bδ normally restricts receptor activity and might act to suppress signalling at very low ligand concentrations.

In the course of investigating whether knockdown of Bα would inhibit Activin/Nodal-induced ALK4 clustering, we noticed that levels of HA-ALK4 were extremely low when Bα was depleted(Fig. 7D). Moreover,overexpression of Bδ, which has the same functional effects as Bαknockdown, similarly reduced HA-ALK4 levels(Fig. 7D).

These data were corroborated in HaCaT cells, where knockdown of Bαled to a strong decrease in levels of endogenous ALK5 protein(Fig. 7E). This effect was specific, as knockdown of Bα or Bδ did not affect endogenous TβR-II levels (Fig. 7F). We conclude that the effect on receptor levels is post-transcriptional in both model systems. In tissue culture cells, knockdown of Bα had no effect on endogenous ALK5 mRNA levels (see Fig. S9A in the supplementary material) and also substantially reduced protein levels of exogenously expressed HA-tagged ALK5 (see Fig. S9B in the supplementary material). Similarly in Xenopus embryos, the effects of Bα knockdown or Bδoverexpression on HA-ALK4 are not at the level of transcription as the receptor is expressed from an injected synthetic mRNA. In principle, Bαcould affect translation of the type I receptors, or their stability. We favour the latter possibility as the mRNAs used in both systems have no 3′ or 5′ UTRs, which are usually required for translational regulation. Furthermore, we could detect a weak interaction between HA-ALK4 and Flag-tagged Bα in a co-immunoprecipitation (see Fig. S10 in the supplementary material) as has been reported previously for ALK5(Griswold-Prenner et al.,1998), suggesting that Bα might affect ALK4 and ALK5 at the protein level.

To identify the pathway of degradation, we asked whether inhibitors of the lysosome (bafilomycin A1) or the proteasome (lactacystin or MG132) could rescue the effects of Bα knockdown in tissue culture cells. We found that in HeLa cells overexpressing ALK5, the effects of Bα knockdown were rescued by bafilomycin A1, but not MG132 (see Fig. S11A in the supplementary material). Similarly in HaCaT cells, a partial rescue of endogenous ALK5 levels was observed with bafilomycin A treatment, but not by treatment with lactacystin or MG132 (see Fig. S11B in the supplementary material).

Taken together, these data suggest that knockdown of Bα (and in Xenopus, also overexpression of Bδ) promotes degradation of the type I receptors ALK4 and ALK5 via a lysosomal pathway. We speculated that PP2A might regulate Dapper2, which is known to promote lysosomal degradation of ALK4 and ALK5 (Zhang et al.,2004). However, knockdown of Dapper2 did not ameliorate the decrease in ALK5 levels resulting from Bα knock-down, even though Dapper2 knockdown alone clearly raised levels of ALK5 (see Fig. S11C in the supplementary material). Thus, it is more likely that Dapper2 regulates PP2A or that the two act independently.

It has long been speculated that the individual, yet highly homologous,members of each subfamily of PP2A regulatory subunits would have unique and specific roles in addition to their established redundant activities. We have demonstrated here for the first time such non-redundant functions of two members of the B subfamily of PP2A regulatory subunits. We have shown that Bα and Bδ perform important functions in modulating the intensity of TGF-β/Activin/Nodal signalling in different species by stabilising basal levels of the ALK4 and ALK5 receptors (Bα) and by restricting receptor activity (Bδ) (Fig. 7G). Perturbation of these mechanisms has dramatic functional consequences in vivo, as demonstrated by altered gene expression and serious developmental defects.

We do not yet know the exact mechanism whereby Bα and Bδregulate receptor levels and activity, respectively. Nevertheless, it is clear from our knockdown experiments that each subunit affects a separate and distinct aspect of receptor biology, ruling out the possibility that Bαand Bδ merely compete with each other for catalytic and scaffolding subunits, so that knockdown of one B subunit increases the levels of complexes containing the other B subunit. The results of our loss-of-function experiments also exclude the possibility that Bα and Bδ have opposing activities on a single common substrate. However, it is likely that subunit competition within the PP2A holoenzyme does explain why Bδoverexpression mimics Bα knockdown (both manipulations causing ALK4 destabilisation in Xenopus embryos). Bδ has a higher affinity for the catalytic subunit than Bα(Fig. 6C,E). Overexpressed Bδ is thus more likely to compete out endogenous Bα than vice versa, and Bδ overexpression would mimic Bα knockdown, whereas Bα overexpression would not necessarily have an effect, as we observe.

We have shown that the PP2A catalytic inhibitor, okadaic acid, can mimic the effects of Bδ knockdown, suggesting that Bδ functions as part of a PP2A holoenzyme complex. However, as okadaic acid is not absolutely specific for PP2A, a mechanism independent of the PP2A catalytic subunit cannot be definitively ruled out. We have shown that knockdown of Bδpromotes receptor clustering at low endogenous ligand concentrations,suggesting that Bδ normally inhibits receptor clustering and thus receptor activation at sub-threshold levels of ligand. Whether it does this by removing (in the context of a PP2A holoenzyme complex) an activating phosphate from serine/threonine residues in the receptors themselves, or via dephosphorylation of another component remains to be investigated.

In contrast to the role ascribed to Bδ, our data demonstrate that Bα regulates the levels of the type I receptors ALK4 and ALK5, most probably by stabilising the receptors and preventing their degradation via the lysosomal pathway. The involvement of phosphatases in protein turnover is not unprecedented. The protein phosphatase Dullard has recently been reported to promote the degradation of BMP type II receptors in Xenopus, thus repressing BMP-dependent phosphorylation of the BMP type I receptor(Satow et al., 2006). More specifically to PP2A, the cycling of the protein Period in Drosophilais dependent upon the activity of Twins and loss of PP2A activity reduces Period expression (Sathyanarayanan et al.,2004). In this case, Twins acts in the context of a PP2A holoenzyme to dephosphorylate Period directly. Moreover, Twins also affects the levels of Armadillo, the Drosophila β-Catenin homologue, as it is required for stabilisation of Armadillo in response to Wingless signalling (Bajpai et al.,2004). Consistent with our demonstration that DrosophilaTwins acts in the same way as vertebrate Bδ, we have observed a reduction in levels of nuclear β-Catenin in Bδ morphant embryos(Fig. 5A). Importantly,however, this effect of Bδ on the Wnt signalling pathway cannot explain the phenotypes we observe in Bδ morphant embryos. In our present study it is not yet clear whether Bα stabilises ALK4 and ALK5 by acting to dephosphorylate the receptors themselves, or whether it acts indirectly. Consistent with a previous report(Griswold-Prenner et al.,1998), we could detect a weak interaction between Bα and ALK4, suggesting that it may act on the receptor directly, or possibly on another component of the receptor complex. We find that the effect of Bαon type I receptor levels occurs in the absence of signalling, indicating that Bα is required for regulating the basal levels of receptor and not responsible for downregulating receptor levels after signal transduction.

In summary, we conclude that Bα- and Bδ-containing phosphatase complexes have distinct substrates, the net effects of which on the TGF-β/Activin/Nodal pathway are opposite. This strongly suggests that the ratio of Bα to Bδ in a particular cell will influence the threshold response to TGF-β/Activin/Nodal ligands, which will in turn determine the levels of target gene transcription and thus developmental programmes.

We thank L. S. Shashidhara, C. Doe, K. Irvine, H. Jaeckle, M. B. O'Connor,U. Schaefer, B. Pellock and the Bloomington Stock Center for fly strains; R. Marais for the HA-Raf1 construct; and B. Wadzinski for the Tet inducible HEK cell lines. We are grateful to Sally Leevers, Peter Parker and members of the Hill laboratory for helpful discussions and comments on the manuscript. The work was funded by Cancer Research UK, by the NIH (grant GM60501 to L.A.R.),by an Erwin Schrödinger Fellowship of the Austrian Science Foundation(J2397-B12 to B.S.) and by an EU Marie Curie Fellowship (515294 to B.S.).