Calnexin Mediates the Maturation of GPI-Anchors Through ER Retention

bioRxiv preprint doi: https://doi.org/10.1101/2020.07.08.192955; this version posted July 8, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

1 Calnexin mediates the maturation of GPI-anchors through ER retention

3 Xin-Yu Guo1, Yi-Shi Liu1, Xiao-Dong Gao1, Taroh Kinoshita2, 3, and Morihisa Fujita1,*

5 1Key Laboratory of Carbohydrate Chemistry and Biotechnology, Ministry of Education,

6 School of Biotechnology, Jiangnan University, Wuxi, Jiangsu 214122, China; 2Research

7 Institute for Microbial Diseases, Osaka University, Suita, Osaka 565-0871, Japan; 3WPI

8 Immunology Frontier Research Center, Osaka University, Suita, Osaka 565-0871, Japan

10 *Corresponding author: Morihisa Fujita, Ph.D.

11 E-mail: [email protected]

12 Tel & Fax: +86-510-851-97071

16 Abstract

17 The protein folding and lipid moiety status of glycosylphosphatidylinositol-anchored proteins

18 (GPI-APs) are monitored in the endoplasmic reticulum (ER), with calnexin playing dual roles

19 in the maturation of GPI-APs. In the present study, we investigated the functions of calnexin

20 in the quality control and lipid remodeling of GPI-APs in the ER. By directly binding the

21 N-glycan on proteins, calnexin was observed to efficiently retain GPI-APs in the ER until

22 they were correctly folded. In addition, sufficient ER retention time was crucial for

23 GPI-inositol deacylation, which is mediated by post-GPI attachment protein 1 (PGAP1). Once

24 the calnexin/calreticulin cycle was disrupted, misfolded and inositol-acylated GPI-APs could

25 not be retained in the ER and were exposed on the plasma membrane. In calnexin/calreticulin

26 deficient cells, endogenous GPI-anchored alkaline phosphatase was expressed on the cell

27 surface, but its activity was significantly decreased. ER stress induced surface expression of

28 misfolded GPI-APs, but proper GPI-inositol deacylation occurred due to the extended time

29 that they were retained in the ER. Our results indicate that calnexin-mediated ER quality

30 control systems for GPI-APs are necessary for both protein folding and GPI-inositol

31 deacylation.

33 Introduction

34 Glycosylphosphatidylinositol (GPI) is a complex glycolipid that acts as a membrane

35 anchor for many cell surface proteins (Ikezawa, 2002; McConville & Menon, 2000; Tiede,

36 Bastisch, Schubert, Orlean, & Schmidt, 1999) and is a highly conserved post-translational

37 modification from yeast to mammals. In mammals, there are at least 150 GPI-anchored

38 proteins (GPI-APs), which serve as receptors, adhesion molecules, enzymes, protease

39 inhibitors, and so on (Kinoshita, 2020). The biosynthesis of GPI and its attachment to protein

40 occur in the endoplasmic reticulum (ER). Nascent GPI-APs synthesized by GPI transamidase

41 are still immature and undergo remodeling reactions to become mature GPI-APs. In the ER,

42 two remodeling reactions occur in many GPI-APs. First, an acyl-chain on an inositol-ring of

43 the GPI-anchor is eliminated by the GPI-inositol deacylase PGAP1 (Chen et al., 1998; Tanaka,

44 Maeda, Tashima, & Kinoshita, 2004). Second, a side-chain ethanolamine-phosphate (EtNP)

45 attached on the second mannose (Man2) of GPI-glycan is removed by the GPI-EtNP

46 phosphodiesterase PGAP5 (Fujita et al., 2009). These remodeling reactions are crucial for the

47 interaction of GPI-APs with p24 protein complexes, which are cargo receptors for GPI-APs

48 (Fujita & Kinoshita, 2012; Fujita et al., 2011), indicating that GPI-AP remodeling in the ER is

49 required for their efficient sorting into transport vesicles at the ER exit sites. Pathogenic

50 homozygous mutations in PGAP1 cause an inherited GPI deficiency, which results in

51 intellectual disability, encephalopathy and hypotonia (Granzow et al., 2015; Kettwig et al.,

52 2016; Murakami et al., 2014). It has been reported that Pgap1 mutant mice exhibit abnormal

53 head development, such as otocephaly (Ueda et al., 2007; Zoltewicz et al., 2009) and

54 holoprosencephaly (McKean & Niswander, 2012), suggesting that Pgap1 function is required

55 for normal forebrain formation. In addition, male Pgap1-knockout mice are infertile (Ueda et

56 al., 2007). Taken together, these findings show that correct processing of GPI-anchors is

57 crucial for the proper functions of these proteins in vivo.

58 Protein asparagine (N)-glycosylation is another type of post-translational modification

59 occurring in the ER (Hammond, Braakman, & Helenius, 1994; Schwarz & Aebi, 2011). An

60 oligosaccharide consisting of Glc3Man9GlcNAc2 (Glc, glucose; Man, mannose, GlcNAc,

61 N-acetylglucosamine) is transferred to the amino group of asparagine within the motif NxS/T

62 (N, asparagine; S/T, serine or threonine; x, any amino acids except proline) of newly

63 synthesized proteins. The folding states of secretory proteins are monitored by the ER quality

64 control systems (Sun & Brodsky, 2019). N-glycans are processed in the ER, which contributes

65 to protein folding. First, two Glc residues on N-glycans are trimmed by -glucosidase I and II.

66 Processed monoglucosylated N-glycan structures are then recognized by calnexin and

67 calreticulin, which are molecular chaperones that possess a lectin domain. Calnexin has a

68 transmembrane domain, whereas calreticulin is a soluble protein (Helenius & Aebi, 2004),

69 and both proteins associate with protein disulfide isomerases such as ERp57 and ERp29,

70 prompting the folding of newly synthesized proteins (Ellgaard & Frickel, 2003; Kozlov,

71 Muñoz-Escobar, Castro, & Gehring, 2017). Once the remaining Glc residues on protein

72 N-glycans are trimmed by -glucosidase II, calnexin/calreticulin dissociate from the proteins.

73 However, proteins remaining in an unfolded state are re-glucosylated by UDP-Glc:

74 glycoprotein glucosyltransferase (UGGT), subsequently becoming bound again to calnexin

75 and calreticulin for refolding. This series of reactions is called the calnexin/calreticulin cycle

76 and is essential for glycoprotein folding.

77 In our previous study, we performed a genetic screening to identify factors that affect

78 GPI-inositol deacylation. In the screening, MOGS, GANAB, and CANX, as well as PGAP1,

79 were identified as candidates (Liu et al., 2018). MOGS, GANAB, and CANX encode

80 α-glucosidase I and II and calnexin, respectively (Tannous, Pisoni, Hebert, & Molinari, 2015).

81 These results suggested that N-glycan-dependent ER quality control systems participate in the

82 lipid remodeling of GPI-APs, whereas it was unclear how calnexin contributes to the

83 processing.

84 When proteins fail to fold correctly, they are recognized as misfolded proteins, the majority

85 of which are retained in the ER and degraded through the ER-associated degradation (ERAD)

86 pathway (Vembar & Brodsky, 2008). However, misfolded GPI-APs do not appear to be

87 suitable substrates for ERAD, possibly because of the presence of GPI-anchors. A fraction of

88 misfolded GPI-APs are degraded through proteasomes (Fujita, Yoko-o, & Jigami, 2006;

89 Ishida et al., 2003), whereas the rest are exported from the ER and delivered to the vacuole for

90 degradation in yeast (Hirayama, Fujita, Yoko-o, & Jigami, 2008; Sikorska et al., 2016). In

91 mammalian cells, although misfolded GPI-APs are retained in the ER, they are rapidly

92 released into the secretory pathway upon acute ER stress despite their misfolding

93 (Satpute-Krishnan et al., 2014). Time-lapse imaging of live cells under acute ER stress

94 conditions showed misfolded prion proteins are transported from the ER to the Golgi and

95 plasma membrane, and subsequently to lysosomes for degradation.

96 In the present study, we investigated the roles of calnexin in the quality control and lipid

97 remodeling of GPI-APs in the ER. In wild-type (WT) cells, GPI-APs were folded and inositol

98 deacylated in the ER and expressed at the cell surface. In calnexin and calreticulin double

99 knockout (CANX&CALR-DKO) cells, protein folding and GPI-inositol deacylation

100 efficiencies were significantly decreased such that misfolded and inositol-acylated GPI-APs

101 were exposed on the plasma membrane. Thus, these results indicate that the

102 N-glycan-dependent calnexin/calreticulin cycle is responsible for the correct folding of

103 GPI-APs and provides sufficient ER retention time for efficient GPI-inositol deacylation.

104

105 Materials and Methods

106 Cells, antibodies, and materials

107 All the cell lines, antibodies, and other reagents are listed in the key resource table. HEK293

108 and HEK293FF6 cells (Hirata et al., 2015) and their knockout (KO) derivatives were cultured

109 in Dulbecco’s Modified Eagle medium (DMEM) containing 10% (vol/vol) FCS (Biological

110 Industries). G418 (400 g/ml), puromycin (1 g/ml), and streptomycin/penicillin (100

111 units/mL of penicillin and 100 µg/mL of streptomycin) were used where necessary. Cells

112 were maintained at 37°C under a humidified atmosphere with 5% CO2. CANX&CALR-DKO

113 cells stably expressing CANX WT or mutants were selected with 10 μg/ml blasticidin.

114 Mouse monoclonal anti-CD59 (clone 5H8), anti-CD55 (clone IA10) (Liu et al., 2018; Maeda

115 et al., 2007), anti-FLAG (M2; Sigma), anti-RGS-His6 (BSA-free; QIAGEN), anti-calnexin

116 (M178-3; MBL), and anti-GAPDH (clone 1E6D9; Proteintech) antibodies were used as

117 primary antibodies. Other primary antibodies used in the present study included rabbit

118 monoclonal anti-calreticulin (clone D3E6; Cell Signaling Technology), polyclonal

119 anti-calnexin (C4731; Sigma), polyclonal anti-ERp57 (Proteintech), polyclonal anti-GRP78

120 (Novus), polyclonal anti-eGFP (Proteintech), monoclonal anti-ALPP (clone SP15; Abcam)

121 and anti-HA-Tag (C29F4; Cell Signaling Technology). Phycoerythrin (PE)-conjugated goat

122 anti-mouse IgG (eBioscience), PE-conjugated donkey anti-rabbit IgG (eBioscience),

123 HRP-conjugated anti-mouse IgG (Cell Signaling Technology) and anti-rabbit IgG (Cell

124 Signaling Technology), F(ab')2-Goat anti-Mouse IgG (H+L) Cross-Adsorbed Secondary

125 Antibody, Alexa Fluor 555 (Thermo Fisher Scientific), and F(ab')2-Goat anti-Mouse IgG

126 (H+L) Cross-Adsorbed Secondary Antibody, Alexa Fluor 488 (Thermo Fisher Scientific)

127 were used as secondary antibodies. PE-conjugated anti-CD109 (HU17; eBioscience) was used

128 for flow cytometry. Thapsigargin (100 nM, TG, Sigma-Aldrich) and doxycycline (1 μg/ml,

129 631311; Clontech Laboratories) were used for drug treatments.

130

131 Plasmids

132 All the plasmids and oligonucleotides used in this study are listed in Table S1 and Table S2,

133 respectively. For the CRISPR-Cas9 system used to KO target genes, guide RNA sequences

134 were designed using the E-CRISP website (Heigwer, Kerr, & Boutros, 2014)

135 (http://www.e-crisp.org/E-CRISP/), and the corresponding DNA fragments were ligated into

136 the BpiI-digested vector pX330-EGFP. The CANX cDNA fragment was amplified from

137 human cDNA and cloned into the retroviral vector pLIB2-BSD to generate

138 pLIB2-BSD-CANX. The lectin-deficient mutant plasmids pLIB2-BSD-CANX (Y164A),

139 pLIB2-BSD-CANX (K166A), pLIB2-BSD-CANX (M188A) and pLIB2-BSD-CANX

140 (E216A) and the PDI interaction mutant plasmids pLIB2-BSD-CANX (D343A),

141 pLIB2-BSD-CANX (E351A), pLIB2-BSD-CANX (D343A, E351A), pLIB2-BSD-CANX

142 (W342A, D343A), and pLIB2-BSD-CANX (D347A, E351A) were constructed by site-direct

143 mutagenesis. The DNA fragments encoding mEGFP-FLAG-CD59 and mEGFP-FLAG-CD55

144 were digested with EcoRI and NotI from pPB-FRT-PGKp-BSD-mEGFP-FLAG-CD59 and

145 pPB-FRT-PGKp-BSD-mEGFP-FLAG-CD55 (Liu et al., 2018), and then cloned into

146 pME-BSD-mEGFP-FLAG-CD59 and pME-BSD-mEGFP-FLAG-CD55 respectively. The

147 misfolded, soluble, and N-glycan-deficient mutants of CD59 were constructed by site-directed

148 mutagenesis. The transmembrane portion of CD46 was amplified from CD46 cDNA and

149 ligated to pME-BSD-EGFP-FLAG-BSD-CD59-(C94S) to generate CD59-(C94S)-TM. For 7

150 localization, pME-Zeo-mRFP-KDEL (Fujita et al., 2009) was used as an ER marker. The

151 LY6D and CD52 cDNA fragment was amplified from human cDNA and cloned into the

152 retroviral vector pLIB2-His6-IRES2-BFP to generate pLIB2-His6-LY6D-IRES2-BFP and

153 pLIB2-His6-CD52-IRES2-BFP. The plasmid pME-VSVGts-FF-mEGFP-GPI (VFG-GPI)

154 encoded a reporter protein consisting of the extracellular domain of a temperature-sensitive

155 vesicular stomatitis virus G (VSVGts) protein, a furin cleavage site, a FLAG tag, mEGFP

156 (modified EGFP) and a GPI attachment signal (Fujita et al., 2009; Fujita et al., 2011; Maeda,

157 Ide, Koike, Uchiyama, & Kinoshita, 2008; Takida, Maeda, & Kinoshita, 2008).

158

159 Establishment of knockout cell lines

160 To generate PDIA3 gene knockout cell lines, HEK293 cells were transiently transfected with

161 two pX330-PDIA3 plasmids (cr1 and cr2) (Table S1) bearing different gRNAs targeting to

162 exon regions of PDIA3 gene. After 3 days, cells with EGFP were sorted using a cell sorter

163 S3e (Bio-Rad). Then, the collected cells were cultured for 8 days and subjected to limiting

164 dilution and obtain the clonal KO cells. Clones lacking wild-type alleles of the target gene

165 were selected, and DNA sequences were analyzed using the Sanger method.

166

167 PI-PLC treatment and flow cytometry analysis

168 For phosphatidylinositol-specific phospholipase C (PI-PLC; Thermo Fisher Scientific)

169 treatment of endogenous GPI-APs, 106 cells were harvested with trypsin/EDTA. Then, the

170 cells were incubated with 5 U/ml of bacterial PI-PLC dissolved in PBS supplemented with

171 0.5% BSA, 5 mM EDTA, and 10 mM HEPES (pH 7.4) plus DMEM medium without FCS for

172 1.5 h at 37°C. After being washed with PBS, the cells were stained with the specific primary

173 antibodies for endogenous CD59, CD55 and CD109 (10 μg/ml) for 25 min on ice and then

174 washed two times with cold FACS buffer (PBS containing 1% BSA and 0.1% NaN3).

175 Subsequently, the cells were incubated with PE-conjugated goat anti-mouse IgG as the

176 secondary antibody for 25 min on ice, washed two times with FACS buffer and then analyzed

177 with an Accuri C6 instrument (BD). The data were analyzed using FlowJo, and the remaining

178 GPI-APs were calculated from mean values of PE-fluorescent intensity treated with or

179 without PI-PLC.

180 For exogenous GPI-APs, pME-HA-GPI-AP (Table S1) was transfected together with

181 pME-BFP, which acts as a transfection control, into WT and CANX&CALR-DKO cells. A

182 retroviral plasmid vector containing His6-tagged GPI-AP-IRES2-BFP sequence was

183 transfected into HEK293T packaging cells, and subsequently infected into WT and

184 CANX&CALR-DKO cells. Lipofectamine 2000 (Thermo Fischer Scientific) was used as the

185 transfection reagent. Three days after transfection, 106 cells were harvested and treated with

186 or without PI-PLC, and anti-HA or anti-His6 was used as the primary antibody. Subsequently,

187 secondary antibody staining and flow cytometry analysis were performed as described above.

188

189 Immunofluorescence analysis

190 To detect the subcellular localization of the EGFP-FLAG-tagged misfolded CD59

191 (EGFP-FLAG-CD59 (C94S)), N-glycan-deficient mutant CD59 (EGFP-FLAG-CD59 (C94S,

192 N43Q)), transmembrane misfolded CD59 (EGFP-FLAG-CD59 (C94S) TM), misfolded CD55

193 (EGFP-FLAG-CD55 (C81A)), and N-glycan mutant CD55 (EGFP-FLAG-CD55 (C81A,

194 N95Q)), HEK293FF6 and its KO derivative cells were transfected with plasmids expressing

195 these proteins together with the plasmid pME-mRFP-KDEL as an ER marker. Then, 36 h

196 after transfection, cells were replated onto glass coverslips pretreated with 1% gelatin and

197 cultured for another 2 days. Subsequently, the cells were washed with PBS, fixed in 4%

198 paraformaldehyde for 10 min at roon temperature, washed again with PBS, and then

199 incubated with 40 mM ammonium chloride for 10 min at room temperature. Finally, the

200 coverslips were mounted onto slides using a mounting solution containing DAPI for 5 min.

201 To assess the localization of VFG-GPI, HEK293FF6, CANX&CALR-DKO, and

202 CANX&CALR-DKO+CANX cells were transfected with pME-mRFP-KDEL. Subsequently,

203 36 h after transfection, the cells were harvested and replated onto glass coverslips pretreated

204 with 1% gelatin at 37°C for 1 day and then incubated with 1 µg/ml doxycycline at 40°C for

205 24 h to induce VFG-GPI expression. Then, the cells were quickly fixed with 4%

206 paraformaldehyde for 10 min before being washed with PBS and then 40 mM

207 ammonium chloride.

208 To assess the localization of endogenous calnexin, the cells were fixed and permeabilized

209 with -20°C methanol for 5 min at 4°C before being blocked with PBS containing 5% FCS

210 (blocking buffer) for 1 h. Subsequently, the cells were incubated for 1 h with

211 mouse-anti-calnexin (M178-3; MBL) as primary antibody diluted in blocking buffer.

212 Subsequently, the cells were gently washed with PBS twice. Then the cells were incubated for

213 1 h with an F(ab')2-Goat anti-Mouse IgG (H+L) cross-adsorbed secondary antibody, Alexa

214 Fluor 555 (Invitrogen, Thermo Fisher Scientific) diluted in blocking buffer, after which the

215 cells were gently washed with PBS twice.

216 To assess the localization of HLA1-1-147-FLAG, pME-HLA1-1-147-FLAG and

217 pME-mRFP-KDEL were transfected together into WT cells. Then, 36 h after transfection, the

218 cells were harvested and replated onto glass coverslips pretreated with 1% gelatin and

219 incubated at 37°C for 1 day followed by an incubation at 37°C with 100 nM TG for 6 h. The

220 cells were then fixed with 4% paraformaldehyde, washed with PBS, and incubated with 40

221 mM ammonium chloride for 10 min at room temperature. Subsequently, the cells were

222 permeabilized with 0.2% Triton-X 100 diluted in blocking buffer for 15 min at room

223 temperature. Mouse-anti-FLAG (M2; Sigma-Aldrich) was used as the primary antibody and

224 was diluted in blocking buffer for 1 h, after which the cells were gently washed with PBS

225 twice. Then, the cells were incubated for 1 h with an F(ab')2-Goat anti-Mouse IgG (H+L)

226 cross-adsorbed secondary antibody conjugated with Alexa Fluor 488 (Invitrogen, Thermo

227 Fisher Scientific) diluted in blocking buffer. Subsequently, the cells were gently washed with

228 PBS twice.

229 The cells were visualized using a confocal microscope (C2si; Nikon) with a CFI Plan

230 Apochromat VC oil objective lens (100× magnification and 1.4 NA). Pearson’s correlation

231 coefficient (GFP/RFP) was analyzed by ImageJ using the JACoP plug-in.

232

233 Immunoprecipitation

234 HEK293 cells cultured in 10-cm dishes were transfected with 16 g of plasmids using

235 Lipofectamine 2000 and incubated for 48 h. Then, the cells were harvested with

236 trypsin/EDTA, washed with cold PBS at 4°C, and then incubated with 550 l of lysis buffer

237 (25 mM HEPES pH 7.4, 150 mM NaCl, 0.5% CHAPS, protein inhibitor cocktail (EDTA-Free,

238 MCE) and 1 mM PMSF) on ice for 30 min. After incubation, the tube was centrifuged at

239 10,000 g for 10 min at 4°C to remove insoluble fractions. Then, a portion of the supernatant

240 was mixed with sample buffer, the rest was transferred to a new tube, and then 20 l of

241 prewashed anti-FLAG affinity gel (A2220; Sigma-Aldrich) was added. After rotating the

242 tubes at 4°C for 1-1.5 h, the gel with tagged proteins were washed 4 times with lysis buffer.

243 The proteins were eluted by boiling in SDS sample buffer and analyzed by Western blotting.

244

245 Cell surface alkaline phosphatase (ALP) activity assay

246 For both HEK293FF6 WT and CANX&CALR-DKO cells, 8,000 cells/well were seeded in

247 96-well plates and cultured overnight at 37°C. After the medium was removed, the cells were

248 gently washed with 1×HBSS buffer (Ca2+ and Mg2+ free, pH 7.4; Sangon Biotech) twice,

249 after which 150 μl/well of a pNPP solution (Alkaline Phosphatase Yellow (pNPP) Liquid

250 Substrate System for ELISA, P7998; Sigma) was added. Then, the plates were incubated at

251 37°C for 6.5 h, the reaction was stopped by adding 50 μl/well of 3 M NaOH, and the

252 absorbance at 405 nm was measured using an Enspire 2300 instrument. The calf intestinal

253 alkaline phosphatase (TaKaRa) was used as a standard for the reaction.

254

255

256 Results

257 Calnexin/calreticulin cycle is required for efficient GPI-inositol deacylation for both

258 endogenous and exogenous N-glycosylated GPI-APs

259 After GPI attachment to proteins, an acyl-chain linked to inositol on many GPI-APs is

260 removed by the GPI-inositol deacylase PGAP1. The inositol-deacylation status of GPI-APs

261 can be determined by assessing their sensitivity to bacterial PI-specific phospholipase C

262 (PI-PLC) (Ferguson, Kinoshita, & Hart, 2009; Kmenon, 1994). PI-PLC reacts with the

263 hydroxyl group on the 2-position of the inositol ring, producing inositol-1,2-cyclic phosphate

264 (Heinz, Ryan, Bullock, & Griffith, 1995). If an acyl-chain is attached to the 2-position of the

265 inositol, PI-PLC cannot cleave the substrate (Figure 1A). We previously demonstrated that

266 GPI-inositol deacylation of CD59 and CD55 is partially impaired in cells defective for

267 calnexin (CANX) and is more strongly affected in cells defective for both CANX and

268 calreticulin (CALR) (Liu et al., 2018). To determine if the CANX/CALR-dependency of

269 GPI-inositol diacylation is a general phenomenon of various GPI-APs, we treated

270 CANX&CALR-DKO HEK293 cells with PI-PLC, and endogenous CD59, CD55, and CD109

271 in the CANX&CALR-DKO cells showed PI-PLC resistance (Figure 1B and C). To confirm

272 whether the PI-PLC resistance of GPI-APs is a general phenomenon, exogenous human

273 GPI-APs CD48, CD14, ART3, ART4, LY6E, LY6G6C, and CD52 were transiently expressed

274 and assessed by PI-PLC treatment. Consistent with the endogenous GPI-APs, the exogenous

275 GPI-APs also showed PI-PLC resistance in CANX&CALR-DKO cells (Figure 1D and Figure

276 1-figure supplement). These results suggest that disruption of the calnexin/calreticulin cycle

277 causes inefficient GPI-inositol deacylation in GPI-APs.

278

279

280 14

281 Figure 1. GPI-inositol-acylated GPI-APs are expressed on the plasma membrane in CANX&CALR-DKO 282 cells. 283 (A) After lipid remodeling, mature GPI-APs on the cell surface can be cleaved and released from the plasma 284 membrane by PI-PLC (left). When an acyl-chain is modified to the 2-position of the inositol ring on GPI-APs, 285 PI-PLC cannot cleave the GPI-APs (right). 286 (B) WT and CANX&CALR-DKO cells were treated with or without PI-PLC. Endogenous CD59, CD55 and 287 CD109 were stained with the appropriate antibodies and analyzed by flow cytometry. Gray shaded areas indicate 288 cells treated with PI-PLC, heavy solid lines indicate cells without PI-PLC treatment, and dashed lines show the 289 background of the secondary antibody. 290 (C) Remaining endogenous GPI-APs after PI-PLC treatment. The fluorescence intensity of GPI-APs after 291 PI-PLC treatment was divided by that observed without PI-PLC treatment. The values in WT cells were set as 1, 292 and the relative values for CANX&CALR-DKO were plotted. The data are presented as the means ± SD of three 293 independent measurements. P-values (one-tailed, student’s t-test) are shown. 294 (D) Remaining exogenous GPI-APs after PI-PLC treatment. HA-tagged GPI-AP constructs were co-transfected 295 with a BFP-expressing plasmid into WT and CANX&CALR-DKO cells. A retroviral vector containing 296 His6-tagged CD52-IRES2-BFP was stably expressed in WT and CANX&CALR-DKO cells. The BFP-positive 297 regions were gated, and cell surface expression of GPI-APs were analyzed by flow cytometry. The values were 298 calculated as described in (C). 299

300 Approximately 95% of GPI-APs in mammalian cells contain at least one N-glycan (Liu et

301 al., 2018). To determine whether calnexin/calreticulin also affects the inositol-deacylation of

302 non-N-glycosylated GPI-APs, we analyzed LY6D and GFP-GPI, which have no N-glycan, by

303 PI-PLC treatment in WT and CANX&CALR-DKO cells. In WT cells, both LY6D and

304 GFP-GPI were sensitive to PI-PLC cleavage (Figure 2A, left top), whereas in

305 CANX&CALR-DKO cells, unlike N-glycosylated GPI-APs, LY6D and GFP-GPI did not

306 show PI-PLC resistance (Figure 2A, left bottom, right). We next constructed HA-tagged

307 CD59 with (WT) or without N-glycan (N43Q) to compare the PI-PLC sensitivity between

308 WT and CANX&CALR-DKO cells. HA-CD59 WT showed PI-PLC resistance in

309 CANX&CALR-DKO cells, compared to that observed in WT cells (Figure 2B, left). The

310 PI-PLC sensitivity of HA-CD59 (N43Q) was comparable between the WT and

311 CANX&CALR-DKO cells (Figure 2B, right). The observed difference between

312 N-glycosylated and non-N-glycosylated GPI-APs suggests that GPI-inositol deacylation

313 regulation by calnexin/calreticulin is dependent on the N-glycans on GPI-APs.

314

315 Figure 2. Non-N-glycosylated GPI-APs use a calnexin-independent mechanism for GPI-inositol 316 deacylation. 317 (A) The plasmids pLIB2-His6-LY6D-IRES-BFP or pLIB2-His6-GFP-GPI-IRES-BFP were packaged into 318 retroviruses and infected into WT and CANX&CALR-DKO cells. Cells were treated with or without PI-PLC 319 after 3 d of transfection. Anti-His6 was used as the primary antibody to detect the surface levels of His6-LY6D 320 and His6-GFP-GPI. The cells showing the same BFP intensities were gated, and the surface expression was 321 analyzed by flow cytometry (left). The values for the remaining GPI-APs after PI-PLC treatment in WT cells 322 were set as 1, and the relative values in CANX&CALR-DKO cells were plotted (right). The data are presented 323 as the means ± SD of three independent measurements. P-values (one-tailed, student’s t-test) are shown. 324 (B) WT or non-N-glycosylated mutant constructs (pME-HA-CD59 and pME-HA-CD59 (N43Q), respectively) 325 were transfected into cells together with pME-BFP (transfection control). Three days after transfection, the cells 326 were treated with or without PI-PLC, then HA-tagged CD59 was analyzed by staining with an anti-HA antibody 327 (left). Cells showing the same BFP intensities were gated for the same transfection level. The values for the

328 remaining HA-CD59 WT or N43Q after PI-PLC treatment in WT cells were set as 1, and the relative values in 329 CANX&CALR-DKO cells were plotted (right). The data are presented as the means ± SD of three independent 330 measurements. P-values (one-tailed, student’s t-test) are shown. 331

332 Glycan binding of calnexin is required for efficient GPI-inositol deacylation

333 The structure of the ER-luminal portion of calnexin has two domains: a glycan binding

334 domain, consisting of a globular β-sandwich structure, and an extended arm domain

335 (P-domain), consisting of two β-strands (Schrag et al., 2001). The glycan binding domain

336 preferentially recognizes an N-glycan containing Glc1Man9GlcNAc2 (Spiro, Zhu, Bhoyroo, &

337 Soling, 1996; Vassilakos, Michalak, Lehrman, & Williams, 1998). The amino acids Y164,

338 K166, M188, and E216 (Michael R Leach & Williams, 2004) have been reported to be

339 important for the lectin activity of calnexin. To assess whether this activity affects

340 GPI-inositol deacylation, we constructed calnexin constructs with mutations in the amino

341 acids responsible for glycan binding (Y164A, K166A, M188A, and E216A). Compared with

342 a wild-type calnexin construct, the PI-PLC sensitivity of CD59 in CANX&CALR-DKO cells

343 transfected with glycan binding-deficient calnexin constructs was rarely rescued, which was

344 consistent with previous results (Y164A and E216A) (Liu et al., 2018) (Figure 3A and Figure

345 3-figure supplement A and B). The results indicated that lectin activity of calnexin is

346 necessary for efficient GPI-inositol deacylation.

347 Nuclear magnetic resonance analysis and in vitro experimental results have indicated that

348 the P-domain of calnexin is important for its association with the protein disulfide isomerases

349 ERp57 (Frickel et al., 2002; M. R. Leach, Cohen-Doyle, Thomas, & Williams, 2002; Pollock

350 et al., 2004) and ERp29 (Kozlov et al., 2017) through the tip of its arm domain. Furthermore,

351 tryptophan at position 342 (W342), aspartic acid residues at positions 343 (D343) and 347

352 (D347), and glutamic acid at position 351 (E351) of calnexin are thought to be important for

353 this association (Kozlov et al., 2017; Pollock et al., 2004). To assess the importance of these

354 residues in the association of calnexin with ERp57 and ERp59, single (D343A, E351A) or

355 double mutant calnexin constructs (W342A D343A, D343A E351A, D347A E351A) were

356 constructed to disrupt the interaction. In cells transfected with these mutants, the PI-PLC

357 sensitivity of CD59 was restored, whereas some of the constructs affected in their protein

358 stability (Figure 3B and C), suggesting that calnexin participates in GPI-inositol deacylation

359 independent of its binding with ERp57. To assess this possibility, PDIA3, which encodes

360 ERp57, was knocked out in HEK293 cells, and the PI-PLC sensitivity of GPI-APs was

361 analyzed. Interestingly, the PI-PLC sensitivity of both CD59 and CD55 was increased in

362 PDIA3-KO cells compared to that observed in the parental WT cells (Figure 3D and E),

363 suggesting that GPI-inositol deacylation occurred more efficiently in the absence of ERp57.

364 Calnexin was observed to associate with CD59 in an N-glycan- and GPI-dependent

365 manner (Figure 3-figure supplement C), as previously observed (Liu et al., 2018). ERp57

366 weakly co-precipitated with CD59 and misfolded CD59, whereas interactions with other ER

367 chaperones, such as calreticulin and BiP, were not detected with any CD59 construct (Figure 3

368 supplement C), indicating that the N-glycan and membrane-associated domain of GPI-APs

369 are important for calnexin interaction. To assess whether ERp57 is required for calnexin to

370 bind CD59, we expressed misfolded CD59 (C94S) and misfolded and soluble CD59 (C94S

371 G103stop) in WT and PDIA3-KO cells and analyzed their ability to bind calnexin. Calnexin

372 was co-precipitated with misfolded CD59 (C94S) but not misfolded and soluble CD59 (C94S

373 G103stop). Furthermore, knockout of PDIA3 did not affect the interaction between misfolded

374 CD59 (C94S) and calnexin (Figure 3F), suggesting that calnexin binds with misfolded

375 GPI-APs in an ERp57-independent manner.

376

377 Figure 3. Lectin but not ERp57 binding of calnexin is required for efficient GPI-inositol deacylation. 378 (A) Calnexin mutants defective in lectin activity did not rescue the PI-PLC resistance in CANX&CALR-DKO 379 cells. WT and CANX&CALR-DKO cells stably expressing WT CANX or lectin activity-deficient CANX 380 mutants (Y164A, K166A, M188A or E216A) were treated with or without PI-PLC. The relative intensities of 381 remaining CD59 were plotted. The value in WT cells was set as 1. The data are presented as the means ± SD of 382 three independent measurements. P-values (two-tailed, student’s t-test) are shown.

383 (B) WT and CANX&CALR-DKO cells stably expressing WT CANX or CANX mutants defective in ERp57 384 binding (D343A, E351A, W342A&D343A, D343A&E351A or D347A&E351A) were lysed, after which 385 calnexin and calreticulin were detected by immunoblotting. GAPDH was used as a loading control. 386 (C) Calnexin mutants defective in ERp57 and ERp29 binding could rescue the PI-PLC resistance in 387 CANX&CALR-DKO cells. WT and CANX&CALR-DKO cells stably expressing WT CANX or CANX mutants 388 defective in ERp57 binding (D343A, E351A, W342A&D343A, D343A&E351A or D347A&E351A) were 389 treated with or without PI-PLC. The surface expression of CD59 was analyzed by flow cytometry as described in 390 Figure 1B. 391 (D and E) GPI-APs in PDIA3-KO cells showed greater PI-PLC sensitivity than that observed for the parental 392 WT cells. WT and PDIA3-KO cells were treated with or without PI-PLC. After treatment, the surface expression 393 of CD59 and CD55 was analyzed (D) as described in Figure 1B. The levels of CD59 and CD55 remaining after 394 PI-PLC treatment were plotted. The values for the remaining CD59 and CD55 in WT cells were set as 1. The 395 relative values were calculated and are presented as the means ± SD from three independent experiments. 396 P-values (one-tailed, student’s t-test) are shown. 397 (F) Misfolded CD59 (EGFP-FLAG-CD59 (C94S)) interacted with calnexin in PDIA3-KO cells. Constructs 398 expressing misfolded CD59 (EGFP-FLAG-CD59 (C94S)) or soluble misfolded CD59 (EGFP-FLAG-CD59 399 (C94S, G103*)) were transiently transfected into WT and PDIA3-KO cells. Three days after transfection, cells 400 were collected and lysed with buffer containing 0.5% CHAPS. The lysates were immunoprecipitated with 401 anti-FLAG affinity gel, and after washing, the precipitated proteins were released by the addition of SDS sample 402 buffer. The input (2.5% of total protein) and immunoprecipitated fractions (IP) were analyzed by 403 immunoblotting with the indicated antibodies. 404

405 Disruption of the calnexin/calreticulin cycle leads to the exposure of misfolded GPI-APs

406 on the plasma membrane

407 Under steady state conditions, the majority of misfolded prion proteins localize in the ER

408 (Satpute-Krishnan et al., 2014). Because it appears that calnexin retains misfolded GPI-APs in

409 the ER until their correct folding, we investigated the role of calnexin in the ER retention of

410 GPI-APs by analyzing the localization of misfolded GPI-APs in the absence of the

411 calnexin/calreticulin cycle. In HEK293 cells, EGFP-FLAG-tagged CD59 (C94S) was

412 primarily localized in the ER (Figure 4A, top). In CANX-KO and CALR-KO cells, CD59

413 (C94S) was retained in the ER, similar to WT cells. In contrast, in CANX&CALR-DKO cells,

414 CD59 (C94S) was not retained in the ER but rather localized at the plasma membrane (Figure

415 4A, bottom). Quantification of CD59 (C94S) co-localization with the ER marker was

416 performed using multiple images (N=28 for WT cells; N=25 for CANX&CALR-KO cells)

417 (Figure 4D). The Pearson’s correlation coefficient for EGFP-FLAG-CD59 (C94S) versus

418 mRFP-KDEL changed from 0.715±0.05 (meanSD) in WT cells to 0.176±0.04 in

419 CANX&CALR-KO cells. Furthermore, another misfolded GPI-AP, EGFP-FLAG-CD55

420 (C81A), was also expressed on the surface of CANX&CALR-DKO cells (Figure 4-figure

421 supplement A).

422 Since calnexin/calreticulin recognizes monoglucosylated N-glycan structures, we next

423 analyzed the localization of misfolded CD59 in cells defective in glucose trimming. In

424 MOGS-KO or GANAB-KO cells, the majority of N-glycan structures on proteins in the ER

425 are retained as Glc3Man9GlcNAc2 and Glc2Man9GlcNAc2, which are not recognized by

426 calnexin. In MOGS-KO and GANAB-KO cells, fractions of CD59 (C94S) were expressed on

427 the cell surface (Figure 4B), supporting that misfolded GPI-APs lose their ability to interact

428 with calnexin, causing their expression on the cell surface. The Pearson’s correlation

429 coefficient analysis showed localization of EGFP-FLAG-CD59 (C94S) was changed from the

430 ER in MOGS-KO and GANAB-KO cells. These results indicate that the calnexin/calreticulin

431 cycle mediates the ER retention of misfolded GPI-APs.

432 To verify whether N-glycan structures on GPI-APs are required for their ER retention, the

433 localization of non-N-glycosylated misfolded CD59 (C94S, N43Q) and CD55 (C81A, N95Q)

434 was assessed. The results showed that these proteins were not retained in the ER, but fractions

435 of them were expressed on the cell surface, even in the WT cells (Figure 4C, D and Figure

436 4-figure supplement B). In CANX&CALR-DKO cells, non-N-glycosylated misfolded CD59

437 and CD55 were further expressed on the plasma membrane (Figure 4C and Figure 4-figure

438 supplement B). These results are consistent with the co-IP results showing that

439 non-N-glycosylated misfolded CD59 lost its ability to interact with calnexin such that it could

440 not be retained in the ER.

441 The above results suggest that calnexin retains misfolded GPI-APs in the ER through its

442 binding to N-glycans on these proteins. To further investigate this possibility, we assessed the

443 localization of misfolded CD59 in CANX&CALR-DKO cells expressing lectin-deficient

444 calnexin. The lectin-deficient calnexin could not rescue the phenotype of

445 CANX&CALR-DKO cells (Figure 4-figure supplement C), whereas mutant calnexin

446 defective in ERp57 interaction could rescue and retain misfolded CD59 in the ER (Figure

447 4-figure supplement D). These results provide evidence that the lectin domain of calnexin is

448 responsible for the ER retention of N-glycosylated GPI-APs.

449 To assess the specificity of the localization changes in CANX&CALR-DKO cells, a

450 transmembrane form of CD59 (EGFP-FLAG-CD59-TM (C94S)), in which the

451 GPI-attachment signal was replaced with a transmembrane domain of CD46, was constructed,

452 and its localization was analyzed in WT and CANX&CALR-DKO cells. The transmembrane

453 form of misfolded CD59 was localized in the ER in WT cells, whereas the proteins were still

454 retained in the ER of CANX&CALR-DKO cells (Figure 4-figure supplement E), suggesting

455 that the GPI-anchor is responsible for the change in misfolded protein localization in

456 CANX&CALR-DKO cells.

457

458 Figure 4. N-glycan-dependent calnexin/calreticulin cycle mediates the efficient ER retention of misfolded 459 GPI-APs. 460 (A-C) Localization of misfolded CD59 (EGFP-FLAG-CD59 (C94S)) was analyzed in WT, CANX-KO, 461 CALR-KO, and CANX&CALR-DKO cells (A) and in MOGS-KO and GANAB-KO cells (B). Localization of 462 misfolded and non-N-glycosylated CD59 (EGFP-FLAG-CD59 (C94S, N43Q)) was performed as described in 463 (C). mRFP-KDEL was used as an ER marker. Three days after transfection, images were obtained using confocal

464 microscopy. DAPI staining is shown as blue in merged images. Scale bar, 10 μm. 465 (D) Pearson’s correlation coefficient values between EGFP-FLAG-CD59 (C94S) or (C94S, N43Q) and 466 mRFP-KDEL were calculated using the ImageJ plugin JACoP. The data are presented as the means ± SD of the 467 measurements. “N=” represents cell number used for the calculation. P-values (two-tailed, student’s t-test) are 468 shown. 469

470 Misfolded and inositol-acylated GPI-APs are exposed on the plasma membrane in

471 CANX&CALR-DKO cells

472 The results of a previous study confirmed that misfolded prions resulting from acute ER

473 stress are released to the secretory pathway and can reach the plasma membrane

474 (Satpute-Krishnan et al., 2014). In the present study, by treating cells with 100 nM

475 thapsigargin, the localization of misfolded CD59 (C94S) began to shift from the ER to the cell

476 surface (Figure 5A and B). After thapsigargin (TG) treatment, misfolded CD59 was expressed

477 on the cell surface, whereas calnexin was intracellularly retained (Figure 5-figure supplement

478 A).

479 To confirm the specificity of this phenomena that were observed in misfolded GPI-APs,

480 we analyzed the localization of an ERAD substrate, HLA1-1-147 (Zhang, Xu, Liu, & Ye,

481 2015), and that of the transmembrane form of misfolded CD59 (C94S)-TM with or without

482 ER stress induction. Both HLA1-1-147 and CD59 (C94S)-TM remained localized in the ER,

483 even after cells were treated with thapsigargin for 6 h (Figure 5-figure supplement B and C),

484 suggesting that the observed localization changes are specific for GPI-APs.

485 By induction of acute ER stress, misfolded GPI-APs are released into the secretory

486 pathway and can be expressed on the plasma membrane. Flow cytometry results showed that

487 fractions of misfolded CD59 and CD55 were expressed on the cell surface in WT cells

488 without the ER stress, while this surface expression was significantly increased after

489 treatment with thapsigargin for 6 h (by 5.3- and 12.6-fold for CD59 and CD55, respectively)

490 (Figure 5C, left panels, heavy solid line and Figure 5D). The misfolded CD59 and CD55 on

491 the surface of WT cells showed PI-PLC sensitivity (Figure 5C, left panels, gray peak). On the

492 other hand, on the surface of CANX&CALR-DKO cells, the levels of misfolded CD59 and

493 CD55 were 4.7- and 13.5-fold higher than those observed in WT cells without thapsigargin

494 treatment, respectively (Figure 5C, right panels, heavy solid line and Figure 5D). By PI-PLC

495 treatment, misfolded CD59 and CD55 showed PI-PLC resistance (2.6 and 2.4 times higher) in

496 CANX&CALR-DKO cells, compared with WT cells (Figure 5C, top panels, gray peak and

497 Figure 5E). ER stress induction by thapsigargin further increased the PI-PLC resistance of

498 CD59 and CD55 in CANX&CALR-DKO cells compared with that observed in WT cells

499 (11.2- and 4.7-fold higher, respectively) (Figure 5C, bottom panels, gray peak and Figure 5F).

500 These results indicate that misfolded GPI-APs in the ER of WT cells are already

501 inositol-deacylated, and express on the cell surface by ER stress induction. On the other hand,

502 misfolded GPI-APs in CANX&CALR-DKO cells are expressed regardless of acute ER stress,

503 parts of whose GPI-moieties are not processed. The difference in PI-PLC sensitivity of

504 GPI-APs between CANX&CALR-DKO cells and thapsigargin-treated WT cells suggests that

505 ER retention time is important for efficient GPI-inositol deacylation. In the WT cells,

506 misfolded GPI-APs were retained in the ER for a long time, providing them sufficient time to

507 interact with the GPI-inositol deacylase PGAP1. In contrast, in CANX&CALR-DKO cells,

508 misfolded GPI-APs could not be retained in the ER.

509

510 Figure 5. Misfolded and inositol-deacylated GPI-APs are exposed on the plasma membrane under acute 511 ER stress conditions. 512 (A and B) EGFP-FLAG-CD59 (C94S) and mRFP-KDEL constructs were transiently transfected into WT cells. 513 Three days after transfection, the cells were treated with 100 nM thapsigargin (TG) for 1, 3 and 6 h. The cells 514 were fixed and imaged by confocal microscopy (A). Scale bar, 10 μm. Pearson’s correlation coefficient values 515 between EGFP-FLAG-CD59 (C94S) and mRFP-KDEL were calculated using the ImageJ plugin JACoP (B). The

516 data are presented as the means ± SD of the measurements. “N=” represents cell number used for the calculation. 517 (C) EGFP-FLAG-CD59 (C94S) or EGFP-FLAG-CD55 (C81A) constructs together with a BFP-expressing 518 plasmid were transiently transfected into WT and CANX&CALR-DKO cells. Three days after transfection, the 519 cells were incubated with or without 100 nM TG for 6 h and then treated with or without PI-PLC. The surface 520 expression of EGFP-FLAG-CD59 (C94S) and EGFP-FLAG-CD55 (C81A) was analyzed by flow cytometry. 521 Cells showing the same BFP intensities were gated for the same transfection level. 522 (D) Surface expression of misfolded CD59 (EGFP-FLAG-CD59 (C94S)) or CD55 (EGFP-FLAG-CD55 (C81A)) 523 in WT or CANX&CALR-DKO cells with or without TG treatment for 6 h shown in Figure 5C (heavy solid line) 524 were plotted. The data are presented as the means ± SD of three independent measurements. P-values (one-tailed, 525 student’s t-test) are shown. 526 (E and F) After PI-PLC treatment, surface misfolded CD59 (EGFP-FLAG-CD59 (C94S)) and CD55 527 (EGFP-FLAG-CD55 (C81A)) in WT or CANX&CALR-DKO cells without (E) or with (F) TG treatment for 6 h 528 shown in Figure 5C (gray shaded) were plotted. The data are presented as the means ± SD of three independent 529 measurements. P-values (one-tailed, student’s t-test) are shown. 530

531 Alkaline phosphatase activity is greatly decreased in CANX&CALR-DKO cells

532 compared to that observed in WT cells

533 Given the results showing that misfolded GPI-APs were expressed on the plasma

534 membrane in CANX&CALR-DKO cells, we hypothesized that misfolded/unfolded

535 endogenous GPI-APs are expressed on the plasma membrane in CANX&CALR-DKO cells

536 under normal conditions. Alkaline phosphatases are GPI-APs, for which there are four types,

537 including tissue non-specific, intestinal, placental, and placental-like typestypes in human.

538 Among these different types, the expression of the placental alkaline phosphatase (ALPP,

539 contain 2 N-glycosylation sites) is the highest in HEK293 cells (Figure 6-figure supplement).

540 The surface expression of endogenous ALPP was almost the same in WT and

541 CANX&CALR-DKO cells (Figure 6A, heavy solid lines and Figure 6B). The ALPP showed

542 2.3-fold higher PI-PLC resistance in CANX&CALR-DKO cells compared to that observed in

543 WT cells (Figure 6A, gray peaks and Figure 6C). We then assessed the ALP activity on the

544 cell surface and compared with WT cells, the activity in CANX&CALR-DKO cells was

545 significantly decreased by 60% (Figure 6D). These results indicate that GPI-APs that are not

546 fully functional are expressed on the plasma membrane in CANX&CALR-DKO cells.

547

548 Figure 6. ALPP activity decreases in CANX&CALR-DKO cells. 549 (A - C) Sensitivity of PI-PLC to placental alkaline phosphatase (ALPP) in WT and CANX&CALR-DKO cells 550 was analyzed by flow cytometry (A). An anti-ALPP antibody was used as the primary antibody. The dashed lines 551 show the background of the isotype control. The surface expression of ALPP in WT or CANX&CALR-DKO 552 cells was plotted (B). The data are presented as the means ± SD of three independent measurements. The levels 553 of ALPP remaining after PI-PLC treatment are plotted (C). The remaining ALPP values in WT cells were set as 1. 554 The relative values were calculated and are presented as the means ± SD from three independent experiments. 555 P-values (one-tailed, student’s t-test) are shown. 556 (D) The relative activity of surface ALP in WT and CANX&CALR-DKO cells are shown. The activity of 557 surface ALP was measured as described in the Materials and Methods. The data are presented as the means ± SD 558 of three independent measurements. P-values (two-tailed Student’s t-test) are shown. 559

560

561 Sufficient ER retention time is required for efficient inositol deacylation by PGAP1

562 To confirm whether sufficient ER retention time is required for GPI-inositol deacylation,

563 a doxycycline (DOX)-inducible VSVGts-FLAG-GFP-GPI (VFG-GPI) reporter system (Maeda

564 et al., 2008) was used. VFG-GPI is a temperature-sensitive glycoprotein, and the reporter

565 causes misfolding at 40°C due to a mutation in the VSVG portion and is retained in the ER,

566 whereas it is folded at 32°C and transported to the plasma membrane (Fujita et al., 2009;

567 Fujita et al., 2011; Hirata et al., 2015). Since CANX&CALR-DKO cells cannot retain

568 misfolded GPI-APs such as CD59 and CD55 in the ER under normal culture conditions

569 (37°C), the localization of VFG-GPI in CANX&CALR-DKO cells at 40°C was assessed.

570 VFG-GPI was localized in the ER in WT, CANX&CALR-DKO, and CANX-rescued

571 CANX&CALR-KO cells (Figure 7A) under 40°C.

572 We then analyzed surface-expressed VFG-GPI under different conditions. When

573 VFG-GPI expression was induced at 32°C, VFG-GPI in WT and CANX-rescued

574 CANX&CALR-DKO cells was sensitive to PI-PLC (Figure 7B, top), whereas partial

575 VFG-GPI resistance to PI-PLC was detected in CANX&CALR-DKO cells (3.6-fold higher

576 than that measured for WT cells) (Figure 7B, top panels, middle and Figure 7C, left), which is

577 consistent with that observed for other GPI-APs.

578 We next analyzed VFG-GPI using a chase experiment. VFG-GPI was expressed at 40°C

579 to accumulate in the ER, after which the temperature was shifted to 32°C to allow its transport

580 to the plasma membrane. Under these condition, surface VFG-GPI in CANX&CALR-DKO

581 became sensitive to PI-PLC cleavage, similar to that observed in WT cells (Figure 7B, bottom

582 panels and Figure 7C, right). These results proved that sufficient ER retention time is

583 necessary for GPI-inositol deacylation.

584

585 Figure 7. ER retention time regulates GPI-inositol deacylation. 586 (A) Localization of VSVGts-FLAG-GFP-GPI (VFG-GPI) at 40°C in WT, CANX&CALR-DKO, and 587 CANX&CALR-DKO+CANX WT cells. The construct expressing mRFP-KDEL, an ER marker, was transfected 588 into WT, CANX&CALR-DKO and CANX&CALR-DKO+CANX WT cells. At 36 h after transfection, cells 589 were incubated with 1 µg/ml doxycycline at 40°C for 24 h to induce VFG-GPI expression. Subsequently, the 590 cells were quickly fixed with 4% paraformaldehyde and then imaged by confocal microscopy. Scale bar, 10 μm.

591 (B) HEK293FF6WT, CANX&CALR-DKO and CANX&CALR-DKO+CANX WT cells were incubated with 592 with 1 µg/ml doxycycline at 32°C for 24 h (upper panels). Alternatively, cells were incubated with 1 µg/ml 593 doxycycline at 40°C for 24 h followed by a 32°C incubation for 1 h (lower panels). After induction, the cells 594 were treated with or without PI-PLC, and surface VFG-GPI was stained with an anti-FLAG antibody and 595 analyzed by flow cytometry. The region of same GFP intensity was gated to normalize VFG-GPI expression. 596 (C) Remaining of VFG-GPI after PI-PLC treatment in (B) were plotted. The values in WT cells were set as 1, 597 and relative values in CANX&CALR-DKO were plotted. The data are presented as the means ± SD of three 598 independent measurements. P-values (one-tailed, student’s t-test) are shown. 599

600 Discussion

601 Both calnexin-mediated ER quality control and GPI-inositol deacylation by PGAP1 are

602 performed in the ER. The calnexin/calreticulin cycle is essential for N-glycosylated protein

603 folding (Tannous et al., 2015), and GPI-inositol deacylation is required for the efficient

604 sorting of GPI-APs into transport vesicles and their subsequent transport from ER to the Golgi

605 apparatus (Fujita & Kinoshita, 2012; Muñiz & Riezman, 2016). In the present study, we

606 confirmed that calnexin not only contributes to protein folding but also GPI-inositol

607 deacylation, playing dual roles in the maturation of GPI-APs in the ER.

608 In WT cells, calnexin was observed to retain GPI-APs in the ER until achieving correct

609 folding by directly binding to the N-glycans on the GPI-APs (Figure 8, top). The results of

610 our previous study showed that PGAP1 associates with calnexin (Liu et al., 2018).

611 Interestingly, the temporal ER retention of GPI-APs and their association with PGAP1

612 mediated by calnexin promoted the efficient cleavage of acyl chains from GPI-anchors.

613 Correctly folded and inositol deacylated GPI-APs are further processed by PGAP5 and

614 recognized by p24 cargo receptors to become incorporated into transport vesicles. In

615 CANX&CALR-DKO cells, newly synthesized GPI-APs could not be retained in the ER, and

616 fractions of GPI-APs were transported without correct folding and GPI processing (Figure 8,

617 middle). Since structural remodeling of GPI-APs is crucial for their binding to p24 cargo

618 receptors, inositol-acylated GPI-APs would exit from the ER through bulk flow pathway.

619 Since ER quality control systems are impaired in CANX&CALR-DKO cells, both correctly

620 folded and misfolded GPI-APs are expressed on the cell surface. The observation that more

621 than 90% of GPI-APs are N-glycosylated (Liu et al., 2018) indicates that calnexin-dependent

622 GPI-anchor maturation is a commonly used pathway. In addition, calnexin-independent

623 GPI-inositol deacylation occurs for non-N-glycosylated GPI-APs. Other molecular

624 chaperones such as BiP and protein disulfide isomerases, which possess ER-retention signals,

625 may be involved in the folding and GPI-processing of non-N-glycosylated proteins. When

626 misfolded GPI-APs are expressed under basal conditions, they are retained in the ER by

627 calnexin. During the long ER retention time, an acyl-chain on the GPI-inositol is cleaved by

628 PGAP1 (Figure 8, bottom). Since the GPI moiety of misfolded ER-resident GPI-APs signals

629 that they are ready to exit, they would be quickly transported from the ER via acute ER stress

630 induction.

631 Calnexin/calreticulin interact with newly synthesized proteins in an N-glycan-dependent

632 manner. However, once high molecular weight aggregates of the misfolded protein are

633 generated, calnexin can interact with them in an N-glycan-independent manner (Cannon,

634 Hebert, & Helenius, 1996; Thammavongsa, Mancino, & Raghavan, 2005).

635 N-glycan-dependent calnexin binding is a mechanism to protect against misfolded GPI-APs

636 becoming insoluble aggregates, while N-glycan-independent binding will lead to the

637 degradation of misfolded proteins (Seckler & Jaenicke, 1992; Swanton, High, & Woodman,

638 2003). Calnexin binding to GPI-APs was observed to occur in an N-glycan-dependent and

639 PDI-independent manner. In PDIA3-KO cells, GPI-APs showed greater PI-PLC sensitivity,

640 suggesting that ERp57 is dispensable for their binding and that calnexin could retain GPI-APs

641 for longer time in the ER to allow their correct folding and GPI-processing.

642 Rapid ER stress-induced export (RESET) is a degradation pathway for misfolded

643 GPI-APs under acute ER stress conditions (Satpute-Krishnan et al., 2014). Misfolded

644 GPI-APs are retained in the ER under normal conditions. However, through the induction of

645 acute ER stress, misfolded GPI-APs are transported from the ER to the plasma membrane and

646 degraded in lysosomes. Our results are consistent with this process. Misfolded GPI-APs are

647 recognized by calnexin, which is dependent on N-glycan for retention in the ER. We

648 demonstrated that interaction of GPI-APs with calnexin contributes to GPI-anchor processing

649 by PGAP1. Recently, misfolded prions have been reported to be expressed on the plasma

650 membrane in association with calnexin and p24 family proteins by the induction of acute ER

651 stress (Zavodszky & Hegde, 2019). In the present study, we showed that misfolded GPI-APs

652 were stably expressed on the cell surface in the absence of calnexin and calreticulin,

653 indicating that at least calnexin is not required for the exit of GPI-APs from the ER under

654 stress conditions. Disruption of the calnexin/calreticulin cycle leads to the exit of misfolded

655 GPI-APs from the ER due to a lack of ER retention mechanisms. In the RESET pathway,

656 misfolded GPI-APs are subsequently endocytosed and delivered to lysosomes. In

657 CANX&CALR-DKO cells, misfolded GPI-APs were stably expressed on the cell surface. In

658 future studies, it would be worth identifying determinants for the endocytosis of misfolded

659 GPI-APs.

660 Misfolded CD59 (C94S) and CD55 (C81A) have been shown to be the substrates for the

661 RESET pathway (Satpute-Krishnan et al., 2014). When we replaced the GPI-attachment

662 signal at the C-terminus to the transmembrane domain of CD46 (CD59 (C94S) TM) or used

663 an ERAD substrate HLA1-1-147, their localization from the ER to the plasma membrane was

664 not altered in CANX&CALR-DKO cells or under thapsigargin treatment. Therefore, it

665 appears that the change in the localization of misfolded proteins in CANX&CALR-DKO cells

666 is specific to GPI-APs. Since GPI-APs are tethered to the membrane through a phospholipid,

667 their behaviors would be close to soluble proteins or phospholipids rather than transmembrane

668 proteins. Some soluble ER-resident proteins having ER-retention signals have been reported

669 to be secreted by the depletion of ER calcium (Trychta, Back, Henderson, & Harvey, 2018),

670 which is a trigger of the unfolded stress response. It remains unclear how such ER-resident

671 proteins are secreted in response to ER stress conditions. Thus, analysis of misfolded

672 GPI-APs may allow for a better understanding of the associated mechanisms, which should be

673 addressed in future studies.

674

675

676 Figure 8. Summary models for folding and inositol deacylation of GPI-APs regulated by 677 calnexin/calreticulin. 678 Top: Under normal conditions, N-glycans and GPI are transferred to the newly synthesized GPI-APs. After 679 glucose trimming by α-glucosidase I and II, the N-glycan structure becomes Glc1Man9GlcNAc2, which is 680 specifically recognized by calnexin/calreticulin. By directly binding with calnexin, immature GPI-APs enter the 681 ER quality control system and finally become mature GPI-APs. In addition, calnexin retains GPI-APs in the ER 682 and associates with PGAP1, which is required for the efficient GPI-inositol deacylation. After the GPI moiety is 683 remodeled by PGAP1 and PGAP5, and GPI-APs are transported from the ER to the Golgi by the cargo receptors, 684 p24 family of proteins. 685 Middle: In CANX&CALR-DKO cells, there is no calnexin/calreticulin-dependent ER quality control system and 686 ER retention system. GPI-APs exit the ER with incomplete protein folding and GPI remodeling. 35

687 Bottom: In WT cells, under basal conditions, misfolded GPI-APs are retained in the ER by calnexin. During the 688 long time being retained in the ER, an acyl-chain on the GPI-inositol is removed by PGAP1. Once acute ER 689 stress is induced, deacylated and misfolded GPI-APs are quickly transported by p24 family members and 690 exposed on the plasma membrane. 691

692 Acknowledgements

693 We thank Drs Hideki Nakanishi, Ganglong Yang, Ning Wang (Jiangnan University) for their

694 critical reading of the manuscript and comments. This work was supported by grants-in-aid

695 from the National Natural Science Foundation of China (31770853), the Program of

696 Introducing Talents of Discipline to Universities (111-2-06), National first-class discipline

697 program of Light Industry Technology and Engineering (LITE2018-015), Top-notch

698 Academic Programs Project of Jiangsu Higher Education Institutions, the International Joint

699 Research Laboratory for Investigation of Glycoprotein Biosynthesis at Jiangnan University,

700 and a grant for Joint Research Project of the Research Institute for Microbial Diseases, Osaka

701 University.

702

703 Conflict of interest

704 The authors declare that they have no conflicts of interest with the contents of this article.

705

706

707 References 708 Cannon, K. S., Hebert, D. N., & Helenius, A. (1996). Glycan-dependent and -independent 709 association of vesicular stomatitis virus G protein with calnexin. Journal of Biological 710 Chemistry, 271(24), 14280-14284. doi:10.1074/jbc.271.24.14280 711 Chen, R., Walter, E. I., Parker, G., Lapurga, J. P., Millan, J. L., Ikehara, Y., Udenfriend, S., & 712 Medof, M. E. (1998). Mammalian glycophosphatidylinositol anchor transfer to 713 proteins and posttransfer deacylation. Proceedings of the National Academy of 714 Sciences of the United States of America, 95(16), 9512-9517. 715 doi:10.1073/pnas.95.16.9512 716 Ellgaard, L., & Frickel, E. M. (2003). Calnexin, calreticulin, and ERp57: teammates in 717 glycoprotein folding. Cell Biochemistry and Biophysics, 39(3), 223-247. 718 doi:10.1385/CBB:39:3:223 719 Ferguson, M. A. J., Kinoshita, T., & Hart, G. W. (2009). Glycosylphosphatidylinositol 720 Anchors. In nd, A. Varki, R. D. Cummings, J. D. Esko, H. H. Freeze, P. Stanley, C. R. 721 Bertozzi, G. W. Hart, & M. E. Etzler (Eds.), Essentials of Glycobiology. Cold Spring 722 Harbor (NY): Cold Spring Harbor Laboratory Press. 723 Frickel, E.-M., Riek, R., Jelesarov, I., Helenius, A., Wüthrich, K., & Ellgaard, L. (2002). 724 TROSY-NMR reveals interaction between ERp57 and the tip of the calreticulin 725 P-domain. Proceedings of the National Academy of Sciences of the United States of 726 America, 99(4), 1954-1959. doi:10.1073/pnas.042699099 727 Fujita, M., & Kinoshita, T. (2012). GPI-anchor remodeling: potential functions of 728 GPI-anchors in intracellular trafficking and membrane dynamics. Biochimica et 729 Biophysica Acta, 1821(8), 1050-1058. doi:10.1016/j.bbalip.2012.01.004 730 Fujita, M., Maeda, Y., Ra, M., Yamaguchi, Y., Taguchi, R., & Kinoshita, T. (2009). GPI glycan 731 remodeling by PGAP5 regulates transport of GPI-anchored proteins from the ER to 732 the Golgi. Cell, 139(2), 352-365. doi:10.1016/j.cell.2009.08.040 733 Fujita, M., Watanabe, R., Jaensch, N., Romanova-Michaelides, M., Satoh, T., Kato, M., 734 Riezman, H., Yamaguchi, Y., Maeda, Y., & Kinoshita, T. (2011). Sorting of 735 GPI-anchored proteins into ER exit sites by p24 proteins is dependent on remodeled 736 GPI. Journal of Cell Biology, 194(1), 61-75. doi:10.1083/jcb.201012074 737 Fujita, M., Yoko-o, T., & Jigami, Y. (2006). Inositol deacylation by Bst1p is required for the 738 quality control of glycosylphosphatidylinositol-anchored proteins. Molecular Biology 739 of the Cell, 17(2), 834-850. doi:10.1091/mbc.e05-05-0443 740 Granzow, M., Paramasivam, N., Hinderhofer, K., Fischer, C., Chotewutmontri, S., Kaufmann, 741 L., Evers, C., Kotzaeridou, U., Rohrschneider, K., & Schlesner, M. (2015). Loss of 742 function of PGAP1 as a cause of severe encephalopathy identified by Whole Exome 743 Sequencing: Lessons of the bioinformatics pipeline. Molecular and Cellular Probes, 744 29(5), 323-329. doi:10.1016/j.mcp.2015.05.012 745 Hammond, C., Braakman, I., & Helenius, A. (1994). Role of N-linked oligosaccharides, 746 glucose trimming and calnexin during glycoprotein folding in the endoplasmic 747 reticulum. Proceedings of the National Academy of Sciences of the United States of 748 America, 91, 913-917. doi:10.1073/pnas.91.3.913 749 Heigwer, F., Kerr, G., & Boutros, M. (2014). E-CRISP: fast CRISPR target site identification. 750 Nature Methods, 11(2), 122-123. doi:10.1038/nmeth.2812

751 Heinz, D., Ryan, M., Bullock, T., & Griffith, O. (1995). Crystal structure of the 752 phosphatidylinositol-specific phospholipase C from Bacillus cereus in complex with 753 myo-inositol. EMBO Journal, 14(16), 3855-3863. 754 doi:doi.org/10.1002/j.1460-2075.1995.tb00057.x 755 Helenius, A., & Aebi, M. (2004). Roles of N-linked glycans in the endoplasmic reticulum. 756 Annual Review of Biochemistry, 73(1), 1019-1049. 757 doi:10.1146/annurev.biochem.73.011303.073752 758 Hirata, T., Fujita, M., Nakamura, S., Gotoh, K., Motooka, D., Murakami, Y., Maeda, Y., & 759 Kinoshita, T. (2015). Post-Golgi anterograde transport requires GARP-dependent 760 endosome-to-TGN retrograde transport. Molecular Biology of the Cell, 26(17), 761 3071-3084. doi:10.1091/mbc.E14-11-1568 762 Hirayama, H., Fujita, M., Yoko-o, T., & Jigami, Y. (2008). O-mannosylation is required for 763 degradation of the endoplasmic reticulum-associated degradation substrate Gas1*p via 764 the ubiquitin/proteasome pathway in Saccharomyces cerevisiae. Journal of 765 Biochemistry, 143(4), 555-567. doi:10.1093/jb/mvm249 766 Ikezawa, H. (2002). Glycosylphosphatidylinositol (GPI)-anchored proteins. Biological and 767 Pharmaceutical Bulletin, 25(4), 409-417. doi:10.1248/bpb.25.409 768 Ishida, Y., Komaru, K., Ito, M., Amaya, Y., Kohno, S., & Oda, K. (2003). Tissue-nonspecific 769 alkaline phosphatase with an Asp(289)-->Val mutation fails to reach the cell surface 770 and undergoes proteasome-mediated degradation. Journal of Biochemistry, 134(1), 771 63-70. doi:10.1093/jb/mvg114 772 Kettwig, M., Elpeleg, O., Wegener, E., Dreha-Kulaczewski, S., Henneke, M., Gartner, J., & 773 Huppke, P. (2016). Compound heterozygous variants in PGAP1 causing severe 774 psychomotor retardation, brain atrophy, recurrent apneas and delayed myelination: a 775 case report and literature review. BMC Neurology, 16(1), 74. 776 doi:10.1186/s12883-016-0602-7 777 Kinoshita, T. (2020). Biosynthesis and biology of mammalian GPI-anchored proteins. Open 778 Biology Journal, 10(3), 190290. doi:10.1098/rsob.190290 779 Kmenon, A. (1994). Structural analysis of glycosylphosphatidylinositol anchors. In Methods 780 in Enzymology (Vol. 230, pp. 418-442): Elsevier. 781 Kozlov, G., Muñoz-Escobar, J., Castro, K., & Gehring, K. (2017). Mapping the ER 782 interactome: The P domains of calnexin and calreticulin as plurivalent adapters for 783 foldases and chaperones. Structure, 25(9), 1415-1422. e1413. 784 doi:10.1016/j.str.2017.07.010 785 Leach, M. R., Cohen-Doyle, M. F., Thomas, D. Y., & Williams, D. B. (2002). Localization of 786 the lectin, ERp57 binding, and polypeptide binding sites of calnexin and calreticulin. 787 Journal of Biological Chemistry, 277(33), 29686-29697. 788 doi:10.1074/jbc.M202405200 789 Leach, M. R., & Williams, D. B. (2004). Lectin-deficient calnexin is capable of binding class 790 I histocompatibility molecules in vivo and preventing their degradation. Journal of 791 Biological Chemistry, 279(10), 9072-9079. doi:10.1074/jbc.M310788200 792 Liu, Y.-S., Guo, X.-Y., Hirata, T., Rong, Y., Motooka, D., Kitajima, T., Murakami, Y., Gao, 793 X.-D., Nakamura, S., & Kinoshita, T. (2018). N-Glycan–dependent protein folding 794 and endoplasmic reticulum retention regulate GPI-anchor processing. Journal of Cell

795 Biology, 217(2), 585-599. doi:10.1083/jcb.201706135 796 Maeda, Y., Ide, T., Koike, M., Uchiyama, Y., & Kinoshita, T. (2008). GPHR is a novel anion 797 channel critical for acidification and functions of the Golgi apparatus. Nature Cell 798 Biology, 10(10), 1135. doi:10.1038/ncb1773 799 Maeda, Y., Tashima, Y., Houjou, T., Fujita, M., Yoko-o, T., Jigami, Y., Taguchi, R., & 800 Kinoshita, T. (2007). Fatty acid remodeling of GPI-anchored proteins is required for 801 their raft association. Molecular Biology of the Cell, 18(4), 1497-1506. 802 doi:10.1091/mbc.e06-10-0885 803 McConville, M. J., & Menon, A. K. (2000). Recent developments in the cell biology and 804 biochemistry of glycosylphosphatidylinositol lipids (review). Molecular Membrane 805 Biology, 17(1), 1-16. doi:10.1080/096876800294443 806 McKean, D. M., & Niswander, L. (2012). Defects in GPI biosynthesis perturb Cripto 807 signaling during forebrain development in two new mouse models of 808 holoprosencephaly. Biology Open, 1(9), 874-883. doi:10.1242/bio.20121982 809 Muñiz, M., & Riezman, H. (2016). Trafficking of glycosylphosphatidylinositol anchored 810 proteins from the endoplasmic reticulum to the cell surface. Journal of Lipid Research, 811 57(3), 352-360. doi:10.1194/jlr.R062760 812 Murakami, Y., Tawamie, H., Maeda, Y., Buttner, C., Buchert, R., Radwan, F., Schaffer, S., 813 Sticht, H., Aigner, M., Reis, A., Kinoshita, T., & Jamra, R. A. (2014). Null mutation in 814 PGAP1 impairing Gpi-anchor maturation in patients with intellectual disability and 815 encephalopathy. Plos Genetics, 10(5), e1004320. doi:10.1371/journal.pgen.1004320 816 Pollock, S., Kozlov, G., Pelletier, M. F., Trempe, J. F., Jansen, G., Sitnikov, D., Bergeron, J. J., 817 Gehring, K., Ekiel, I., & Thomas, D. Y. (2004). Specific interaction of ERp57 and 818 calnexin determined by NMR spectroscopy and an ER two‐hybrid system. EMBO 819 Journal, 23(5), 1020-1029. doi:10.1038/sj.emboj.7600119 820 Satpute-Krishnan, P., Ajinkya, M., Bhat, S., Itakura, E., Hegde, R. S., & Lippincott-Schwartz, 821 J. (2014). ER stress-induced clearance of misfolded GPI-anchored proteins via the 822 secretory pathway. Cell, 158(3), 522-533. doi:10.1016/j.cell.2014.06.026 823 Schrag, J. D., Bergeron, J. J., Li, Y., Borisova, S., Hahn, M., Thomas, D. Y., & Cygler, M. 824 (2001). The Structure of calnexin, an ER chaperone involved in quality control of 825 protein folding. Molecular Cell, 8(3), 633-644. doi:10.1016/s1097-2765(01)00318-5 826 Schwarz, F., & Aebi, M. (2011). Mechanisms and principles of N-linked protein glycosylation. 827 Current Opinion in Structural Biology, 21(5), 576-582. doi:10.1016/j.sbi.2011.08.005 828 Seckler, R., & Jaenicke, R. (1992). Protein folding and protein refolding. FASEB Journal, 6(8), 829 2545-2552. doi:10.1096/fasebj.6.8.1592207 830 Sikorska, N., Lemus, L., Aguilera-Romero, A., Manzano-Lopez, J., Riezman, H., Muniz, M., 831 & Goder, V. (2016). Limited ER quality control for GPI-anchored proteins. Journal of 832 Cell Biology, 213(6), 693-704. doi:10.1083/jcb.201602010 833 Spiro, R. G., Zhu, Q., Bhoyroo, V., & Soling, H. D. (1996). Definition of the lectin-like 834 properties of the molecular chaperone, calreticulin, and demonstration of its 835 copurification with endomannosidase from rat liver Golgi. Journal of Biological 836 Chemistry, 271(19), 11588-11594. doi:10.1074/jbc.271.19.11588 837 Sun, Z., & Brodsky, J. L. (2019). Protein quality control in the secretory pathway. Journal of 838 Cell Biology, 218(10), 3171-3187. doi:10.1083/jcb.201906047

839 Swanton, E., High, S., & Woodman, P. (2003). Role of calnexin in the glycan‐independent 840 quality control of proteolipid protein. EMBO Journal, 22(12), 2948-2958. 841 doi:10.1093/emboj/cdg300 842 Takida, S., Maeda, Y., & Kinoshita, T. (2008). Mammalian GPI-anchored proteins require p24 843 proteins for their efficient transport from the ER to the plasma membrane. 844 Biochemical Journal, 409(2), 555-562. doi:10.1042/BJ20070234 845 Tanaka, S., Maeda, Y., Tashima, Y., & Kinoshita, T. (2004). Inositol deacylation of 846 glycosylphosphatidylinositol-anchored proteins is mediated by mammalian PGAP1 847 and yeast Bst1p. Journal of Biological Chemistry, 279(14), 14256-14263. 848 doi:10.1074/jbc.M313755200 849 Tannous, A., Pisoni, G. B., Hebert, D. N., & Molinari, M. (2015). N-linked sugar-regulated 850 protein folding and quality control in the ER. Paper presented at the Seminars in Cell 851 and Developmental Biology. 852 Thammavongsa, V., Mancino, L., & Raghavan, M. (2005). Polypeptide substrate recognition 853 by calnexin requires specific conformations of the calnexin protein. Journal of 854 Biological Chemistry, 280(39), 33497-33505. doi:10.1074/jbc.M503648200 855 Tiede, A., Bastisch, I., Schubert, J., Orlean, P., & Schmidt, R. E. (1999). Biosynthesis of 856 glycosylphosphatidylinositols in mammals and unicellular microbes. Biological 857 Chemistry, 380(5), 503-523. doi:10.1515/BC.1999.066 858 Trychta, K. A., Back, S., Henderson, M. J., & Harvey, B. K. (2018). KDEL Receptors Are 859 Differentially Regulated to Maintain the ER Proteome under Calcium Deficiency. Cell 860 Reports, 25(7), 1829-1840 e1826. doi:10.1016/j.celrep.2018.10.055 861 Ueda, Y., Yamaguchi, R., Ikawa, M., Okabe, M., Morii, E., Maeda, Y., & Kinoshita, T. (2007). 862 PGAP1 knock-out mice show otocephaly and male infertility. Journal of Biological 863 Chemistry, 282(42), 30373-30380. doi:10.1074/jbc.M705601200 864 Vassilakos, A., Michalak, M., Lehrman, M. A., & Williams, D. B. (1998). Oligosaccharide 865 binding characteristics of the molecular chaperones calnexin and calreticulin. 866 Biochemistry, 37(10), 3480-3490. doi:10.1021/bi972465g 867 Vembar, S. S., & Brodsky, J. L. (2008). One step at a time: endoplasmic reticulum-associated 868 degradation. Nature Reviews: Molecular Cell Biology, 9(12), 944-957. 869 doi:10.1038/nrm2546 870 Zavodszky, E., & Hegde, R. S. (2019). Misfolded GPI-anchored proteins are escorted through 871 the secretory pathway by ER-derived factors. eLife, 8, e46740. 872 doi:10.7554/eLife.46740 873 Zhang, T., Xu, Y., Liu, Y., & Ye, Y. (2015). gp78 functions downstream of Hrd1 to promote 874 degradation of misfolded proteins of the endoplasmic reticulum. Molecular Biology of 875 the Cell, 26(24), 4438-4450. doi:10.1091/mbc.E15-06-0354 876 Zoltewicz, J. S., Ashique, A. M., Choe, Y., Lee, G., Taylor, S., Phamluong, K., Solloway, M., 877 & Peterson, A. S. (2009). Wnt signaling is regulated by endoplasmic reticulum 878 retention. PloS One, 4(7), e6191. doi:10.1371/journal.pone.0006191

879

880

881 Supplemental Figures

882 883 Figure 1-supplement. Calnexin/calreticulin-dependent GPI-inositol deacylation is commonly observed in 884 N-glycosylated GPI-APs. 885 HA-tagged or His6-tagged exogenous GPI-APs were expressed in WT and CANX&CALR-DKO cells, and their 886 PI-PLC sensitivity was analyzed. The pME-HA-GPI-AP plasmid was transfected together with pME-BFP into 887 WT and CANX&CALR-DKO cells. The plasmid pLIB2-His6-GPI-AP-IRES-BFP was transfected into 888 HEK293T packaging cells, and retroviral vectors were infected into WT and CANX&CALR-DKO cells. Three 889 days after transfection, cells were treated with or without PI-PLC. Anti-HA or anti-His6 were used as the 890 primary antibodies for staining the surface proteins, which were analyzed by flow cytometry. The cells showing 891 the same BFP intensities were gated. The values are shown in Figure 1C.

892 893 Figure 3-supplement. N-glycan-binding activity of calnexin is necessary for efficient GPI-inositol 894 deacylation. 895 (A and B) WT and CANX&CALR-DKO cells stably expressing WT CANX or CANX mutants defective in 896 lectin activity (Y164A, K166A, M188A or E216A) were treated with or without PI-PLC. Surface CD59 was 897 detected by flow cytometry (A) as described in Figure 1B. CANX and CALR from cell lysates in (A) were 898 analyzed by immunoblotting (B). GAPDH was used as a loading control. 899 (C) Cells were transiently transfected with constructs expressing WT EGFP-FLAG-CD59, misfolded CD59 900 (EGFP-FLAG-CD59 (C94S)), misfolded CD59 lacking GPI (EGFP-FLAG-CD59 (C94S, G103*)), or misfolded 901 CD59 lacking an N-glycan (EGFP-FLAG-CD59 (C94S, N43Q)) and then were lysed with lysis buffer containing 902 0.5% CHAPS. The lysates were then subjected to immunoprecipitation (IP) with anti-FLAG affinity gel, and 903 after washing, the precipitated proteins were released by the addition of SDS sample buffer. The input (10% of 904 total protein) and immunoprecipitated fractions were analyzed by immunoblotting with the indicated antibodies.

905 906 Figure 4-supplement. Localization of misfolded GPI-APs in CANX&CALR-DKO cells. 907 (A and B) EGFP-FLAG-CD55 (C81A) or EGFP-FLAG-CD55 (C81A, N95Q) constructs were transiently 908 transfected together with mRFP-KDEL into WT and CANX&CALR-DKO cells. The images were obtained 909 using confocal microscopy at 3 days after transfection. DAPI staining is shown as blue in merged images. Scale 910 bar, 10 μm.

911 912 (C and D) EGFP-FLAG-CD59 (C94S) and mRFP-KDEL constructs were transiently transfected into 913 CANX&CALR-DKO cells carrying CANX mutants defective in N-glycan binding activity (Y164A, K166A, 914 M188A or E216A) or defective in ERp57 binding (W342A&D343A, D343A&E351A or D347A&E351A). 915 Confocal images were obtained at 3 days after transfection. Scale bar, 10 μm.

916 917 (E) The GPI attachment signal of CD59 (C94S) was replaced with the transmembrane sequence of CD46 to 918 generate a transmembrane form of misfolded CD59 (C94S) (EGFP-FLAG-CD59 (C94S) TM) and was 919 co-transfected with mRFP-KDEL into WT and CANX&CALR-DKO cells. The localization of 920 EGFP-FLAG-CD59 (C94S) TM was imaged by confocal microscopy. Scale bar, 10 μm.

921 922 Figure 5-supplement. Protein localization change from the ER to the plasma membrane is specific for 923 GPI-APs. 924 (A) Localization of endogenous calnexin under acute ER stress. HEK293 cells transfected with 925 EGFP-FLAG-CD59 (C94S) were incubated with or without 100 nM TG for 6 h. Cells were then fixed, stained 926 with an anti-calnexin antibody followed by an anti-mouse-Alexa Fluor 555, and detected by confocal microscopy. 46

927 Scale bar, 10 μm. 928 (B and C) Localization of HLA1-1-147-FLAG (B) and EGFP-FLAG-CD59 (C94S) TM (C) under acute ER 929 stress. HLA1-1-147-FLAG or EGFP-FLAG-CD59 (C94S) TM and mRFP-KDEL constructs were transfected 930 into WT cells. Cells were then incubated with or without 100 nM TG for 6 h 3 days after transfection. 931 HLA1-1-147-FLAG was stained with anti-FLAG antibody follow by anti-mouse-Alexa Fluor 488, and detected 932 by confocal microscopy. EGFP-FLAG-CD59 (C94S) TM, and mRFP-KDEL were imaged by confocal 933 microscopy. Scale bar, 10 μm. 934

935 936 Figure 6-supplement. Expression of genes encoding alkaline phosphatases in HEK293 cells. 937 The RNA-seq data of HEK293 cells (Huang et al., manuscript in preparation) were used to assess gene 938 expression. Transcription per million (TPM) values are presented as the means ± SD of three independent 939 measurements. ALPI, intestinal-type alkaline phosphatase; ALPL, tissue-nonspecific alkaline phosphatase; ALPP, 940 placental type alkaline phosphatase; ALPPL2, germ cell type alkaline phosphatase.

941