A GPI processing phospholipase A2, PGAP6, modulates Nodal signaling in embryos by shedding CRIPTO

Glycosylphosphatidylinositol (GPI) is a glycolipid widely used among eukaryotes as membrane anchors for many cell surface proteins (Ferguson et al., 2009). GPI is biosynthesized and transferred to proteins in the ER. After GPI attachment to proteins, remodeling of GPI moieties on GPI-anchored proteins (GPI-APs) occurs during transport (Tanaka et al., 2004; Tashima et al., 2006; Maeda et al., 2007; Fujita et al., 2009). Those remodeling reactions confer unique features on GPI-APs, including lipid raft association on the membrane (Maeda et al., 2007). Another feature of GPI-APs is their cleavage within the GPI moiety by GPI-cleaving enzymes and shedding from the membrane (Fujihara and Ikawa, 2016). The number of proven examples of GPI-cleaving enzyme–mediated GPI-AP shedding under physiological conditions is small (Fujihara and Ikawa, 2016). Theoretically, the shedding of GPI-APs has two biological modes of action: (1) shed GPI-APs act at sites remote from the original cells and affect the fate of other cells; and (2) the initiation of specific activities suppressed by inhibitory GPI-APs on the same cells (Fujihara and Ikawa, 2016). There are two examples of the latter mode; shedding of TEX101 from sperm by testis type of angiotensin-converting enzyme is involved in sperm maturation (Kondoh et al., 2005; Fujihara et al., 2013b) and shedding of a metalloprotease inhibitor RECK by glycerophosphodiester phosphodiesterase 2, resulting in metalloprotease-mediated degradation of a Notch ligand DLL1, which, in turn, reduces Notch signaling in adjacent progenitor cells to induce differentiation into neurons (Park et al., 2013). A clear example of the former mode has not been demonstrated (Fujihara and Ikawa, 2016).

During embryonic development, Nodal signaling is required for many aspects, including anterior–posterior axis patterning, mesoderm induction, and left–right axis specification (Tian and Meng, 2006; Shen, 2007). CRIPTO, a GPI-AP (Minchiotti et al., 2000), forms a complex with type I and II activin receptors for Nodal on the membrane and induces cell-autonomous signaling (Yan et al., 2002). In addition, a biologically active, soluble form of CRIPTO is generated by GPI cleavage (Watanabe et al., 2007). The soluble form of CRIPTO is required for non–cell autonomous CRIPTO-Nodal signaling in mammalian cells (Yan et al., 2002; Parisi et al., 2003), which is critical for axial mesendoderm formation (Chu et al., 2005). However, molecular mechanisms of CRIPTO shedding have not been clarified (Minchiotti et al., 2001; Yan et al., 2002).

We previously reported that post-glycosylphosphatidylinositol attachment to proteins 3 (PGAP3) is a Golgi-resident, GPI-specific phospholipase A2 (GPI-PLA2) involved in fatty acid remodeling of GPI-APs (Fujita et al., 2006; Maeda et al., 2007). In the remodeling process, PGAP3 is required for the removal of an unsaturated fatty acid from the sn-2 position. A bioinformatics approach revealed that PGAP3, Per1p (a yeast homologue of PGAP3), and alkaline ceramidase belong to a membrane-bound hydrolase superfamily, termed CREST (alkaline ceramidase, PAQR receptor, Per1, SID-1, and TMEM8) (Pei et al., 2011). In this study, we identified an uncharacterized gene, TMEM8A, here renamed PGAP6, which has close similarity to PGAP3 in the superfamily members. Our data show that PGAP6 is a GPI-PLA2 expressed on the cell surface, sheds CRIPTO as an active Nodal coreceptor, and is critical for anterior–posterior axis formation in embryonic development through modulating CRIPTO-Nodal signaling.

CREST members share seven predicted transmembrane segments containing five conserved residues (three His, Asp, and Ser). TMEM8 family proteins, which are located near the PGAP3 gene cluster, are conserved in vertebrates and include three members (TMEM8A, B, and C) in mammals (Pei et al., 2011). Among them, TMEM8A has conserved putative catalytic amino acids within transmembrane domains at the C-terminal regions (Fig. 1 A). In addition, TMEM8A has an unannotated N-terminal region and an EGF-like domain after the signal sequence for ER insertion. TMEM8A was detected by anti-TMEM8A mAb on human embryonic carcinoma NTERA2 cells and HEK293T cells by flow cytometry, indicating that TMEM8A, unlike Golgi-resident PGAP3, is expressed on the cell surface (Fig. 1 B). To further characterize TMEM8A, TMEM8A or HA-tagged TMEM8A (HA-TMEM8A) was expressed in CHO 3B2A cells. Nontagged TMEM8A was detected on the cell surface (Fig. 1 C). HA-TMEM8A was also detected on the cell surface (Fig. 1 D) and appeared in 120- and 90-kD bands (Fig. 1 E, lane NT). The 120-kD band was sensitive to PNGase and resistant to EndoH, indicating that it represents a mature cell surface TMEM8A protein (Fig. 1 E, lanes EnH and PNG). The 90-kD band was sensitive to both PNGase and EndoH, indicating that it has immature N-glycans. Subcellular fractionation and microscopic observation confirmed that the 120-kD protein is localized in the plasma membrane (PM) and that the 90-kD protein is localized in the ER (Fig. 1, F and G). The ER form might be an intermediate during transport.

We tested whether TMEM8A is involved in GPI-AP processing using CHO 3B2A cells overexpressing HA-TMEM8A. TMEM8A significantly reduced amounts of two GPI-APs, CD59 and decay-accelerating factor (DAF), whereas another GPI-AP urokinase-type plasminogen activator receptor (uPAR) was only slightly decreased (Fig. 2 A). Because TMEM8A has similarity to PGAP3 and decreased the levels of GPI-APs, we renamed it PGAP6 (for post-GPI attachment to proteins 6) as a protein involved in the processing of GPI-APs.

Because different GPI-APs showed different sensitivity to PGAP6, we searched for a sensitive substrate GPI-AP, especially among those known to be released from the cell surface, and found CRIPTO is highly sensitive (Fig. 2 B). Strikingly, CRYPTIC, another GPI-AP and a close homologue of CRIPTO, was not decreased from the cell surface even with overexpression of PGAP6 (Fig. 2 B), indicating that PGAP6 has specificity and selectivity against substrate GPI-APs.

We next addressed the molecular mechanisms of decreased GPI-APs on the cell surface. CRIPTO exists in soluble and membrane-bound forms (Yan et al., 2002; Watanabe et al., 2007). Because CRIPTO was constantly released into the medium when CRIPTO-transfected HEK293T cells were cultured in the presence of 10% FCS, we analyzed the effect of PGAP6 using serum-free medium. Western blotting revealed 23- and 21-kD bands of HA-CRIPTO (Fig. 2 C). The 23-kD band represents a mature protein bearing a complex type N-glycan, whereas the 21-kD band represents the immature, ER form of HA-CRIPTO bearing EndoH-sensitive N-glycan (Fig. S1). When PGAP6 was overexpressed, the mature form of CRIPTO was nearly completely lost, whereas the ER form of CRIPTO remained (Fig. 2 C, left, top panel). The surface expression of CRIPTO was strongly reduced by PGAP6 overexpression (Fig. 2 C, top right) and secretion of HA-CRIPTO was enhanced (Fig. 2 C, left, bottom panel).

NTERA2 cells express both CRIPTO and PGAP6 (Fig. 2 D, top). CRIPTO was released from NTERA2 cells into medium (Fig. 2 D, bottom left, lane siCon), and knockdown of PGAP6 almost completely inhibited the release (lane siPGAP6), indicating that endogenous PGAP6 released endogenous CRIPTO from the cell surface. After knockdown of PGAP6, the surface expression of endogenous CRIPTO increased by ∼40% (Fig. 2 D, top and bottom right, histogram and quantified data, respectively).

Knockdown of PGAP6 in HEK293T cells also greatly decreased the release of HA-CRIPTO into the medium (Fig. 2 E, bottom left and right, Western and quantified data, respectively). After knockdown of PGAP6, the surface expression of HA-CRIPTO increased by ∼50% (Fig. S1 B). Specificity of PGAP6 knockdown was confirmed by rescue of CRIPTO release by siRNA-resistant PGAP6 but not by siRNA-resistant catalytic-dead PGAP6 (D610A; Fig. S1 C). The secretion of HA-CRIPTO was also inhibited by PGAP6 knockdown even in medium containing 10% FCS (Fig. S1 D), suggesting that the effect of serum on CRIPTO release might be dependent on PGAP6 activity. We next analyzed whether amino acid residues of PGAP6 conserved in the CREST family are required for the release of GPI-APs. All of the point mutants except the S584A mutant lost CRIPTO-releasing activity (Fig. S2), suggesting that the conserved amino acids are important for PGAP6 enzymatic activity.

To determine whether PGAP6 is GPI-PLA2 as expected from similarity to PGAP3, we used partitioning with Triton X-114. Because conditions of mass spectrometry (MS) analysis of the C-terminal peptides bearing lipid-containing GPI have not been established, Triton X-114 partitioning is useful to analyze whether GPI-APs possess lipid moieties. Most CRIPTO released by PGAP6 was partitioned into the aqueous phase, and only a small amount was partitioned into the detergent phase, suggesting released GPI-APs did not have a lipid chain (Fig. S3 A). We then used His-Flag-GST-Flag–tagged CD59 (HFGF-CD59) to determine the structure of the GPI moiety in released GPI-APs. HFGF-CD59 was released by overexpression of PGAP6 (Fig. S3 B, right). The GPI structures of released HFGF-CD59 were analyzed by MS. Four peaks containing a C-terminal peptide bearing GPI moiety were detected. In the second MS, all four peaks gave characteristic fragments derived from a GPI anchor, such as mass-to-charge ratio (m/z) 447⁺ and 609⁺. Based on the profile of the fragment ions, we concluded that the C-terminal peptide of the parent mass (m/z 1,226.44²⁺) had a GPI glycan core structure without a phospholipid (Fig. 3 A). Three other peaks contained the same GPI core structure and a side chain linked to the first mannose with different sugar extensions known to occur in mammalian GPIs (Paulick and Bertozzi, 2008; Fig. S3 C). All four peaks were generated by hydrolysis between inositol and phosphate. Therefore, the released HFGF-CD59 was the product of phospholipase D (PLD)–mediated cleavage.

These results suggest that if PGAP6 is a GPI-PLA2, lyso–GPI-AP generated by PGAP6 is subsequently cleaved by PLD. We next assessed this scenario by determining the number of fatty chains linked to GPI-APs in cells overexpressing PGAP6. For this, we used hydrophobic chromatography with octyl-Sepharose. The reporter GPI-AP VSVGts-FF-mEGFP-GPI consists of the extracellular domain of temperature-sensitive vesicular stomatitis virus G protein, a furin cleavage site, a Flag tag, modified EGFP, and a GPI attachment signal (Tashima et al., 2006). Upon arrival at the trans-Golgi network, the reporter becomes a smaller molecule termed Flag-mEGFP-GPI by furin cleavage. Sufficient amounts of Flag-mEGFP-GPI were isolated from PGAP6-overexpressed cell lysates for chromatographic analysis. Flag-mEGFP-GPI harboring two fatty chains isolated from wild-type cells was eluted in fractions 15 and 16 (Fig. 3 B, a), whereas Flag-mEGFP-GPI from PGAP6-overexpressing cells was eluted in fractions 13 and 14, one fraction earlier than the wild-type protein (Fig. 3 B, b). This suggested that one fatty acid was removed by PGAP6. To confirm this, Flag-mEGFP-GPI from wild-type cells was treated with honeybee PLA2 and subjected to the same column either alone or after mixing with Flag-mEGFP-GPI from PGAP6-overexpressing cells. As shown in Fig. 3 B (c and b + c), they were coeluted. These results suggest that PGAP6 has GPI-PLA2 activity.

We next analyzed whether GPI-PLD (GPLD1) is involved in the PGAP6-dependent release process. We analyzed GPLD1-knockout (KO) HEK293 cells in serum-free medium to eliminate the effects of GPI-PLD in the serum. Flow cytometric analysis showed the surface expression of CRIPTO in GPLD1-KO expressing PGAP6 was reduced similar to that in wild-type cells expressing PGAP6 (Fig. 3 C). Therefore, GPI-PLD was not required for PGAP6-dependent CRIPTO release itself. However, a significant fraction of CRIPTO released by PGAP6 in the absence of GPI-PLD was partitioned to the detergent phase (Fig. 3 D, right, middle panel), indicating that CRIPTO bearing lyso-GPI released from the cell membrane is eventually converted to the lipid-free form by the action of GPI-PLD or other PLD active on lyso-GPI (Fig. 4 A).

CRIPTO functions as an essential coreceptor of a heteromeric receptor for TGF-β ligands, including Nodal (Shen and Schier, 2000; de Castro et al., 2010). CRIPTO could act as a Nodal coreceptor in two ways, either in cis or in trans. We first analyzed the effects on the trans functions of CRIPTO. CRIPTO-expressing HEK293 cells were cultured, and activity of CRIPTO released into the culture media was assessed by incubating the culture media with responder cells expressing Nodal receptor and (n2)7-luciferase reporter (Saijoh et al., 2000). CRIPTO released in the absence of exogenous PGAP6 showed Nodal signaling activity in trans, which was enhanced by the expression of PGAP6 (Fig. 5 A). In contrast, culture medium collected from PGAP6 knockdown cells did not activate Nodal signaling in trans (Fig. 5 B). Endogenous CRIPTO released from NTERA2 cells also had trans Nodal coreceptor activity, which was inhibited by knockdown of PGAP6 (Fig. 5 C). These results indicate that CRIPTO released by PGAP6 has an ability to function as a Nodal coreceptor on other cells.

Because PGAP6 expression reduced cell surface CRIPTO, we investigated whether cis Nodal signaling was affected by PGAP6. There was a 2.5-fold increase in reporter response with HA-CRIPTO expression. This increased response was decreased by cotransfection of PGAP6 with HA-CRIPTO (Fig. 5 D), although this inhibition was not complete, probably because of the trans activity of secreted HA-CRIPTO. Moreover, HA-CRIPTO induced the phosphorylation of Smad2 in HEK293T cells, and this phosphorylation was impaired in cells cotransfected with PGAP6 (Fig. 5 E). These results indicate that PGAP6 modulates both cis and trans CRIPTO-dependent Nodal signaling through cleavage of the GPI anchor of CRIPTO.

It was reported that two recessive mutations in GPI biosynthetic genes Pign and Pgap1 in mice caused defective Cripto signaling and holoprosencephaly, a form of impaired development of the forebrain (McKean and Niswander, 2012). PIGN (Pign in mice) is required for the addition of an ethanolamine phosphate side chain to the first mannose on GPI during biosynthesis in the ER (Hong et al., 1999), and PGAP1 (Pgap1 in mice) acts as a GPI-inositol deacylase after GPI transfer to proteins (Tanaka et al., 2004; Fig. 4). These observations prompted us to analyze whether PGAP6 processes and releases CRIPTO in the PIGN and PGAP1 mutant backgrounds. CRIPTO transfected into PGAP1 mutant cells was expressed on the cell surface at levels substantially lower than those on wild-type cells (Fig. 6 A). The slow transport of CRIPTO from the ER might explain the observed phenotypes (Tanaka et al., 2004). When PGAP6 was cotransfected with CRIPTO, the expression of CRIPTO on the surface of PGAP1 mutant CHO C10 cells was only slightly decreased (statistically not significant), whereas CRIPTO on parental CHO 3B2A cells was reduced significantly (Fig. 6 A). Western blotting also showed that CRIPTO was not released into the culture medium in PGAP1 mutant cells expressing PGAP6 (Fig. 6 B, bottom). The lipid properties of GPI-APs in PGAP1 mutant cells were then analyzed by octyl-Sepharose chromatography (Fig. 6 C). Flag-mEGFP-GPI proteins from PGAP1 mutant cells were mainly eluted in fraction 16, indicating that they were more hydrophobic than those from wild-type cells because of an additional acyl chain (Fig. 3 B [a] vs. Fig. 6 C [a]). PLA2 treatment of Flag-mEGFP-GPI prepared from PGAP1 mutant cells resulted in elution in more hydrophilic fractions (Fig. 6 C, c). Flag-mEGFP-GPI from PGAP1 mutant cells expressing PGAP6 was eluted earlier than that from PGAP1 mutant cells with an empty vector (Fig. 6 C, b). The eluted position was similar to that of wild type (Fig. 3 B, a), suggesting PGAP6 cleaved inositol-acylated GPI-APs to generate inositol-acylated lyso–GPI-APs. These results indicate that inositol-acylated GPI-AP is sensitive to PGAP6 but is not secreted into the medium because of the additional acyl-chain linked to inositol (Fig. 4, A and B). Consistent with a lack of CRIPTO secretion, Nodal signaling was not induced by culture media from PGAP1 mutant cells (Fig. 6 D).

It is known that GPI-APs, lacking a first mannose–linked ethanolamine phosphate side chain, are expressed on the PIGN-null cell surface, although usually at decreased levels than normal (Hong et al., 1999; Ohba et al., 2014). When CRIPTO was transfected into PIGN-KO and wild-type HEK293 cells, CRIPTO on the PIGN-KO cells, although its surface expression was slightly lower than that on wild-type cells, was resistant to PGAP6 (Fig. 6 E, bottom). Western blotting also showed that CRIPTO was not secreted into the medium by PGAP6 (Fig. 6 F), suggesting that the ethanolamine phosphate side chain was critical for cleavage by PGAP6. Culture media from PIGN-KO cells expressing CRIPTO with PGAP6 did not induce Nodal signaling (Fig. 6 G). These results indicate that mutations in PGAP1 and PIGN impaired PGAP6-dependent CRIPTO secretion (Fig. 4, B and C).

Cripto plays a key role in mouse embryo by acting as a Nodal coreceptor (Ding et al., 1998; Xu et al., 1999). To confirm that Pgap6 is expressed in embryo when Cripto functions, we determined Pgap6 transcripts in day 6.2 embryos (E6.2) by in situ hybridization. Pgap6 was weakly but widely expressed in epiblasts (Fig. 7, A and B). Cripto was expressed in E6.2 epiblasts, forming a gradient from high expression in the proximal region to weaker expression in the distal region (Fig. 7, E and F; Ding et al., 1998). In E6.7 embryos, Pgap6 transcript levels decreased (Fig. 7, G and H), and in E7.7, they were below the detection limit of in situ hybridization (not depicted), suggesting that Pgap6 is transiently coexpressed with Cripto in epiblasts at 6.2 d postcoitum.

To determine roles of Pgap6 in vivo, we generated and analyzed Pgap6 KO embryos (Fig. S4). All of the Pgap6 KO homozygotes except one were embryonic lethal. In wild-type day 6.7 embryos, Cerberus-like (Cerl) was expressed in the anterior visceral endoderm during early gastrulation, and Brachyury, a marker of the primitive streak, was expressed in primitive streak (Belo et al., 1997; Inman and Downs, 2006; Fig. 7 M). In contrast, 3 of 13 Pgap6 KO embryos at E6.7 showed anterior–posterior axis defect; the Cerl was expressed in distal visceral endoderm, and Brachyury was broadly localized in a proximal area near the embryonic/extraembryonic constriction (Fig. 7 N). These phenotypes, especially impaired anterior shift of Cerl, are common with Cripto KO embryos (Ding et al., 1998). In E8.0, growth of Pgap6 KO embryos was not seen, and their regression was observed. The rest of Pgap6 KO embryos at E6.7 did not show anterior–posterior axis defect (Fig. 7 O). In E9.0 and E10.0 of most Pgap6 KO, embryonic development was ceased (Fig. S4). These data strongly suggest that Pgap6 is critical for Cripto function and regulates the anterior–posterior axis formation in embryos.

GPI is a cleavable membrane anchor that regulates localization and function of proteins modified with GPI. The major findings of this study were the identifications of PGAP6 as a cell-surface GPI anchor–processing enzyme with GPI-PLA2 activity and of CRIPTO, a GPI-AP required for Nodal signaling, as a sensitive substrate of PGAP6. CRIPTO was processed and released by PGAP6 expressed endogenously in human embryonic carcinoma cell NTERA2. In contrast, CRYPTIC, a GPI-AP that makes the EGF-CFC family with CRIPTO (de Castro et al., 2010), was almost completely resistant to PGAP6 even at overexpressed levels. CD59 and DAF were partially sensitive, and uPAR was nearly completely resistant to overexpressed PGAP6. Therefore, PGAP6 appeared to be a highly selective GPI-PLA2 against CRIPTO. It seems that PGAP6 recognizes not only GPI moiety but also protein structure. We found PGAP6 bearing immature N-glycans in the ER. Because a fraction of CRIPTO had immature N-glycans suggesting an ER form, it should be rigorously established whether processing of CRIPTO by PGAP6 occurs only at the cell surface or if it could also occur in other locations within the secretory pathway.

We showed that CRIPTO shed by PGAP6 was active as a Nodal coreceptor and that PGAP6-mediated release of CRIPTO reduced Nodal signaling in cells on which CRIPTO was expressed. Furthermore, Pgap6 was involved, similarly to Cripto, in correct formation of the anterior–posterior axis in mouse embryos, indicating that at least one critical function of PGAP6 is to shed CRIPTO. This is a clear example of one of the two biological functions of GPI-AP shedding, namely, shed GPI-AP functions at cells different from the cell on which the GPI-AP was originally expressed.

We used an exon-trap construct to disrupt Pgap6 in making the Pgap6-KO mice. A leakiness of the exon trap seems to be the basis of the variable phenotypes of the Pgap6-KO mice, namely, of 244 pups born from crosses of heterozygous mice, 1 was homozygous (Fig. S4), and only 3 of 13 homozygous embryos had abnormal anterior–posterior axis formation (Fig. 7). All of the studied homozygous Pgap6 KO that passed the stage of anterior–posterior axis formation died before day 9. It seems, therefore, that Pgap6 has roles in other Nodal-dependent processes.

It has been reported that defects in GPI–ethanolamine phosphate transferase I, Pign, and GPI–inositol deacylase Pgap1 perturbed Cripto signaling during forebrain development and caused holoprosencephaly (McKean and Niswander, 2012). We found that the PGAP6-mediated release of CRIPTO was negligible in cells defective in PGAP1 and PIGN. It is possible that the phenotypes of Pgap1 and Pign mutant mice could be caused by inefficient Cripto secretion by Pgap6.

We identified three steps in the generation of hydrophilic CRIPTO from the surface (Fig. 4). First, an acyl-chain at the sn-2 position of GPI in CRIPTO is cleaved by PLA2 activity of PGAP6, generating a lyso-form of CRIPTO (step 1). Lyso-CRIPTO is not stable on the membrane and is shed into the medium (step 2). Shed lyso-CRIPTO is further cleaved by PLD in the medium (step 3). GPI-PLD accounts for a part of the PLD activity acting in step 3. GPI-PLD cannot efficiently cleave GPI-APs associated with the membrane and requires detergents for efficient cleavage (Low and Huang, 1991). However, if GPI anchors become lyso-forms, they could be cleaved more efficiently. It is possible that cleavage of a lyso-form of CRIPTO by GPI-PLD might occur even before release of the lyso-form from the PM. Because many GPI-APs form transient homodimers (Suzuki et al., 2012; Seong et al., 2013), CRIPTO might also form homodimers transiently. A homodimer of lyso–GPI-CRIPTO has two acyl chains and would be stably associated with the membrane. However, a reported t_1/2 of transient homodimers of GPI-APs is short (in a range of 100–200 ms), and a release from the membrane is expected to occur during the monomeric state. It was also reported that shedding of CRIPTO was mediated by GPI-PLD activity (Watanabe et al., 2007). CRIPTO is constantly released into the medium in the presence of 10% serum, and serum growth factors induce the release of CRIPTO via activation of PLCγ and PKC signaling. Thus, PGAP6 might be the target of regulation. Indeed, shedding of CRIPTO was inhibited by PGAP6 knockdown even in the presence of 10% serum. The regulation of PGAP6 activity needs to be addressed to better understand the spatiotemporal regulation of CRIPTO functions.

HEK293T HA-CRIPTO stably expressing HA-tagged CRIPTO was established by transfection of linearized pME-pgkpuro-HA-CR1 and cultured in DMEM containing 10% FCS and 10 µg/ml puromycin. To prepare pME-pgkpuro-HA-CR1, we obtained human CRIPTO cDNA (Kazusa DNA Res. Inst.) and cloned it into a mammalian expression plasmid pME bearing a puromycin resistance gene and a segment of signal sequence plus HA tag (Table S1). 3B2A cells are derived from CHO-K1 cells and stably express two human GPI-APs, CD59 and DAF (Nakamura et al., 1997). All CHO cell lines, including 3B2A wild-type and 3B2A cells stably expressing HFGF-CD59, HA-TMEM8A, HF-TMEM8A, TMEM8A-Flag, or HA-CRYPTIC (human CRYPTIC HA-tagged at the N terminus after the ER signal sequence), were cultured in Ham’s F-12 medium containing 10% FCS with 6 µg/ml blasticidin, 10 µg/ml puromycin, or 500 µg/ml hygromycin, if necessary. Human CRYPTIC cDNA (clone AK315326) was obtained from NITE Biological Resource Center. NTERA2 cells (clone D1) were obtained from ATCC and cultured in DMEM containing 10% FCS. PIGN and GPLD1 genes were disrupted by the CRISPR-Cas9 gene editing system (Cong et al., 2013) to generate HEK293-PIGN-KO (Ohba et al., 2014) and HEK293-GPLD-KO cells, respectively. Antibodies used were monoclonal rabbit anti–CRIPTO, polyclonal rabbit anti–Na/K ATPase 1, anti-Smad2, and anti–phospho-Smad2 S465/467 (Cell Signaling Technology); anti-GPP130 (Covance); monoclonal mouse anti–Flag (M2) and anti–HA-tag (H7; Sigma-Aldrich); anti-GAPDH (Ambion); anti-tubulin and anti-transferrin receptor (Invitrogen); anti–caveolin-1 and anti–syntaxin 6 (BD); anti–ribophorin I (Santa Cruz Biotechnology, Inc.); and anti-DAF (1A10), anti-CD59 (5H8), and anti-uPAR (5D6; Maeda et al., 2007). Mouse anti–human TMEM8A mAb (clone 5C3; IgG2b) was a gift from BioLegend. Polyclonal rabbit anti–PGAP6 was raised by immunization with a peptide corresponding to the C-terminal cytosolic tail of mouse PGAP6 (SCRUM Inc.). Secondary antibodies used were HRP-conjugated anti–mouse IgG (GE Healthcare); anti–rabbit IgG (Cell Signaling Technology); phycoerythrin-conjugated goat anti–mouse IgG (BD); Alexa Fluor 488–conjugated goat anti–mouse IgG, goat anti–rabbit IgG, and Alexa Fluor 594–conjugated goat anti–rabbit IgG (Thermo Fisher Scientific).

3B2A cells (5 × 10⁶) were electroporated with 12 µg pME-HA-CRIPTO plasmid at 250 V and 1,000 µF using a Gene Pulser (Bio-Rad Laboratories). For HEK293T cells (2 × 10⁵), plasmids were transfected using Lipofectamine 2000 (Invitrogen) and incubated for 24 h. At 24 or 52 h after transfection, cells were incubated with serum-free medium for 6 h. The culture medium was collected, and anti-HA beads (clone H7) were added and incubated with rotation for 1 h at 4°C. Beads were washed four times with PBS. Immunoprecipitated proteins were eluted using 4× SDS-PAGE sample buffer and boiled for 5 min. Samples were subjected to SDS-PAGE, and proteins were probed by Western blotting.

Triton X-114 partitioning of GPI-APs was performed as described previously with some modification (Doering et al., 2001). Cells (10⁶) were lysed in lysis buffer (20 mM Tris-HCl, pH 7.4, 150 mM NaCl, 5 mM EDTA, complete protease inhibitors [Roche], and 2% Triton X-114) for 30 min on ice. Supernatants were collected to new tubes and proteins precipitated by cold acetone (eight times volume) after incubation for 1 h at −80°C. After centrifugation, the precipitated proteins were solubilized in lysis buffer, and phase separation was performed by warming up to 37°C and subsequent centrifugation at 37°C and 1,000 g for 10 min. The upper aqueous phase and lower detergent phase were collected and subjected to SDS-PAGE followed by Western blotting to determine partitioned proteins.

HEK293T and 3B2A cells (5 × 10⁶) transfected with plasmids and NTERA2 cells were harvested with PBS containing 5 mM EDTA and 0.1% BSA. After washing with FACS buffer (PBS containing 1% BSA and 0.1% sodium azide), cells were incubated for 25 min with anti-CRIPTO, anti-TMEM8A, anti-DAF, anti-CD59, or anti-HA antibodies (20 µg/ml) on ice. In some cases, cells were treated with phosphatidylinositol-specific PLC (PI-PLC) before staining by antibodies. For this, four volumes of cells (1.25 × 10⁷/ml in F12 medium without FCS) were mixed with one volume of 5 U/ml PI-PLC (from Bacillus cereus; Invitrogen) in PBS supplemented with 0.5% BSA, 5 mM EDTA, and 10 mM Hepes and incubated for 1 h at 37°C. After incubation with antibodies, cells were washed with FACS buffer three times. Cells were incubated for 25 min with anti–rabbit Alexa Fluor 488–conjugated antibody or anti–mouse IgG phycoerythrin-conjugated antibody (1:100) on ice. Cells were then pelleted, suspended in 300 µl FACS buffer, and analyzed using an FACS Canto II (BD).

Cells were washed twice with buffer (0.25 M sucrose in 20 mM Hepes-NaOH, pH 7.4), homogenized three times at 4°C in the same buffer using a tight-fitting Dounce pestle, and debris removed by centrifugation at 3,310 g for 10 min. The supernatant was placed on a 2-M sucrose cushion and ultracentrifuged at 200,000 g for 1 h. The pellet was resuspended and sucrose concentration was adjusted to 1.5 M. A discontinuous sucrose gradient was generated in an SW41 tube (Beckman Coulter) by overlaying the following sucrose solutions all in buffer consisting of 150 mM NaCl and 20 mM Hepes-NaOH, pH 7.4: 1.5 ml of 2 M, 2 ml of 1.7 M, 3 ml supernatant, 2 ml of 1.25 M, 2 ml of 0.9 M, and 1.5 ml of 0.5 M. The gradient was centrifuged at 110,000 g in an SW41 rotor (Beckman Coulter) for 2 h. Fractions were collected from the top. Equal aliquots of fractions were subjected to SDS-PAGE and proteins probed by Western blotting. Organelles in each fraction were identified by determining the PM marker Na/K ATPase 1, ER marker ribophorin I, and Golgi marker syntaxin 6. The His-Flag–tagged PGAP6 was determined by anti-PGAP6 and anti-Flag antibodies.

3B2A cells stably transfected with vector or PGAP6 (2 × 10⁷) were electroporated with pME-Neo-VSVGts-FF-mEGFP-GPI (Tashima et al., 2006). Cells were cultured for 12 h at 37°C followed by 1.5 d at 32°C and harvested using PBS containing 2.5 mM EDTA and 0.5% BSA. After centrifugation, the cell pellet was suspended in 2.5 ml buffer (20 mM Tris-HCl, pH 7.4, 150 mM NaCl, and 1 mM EDTA) and incubated for 1 h on ice. The supernatants were collected and incubated with anti-Flag (M2) beads overnight. Beads were washed two times with 0.5 ml wash buffer 1 (30 mM 1-octyl-β-d-glucoside, 20 mM Tris-HCl, pH 7.4, 150 mM NaCl, and 1 mM EDTA), two times with 0.5 ml wash buffer 2 (20 mM Tris-HCl, pH 7.4, and 0.1% NP-40), and two times with 0.5 ml wash buffer 3 (20 mM Tris-HCl, pH 7.4, and 0.03% NP-40). Flag-mEGFP-GPI was eluted three times with 150 µl elution buffer (0.1 M ammonium acetate containing 1 mg/ml Flag-peptide [Sigma-Aldrich] and 0.03% NP-40). For PLA2 treatment, immunoprecipitates bound to anti-Flag beads were washed, suspended in buffer (100 mM Tris-HCl, pH 7.4, 6 mM CaCl₂, and 0.1% NP-40), and treated with 75 U honeybee PLA2 (Sigma-Aldrich) at 37°C for 6 h. The beads were washed with wash buffers 2 and 3 and eluted as in a similar way. A 100-µl aliquot of the eluate was mixed with 100 µl buffer (0.1 M ammonium acetate, 0.03% NP-40, and 10% 1-propanol), loaded onto an Octyl-FF column (GE Healthcare), and chromatographed in ÄKTA (GE Healthcare) at a flow rate of 0.5 ml/min using a 54–100% gradient of 1-propanol. Fractions were collected, dried, subjected to SDS-PAGE, and proteins were probed by Western blotting using an anti-Flag (M2) antibody.

3B2A cells (5 × 10⁶) electroporated with HF-PGAP6 together with Lyn11-mRFP or RFP-KDEL constructs were cultured for 2 d and fixed with 4% PFA in PBS for 25 min at room temperature. After quenching with 40 mM ammonium chloride in PBS, the cells were permeabilized with or without PBS containing 0.1% Triton X-100 for 1 h at room temperature. The cells were stained with mouse anti–TMEM8A or anti–Flag with or without rabbit anti–GPP130 followed by Alexa Fluor 488–conjugated goat anti–mouse IgG with or without Alexa Fluor 594–conjugated goat anti–rabbit IgG antibodies. Confocal images were acquired on a FluoView FV1000 (Olympus).

HEK293T cells (8 × 10⁴) were transfected with HA-tagged CRIPTO either with an empty vector or PGAP6 plasmid, plus Nodal-responsive luciferase reporter construct (Saijoh et al., 2000), FAST2, ALK4, and pTK-Renilla using FuGene transfection reagent (Promega). At 24 h after transfection, medium was replaced with serum-free medium with or without 500 ng/ml recombinant human Nodal (R&D Systems) and incubated for 6 h. Luciferase activity was measured using a dual luciferase reporter assay kit (Promega). These experiments were repeated three times independently with triplicate samples.

We generated 3B2A cells stably expressing CD59 with tandem His, Flag, GST, and Flag-tags at the N terminus after the signal sequence (HFGF-CD59; Tashima et al., 2006). The cells were stably transfected with an empty vector or PGAP6 and selected in 1 µg/ml puromycin. Expressed and released HFGF-CD59 was purified from the culture medium by glutathione Sepharose 4B and eluted with a solution (20 mM reduced glutathione, 30 mM Tris-HCl, pH 8.0, and 150 mM NaCl). Purified HFGF-CD59 was subjected to SDS-PAGE, stained with Imperial CBB (Thermo Fisher Scientific), and digested by trypsin. The extracted peptides from the SDS-PAGE gel were directly analyzed by liquid chromatography–tandem MS (LC-MS/MS) using an L-column (C18, 0.1 × 150 mm; Chemicals Evaluation and Research Institute) on a nanoflow LC system (Advance UHPLC; Bruker) coupled to a mass spectrometer (TQ Orbitrap Velos; Thermo Fisher Scientific). The mass spectrometer was operated in the positive ion mode with data-dependent MS/MS. Details were described previously (Tashima et al., 2006; Fujita et al., 2009).

HEK293T and NTERA2 cells (2 × 10⁶ cells) were transfected with siPGAP6 (sense, 5′-CCUCCAUGAUGACUAGCGATT-3′; and antisense, 5′-UCGCUAGUCAUCAUGGAGGTG-3′; Ambion) or siControl by RNAi MAX reagent (Invitrogen) on day 0 with 30 pmol siRNA and on day 1 with 60 pmol siRNA. On day 3, cells were assessed for knockdown efficiency by either quantitative PCR or flow cytometry and for phenotypic changes.

Embryos were carefully staged based on their morphology. They were fixed overnight at 4°C in PBS containing 4% PFA, washed twice with PBS, and dehydrated with methanol series. Whole-mount in situ hybridization was performed using standard procedures (Wilkinson, 1992) with digoxigenin-labeled riboprobes specific for Cripto (Ding et al., 1998; Xu et al., 1999) and Pgap6. The full-length antisense and sense riboprobes for Pgap6 were synthesized from FANTOM2 cDNA clone (AK044384; 2.6 kb). E6.2 and E6.7 embryos were from the ICR strain.

To generate Pgap6/Tmem8a (Mouse Genome Informatics identification number 1926283) KO mice, a targeting construct was obtained from European Conditional Mouse Mutagenesis. The construct was electroporated into EGR-G101 embryonic stem (ES) cells (Fujihara et al., 2013a), followed by selection. The ES cells with the construct correctly targeted into the Pgap6 genomic locus were injected into blastocysts, and chimeric males were obtained. Using the chimeric male mice and C57BL/6 female mice, heterozygous KO mice were generated. Genotyping was performed with a mixture of three PCR primers (Fig. S4). The Pgap6 heterozygous KO strains were maintained in a C57BL/6 genetic background. All animal experiments were performed according to the guidelines for the Care and Use of Laboratory Animals of Osaka University.

Embryos were recovered in PBS and carefully staged based on their morphology and cell numbers. They were fixed overnight at 4°C in PBS containing 4% PFA, washed twice with PBS, permeabilized with 0.2% Triton X-100 in PBS for 30 min at room temperature, incubated for 1 h at room temperature with TSA blocking reagent (PerkinElmer), and incubated overnight at 4°C with the following primary antibodies diluted in blocking reagent. First immunoreaction was performed with goat antibodies to Brachyury (1/50 dilution; N-19, #sc-17743; Santa Cruz Biotechnology, Inc.) and rat antibodies to Cerl1 (1/100 dilution; MAB1986; R&D Systems). After washing three times with PBS, they were incubated with the following secondary antibodies diluted in TSA blocking reagent. Second antibodies were Alexa Fluor 488 donkey anti–mouse IgG (1/200 dilution; A21206; Molecular Probes), Alexa Fluor 594 donkey anti–rat IgG (1/200 dilution; A21209; Molecular Probes), Alexa Fluor 594 donkey anti–goat IgG (1/200 dilution; A11057; Molecular Probes), and Alexa Fluor 633 donkey anti–goat IgG (1/200 dilution; A21082; Molecular Probes). Nuclei were stained by incubation for 30 min with DAPI (1/2,000 dilution in PBS; FQ023; Wako Pure Chemical Industries). Images were obtained with a laser-scanning confocal microscope system (FV1000; Olympus) and a UPLSAPO 20× (NA 0.75; Olympus) objective lens. Embryos were genotyped by PCR analysis with their parts.

Fig. S1 shows that PGAP6 knockdown inhibits release of mature CRIPTO from the cell surface even in the presence of serum. Fig. S2 shows functional analysis of PGAP6 point mutants demonstrating that amino acids conserved in the CREST superfamily are functionally important, except S584A. Fig. S3 shows LC/electrospray ionization (ESI)-MS/MS analysis of GPI moiety in HFGF-CD59 shed by PGAP6. Fig. S4 shows the procedure and some characteristics of Pgap6 KO mice. Table S1 contains primers used in this study.

The authors thank Drs. Ryo Taguchi, Hidetaka Shiratori, and Tatsuo Michiue for helpful discussion; BioLegend (Osaka, Japan) for anti-TMEM8A mAb; and Keiko Kinoshita, Kana Miyanagi, and Yukari Hirata for technical help.

This work was supported by grants-in-aid from the Japan Society for the Promotion of Science (to T. Kinoshita and M. Fujita), the Takeda Science Foundation, the National Science Foundation for Young Scientists of China (grant 31400693), the Natural Science Foundation of Jiangsu Province (grant BK20140141), and Research Funds for the Central Universities (grant JUSRP11540 to M. Fujita).

The authors declare no competing financial interests.

Author contributions: G.-H. Lee performed the main experiments and wrote an original draft; M. Fujita proposed the project, performed initial experiments, and wrote a paper; K. Takaoka, E. Kajikawa, and H. Hamada analyzed mouse embryos; Y. Murakami analyzed KO mice; Y. Fujihara and M. Ikawa generated and analyzed KO mice; N. Kanzawa, Y. Takada, and K. Saito determined GPI structure by MS analysis; K. Murakami generated GPI-PLD KO cells and performed experiments with them; Y. Maeda designed cell biological experiments and edited a paper; and T. Kinoshita supervised the project and wrote a paper.