Human Smp3p Adds a Fourth Mannose to Yeast and Human Glycosylphosphatidylinositol Precursors in Vivo*

Yeast and human glycosylphosphatidylinositol (GPI) precursors differ in the extent to which a fourth mannose is present as a side branch of the third core mannose. A fourth mannose addition to GPIs has scarcely been detected in studies of mammalian GPI synthesis but is an essential step in the Saccharomyces cerevisiae pathway. We report that human SMP3 encodes a functional homolog of the yeast Smp3 GPI fourth mannosyl-transferase. Expression of hSMP3 in yeast complements growth and biochemical defects of smp3 mutants and permits in vivo mannosylation of trimannosyl (Man3)-GPIs. Immunolocalization shows that hSmp3p resides in the endoplasmic reticulum in human cells. Northern analysis of mRNA from human tissues and cell lines indicates that hSMP3 is expressed in most tissues, with the highest levels in brain and colon, but its mRNA is nearly absent from cultured human cell lines. Correspondingly, increasing expression of hSMP3 in cultured HeLa cells causes abundant formation of three putative tetramannosyl (Man4)-GPIs. Our data indicate that hSmp3p functions as a mannosyltransferase that adds a fourth mannose to certain Man3-GPIs during biosynthesis of the human GPI precursor, and suggest it may do so in a tissue-specific manner.

Glycosylphosphatidylinositols (GPIs) 1 are essential glycolipids synthesized by all eukaryotes. Many GPIs become covalently attached to the carboxyl termini of various secretory proteins and serve to anchor them to the exterior face of the plasma membrane (1,2). Others remain protein-free and are distributed in the membranes of major cellular organelles and the plasma membrane (3,4). GPIs are synthesized in the endoplasmic reticulum (ER) by stepwise addition of components to phosphatidylinositol. The end product of GPI synthesis is a "complete precursor" (5) that is substrate for the GPI transamidase complex that attaches it to proteins. Many of the steps and enzymes of GPI precursor assembly are conserved between humans and Saccharomyces cerevisiae and produce precursors with a common core structure of EthN-PO 4 -6Man␣1,2Man␣1,6Man␣1,4GlcN␣1,6Ins-PO 4lipid. In both pathways, the glycan portion of the GPI may be modified further with side branching ethanolamine phosphate (EthN-P) moieties on the first and second mannoses (6 -19).
A notable difference in GPI structure between yeast and mammals is the extent to which a fourth mannose (Man-4) is present as a ␣1,2-linked side branch of the third mannose (Man-3). In S. cerevisiae, late stage intermediates in GPI precursor synthesis (17, 19 -21), the presumed GPI transamidase substrates (5), and protein-bound GPIs (22) all contain four mannoses. Addition of Man-4 to trimannosyl-GPIs (Man 3 -GPIs) by the essential Smp3 mannosyltransferase is a mandatory step in yeast GPI precursor assembly which precedes the addition of the terminal EthN-P to Man-3 through which the GPI is ultimately attached to protein (23). Thus, it is probable that all yeast GPI transamidase substrates bear four mannoses. Conversely, studies of the synthesis of mammalian GPI precursors have uncovered little evidence that a similar pathway step occurs in mammals. The largest characterized mammalian GPI precursors and presumed GPI transamidase substrates contain only three mannoses (10,14,24). Only trace amounts of GPI intermediates with chromatographic mobilities consistent with the possible presence of Man-4 have been observed in some mammalian cell lines (10,15,25). However, Man-4 is present on the GPIs of many purified mammalian GPI-anchored proteins (6, 26 -31), raising questions as to how and when transfer of Man-4 to mammalian GPIs occurs.
In this study, we show that the human Smp3 protein (hSmp3p), a sequence homolog of S. cerevisiae Smp3, is a candidate human GPI fourth mannosyltransferase. We demonstrate that hSmp3p is a resident ER protein that adds Man-4 to certain human Man 3 -GPIs in vivo during synthesis of GPI precursors. Our data suggest that the addition of Man-4 to both human and yeast GPIs occurs via similar mechanisms but that Man 4 -GPI formation is likely not necessary for transfer of GPIs to proteins in mammals. Our results also suggest that although the addition of Man-4 is rarely detectable in cultured cell lines, this modification may be more common in the cells of many human tissues.
Yeast Strains and Media-S. cerevisiae strains used in this study are listed in Table I. Heterozygous diploids with disrupted alleles of smp3, gpi10, and yjr013w were from Research Genetics (Huntsville, AL). YPD and SD (also referred to as SGlc) medium were as described previously (32). YPGal medium has the same composition as YPD but with 2% galactose in place of glucose. Inositol-free synthetic medium and synthetic medium containing 0.2% yeast extract and glycerol (SGlyYE), galactose (SGalYE) or glucose (SGlcYE) were prepared as described previously (19). Sporulation of diploid cells was performed on 1% potassium acetate agar containing 5 g/ml of each supplement required to complement strain auxotrophies.
Expression of hSMP3 and ScSMP3 in S. cerevisiae-The expression vector p416-PGK was made by PCR amplifying a 501-bp fragment of the yeast PGK1 promoter and cloning it into the NotI-EcoRI sites of pRS416 (33). Plasmid p416-PGK-hSMP3-HA, in which hSMP3-HA is expressed from the PGK1 promoter, was made by cloning PCR-amplified hSMP3-HA into the EcoRI-XhoI sites of p416-PGK. Plasmid p425-PGK-hSMP3-HA was created by subcloning a SacI-XhoI fragment containing the PGK1 promoter and hSMP3-HA from p416-PGK-hSMP3-HA into the SacI-XhoI sites of pRS425. Plasmid pGAL-hSMP3-HA, in which expression of hSMP3-HA is controlled by the glucose-repressible GAL10 promoter, was made by cloning hSMP3-HA into pMW20 (34). Plasmids for yeast expression of ScSMP3 were constructed as follows. p425-ScSMP3 was made by cloning a 2.4-kb BglII-HindIII genomic fragment containing ScSMP3 and 513 bp of its native promoter from a YCp50-LEU2 library clone (library obtained from P. Hieter, University of British Columbia) into pRS425. A PCR-amplified fragment containing the ScSMP3 coding region, 513 bp of its native promoter, and 346 bp immediately downstream of its stop codon was cloned into pRS416 to make p416-ScSMP3. Yeast cells were transformed using lithium acetate (35).
Mammalian Cell Culture and Transfection Conditions-HeLa cells were obtained from the American Type Culture Collection (ATCC; Manassas, VA) and cultivated in minimum Eagle's medium (ATCC) or RPMI 1640 medium (Invitrogen) supplemented with glutamine, 100 IU of penicillin, 100 g of streptomycin/ml, and 10% fetal bovine serum. Cells were grown at 37°C in an atmosphere containing 5% CO 2 . For transient transfection of cells for metabolic labeling experiments, 1.5-3 ϫ 10 6 cells were plated/13.5-cm dish and transfected with FuGENE 6 Transfection Reagent (Roche Applied Science) the next day. The transfected cells were incubated at 37°C in 5% CO 2 for 24 -48 h prior to labeling (see below). Stable hSMP3-expressing HeLa cell lines were made by transfecting HeLa cells in 60-mm dishes with 2 g of pcDNA3.1-hSMP3-HA using FuGENE 6. Stable transfectants were selected in minimum Eagle's medium containing 500 g of G418/ml.
Expression of hSMP3, ScSMP3, and PIG-B in HeLa Cells-A human SMP3 cDNA was amplified by PCR from a human brain cDNA library using the primers 5Ј-CCGGAATTCATGCAGATCTGTGGATCCAGC-3Ј and 5Ј-CCGCTCGAGTTAAGCATAATCTGGAACATCATATGGATAG-GTTTCTTCCCCCAGCTCCACAAT-3Ј. The product, encoding hSmp3p with a carboxyl-terminal HA epitope, was cloned into the EcoRI-XhoI sites of the mammalian expression vector pcDNA3.1 (Invitrogen). A human PIG-B cDNA was amplified by PCR from a human liver cDNA library using the primers 5Ј-GCTGGATCCGGGATGAGGAGGCCCCT-AAGC-3Ј and 5Ј-TTCCCTCGAGTCACTTATCGTCGTCATCCTTGTAA-TCGAATTTCATCTTCATGTTGAATTTCC-3Ј. The product, encoding PIG-Bp with a carboxyl-terminal FLAG epitope, was cloned into the BamHI-XhoI sites of pcDNA3.1. ScSMP3 was amplified from p416-Sc-SMP3 by PCR using the primers 5Ј-GTAGGATCCCCACCATGATGA-GGTATCAATGG and 5Ј-CCGCTCGAGTCAAGCATAATCTGGAACAT-CATATGGATATAGTAGTTCGATGGAGTACACTGT. The product, encoding ScSmp3p with a carboxyl-terminal HA epitope, was cloned into the BamHI-XhoI sites of pcDNA3.1.
Immunolocalization Microscopy-Approximately 2 ϫ 10 5 HeLa cells were grown on glass coverslips in 6-well dishes. Cells were transfected with 1-2 g of an expression plasmid and incubated at 37°C in 5% CO 2 for 16 -24 h, after which cells were washed twice with 5 ml of PBS and fixed in 5 ml of PBS containing 4% paraformaldehyde for 10 min at room temperature. The cells were washed twice with 5 ml of PBS then permeabilized in 5 ml of PBS containing 0.2% Triton X-100 for 10 min at room temperature. Permeabilized cells were washed three times with 5 ml of PBS and twice with 5 ml of PBS containing 5% bovine serum albumin. Cells were incubated for 1 h at room temperature with one of the following monoclonal primary antibodies in PBS containing 5% bovine serum albumin: 1 g/ml anti-HA (Roche Applied Science), 4.9 g/ml anti-FLAG M2 (Sigma), 2 g/ml anti-human golgin-97 (Molecular Probes, Eugene, OR), or 10 g/ml anti-calreticulin (BD Biosciences, Palo Alto, CA). Cells were washed five times with 5 ml of PBS, then incubated with either an Alexa Fluor 488 goat anti-rat IgG (1:500 dilution) or Alexa Fluor 568 goat anti-mouse IgG (1:500 dilution, Molecular Probes) secondary antibody in 10% goat serum for 1 h at room temperature. In double staining experiments, this process was repeated using a second combination of primary and secondary antibodies. After antibody incubations, cells were subjected to three washes in 5 ml of This study Research Genetics Research Genetics PBS, after which coverslips containing immunostained cells were mounted onto glass slides in 6 l of 20 mM Tris, pH 8.0, containing 0.5% N-propyl gallate and 80% glycerol and analyzed using a Nikon TE2000-U fluorescence microscope and SPOT software (Nikon, Melville, NY).  (19). Because smp3::Kan R /pGAL-hSMP3-HA does not grow well in SGlyYE medium, SGalYE medium was used in its place. To radiolabel lipids of smp3-2/gpi13-pGAL-GPI13 harboring p425-ScSMP3 or p425-PGK-hSMP3-HA, cells were grown in SGlyYE medium at 25°C, then shifted to SGlc medium for 16 h at 25°C to repress GPI13. Cells were then maintained at 25°C or shifted to 37°C for 25 min to arrest smp3-2 before labeling with 15 Ci of [ 3 H]inositol for 2 h.

In Vivo Radiolabeling of Lipids and Thin Layer Chromatography (TLC)-[ 3 H]Inositol labeling of lipids in
For in vivo labeling of GPI lipids in HeLa cells, ϳ0.5-1 ϫ 10 7 transfected cells were incubated for 1 h at 37°C in 5 ml of glucose-free RPMI 1640 medium buffered with 20 mM HEPES, pH 7.3, and supplemented with 10% dialyzed fetal calf serum, 10 g of tunicamycin/ml, and 100 g of glucose/ml. [ 3 H]Mannose was added to 25 Ci/ml, and the cells were incubated for 2 h at 37°C in 5% CO 2 . After labeling, the cells were washed twice with 5 ml of PBS and released from culture plates by incubation in 1 mM EDTA containing 0.25% trypsin at room temperature for 10 min. Cells were pelleted and washed with 10 ml of PBS.
Northern Analysis-PCR was used to amplify a 429-bp probe spanning bases 74 -503 of the human hSMP3 cDNA. To radiolabel the probe, 100 ng of the product was used as template in a second PCR containing a 1 M concentration of each primer, 50 M each dCTP, dGTP, and dTTP, 0.5 M dATP, 10 pmol [␣-32 P]dATP, and 1 unit of platinum Pfx DNA polymerase. Thermocycling consisted of incubation at 94°C for 30 s followed by 15 rounds of successive incubations at 94°C for 30 s and 68°C for 2 min, then a final incubation at 68°C for 5 min. Radiolabeled probe was purified by passage through a ChromaSpin column (Clontech, Palo Alto, CA). A multiple human tissue mRNA blot (BD Biosciences) was prehybridized at 65°C for 1 h in 10 ml of ExpressHyb   FIG. 1. hSMP3 restores viability and full cell wall synthesis to Scsmp3 mutants. A, heterozygous smp3::Kan R / SMP3 diploid S. cerevisiae cells harboring pGAL-hSMP3-HA, pGAL-ScSMP3-HA, or an empty vector control were sporulated. Haploid ascospores were dissected from asci onto YPGal agar medium. Viable spores grew into colonies after 3 days at 30°C. B, four viable haploids from a tetrad of sporulated smp3::Kan R /SMP3 cells harboring pGAL-hSMP3-HA were grown on YPGal agar medium containing 200 g of G418/ml or SGal agar medium with 1 mg of 5-fluoroorotic acid (5-FOA)/ml for 3 days at 30°C. C, 10-fold serial dilutions of smp3-2 cells harboring pPGK-hSMP3-HA, p416-ScSMP3, or an empty vector were spotted onto YPD agar medium with or without 16 g of Calcofluor white (CFW)/ml and grown for 4 days at 25°C.
(Clontech) containing 1 mg of denatured salmon testes DNA (Sigma). Labeled probe was mixed with 30 g of C o t-1 DNA (Roche Applied Science), 150 g of denatured salmon testes DNA, and 50 l of 20ϫ SSC (3 M NaCl, 0.3 M sodium citrate, pH 7.0). This mixture was incubated at 95°C for 5 min, at 68°C for 30 min, and then added to the blot. Hybridization was performed overnight at 65°C followed by five 20-min washes at 65°C with 200 ml of 2ϫ SSC containing 1.0% SDS and two 20-min washes at 55°C with 200 ml of 0.1ϫ SSC, containing 0.5% SDS. Hybridized probe was quantitated using a Storm 860 PhosphorImager and ImageQuant software (Amersham Biosciences).

RESULTS
Identification and Cloning Human SMP3-We reasoned that the protein responsible for Man-4 addition to human GPIs might resemble the fourth GPI mannosyltransferase of S. cerevisiae, Smp3p. Locus 80235 of human chromosome 3 encodes a predicted 579-amino acid protein (FLJ12769) that has been grouped into a family of dolichol phosphate mannose-utilizing mannosyltransferase sequences by its homology with the asparagine glycosylation pathway mannosyltransferases Alg9p and Alg12p and the GPI mannosyltransferases PIG-Bp and S. cerevisiae Smp3p (23,36). Human FLJ12769 contains a HQEXRF amino acid motif that is characteristic of the SMP3 group within this sequence family. Additionally, it exhibits 30% identity and 56% similarity to S. cerevisiae Smp3p and 26% identity and 50% similarity to Schizosaccharomyces pombe Smp3p, thus making it a plausible candidate human GPI fourth mannosyltransferase. We refer to this protein as hSmp3p in this study. Because Man-4 has been identified on the GPIs of two purified brain proteins (6, 26), we used a human brain cDNA library as template for PCR amplification of a hSMP3 cDNA for cloning. The deduced protein sequence of the cloned hSMP3 cDNA was identical to that of FLJ12769.
Human SMP3 Complements the Growth Defects of Yeast smp3 Mutants-Haploid S. cerevisiae cells harboring a deleted smp3 locus are inviable (23). Therefore, to determine whether hSMP3 functions similarly to ScSMP3, we tested its ability to restore viability to a yeast smp3 null mutant. A hSMP3 cDNA was expressed from the S. cerevisiae GAL10 promoter (pGAL-hSMP3) in a heterozygous SMP3/smp3::Kan R diploid strain. Diploid cells were sporulated, and complementation was assessed via tetrad analysis. Asci from diploids harboring pGAL-hSMP3 gave rise to four viable haploid progeny (Fig. 1A). Additionally, two haploids from each tetrad were resistant to G418 and sensitive to 5-fluoroorotic acid, indicating that they harbored the smp3::Kan R allele and that their viability was dependent upon the complementing URA3-containing pGAL-hSMP3 plasmid (Fig. 1B). Expression of hSMP3 also restored viability to a temperature-sensitive (t-s) smp3 mutant S. cerevisiae strain at nonpermissive temperature (data not shown) and a haploid strain of the fission yeast S. pombe harboring a lethal smp3 null mutation. 2 However, hSMP3 did not restore viability to S. cerevisiae cells harboring lethal null mutations in YJR013w or GPI10, genes encoding mannosyltransferases that add Man-1 (37) and Man-3 (16) to yeast GPIs, respectively (data not shown), or to a strain harboring a null mutation of a putative GPI Man-2 mannosyltransferase. 3 Finally, hSMP3 expression complemented the cell wall synthesis defect of smp3-2 cells by restoring their ability to grow in the presence of Calcofluor white (Fig. 1C). Considered together, these data show that hSMP3 can specifically replace the ScSMP3 GPI Man-4 transferase in vivo in yeast and suggest that hSMP3 does not encode a protein with promiscuous GPI mannosyltransferase activity.
Human Smp3p Adds a Fourth Mannose to Yeast GPIs in Vivo-Depletion of Smp3p in S. cerevisiae cells causes a block in GPI synthesis which results in accumulation of a Man 3 -GPI that can be observed after TLC separation of lipids metabolically labeled with [ 3 H]inositol (23). Therefore, we tested the ability of hSMP3 expression to abolish Man 3 -GPI accumulation in a smp3 mutant strain. We constructed a haploid S. cerevisiae ⌬smp3 strain harboring pGAL-hSMP3 from which hSMP3 expression is regulated by the GAL10 promoter. When this strain is grown in medium containing glucose, hSMP3 expression is repressed, cells become depleted of hSmp3p, and Man 3 -GPIs accumulate ( Fig. 2A, lane 4). However, only traces of Man 3 -GPIs accumulate in cells grown in galactose where hSMP3 expression is induced (Fig. 2A, lane 5). Thus, production of hSmp3p abolishes accumulation of Man 3 -GPIs in smp3 cells, presumably by facilitating their conversion to downstream GPI intermediates.
To expand on this notion, we examined whether hSMP3 expression facilitated conversion of a Man 3 -GPI to a Man 4 -GPI in an in vivo mannosyltransferase assay. In this experiment, we utilized a haploid S. cerevisiae strain with conditional defects in two GPI biosynthetic genes: (i) a t-s smp3-2 allele that causes accumulation of a Man 3 -GPI and (ii) a glucose-repressible wild type allele of GPI13, a downstream gene involved in EthN-P transfer to Man-3, which accumulates a Man 4 -GPI when grown in medium containing glucose. Cells grown and [ 3 H]inositol labeled at 37°C in medium containing glucose accumulate only the Man 3 -GPI attributed to the upstream smp3 mutation (Fig. 2B, lane 1). However, under the same conditions, double mutant cells expressing either hSmp3p or a ScSmp3p control from a plasmid regained their ability to form Man 4 -GPIs (Fig. 2B, lanes 2 and 3). This finding is consistent with the notion that hSmp3p is a GPI mannosyltransferase capable of adding Man-4 to Man 3 -GPIs in a manner similar, likely identical, to ScSmp3p. Furthermore, these data show that hSmp3p can recognize a Man 3

-GPI intermediate that does not yet bear EthN-P on Man-3 as a substrate for Man-4 transfer.
Immunolocalization of hSmp3p in the Mammalian ER-The ability of hSmp3p to function in place of ScSmp3p in yeast suggests that the addition of Man-4 to human GPIs may occur during GPI synthesis in the ER. Therefore, we used immunolocalization microscopy to determine the native subcellular location of hSmp3p in human HeLa cells. We first compared localization of hSmp3p to PIG-Bp, a mannosyltransferase involved in the addition of Man-3 to human GPIs during their synthesis in the ER (38). In transiently transfected HeLa cells producing both FLAG-tagged PIG-Bp (PIG-Bp-FLAG) and HAtagged hSmp3p (hSmp3p-HA), immunostaining revealed colocalization of the two proteins (Fig. 3A). Additionally, we compared the localization of hSmp3p-HA with the endogenous ER and Golgi marker proteins calreticulin and golgin-97, respectively. The immunostaining pattern of hSmp3p-HA was iden-tical to that of ER-localized calreticulin (Fig. 3B) and distinct from Golgi-localized golgin-97 (Fig. 3C). We conclude that hSmp3p is targeted to the ER in human HeLa cells.
Distribution of hSMP3 mRNA in Human Tissues and Cell Lines-The distribution of hSMP3 mRNA in various human tissues and cell types was determined by Northern analysis. A commercial RNA expression array containing human mRNA from 58 adult tissues, 7 fetal tissues, and 8 cell lines was probed with a 32 P-labeled hSMP3 cDNA fragment. Expression of hSMP3 was detected in most adult tissues and all fetal tissues (Fig. 4, A and B). In adult tissues, hSMP3 mRNA was most abundant in brain tissues with the highest levels in the cerebellum (Fig. 4A, 17 and 18). This pattern was also seen in fetal samples, where hSMP3 mRNA was most abundant in the brain (Fig. 4B, 67). Expression of hSMP3 was notably high in transverse colon, descending colon, and rectal tissues (Fig. 4A, 36 -38). No hSMP3 expression was detected in spleen tissue (Fig. 4B, 49) or peripheral blood leukocytes (Fig. 4B, 51). Interestingly, hSMP3 mRNA was only weakly detected in each cultured cell line sample represented in the array (Fig. 4A,  1-8).
The finding that hSMP3 mRNA levels are higher in some tissues (e.g. brain and colon) leads to the prediction that levels of hSmp3 protein are also elevated in these tissues. However, Western blotting of standardized lysates from colon, brain, thymus, and spleen cells using a polyclonal antiserum raised to amino acids 443-579 of hSmp3p failed to detect hSmp3p (data not shown). Therefore, native levels of hSmp3p appear to be very low.  FIG. 4. Distribution of hSMP3 mRNA in human tissues and cell lines. A and B, a commercial human multiple tissue mRNA array was probed with a 32 P-labeled hSMP3 cDNA. The amount of mRNA in each sample has been adjusted by the vendor to produce normalized signals for various housekeeping genes. Thus, hSMP3 mRNA abundance can be compared between samples. The hybridized array was exposed for 69 h and then quantitated using a PhosphorImager. The numbered samples represent mRNA from the following tissues and cell lines: 1, leukemia (undifferentiated HL-60); 2, HeLa S3; 3, leukemia (K-562); 4, leukemia (MOLT-4); 5, Burkitt's lymphoma (Raji); 6, Burkitt's lymphoma (Daudi); 7, colorectal adenocarcinoma (SW480); 8, lung carcinoma (A549); 9, whole brain; 10, cerebral cortex; 11, frontal lobe; 12, parietal lobe; 13 brightly with an anti-HA antibody, indicating abundant hSmp3p-HA production. Thus, when incubated with [ 3 H]mannose, GPIs from both transfected and untransfected cells were labeled. Stably transfected cell lines comprised homogeneous populations of hSmp3p-HA-producing cells but made less hSmp3p-HA than transiently transfected cells.

Human Smp3p Expression Causes Man 4 -GPI Formation in Vivo in
TLC separation of lipids isolated from transiently transfected cells revealed formation of three new lipids (M4A, M4B, and M4C), each with a mobility consistent with it being a Man 4 form of a characterized Man 3 -GPI (Fig. 5A, lane 2). The most abundant lipid (M4A) ran as a slightly more polar species than H8, a Man 3 -GPI with three EthN-P residues (14,15,39), consistent with the reported mobility of H8Ј, a putative Man 4 form of H8 which very weakly accumulates in mouse F9 cells that have a deletion of the GPI transamidase subunit GAA1 (15). M4B was slightly more polar than H7Ј and H7, two Man 3 -GPIs that each bear two EthN-Ps (10,14,15,39). Its mobility was consistent with H7Љ, a second putative Man 4 -GPI that weakly forms in mouse ⌬gaa1 cells (15). Finally, M4C ran as a slightly more polar lipid than H6, a Man 3 -GPI intermediate with EthN-P on Man-1 (10). Interestingly, the same three lipids also accumulated in a stable hSMP3-expressing HeLa cell line (Fig.  5A, lane 5) and in HeLa cells transiently expressing S. cerevisiae SMP3 (Fig. 5A, lane 3).
[ 3 H]Mannose-labeled lipids from hSMP3-expressing transiently transfected HeLa cells were treated with JB␣M, PIphospholipase C, and GPI phospholipase D. M4A, M4B, and M4C showed resistance to PI-phospholipase C (Fig. 5B, lanes 4 and 5) and sensitivity to JB␣M (Fig. 5B, lanes 2 and 3) and GPI-phospholipase D (Fig. 5B, lanes 6 and 7), suggesting that each is a GPI intermediate with terminal mannoses not substituted with EthN-P, consistent with the presence of Man-4. The abundance of M4A allowed us to test further whether it was a Man 4 form of H8. Upon TLC purification and JB␣M digestion, M4A shifted mobility to that of H8 (Fig. 5C, lanes 2  and 3), indicating that M4A likely contains the same core Man 3 -GPI as H8 but with at least one additional mannose. The low abundance and proximity of M4B and M4C to adjacent lipids complicated a similar characterization. However, preliminary observations with partially pure M4C and H6 suggest that JB␣M digestion releases the same lipid product from both lipids (data not shown). Additionally, the ability of hSmp3p to mannosylate a yeast lipid with a head group identical to human H6 in vivo suggests that it would likely recognize H6 as a substrate for Man-4 transfer. Thus, it is plausible that M4C is a Man 4 form of H6.

DISCUSSION
Yeast and human GPI precursors differ in the extent to which a fourth mannose is present as a side branch on the third core mannose. In yeast, all GPI precursors likely receive four mannoses because the addition of Man-4 is a mandatory step in GPI biosynthesis which precedes GPI attachment to proteins (23). However, to date, studies of mammalian GPI synthesis have uncovered little evidence that an analogous pathway step occurs in mammals. In the present study, we report the characterization of hSmp3p, a member of a mannosyltransferase family that includes the S. cerevisiae Smp3p GPI fourth mannosyltransferase. First, we demonstrated that hSMP3 could replace ScSMP3 in vivo in yeast complementation experiments, and that expression of hSMP3 in yeast led to in vivo mannosylation of a Man 3 -GPI. Second, we showed that constitutive expression of hSMP3 in HeLa cells results in a dramatic increase in formation of human Man 4 -GPIs. Third, we showed that hSmp3p immunolocalizes to the human ER. Together, these studies indicate that hSmp3p functions as a mannosyltransferase that is capable of adding a fourth ␣-linked mannose to human Man 3 -GPI intermediates during GPI precursor synthesis in the ER. In addition, we demonstrated that hSMP3 is expressed in most human tissues but is only weakly expressed in many cultured mammalian cell lines. Our findings lead us to modify current models for the mammalian GPI assembly pathway. They also raise the possibility that Man 3 -and Man 4 -GPIs may be expressed in a tissue-specific manner and have impli- cations for the specificity of the human GPI transamidase for its GPI substrate.
Specificity of Man-4 Addition to Human GPIs-Five mammalian Man 3 -GPI intermediates (H4, H6, H7, H7Ј, and H8) that differ in the number and position of EthN-P residues have been characterized structurally (10, 12-15, 24, 39, 40). In principle, each could be an acceptor lipid for the transfer of Man-4. However, our observations of the lipids that form in BT7 cells, a HeLa cell line that stably expresses hSmp3p, suggest that only certain Man 3 -GPIs receive a fourth mannose. In BT7 cells, the most abundant late stage GPI intermediates are the putative Man 4 -GPIs M4A, M4B, and M4C, suggesting that the presence of hSmp3p has nearly completely biased the GPI synthetic pathway toward production of Man 4 -GPIs (Fig. 5A,   lane 5). These cells form normal levels of only two Man 3 -GPIs (H4 and H7Ј) and are nearly devoid of lipids H6, H7, and H8. Thus, it is likely that pools of H6, H7, and H8 have become depleted because they are converted efficiently to Man 4 -GPIs, whereas H4 and H7Ј are either not acceptors for Man-4 transfer or are not accessible to hSmp3p.
The most recent scheme for mammalian GPI anchor synthesis suggests that the pathway is split into two branches (15). We have amended this model to reflect our current data (Fig.  6). Because there is no evidence from metabolic labeling studies that Man-4 is added to H4 or H7Ј, we propose that Man-4 transfer to human GPIs likely occurs only within Pathway I of GPI synthesis. Thus, the end product of Pathway I is either the Man 3 -GPI H8 or the Man 4 -GPI M4A when hSMP3 is being expressed (see next section). Additionally, because M4A and M4B likely have EthN-P on Man-3, both could be potential Man 4 -GPI substrates for GPI transamidase.
Formation of the putative Man 4 -GPIs M4A, M4B, and M4C could occur two ways: (i) by direct addition of Man-4 to lipids H8, H7, and H6, respectively; or (ii) by addition of Man-4 to H6 to form M4C, after which M4C is converted to M4B and then to M4A by the sequential addition of two EthN-P groups. Several observations support the latter model. First, hSmp3p can replace ScSmp3p in vivo in yeast, and expression of either Sc-SMP3 or hSMP3 in HeLa cells causes formation of the same three Man 4 -GPIs, suggesting that these proteins have similar, if not identical specificities for their acceptor GPIs. Because ScSmp3p likely transfers Man-4 to yeast Man 3 -GPIs bearing a single EthN-P (23), GPI lipids with similar head groups, including human H6, may be preferred substrates for Smp3 mannosyltransferases. Second, if M4C is a Man 4 form of H6, then it does not bear EthN-P on Man-3 and is not able to be transferred to proteins. Thus, M4C is likely a precursor of M4A and M4B. Finally, BT7 HeLa cells form predominantly M4A and much lower quantities of M4B and M4C, suggesting that M4A is an end product of Man 4 -GPI synthesis. Our current efforts to address the specificity of hSmp3p in vitro will help further determine which Man 3 -GPIs can serve as lipid acceptors for Man-4 transfer.
Abundance of Man 4 -GPIs in Human Cells-To date, the importance of Man-4 addition to mammalian GPI intermediates has been unclear because Man 4 -GPI formation has scarcely been observed in studies of GPI biosynthesis in cultured mammalian cell lines. However, our data suggest that cultured cell lines may not provide an accurate picture of the abundance of Man 4 -GPIs. Using cultured HeLa cells, we showed that: (i) expression of hSMP3 is barely detectable by Northern analysis of HeLa cell mRNA, (ii) HeLa cells form predominantly Man 3 -GPIs, and (iii) increasing hSMP3 expression in HeLa cells causes abundant formation of Man 4 -GPIs. Thus, the ability of a cell to form Man 4 -GPIs likely depends on the level of hSMP3 expression. Our finding that hSMP3 is barely expressed in many cell lines may account for the absence of Man 4 -GPI formation in studies of GPI synthesis. However, formation of Man 4 -GPIs may be far more common and perhaps predominate in most human tissues where hSMP3 mRNA levels are significantly higher (Fig. 4). In support of this notion, a fourth mannose is present on 7 of 10 characterized mammalian protein-bound GPIs (6, 7, 26 -31, 41, 42). Six of the 7 Man 4 -GPIs were isolated from proteins purified directly from primary sources like homogenized organ tissue (6,26,28,29,31) or urine (27). Only 1 was from a protein purified from a cultured mouse cell line (30). In all cases, however, mixtures of both Man 3 -and Man 4 -GPIs were observed.
We would predict Man 4 -GPIs to be most prevalent in brain and colorectal tissues where hSMP3 expression is greatest, and this notion is supported by structural data indicating that Man 4 -GPIs are present on various GPI-anchored brain proteins (6,26). Additionally, hSMP3 expression is barely detectable in the spleen and peripheral blood leukocytes, suggesting that Man 3 -GPIs may predominate in these tissues. Thus, it is possible that the presence or absence of Man-4 on GPI precursors is regulated according to cell type, and we speculate that these differences in GPI glycan structure affect the function or localization of specific GPI-anchored proteins.
Implications for human GPI Transamidase Specificity-Our demonstration that human cells produce both Man 3 -and Man 4 -GPI precursors raises the possibility that the human GPI transamidase complex recognizes both forms as substrates for transfer to proteins. Various observations support this notion.
First, mammalian cells form both Man 3 and Man 4 forms of GPI intermediates that appear structurally competent for transfer to proteins because of the presence of EthN-P on Man-3 (e.g. H7, H7Ј, M4A, and M4B). Second, a mammalian cell line defective in GPI transamidase accumulates both Man 3 -GPIs and trace amounts of lipids with chromatographic mobilities consistent with their being the Man 4 -GPIs M4A and M4B (15). Third, the Man 3 -GPI H8 and a lipid whose chromatographic mobility is consistent with its being the Man 4 -GPI M4A physically associate with human GPI transamidase in immunoprecipitation experiments (43). Finally, mammalian GPIs having either three or four mannoses have been isolated from purified GPI-anchored proteins (6, 7, 26 -31, 41, 42). Thus, in mammalian cells, it is likely that transfer of GPIs to proteins occurs irrespective of the presence or absence of Man-4 on GPI precursors.
In contrast, the addition of Man-4 to S. cerevisiae GPIs is a mandatory step in GPI precursor synthesis which precedes EthN-P addition to Man-3 and subsequent attachment of complete GPI precursors to proteins (23). To date, there is no evidence that yeast can form Man 3 -GPIs that bear EthN-P on Man-3. Additionally, yeast GPI transamidase mutants accumulate exclusively Man 4 -GPI intermediates (20,21,44). Therefore, it is likely that yeast GPI transamidase exclusively requires Man 4 -GPIs as substrates for transfer to proteins. Thus, fungal and mammalian GPI transamidases may have important differences in specificity which could potentially be exploited in the development of novel anti-fungal strategies.
In summary, we have shown (i) that the human genome encodes a fourth GPI mannosyltransferase that is functional and ER-localized when expressed in human cells and (ii) that although the activity of a fourth GPI mannosyltransferase is scarcely detectable in cultured HeLa cells, hSMP3 is expressed widely in human tissues. Our findings also suggest that Man 4 -GPIs may be formed abundantly in certain human tissues, whereas Man 3 -GPIs may predominate in others, raising the possibility that GPI glycans may differentially influence localization and function of GPI-anchored proteins. Further, our results support the notion that human and fungal GPI transamidases differ in their specificity for GPI precursors, with the human transamidase recognizing both Man 3 -and Man 4 -GPIs and the fungal transamidase requiring exclusively Man 4 -GPIs.