Precursors of select GPI-anchored proteins positively regulate GPI biosynthesis under the control of ERAD

Yi-Shi Liu Jiangnan University Yicheng Wang Osaka University https://orcid.org/0000-0001-7789-1938 Xiaoman Zhou Jiangnan University LinPei Zhang Jiangnan University Ganglong Yang Jiangnan University Xiao-Dong Gao Jiangnan University https://orcid.org/0000-0002-8575-5095 Yoshiko Murakami Osaka University https://orcid.org/0000-0002-4870-5734 Morihisa Fujita Jiangnan University https://orcid.org/0000-0002-0344-2408 Taroh Kinoshita (  tkinoshi@biken.osaka-u.ac.jp ) Osaka University https://orcid.org/0000-0001-7166-7257


Introduction
The endoplasmic reticulum (ER) is a specialized eukaryotic organelle for protein synthesis and processing, lipid synthesis, and calcium storage and release [1][2][3] . Various posttranslational modi cations of proteins that reside in or traverse secretory pathways also occur in the ER. For example, N-linked glycans are attached in the ER for e cient folding and proper functions 2,4,5 .
Glycosylphosphatidylinositol (GPI) anchoring is one of the conserved posttranslational modi cations in eukaryotic cells. More than 150 human proteins have been con rmed as GPI-anchored proteins (GPI-APs) 6,7 . GPIs confer characteristic features to the modi ed proteins that are essential for embryogenesis, neurogenesis, immune response, and fertilization [8][9][10][11] . GPI-APs are biosynthesized in the ER and transported to the cell surface through the Golgi apparatus. The common backbone of GPI, EtNP-6Manα-2Manα-6Manα-4GlcNα-6inositol-phospholipid (where EtNP, Man, and GlcN are ethanolamine phosphate, mannose and glucosamine, respectively), is generated by sequential additions of components to the phosphatidylinositol (PI) moiety 12,13 . Once assembled, GPI is attached to the proteins by GPI transamidase and then undergoes structural remodeling in both glycan and lipid portions [14][15][16] . The structural variations of GPI anchors are introduced by modi cation of the core by side chains. The rst, α1,6-linked mannose, is often modi ed by β1,4-linked N-acetylgalactosamine (GalNAc) by PGAP4 17 . The GalNAc side chain can be further modi ed by β1,3 galactose (Gal) and then with sialic acid 17,18 . Sialic acid in the GPI side chain of prion proteins, α2-3-linked N-acetylneuraminic acid 18 , affects process of prion disease progression 18 ; however, the biological roles of the GPI side chains in various GPI-APs are largely unknown.
The initial GPI identi ed is a type of glycolipid that anchors various proteins on the cell membrane. In certain parasites, nonprotein-linked GPIs (free GPIs) exist on the membrane [19][20][21] . In addition, recent works have indicated that free GPIs are also found in some tissues and cell types in mammals [22][23][24][25] . The T5-4E10 monoclonal antibody (T5 mAb), originally established against free GPI from the parasite Toxoplasma gondii, recognizes a GalNAc side chain only when GalNAc is at the nonreducing end 20,21 . We found that the T5 mAb also recognizes mammalian free GPIs with a GalNAc side chain and that when the GalNAc side chain is modi ed by Gal, the T5 mAb no longer binds 17 (Fig. 1a). We also found that when GPI transamidase is defective, GPIs are expressed as free GPIs on the cell surface 22 .
Nevertheless, in HEK293 cells, GPI transamidase defects caused by knockout (KO) of PIGS, one of the GPI transamidase components, did not induce positive T5 mAb staining 26 , because all the GalNAc side chains were modi ed by Gal. This phenomenon allowed us to perform genome-wide CRISPR screening for genes involved in GPI side-chain galactosylation. Among genes with KO resulting in positive T5 mAb staining of PIGS-KO HEK293 cells, we identi ed B3GALT4, which transfers Gal to the GalNAc side chain 26 . In the same screening, we found that KO of ER-associated degradation (ERAD) pathway genes, such as HRD1, UBE2J1 and UBE2G2, caused positive T5 mAb staining and demonstrated that GPI biosynthesis was upregulated in ERAD-defective PIGS-KO cells, so that the amount of free GPI surpassed the galactosylation capacity of the cells resulting in positive staining of the cell by T5 mAb 26 .
ERAD plays a critical role in proteostasis under normal or stress conditions. Proteins that fail to fold or assemble are recognized by chaperones and lectins, transferred to membrane-integral adaptors and retrotranslocated through E3 ubiquitin ligases, followed by ubiquitination for cytosolic proteasomal degradation [27][28][29][30] . The ERAD system consists of several parallel pathways, ERAD-L, ERAD-M, and ERAD-C, handling the substrate proteins in the ER lumen, ER membrane, and ER cytoplasmic surface, respectively 28,31−34 . Various E3 ligases, including HRD1, GP78, TRC8, and MARCH6, are utilized in these pathways 28, 30 . HRD1 is the best characterized and the major retrotranslocation component for luminal ERAD-L substrates 35,36 . Both in vitro and in vivo studies have revealed that multi-membrane spanning HRD1 serves as a channel for luminal ERAD-L substrates. HRD1 recognizes a variety of substrates to maintain ER homeostasis [35][36][37] .
To reveal the mechanism by which the ERAD pathway regulates GPI biosynthesis, we performed genomewide CRISPR screening to identify factors involved in GPI biosynthesis upregulation under the control of the ERAD pathway. We identi ed two GPI precursor proteins, CD55 (also known as complement decayaccelerating factor, DAF) and CD48, which are recognized as HRD1-dependent ERAD-L substrates and act as positive regulators of GPI biosynthesis. Our results suggest that GPI biosynthesis is regulated on demand by the status of GPI precursor proteins.

Results
Biosynthesis of GPI is upregulated in ERAD-L de cient cells As mentioned above, B3GALT4 transfers Gal to the GalNAc side chains of GPI-APs and free GPI 26 (Fig.   1a). As T5 mAb detects the nonmodi ed GalNAc side chain of free GPI, PIGS-KO cells are T5 mAb staining-negative, whereas PIGS-B3GALT4-double KO (DKO) HEK293 cells become T5 mAb staining-positive (Fig. 1b). We also found that PIGS-HRD1-DKO cells were strongly stained with the T5 mAb even with intact B3GALT4 (Fig. 1b and 1c), suggesting that the amount of free GPI surpassed the Gal-addition capacity of B3GALT4. GPI transamidase consists of ve subunits: PIGS, PIGK, PIGT, PIGU, and GPAA1. To check whether this increase in GPI under HRD1-defective conditions was PIGS KO speci c or common among GPI transamidase defects, we knocked out the HRD1 gene in GPI transamidase gene-de cient HEK293 cells and analyzed free GPI levels on the cell surface in bulk populations. Knocking out HRD1 in PIGK-, PIGT-, PIGU-, and GPAA1-KO cells increased T5 mAb staining, as observed in PIGS-HRD1-DKO cells (Fig. 1d), indicating that defects in HRD1 generally increase free GPIs in cells defective in GPI transamidase. We also knocked out HRD1 in SLC35A2 (encoding UDP-galactose transporter necessary for galactosyltransferases) and PIGT double knockout (SLC35A2-PIGT-DKO) HEK293 cells. HRD1 knockout in SLC35A2-PIGT-DKO cells increased T5 mAb staining levels twofold ( Fig. 1e and 1f). These results indicated that the ERAD-L pathway suppressed GPI biosynthesis. We hypothesized that there must be positive regulator(s) of GPI biosynthesis, which is controlled by the ERAD-L pathway (Fig. 1g).

Identi cation of positive regulators of GPI biosynthesis
To identify genes that upregulate GPI biosynthesis when ERAD-L is inactive, we performed a genomewide CRISPR-Cas9 screen in PIGS-HRD1-DKO HEK293 cells. PIGS-HRD1-DKO cells expressed ~3-fold higher levels of free GPI, as assessed by T5 mAb staining compared with parental PIGS-KO cells, allowing the selection of mutant cells defective in upregulation of GPI biosynthesis. PIGS-HRD1-DKO cells were transduced with lentivirus capable of expressing a human genome-scale CRISPR knockout guide RNA (gRNA) library (GeCKO v2) and Cas9 enzyme. We then cultured the cells for 2 weeks, and cells negatively stained with the T5 mAb were enriched by cell sorting (Fig. 2a). After two rounds of cell sorting and culture, more than half of the cells became negative for T5 mAb staining (Fig. 2b). Genomic DNA was extracted from cells, and the integrated gRNA sequences ampli ed by PCR were analyzed by deep sequencing. The enrichment of gRNAs was determined by model-based analysis of the genome-wide CRISPR-Cas9 knockout (MAGeCK) method 38 . The screening enriched for genes required for GPI biosynthesis (PIG genes) and dolichol-phosphate mannose (DPM) biosynthesis (DPM1, DPM2, DPM3, SRD54A3, and DHDDS), as expected. CLPTM1L (cleft lip and palate transmembrane protein 1-like protein) was highly enriched in the screening. We recently reported that CLPTM1L is a lipid scramblase involved in GPI biosynthesis 39 (Fig. 2c and Supplementary Fig. 1a). SPPL3, which encodes a membranebound aspartic protease that cleaves type-II membrane proteins at the Golgi, was also enriched in the screening. It has been reported that SPPL3 affects the expression of a GPI-AP, CD59, through the regulation of glycosphingolipid biosynthesis 40,41 . Considering that the screen would identify substrates of the ERAD-L pathway, we expanded our attention to other genes encoding ER-localized proteins. Two GPI-related genes, ARV1 and CD55, although they were not highly ranked (ranking #88 and #280, respectively), attracted our attention (Fig. 2c). To validate the screening results, we chose and knocked out ve genes, SRD5A3, ARV1, CD55, SPPL3, and CLPTM1L, in PIGS-HRD1-DKO cells. The KO cells were analyzed by staining with T5 mAb. All KO cells showed a clear decrease in T5 mAb staining compared with parental PIGS-HRD1-DKO cells. In particular, T5 mAb staining decreased to nearly background levels after knocking out ARV1 or CD55 ( Fig. 2d and 2e). Decreased T5 mAb staining was restored by overexpression of relevant cDNAs ( Supplementary Fig. 1b). When SRD5A3, ARV1, SPPL3, and CLPTM1L were knocked out in HEK293 wild-type cells, the surface CD59, a GPI-AP, was only slightly decreased or unchanged compared with parental wild-type cells ( Supplementary Fig. 1c and 1d). These genes were not identi ed from previous genetic screens using cell surface GPI-APs as reporters. These data suggested that free GPI would be a more sensitive reporter, enabling us to identify more genes involved in GPI biosynthesis.
CD55 precursor regulates GPI biosynthesis under ERAD control Among the candidate genes identi ed in the screening, CD55 encodes a GPI-anchored complement regulatory factor that is ubiquitously expressed in the human body. To understand the cellular mechanisms by which knocking out the GPI-AP gene causes a decrease in GPI biosynthesis, we rst generated a PIGS-HRD1-CD55-triple KO (TKO) cell line, in which the expression of surface-free GPIs decreased to the background level ( Fig. 2d and 3a). When we stably expressed HA-tagged CD55 in the TKO cells, the amounts of surface-free GPIs were restored or even higher than those in PIGS-HRD1-DKO cells ( Fig. 3a and 3b). To test whether the change was due to GPI biosynthesis, we metabolically labeled cells with [2-3 H] mannose and analyzed radiolabeled GPI mannolipids by thin-layer chromatography. The GPI precursors (H5, H6, H7, H7' and H8) were decreased by knocking out CD55, whereas they were greatly increased by overexpression of CD55, suggesting that GPI biosynthesis was upregulated by expression of CD55 in PIGS-HRD1-CD55-TKO cells (Fig. 3c).
In normal cells, CD55 protein is synthesized and attached to GPI catalyzed by GPI transamidase in the ER, whereas in PIGS-KO cells, CD55 fails to link to GPI and remains in a precursor form that possesses a GPI attachment signal peptide at the C-terminus. We next analyzed the localization of CD55 in PIGS-HRD1-KO HEK293 cells by immuno uorescence. The precursor of CD55 was mainly colocalized with an ER marker, calnexin (CANX), but only weakly with a Golgi marker, GM130 (Fig. 3d), suggesting that majority of CD55 precursor proteins remain in the ER. GPI-AP precursor proteins that fail to anchor GPI are thought to be degraded through ERAD pathways 42,43 . To validate whether the CD55 precursor was an ERAD substrate, we analyzed endogenous CD55 levels. Western blots revealed speci c endogenous immature CD55 bands in PIGS-HRD1-DKO, PIGS-UBE2J1-DKO, and PIGS-UBE2G2-DKO cells (Fig. 3e). In contrast, bands did not appear in PIGS-KO, HRD1-rescued PIGS-HRD1-DKO or PIGS-HRD1-CD55-TKO cells. The protein was also undetectable in PIGS-GP78-DKO cells. The E3 ligase GP78 plays a role in another ERAD pathway for membrane proteins, ERAD-M (Fig. 3e), suggesting that the CD55 precursor is mainly degraded by the HRD1-dependent ERAD-L pathway. We further transiently overexpressed HA-CD55 in PIGS-KO and PIGS-HRD1-DKO cells. The amount of HA-CD55 was much higher when HRD1 was defective (Fig. 3f). To determine the stability of HA-CD55, we treated PIGS-KO and PIGS-HRD1-KO cells with cycloheximide (CHX) to stop new synthesis of proteins. In PIGS-KO cells, CD147, one of the classic ERAD-L substrates, was degraded in a time-dependent manner. HA-CD55 was undetectable under similar conditions. However, both CD147 and HA-CD55 proteins became stable in PIGS-HRD1-DKO cells (Fig. 3g).
These results indicated that the CD55 precursor was a substrate of the HRD1-dependent ERAD pathway and that its accumulation in PIGS-HRD1-DKO cells caused the upregulation of GPI biosynthesis.
The expression pro le of GPI biosynthesis-related genes was not signi cantly affected by CD55 ( Supplementary Fig. 2a). To directly test whether PIG genes were involved in the GPI biosynthesis upregulation, we transiently transfected PIG genes or a mixture of DPM-related genes into PIGS-HRD1-CD55-TKO cells, but no effects on these transfected cells were found ( Supplementary Fig. 2b). Together, these ndings suggested that CD55 did not regulate GPI biosynthesis through changing the protein expression levels of or stabilizing PIG gene-encoded proteins.
Two GPI attachment signal peptides regulate GPI biosynthesis Mature CD55 (GPI form) contains four short consensus repeats (SCRs), one N-linked glycan between SCR 1 and 2, and a membrane-proximal O-linked polysaccharide rich region (S/T region) [44][45][46] (Fig. 4a). The CD55 precursor (ER form) is synthesized and translocated into the ER lumen through an N-terminal signal peptide and probably localized in the ER membrane through the hydrophobic region in the C-terminal GPI attachment signal peptide before GPI attachment (Fig. 4a). To determine which part of CD55 is critical for the regulation of GPI biosynthesis, we constructed truncated variants (Fig. 4a) and transiently transfected them into PIGS-HRD1-CD55-TKO cells. Cells expressing CD55 mutants lacking SCR domains or the S/T region e ciently rescued surface-free GPIs (Fig. 4b). However, CD55 lacking a GPI attachment signal (T349-381) could not rescue the expression of free GPIs. A construct only having the 32-amino acid GPI attachment signal (T35-349) also showed no activity, probably due to instability or di cult expression ( Fig. 4b-4d). To clarify this issue, we fused the GPI attachment signal of CD55 with GFP protein [GFP-CD55(C)] and transiently overexpressed it in PIGS-HRD1-CD55-TKO cells. Chimera GFP-CD55(C) was expressed well and restored surface-free GPIs ( Fig. 4e and 4f), indicating a unique role of the GPI attachment signal of CD55 in regulating GPI biosynthesis.
Considering that more than 150 different human proteins are GPI-APs, we next investigated whether other GPI-AP precursors had the ability to upregulate GPI biosynthesis, similar to CD55. In HEK293 cells, thirteen of 161 GPI-APs were expressed at higher levels than CD55 47 (Fig. 5a). We then chose ve of them (LY6E, GPC4, CD59, CD109, and PRNP) and CD55, which were knocked out by two different gRNAs in PIGS-HRD1-DKO cells. All gene knockouts except CD55 failed to decrease T5 mAb staining ( Supplementary Fig. 3a). Conversely, we overexpressed GPI-APs, the expression of which was low, in PIGS-HRD1-CD55-TKO cells ( Fig. 5a and Supplementary Fig. 3b). Among the 32 GPI-APs tested, only CD48 e ciently restored the expression of free GPIs ( Supplementary Fig. 3b). With the nding that the GPI attachment signal peptide of CD55 regulated GPI biosynthesis, the GPI attachment signal of CD48 was fused with GFP [GFP-CD48(C)], which was expressed in PIGS-HRD1-CD55-TKO cells. The fused protein also restored the expression of free GPIs (Fig. 5b), suggesting that some particular GPI attachment signals, such as those of CD55 and CD48, were functional in the positive regulation of GPI biosynthesis.
We next asked whether CD48 precursor is a substrate of ERAD-L pathway like CD55 precursor. Transiently transfected HA-tagged CD48 (HA-CD48) in PIGS-HRD1-DKO cells was stabilized, compared with that in PIGS-KO cells (Fig. 5c). The level of HA-CD48 in PIGS-HRD1-DKO cells was greatly decreased by rescue of HRD1 (Fig. 5c), indicating that CD48 precursor is also under control of ERAD-L pathway.
According to previous reports that replacement of the w-site serine 353 in CD55 with proline abolishes GPI attachment 48 , we constructed a CD55(S353P) mutant in which serine was replaced with proline and transiently overexpressed it in PIGS-HRD1-CD55-TKO cells. CD55(S353P) was active in restoring T5 mAb staining (Fig. 5d). Interestingly, the CD55(S353P) mutant was more stable than wild-type CD55 in PIGS-KO cells, in which the ERAD system functioned normally (Fig. 5e). This result raised the possibility that stabilization of CD55 in the ER was the only effect of HRD1-KO in the upregulation of GPI biosynthesis. To validate this hypothesis, we stably overexpressed CD55(S353P) in PIGS-KO cells and isolated two clones, #1 and #2, with different levels of CD55(S353P) (Fig. 5f). Even in the presence of HRD1dependent ERAD-L pathway, the free GPIs stained by T5 mAb were detected in those cells (Fig. 5g). The T5 mAb staining levels were correlated with the CD55(S353P) protein levels ( Fig. 5f-5h). These results indicated that GPI biosynthesis was activated by the GPI attachment signal peptide of CD55 in a dosedependent manner.
Key residues in GPI attachment signal for GPI biosynthesis GPI transamidase recognizes and cleaves the GPI attachment signal between ω and ω+1 sites, forming the thioester bond on the catalytic Cys residue to generate an enzyme-substrate intermediate. GPI is then transferred to the newly exposed C-terminus of the protein 47,49 . Although all proteins modi ed by GPI contain a GPI attachment signal peptide at the C-terminus, the sequences are not conserved. We compared the GPI attachment signal peptide sequences of human CD55, mouse CD55, human CD59 and CD48 and tested their activities. Although human CD55 and mouse CD55 sequences showed some identity, mouse CD55 and human CD59 did not have the ability to restore free GPI expression in PIGS-HRD1-CD55-TKO cells (Fig. 6a). Since the C-terminal GPI attachment signal possesses 5-10 hydrophilic amino acids (spacer sequence) and a 15-20 stretch of hydrophobic amino acids, we truncated the hydrophilic and hydrophobic regions of CD55. Neither truncated construct (T353-361 and T362-381) showed activity (Supplementary Fig. 4a and 4b). We also made chimeras by exchanging hydrophilic and hydrophobic regions of functional CD55 and non-functional CD59, respectively, and overexpressed the constructs in PIGS-HRD1-CD55-TKO cells. Chimeric proteins could not restore T5 mAb staining (Supplementary Fig. 4a and 4b), suggesting that both hydrophilic and hydrophobic regions are important for the activity. We then focused on amino acids among GPI attachment signal sequences and constructed point mutants. Since non-functional CD59 and mouse CD55 had histidine within the hydrophobic region, we introduced histidine into CD55 and CD48 by replacing one of the leucines. The CD55 L371H and CD48 L238H mutants lost activity, even if the mutant proteins were expressed at similar levels ( Fig. 6b-6d). Other CD55 mutants (S353G, L370A, G372A, L373V, and G378Y) maintained the functional activity ( Supplementary Fig. 4c and 4d). These ndings imply that leucine in the hydrophobic region in the signal peptides played a functionally important role in the regulatory process.
ARV1 is close to and functionally required for CD55 and CD48 With the nding that the GPI attachment signal peptides of CD55 and CD48 directly regulated GPI biosynthesis, we next investigated the underlying mechanism. To identify proteins that might physically interact with them, we used a proximity-tagging system, Turbo-ID. First, we added two functional GPI attachment signal peptides of CD55 and CD48, and two non-functional ones of CD59 and PRNP to the Cterminus of Turbo-ID (Fig. 7a), which were stably expressed in PIGS-HRD1-CD55-TKO cells. TurboID-CD55(C) and TurboID-CD48(C) restored free GPI expression, as expected, whereas TurboID-CD59(C) and TurboID-PRNP(C) had no activity (Fig. 7b-7d). To identify proteins that speci cally interacted with CD55, we compared biotinylated proteins by TurboID-CD55(C) and TurboID-CD48(C) with those by TurboID-CD59(C). ARV1 was ranked in the top ten after comparing normalized peptide counts of CD55 with CD59 or CD48 with CD59 ( Fig. 7e-7f). To test whether CD55 directly interacted with ARV1, we performed immunoprecipitation experiments by transfecting 3Flag-ARV1 into PIGS-HRD1-CD55-TKO cells stably expressing HA-CD55. We did not detect coprecipitation of ARV1 with CD55 under the tested conditions, suggesting that the interaction was very weak or transient ( Supplementary Fig. 5a). The 3Flag-ARV1 levels were similar in the presence or absence of HA-CD55 in PIGS-HRD1-ARV1-TKO cells; therefore, ARV1 and CD55 precursors did not stabilize each other ( Supplementary Fig. 5b).
Since both CD55 and ARV1 were required for the upregulation of GPI biosynthesis in HRD1-KO cells, we next asked whether ARV1 was required for the upregulation of GPI biosynthesis by CD55 and CD48. The increased T5 mAb staining level induced by the stable CD55 precursor CD55(S353P) in PIGS-KO cells was decreased to the background level by knocking out ARV1 in CD55(S353P)-expressing PIGS-KO cells (clone #2) and restored by overexpression of 3Flag-ARV1 (Fig. 7g). Overexpression of HA-CD55 or HA-CD48 in PIGS-HRD1-ARV1-TKO cells also could not restore the T5 mAb staining ( Supplementary Fig. 5c); therefore, ARV1 was functionally required for CD55 and CD48. These results suggested that CD55/CD48 and ARV1 contributed to the upregulation of GPI biosynthesis in the same pathway.
Finally, we investigated steps in GPI biosynthesis pathway being upregulated by the CD55 precursor. The [2-3 H] mannose labeling showed that all spots of GPI intermediates containing mannoses were increased in PIGS-HRD1-DKO cells and their CD55-overexpressing cells (Fig. 3c), indicating that biosynthesis was upregulated by the CD55 precursor, probably at an earlier step before mannose addition. To determine  (Fig. 7h). Radioactive inositol was incorporated into PI and then used to generate the rst, second, and third GPI intermediates GlcNAc-PI, GlcN-PI, and GlcN-acylPI. After thin-layer chromatography and detection by phosphorimaging, GlcN-PI, which accumulates in PIGW-KO cells, appeared below the abundant PI spots, whereas GlcNAc-PI and GlcN-acylPI did not appear as discrete bands due to overlapping migration with abundant PI and other inositol-containing components 50 . GlcN-PI was increased in PIGS-HRD1-DKO cells compared to PIGS-KO cells, while knocking out CD55 eliminated GlcN-PI increase, which was rescued by CD55 transfection (Fig.  7h). Knocking out ARV1 similarly prevented the GlcN-PI increase (Fig. 7h), suggesting that both CD55 and ARV1 were required for the upregulation of GPI biosynthesis in a very early step (either the generation of GlcNAc-PI or its conversion into GlcN-PI, or both) (Fig. 8).

Discussion
Although the majority of genes required for GPI precursor biosynthesis, GPI transfer to proteins and GPI remodeling have been identi ed, understanding the dynamic regulation of GPI biosynthesis is still a challenge. This is in part due to technical obstacles in directly detecting changes in GPI levels. Here, we designed and implemented a free GPI expression system under GPI transamidase-and ERAD-de cient conditions. Detection of free GPI expression provides a new approach for rapidly determining the status of GPI biosynthesis and can be used to understand the regulatory mechanism. In this work, we used the enhancement of free GPI expression in PIGS-HRD1-DKO cells to identify genes that regulate GPI biosynthesis. The screening facilitated the discovery of CD55, which is the substrate of the ERAD-L pathway and upregulates GPI biosynthesis upon accumulation.
Under normal conditions, a GPI attachment signal of precursor protein is recognized and cleaved by GPI transamidase during GPI modi cation 14 . Most likely, the ratio of GPI and proteins is controlled properly. It is di cult to detect either the accumulation of GPI or precursor proteins in wild-type cells. Normally, misfolded GPI-anchored PrP* (A117V) and Gas1* (G291R) are poor ERAD substrates because of the speci c characteristics of GPI anchors 42,43,51 . In stressed mammalian cells, misfolded GPI-anchored PrP* is transported out of the ER quickly via a pathway termed "rapid ER stress-induced export" (RESET) for degradation in the lysosome 51 . However, misfolded transmembrane type Gas1* (Gas1*-TMD) and PrP maintaining its GPI attachment signal peptide that fail to undergo transamidation are mainly degraded by ERAD 42,43 . In this study, we found that knocking out the PIGS and HRD1 genes caused the accumulation of CD55 precursor proteins, which enabled us to reveal the speci c role of the GPI attachment signal peptide of CD55 in GPI biosynthesis. It is reasonable to postulate that CD55 acts as a regulator because it is widely expressed in diverse organs and cells 52 . Moreover, we found that the GPI attachment signal peptide of CD48 functioned similarly to that of CD55. The expression of CD48 was restricted to hematopoietic cells, suggesting the existence of other GPI-AP precursors regulating GPI biosynthesis in a tissue-speci c manner. Conversely, mouse CD55 could not restore the human CD55 KO phenotype. There are two types of CD55 in mouse cells, the GPI-anchored form and the transmembrane form, which differs from human CD55. We hypothesize that different and diverse GPI-APs are utilized as regulators of GPI biosynthesis among species.
Both our CRISPR-Cas9 genetic screen using PIGS-HRD1-DKO cells and proximity labeling using TurboID-CD55(C) and TurboID-CD48(C) identi ed ARV1, suggesting that ARV1 is spatially close to and functionally required for CD55. Knocking out ARV1 or CD55 in PIGS-HRD1-DKO cells eliminated the accumulation of GlcN-PI, resulting in decreased cell surface-free GPI. It is possible that ARV1 and CD55 both work at early stage of GPI biosynthesis. Several papers have reported that Arv1 plays roles in lipid and cholesterol homeostasis, sphingolipid distribution and GPI biosynthesis in the yeast Saccharomyces cerevisiae 53,54 . It was proposed that in GPI biosynthesis, Arv1 may deliver early GPI intermediate GlcN-acyPI to the ER lumen as the ippase or contribute to the ip reaction, so the loss of Arv1 leads to de ciencies in GPI-AP biosynthesis in S. cerevisiae 55,56 . Furthermore, mutations in human ARV1 cause symptoms similar to inherited GPI de ciencies, indicating that human ARV1 is also involved in GPI biosynthesis 57,58 . A recent study in the African trypanosome Trypanosoma brucei showed that an Arv1like protein (TbArv1) is pulled down by TbGPI3, the mammalian homolog of which is PIGA 59 . Using a combination of RoseTTAFold and AlphaFold, a study to predict protein assemblies with two to ve components in S. cerevisiae also predicted that ARV1 may interact with Gpi1, the yeast homologue of mammalian PIGQ 60 , suggesting that ARV1 may form a complex with GPI Nacetylglucosaminyltransferase (GPI-GnT), the rst enzyme in the pathway ( Figure S1). In this study, we showed that defects in the HRD1-dependent ERAD-L pathway caused the upregulation of GPI biosynthesis in early steps, but further knockout of CD55 or ARV1 eliminated this effect. CD55 and ARV1 might act together, where CD55 served as the signaling molecule and ARV1 might modulate the proper lipid environments to enhance the activity of GPI-GnT or promote conversion to GlcN-PI; however, we did not nd stable interaction between ARV1 and CD55 precursor. It will be interesting to further characterize the relationship between the CD55 precursor and ARV1 in the future.
The expression of free GPIs is sensitive to ARV1 or CD55 KO, indicating that ARV1 is critical for the upregulation of GPI biosynthesis, particularly under conditions where GPI-AP precursor proteins, including CD55, accumulate. However, in HEK293 wild-type cells, knockout of ARV1 or CD55 did not change the expression of GPI-APs, suggesting that they play roles only in GPI biosynthesis under stressed conditions. Since the expression of known PIG genes was not changed in the presence or absence of CD55 and overexpression of PIG genes did not restore free GPI expression in PIGS-HRD1-CD55-TKO cells, the regulation was neither at the transcriptional nor the translational level of GPI biosynthetic genes. Thus, the upregulation of GPI biosynthesis by CD55 precursors is a rather quick and short-lived quantity control system for GPI-APs once the downstream GPI transamidation step fails in wild-type cells. It is possible that CD55 and ARV1 are involved in this regulatory system to balance the biosynthesis of free GPIs and the generation of GPI-APs in the ER. in Supplementary Data 1. For stable expression, PLAT-GP packaging cells were seeded and cultured to 90% con uence and transfected with the pLIB2-BSD plasmid bearing the cDNA of interest using PEI-Max. Viral medium was added to the cells, and these cells were cultured at 32 °C for 12 h. The medium was changed after 24 h and cultured at 37 °C. Three days after infection, the cells were incubated in the medium with 10 μg/mL blasticidin (InvivoGen) for 14 days.

CRISPR-Cas9 screening and FACS
For a large-pooled screen, viral production and functional titration were conducted in the same manner as described previously 26 . Pooled human GeCKOv2 plasmids (lentiCRISPRv2) were cotransfected with the lentiviral packaging plasmids pLP1, pLP2, and pLP/VSVG (Thermo Fisher) into Lenti-X 293T cells (Clontech). Twelve hours later, the medium was changed to 10 mL prewarmed DMEM supplemented with 10% FBS. The viral media was collected 24 h, 48 h, and 72 h after transfection and ltered through a membrane (Mllex 0.45 µm, PVDF, 33 mm). Finally, 30 mL viral media in total was combined and stored at 4 °C for functional titration and pooled screening as quickly as possible.
PIGS-HRD1-DKO cells were plated in 8 × 15 cm dishes (3.5 ×10 6 cells per dish). Approximately 8 × 10 7 cells (1 ×10 7 cells per dish) were transduced with viral supernatant after 36 h of seeding. Cells were selected with 0.5 μg/mL puromycin until the infected cells were expanded to 2.4 × 10 8 to maintain the complexity of the gRNA library. Cells were combined and split 1:4, and a minimum of 6 × 10 7 cells were plated for culture. At 2 weeks posttransduction, a pellet of 5 × 10 7 cells without sorting was stored at -80°C . For cell sorting, approximately 1 × 10 8 cells were harvested and incubated with T5 mAb, followed by staining with anti-mouse IgM. After washing with PBS, the cells were resuspended in Hanks' Balanced Salt Solution (H6648, Sigma-Aldrich), and T5 mAb staining-negative cells were sorted by FACSAria (BD).
We prepared 2 × 10 7 cells for the second sorting. After sorting, the cells were maintained in DMEM supplemented with 0.25 μg/mL puromycin. Pellets of 2 × 10 7 sort2 cells were stored at -80 °C until use. For analysis, cells were cultured in 6-well plates one day before analysis. The cells were harvested and washed once with PBS and then stained with CD59 or T5 mAb in FACS solution (PBS containing 1% BSA and 0.1% NaN 3 ) on ice for 25 min. They were then washed twice in FACS buffer, followed by staining with Alexa Fluor 647-conjugated goat anti-mouse IgM or PE-conjugated goat anti-mouse IgG. After two washes with FACS buffer, the cells were analyzed using a BD FACSCanto II.
Genomic DNA sequencing and analysis Approximately 5 × 10 7 unsorted and 2 × 10 7 twice-sorted PIGS-HRD1-DKO cells were extracted for genomic DNA using a Wizard Genomic DNA Puri cation Kit (Promega). The gRNAs were ampli ed from genomic DNA of unsorted and twice-sorted cells. PCR (25 cycles) was performed to amplify the gRNAs using KOD FX Neo Polymerase (TOYOBO LIFE SCIENCE), making up a total of 65 tubes for unsorted cells and 12 tubes for twice-sorted cells (oligos for ampli cation of gRNAs are shown in Supplementary Data 1). All PCR products were combined and mixed and applied to a 2% agarose gel for puri cation. The PCR products were concentrated, mixed 9:1, and analyzed by paired-end sequencing with a NovaSeq 6000 system (Illumina). Deep sequencing raw data were processed for gRNA counting using Python scripts. The high-throughput sequencing reads were demultiplexed using the 5-bp adapter by cutadapt version v1.18 63 . By using MAGeCK work ow version 0.5.6, the adapters of the demultiplexed reads were trimmed to obtain 20-bp gRNA sequences, and the sgRNA sequences were mapped to the sequences of the Human GeCKO v2 sgRNA library to determine the total number of gRNA counts. The robust rank aggregation (RRA) values and p values were determined using the MAGeCK algorithm 64 .

Plasmids
All primers used in this study are listed in Table S1. To knock out target genes with the CRISPR-Cas9 system, single-guide RNAs (sgRNAs) were designed with the E-CRISP website (http://www.e-crisp.org/E-CRISP/), and the targeting DNA fragments were ligated into the BbsI-digested vector pX330-EGFP. To construct pME-HRD1-3HA and pME-SPPL3-3HA, HRD1 and SPPL3 were ampli ed from a human cDNA library, digested with XhoI and MulI, and were cloned into the same site of pME-B3GALT4-3HA. The HRD1-3HA fragment from pME-HRD1-3HA was digested and ligated to pLIB2-BSD through EcoRI and NotI.

Generation of knockout cell lines
To knock out SRD5A3, ARV1, CD55, SPPL3, and CLPTM1L, cells were transfected with one or two different pX330-EGFPs containing the target sgRNA. Lipofectamine 2000 (11668019, Thermo Fisher Scienti c) was used for transfection. Three days after transfection, the GFP-positive cells were sorted with a cell sorter S3e (Bio-Rad). The sorted cells were cultured for one week or more and then diluted to select clonal KO cells. Clonal cells were determined by rescue experiments and western blotting analysis.
Knockout of HRD1 in PIGK-KO, GPAA1-KO, PIGT-KO, PIGU-KO or CD59, PRNP, GPC4, CD109, LY6E in PIGS-HRD1-DKO cells was carried out within the bulk cells. After transient transfection of sgRNA constructs as described above, the bulk cells were sorted and cultured for 10 days, followed by ow cytometry analysis.
The supernatant was mixed with SDS-PAGE loading buffer and boiled at 95 °C for 5 min or kept at 4 °C overnight (for multitransmembrane proteins). The proteins were separated by SDS-PAGE and transferred onto PVDF membranes, and the membranes were blocked at room temperature (RT) in Tris-buffered saline containing 0.1% Tween-20 (TBS-T) and 5% nonfat milk for 1 h, followed by incubation with primary antibodies at RT for 1 h. After washing three times with TBS-T buffer, the membranes were then incubated with horseradish peroxidase-labeled secondary antibodies for 1 h at RT. After washing with TBS-T, the membrane was visualized with ECL Prime western blotting Detection Reagent (GE Health care).

Immuno uorescence
To detect the subcellular localization of HA-tagged CD55, PIGS-HRD1-CD55-TKO cells stably expressing HA-CD55 were seeded on glass coverslips pretreated with 1% gelatin and cultured for another 2 days. Cells were washed with PBS, xed in 4% paraformaldehyde, washed with PBS, and incubated with 40 mM ammonium chloride. Then, the cells were incubated at RT in blocking buffer A (PBS containing 5% were considered to be of high quality and were processed further and subjected to subsequent analysis. Total RNA-seq libraries were generated using 4 μg of total RNA, which was analyzed using the TruSeq Stranded mRNA LTSample Prep Kit (Illumina, San Diego, CA, USA). These libraries were then sequenced using the Illumina sequencing platform (HiSeqTM 2500 or Illumina HiSeq X Ten), and 125-bp/150-bp paired-end reads were generated. Transcriptome sequencing was conducted by OE Biotech Co., Ltd. (Shanghai, China), and clean reads were provided. The clean reads were mapped to the hg38 reference genome using hisat2 (version 2.1.0). The output BAM les were converted to SAM les using SAMtools 1.9. The nal TPM values were obtained using Stringtie 1.3.5.

In vivo metabolic labeling of GPI intermediates
The process of labeling experiments followed a previous protocol 26  The dried peptides were resuspended in 15 μL of 2% MeCN and 0.1% FA solution and then analyzed using an EASY-nLC 1200 system (Thermo Scienti c, San Jose, CA) coupled with a high-resolution Orbitrap Fusion Lumos spectrometer (Thermo Scienti c, San Jose, CA). Each injection volume was 3 μL. The samples were rst separated on an EASY-nLC 1200 system with an RSLC C18 column (1.9 μm ×100 μm ×20 cm) packed in house.

Statistical analysis
Statistical analysis was determined using the Student's t-test (GraphPad Prism 8.0 software; GraphPad software Inc., La Jolla, CA) to evaluate comparisons between three individual experiments, and p values < 0.05 were considered statistically signi cant.

Data availability
The source data underlying Figs. 1C and F, Fig. 2E, Fig. 4C, Fig. 5H, Fig. 6C    the blue bubble denotes DPM-related and Dol-P-related genes, the yellow bubble denotes SPPL3, and the red bubble denotes functionally unknown genes in GPI biosynthesis. d and e Flow cytometry analysis of the knockout of top-ranking genes identi ed using our CRISPR screen in PIGS-HRD1-DKO cells. Cells were stained with T5 mAb. The relative mean value of the uorescence intensity of PIGS-HRD1-DKO was set to 1, and the relative intensity of free GPI in knockout cells is displayed as the mean ± SD from three independent experiments with p values (unpaired Student's t-test).  The GPI attachment signal peptide of CD55 regulates GPI biosynthesis. were analyzed by western blotting. The expression of fused proteins was detected by GFP antibody.
GAPDH was used as a loading control. and two single clones, #1 and #2, were isolated. The expression of GFP-fused CD55(C) S353P was analyzed using western blotting (f). GAPDH was used as a loading control. g The surface expression of free GPI was stained with T5 mAb and analyzed by ow cytometry. h The relative mean value of the uorescence intensity of PIGS-KO was set to 1, and the relative intensity of free GPI in clones #1 and #2 is displayed as the mean ± SD from three independent experiments with p values (unpaired Student's ttest).  Proximity labeling identi ed ARV1 as spatially close and functionally similar to CD55. a Schematic strategy of biotin labeling and MS-based protein identi cation. A 3x HA-tagged biotin ligase (Turbo-ID) was fused with the GPI attachment signal peptide. b-d PIGS-HRD1-CD55-TKO cells were stably transfected with empty vector and 3x HA-tagged GPI attachment signal peptides of the CD55, CD48, CD59, and PRNP. The surface expression of free GPI was stained with T5 mAb and analyzed by ow cytometry. The relative mean value of the uorescence intensity of PIGS-HRD1-CD55-TKO transfected empty vector was set to 1, and the relative intensity of free GPI in related cells is displayed as the mean ± SD from three independent experiments with p values (unpaired Student's t-test). Cell lysates prepared from the cells used in (b) were analyzed by western blotting. The expression of fused proteins was detected using the HA antibody. CANX was used as a loading control. e The fold change in TurboID-CD55(C) versus TurboID-CD59(C) biotin and TMT-labeled peptide. A fold change ranking at top ten is highlighted in orange, and ARV1 is highlighted in red. f The fold change in TurboID-CD48(C) versus TurboID-CD59(C) biotin and TMT-labeled peptide. A fold change ranking at top ten is highlighted in orange, and ARV1 is highlighted in red. g Flow cytometry of CD55(C) S353P-expressing PIGS-KO cells  Schematic model of the functional roles of the HRD1-dependent ERAD-L substrate CD55 in regulating GPI biosynthesis. CD55 or CD48 precursor proteins are processed and transferred to GPI by the GPI transamidase, generating GPI-anchored forms. Conversely, these precursor proteins that failed to have GPI attachment are degraded normally through the HRD1-dependent ERAD-L pathway. Once the ERAD-L pathway is impaired, CD55 and CD48 precursor proteins accumulate in the ER, in turn upregulating GPI biosynthesis at early step(s). ARV1 is critical for the upregulation of GPI biosynthesis under CD55/CD48 accumulated conditions.