Uncleaved Signals for Glycosylphosphatidylinositol Anchoring Cause Retention of Precursor Proteins in the Endoplasmic Reticulum*

Glycosylphosphatidylinositol (GP1)-anchored pro- teins are generally absent from the surface of cells that are defective in GPI biosynthesis. The current study was undertaken to: (a) examine in detail the intracellular localization and fate of precursors of GPI-an-chored proteins in cells that fail to add GPI groups and fb) define structural characteristics of the precursor proteins that determine their intracellular localization. By examining GPI-deficient cells, we show that the uncleaved precursor of the GPI-anchored protein, Q7b, is retained in the cisternae of the endoplasmic reticulum (ER) and is largely lost intracell~arly with a half- time of 2-4 h. Only a small amount (2-1070) of a proteolytically cleaved form of the protein is secreted into the medium. In cells competent for GPI anchor addition, mutation of the putative cleavage/attachment site for GPI addition in Q7b results in a similar phenotype of ER retention of the uncleaved precursor. An aspartic acid residue (Asp’’’) within the Q7b GPI an- choring signal, previously found to be essential for GPI anchor addition shown be critical

Cross (1990)). GPI-anchored proteins are synthesized as precursors containing cleavable amino-te~inal signal sequences, which direct their translocation across the membrane of the endoplasmic reticulum (ER), and carboxyl-terminal polypeptide extensions, which act as signals for attachment of GPI anchors (Boothroyd et al., 1981;Tse et al., 1985;Caras et al., 1987;Waneck et al., 1988b). Information leading to GPI anchor attachment is believed to be encoded by the global structure or properties of the GPI anchoring signal, since there is no recognizable sequence homology in the carboxyl terminus of the precursor proteins. Recent analyses have begun to define both the general characteristics and the critical elements of the GPI anchoring signals, which include: (a) a sequence of 20-30 predominantly hydrophobic amino acid residues, ( b ) a small number of polar or, in some cases, charged amino acid residues interspersed among the hyd! phobic residues, and ( c ) a site of cleavage and GPI attachme2.c consisting of 1 or 2 amino acid residues with small side chains, located amino-terminal to the hydrophobic sequence (Micanovic et al., 1990;Moran et al., 1991). Signals for GPI anchoring are removed shortly after translocation into the ER lumen and replaced by preformed GPI structures. Cleavage of the polypeptide sequence is in most cases strictly dependent upon the ability to add the GPI moiety, which suggests that the entire reaction is catalyzed by a single enzyme or a tightly associated complex (Kodukula et al., 1992).
GPI anchoring has been shown to be critical for the surface expression of several proteins. Failure to replace the carboxylterminal signals by a GPI structure often results in intracellular retention of the proteins (Ferguson and Williams, 1988;Cross, 1990). This a b n o~a l i~ appears to be the cause of a human disorder known as paroxysmal nocturnal hemoglobinuria (PNH). In this disease, there is a generalized lack of expression of GPI-anchored proteins at the plasma membrane, which causes complement-mediated lysis of red blood cells (Davitz et at. 1986 Medof et aL, 1986Hansch et aL, 1988Wilcox et al., 1991;Selvaraj et al., 1987Selvaraj et al., , 1988Jost et aL, 1991;Mahoney et ai., 1992). A number of murine cell lines have been identified that have phenotypic characteristics similar to cells from PNH patients. These include some mutant T cell lymphomas (Hyman, 1985;Sugiyama et al., 1991) and the fibroblast cell line LM-TK-(hereafter referred to as L) (Singh et a[., 1991). In all of these cases, the precursors of GPI-anchored proteins appear to be synthesized normally inside the cells but fail to acquire GPI anchors and are retained intracellularly.
Whereas the addition of GPI anchors has now been shown to be a requirement for the surface expression of numerous proteins, in most cases the exact intracellular localization of the unprocessed precursors in GPI-deficient cells has not been clearly defined. In neutrophils from PNH patients, morphological studies have demonstrated the presence of precursors 12017 of GPI-anchored proteins in the Golgi complex (Jost et al., 1991). Pulse-chase analyses of the GPI-anchoredprotein Thy-1 in a GPI-deficient thymoma cell line, on the other hand, have suggested an early Golgi or ER localization for the unprocessed precursor (Conzelmann et al., 1988). Recent studies have shown that the precursor of a fusion protein with the GPI anchoring signal of decay accelerating factor accumulates in a compartment situated between the ER and the Golgi system (Moran and Caras, 1992). Studies of the fate of placental alkaline phosphatase in GPI-deficient L cells have shown localization of the enzymatic activity to cytoplasmic vacuoles with the appearance of lysosomes (Singh et al., 1991). The intracellularly retained alkaline phosphatase was found to undergo rapid inactivation in a reaction sensitive to acidotropic agents, suggestive of lysosomal degradation (Singh et al., 1991).
The present study was undertaken to examine systematically the subcellular localization of precursors of GPI-anchored proteins in GPI-deficient cells and the structural characteristics of the precursor proteins that may determine their intracellular retention. Most of the study was done on the GPI-anchored protein, Q7b (Waneck et al., 1987;Stroynowski et al., 1987), whose intracellular fate in GPI-deficient cells had not been previously examined. By both biochemical and morphological analyses of GPI-deficient cells, we show that the unprocessed precursor of the Q7b antigen is largely retained within the cisternae of the ER. The retained protein is progressively lost intracellularly but only a small fraction is recovered in the medium. The phenotype of ER retention can be transferred to other proteins by fusion of the GPI anchoring signals from either Q7b or the GPI-anchored form of CD16 (Simmons and Seed, 1988;Scallon et al., 1989), suggesting that the uncleaved signal can act as a determinant of protein retention in the ER. An acidic amino acid residue within the GPI anchoring signal of Q7b is required for ER retention of both Q7b and a chimera having the Q7b GPI anchoring signal. Sedimentation velocity analyses suggest that the presence of an uncleaved signal for GPI anchoring induces aggregation of the precursor proteins, which may lead to their retention in the ER.

MATERIALS AND METHODS
Recombinant DNAs-A cDNA encoding the Qa-2, Q7b gene product (Waneck et al., 1987;Stroynowski et al., 1987) was obtained from Dr. Richard Flavell (Yale University, New Haven, CT). Plasmids encoding the GPI-anchored form of the (Y chain of the I& Fc receptor type 111 (CD16) (Simmons and Seed, 1988;Scallon et al., 1989) were obtained from Dr. Jeffrey Ravetch (Sloan-Kettering Cancer Institute, New York, NY) and Dr. Brian Seed (Massachusetts General Hospital, Boston, MA). Chimeric proteins were constructed by fusing the extracellular domain of the human Tac antigen (interleukin-2 receptor (Y chain; Leonard et al. (1984)), which was modified to have a BglII site at its carboxyl terminus, with the GPI anchoring signal of either Q7b (nucleotides 898-1005) or CD16 (nucleotides 617-735), using the polymerase chain reaction method of Higuchi et al. (1988). Mutants of Q7b were constructed by M13 mutagenesis (Kunkel, 1985). All recombinant DNAs were subcloned into the expression vector pCDL-SRa (Takebe et al., 1988).
Transfectiorw" cells (LM-TK-) and CHO cells (CHO-K1) were obtained from the American Type Culture Collection (Rockville, MD). Cells were maintained in RPMI 1640 medium (Biofluids Inc., Rockville, MD) with 9% fetal bovine serum (FBS) and 0.09 mg/ml gentamicin (RPMI+). For the development of stable transfectants, cells were plated at a density of 3 X 10' celIs/lOO-mm culture dish and grown overnight at 37 "C. Three hours prior to transfection, the medium was changed to 10 ml of Dulbecco's modified Eagle's medium with 9% FBS and 0.09 mg/ml gentamicin. Cells were incubated -12 h with 20 pg of recombinant DNA and 2 pg of pFNeo (Saito et al., 1987) precipitated by calcium phosphate (Graham and van der Eb (1973), as modified by Gorman et al. (1983b)). Cells were washed with PBS, treated with 10% dimethyl sulfoxide in ice-cold PBS for 5 min, washed with PBS, then incubated in RPMI+ for 24 h before addition of selection medium consisting of 3 mg/ml active Geneticin (G418, GIBCO) in RPMI+. Transfectants were screened by immunofluorescence microscopy (see below).
Antibodies-Monoclonal antibodies to Qa-2, 1.9.9, 1.7.5, and 1.5.9, were obtained from Dr. David Sachs (Massachusetts General Hospital, Boston, MA). The polyclonal B6.Kl anti-B6 alloantiserum was a gift of Dr. Lorraine Flaherty (New York State Department of Health, Albany, NY). The anti-Tac monoclonal antibody, 7G7, was a gift from Dr. David Nelson (National Cancer Institute, Bethesda, MD). Polyclonal anti-Tac, R3134, was a gift of Dr. Warren Leonard (National Heart, Lung, and Blood Institute, Bethesda, MD). The anti-H2-Kk antibody, H100-5, was obtained from the American Type Culture Collection (Rockville, MD).
Metabolic Labeling-Cells were treated overnight with 2 mM sodium butyrate in RPMI' to increase expression of the transfected genes (Gorman et al., 1983a), then labeled with 0.5 mCi/ml [%I methionine (Tran3'S-label, ICN) in methionine-free RPMI+ for 30 min at 37 "C. Cells were then chased in RPMI+ for various times at 37 "C, harvested, and frozen at -70 "C. For [3H]ethanolamine labeling, cells were incubated in 50 ml of 0.02 mCi of [3H]ethanolamine/ ml of RPMI+ (Amersham Corp.) for 24 h at 37 "C, then washed with PBS, harvested, and frozen at -70 "C.
Sodium Carbonate Extraction-The association of normal and mutagenized forms of Q7b with membranes in CHO and L cells was studied by extraction with a sodium carbonate solution (Howell and Palade, 1982;Fujiki et al., 1982). CHO cells expressing Q7b, and L cells expressing either Q7b or the Q7b D316V mutant, were metabolically labeled for 16 h with 0.1 mCi/ml [%3]methionine (Tran3'Slabel) in RPMI 1640 medium containing 1/10 the normal concentration of unlabeled methionine and 10% FBS. Labeled cells were rinsed twice with ice-cold PBS and twice with ice-cold 0.25 M sucrose. Cells were removed from plates and disrupted using a Dounce-type homogenizer fitted with a tight pestle (20 strokes). Homogenates were centrifuged for 10 min at 2,000 rpm to remove intact cells and nuclei. Postnuclear supernatants were spun for 10 min at 230,000 X g in a Beckman TL-100 centrifuge. Membrane pellets were resuspended in 0.5 ml of 0.1 M sodium carbonate, pH 11.3, using a Dounce homogenizer (10 strokes, tight pestle) and incubated for 30 min on ice. The extracted proteins were separated from membranes by an additional centrifugation for 10 min at 230,000 X g. Membranes were again resuspended in 0.5 ml of 0.1 M sodium carbonate, pH 11.3. After neutralization to pH 7 by the addition of 1 N HC1, samples were treated with 0.5% (w/v) Triton X-100 and solubilized proteins isolated by immunoprecipitation with the monoclonal antibody 1.9.9. Immunoprecipitation, Electrophoresis, and Autoradiography-Antibodies to Qa-2 or Tac were bound to protein A-Sepharose beads, which were then washed with lysis buffer (1% (w/v) Triton X-100, 300 mM NaCl, 50 mM Tris-HC1, pH 7.6), and added to cell extracts or cleared cultured supernatants. Supernatants were thawed at 0 "C, centrifuged at 2,000 rpm for 10 min at 4 "C, and then incubated at 4 "C with the appropriate antibody bound to protein A-Sepharose for at least 2 h. Except when otherwise indicated, cell extracts were prepared by thawing cell pellets at 0 "C, then solubilizing in lysis buffer containing 10 mg/ml leupeptin, 0.1 mM phenylmethylsulfonyl fluoride, and 1.8 mg/ml iodoacetamide. After vortexing, extracts were incubated at 4 "C for 15 min, then centrifuged at 14,000 rpm for 15 min to pellet the insoluble material. Cell extracts were incubated with the appropriate antibody bound to protein A-Sepharose for at least 1 h at 4 "C. After binding, the cell extract was removed and the beads washed five times with wash buffer (0.1% (w/v) Triton X-100, 300 mM NaCl, 50 rnM Tris-HC1, pH 7.6), then one time in PBS. For lysis in SDS, cell pellets were incubated for 15 min at 37 "C in 100 pl of 1% SDS with 20 pg/ml DNase I in 300 mM NaCl, 50 mM Tris-HC1, pH 7.6, containing 10 mg/ml leupeptin, 0.1 mM phenylmethylsulfonyl fluoride, and 1.8 mg/ml iodoacetamide. One ml of lysis buffer containing 1% (w/v) Triton X-100 was added to dilute the SDS, the insoluble material was removed by centrifugation at 14,000 rpm for Millex-HA filters (Millipore, Bedford, MA). Immunoprecipitated pro-15 min, and the resulting extract was filtered through 0.45-mm teins were eluted by boiling 5 min in SDS sample buffer with 0.5 M dithiothreitol and analyzed by SDS-PAGE on 13% acrylamide gels. Gels were enhanced with 1 M sodium salicylate, dried, and autoradiographed. Quantitation was performed using either a scanning densitometer (Pharmacia LKB Biotechnology Inc.) or a Molecular Dynamics ImageQuant series 400 system (Sunnyvale, CAI. For endoglycosidase H treatment, samples were incubated at 37 "C for -16 h with 5 milliunits of endoglycosidase H (Endo H) (Boehringer Mannheim) in 0.1 M sodium phosphate buffer, pH 6.1,0.1% Triton X-100, 0.03% SDS, 20 mM EDTA, prior to elution and analysis by SDS-PAGE.
lmmunoelectmn M~mscopy"mmunoelectron microscopy of cells fixed with 2% formaldehyde, 0.075 M lysine, and 0.01 M sodium periodate was done as described by Yuan et al. (1987). Sucrose Gradients-Stably transfected CHO or L cells were pulselabeled with [3sS]methionine for 30 min at 37 "C and chased for 0 or 4 h. Cell extracts were prepared as for metabolic labeling, lysed in 600 pi of lysis buffer (1% (w/v) Triton X-100,300 mM NaCI, 50 mM Tris-HC1, pH 7.6) containing 10 mg/ml leupeptin, 0.1 m M phenylmethylsulfonyl fluoride, and 1.8 mg/ml iodoacetamide, then loaded on a 12-ml 2-15% sucrose gradient, which was centrifuged for 20 h at 40,000 rpm at 10 "C. Gradients were fractionated, fractions were diluted by 42% with lysis buffer, and proteins immunoprecipitated with 1.9.9 (anti-Qa-2) or H100-5 (anti-HZ-K') monoclonal antibody bound to protein A-Sepharose. Material at the bottom of the tubes was solubilized in 100 pl of 1% SDS at 4 "C for 30 min, then diluted with 1 ml of lysis buffer, and immunoprecipitated along with the other fractions, Proteins were eluted in SDS sample buffer and analyzed by SDS-PAGE.

RESULTS
~n t r~e~~~r ~a l i z a t w n and Fate of the Q7b Precursor Protein in CHO and L CeZls-Q7b is a member of the Qa-2 family of non-classical class I molecules of the major histocompatibility complex (reviewed by Stroynowski ( 1 9~) ) .
It is composed of a heavy chain (Mr -40,000) associated noncovalently with microglo globulin ( M , 12,000) (Michaelson et aL, 1981). Unlike other class I molecules, which are bound to membranes by a polypeptide transmembrane sequence, Q7' is anchored to the cell surface by a GPI group (Stroynowski et aL, 1987;Waneck et al., 1987). Q7b is synthesized as a precursor protein that is cleaved in the carboxyl terminus with subsequent addition of GPI in the ER (Fig. 1). In cells with defects in GPI synthesis, Q7b is not expressed on the cell surface (Stroynowski et aL, 1987), but its intracellular localization and fate are not known. In order to examine the intracellular localization, processing, and fate of the Q7b precursor protein in cells competent or incompetent for GPI anchor addition, cells expressing Q7b precursor protein were analyzed by pulse-chase metabolic labeling and immunoprecipitation. CHO cells are known to be capable of efficient GPI addition (Kodukula et ab, 1991). In CHO cells, newly synthesized Q7b molecules were initially detected as Endo H-sensitive precursors of M, 36,000-38,000 that were processed into Endo H-resistant forms of M , -40,000 by 4 h of chase ( Fig. 2 4 ) . Only low levels of an M, -37,000 form of Q7b were released into the culture medium (1-5% after 8-12 h) (Fig. 2B)

... W E P P P Y T V S N M A T I A V V
BZ-Xk . . . " " -S " " " -v " L was observed (Fig. M ) , which was not accounted for by the low levels secreted into the medium (Fig. 2B). The cause for this loss is not known.
In contrast to CHO cells, L cell fibroblasts are defective in GPI anchor addition, probably due to lack of synthesis of glycolipidprecursors (Singh et aL., 1991;Sugiyama et ai., 1991).
In these cells, Q7b was initially detected as an M, 38,000 protein that remained sensitive to Endo H at all times of chase and disappeared with a half-time of 2-4 h (Fig. PC).
The intracellular Q7b was associated with &microglobulin (observed in this experiment as a diffuse band at M, 12,000), suggesting that it was not misfolded (Fig. 2C). We observed that, in general, more &microglobulin was coprecipitated with Q7b in L cells than in CHO cells; this difference may be due to preferential association with mouse rather than Chinese hamster &microglobulin. A small amount of the newly synthesized protein was progressively released into the medium as an M , 36,000 species (Fig. 2 0 ) . The size of the secreted Q7b was smaller than the size of the precursor protein by -2,000 (Fig. 2E).
Similar observations were made in L cells upon solubilization with 1% SDS and immunoprecipitation with three monoclonal antibodies, 1.9.9, 1.5.9, and 1.7.5, recognizing three distinct epitope clusters in the Q7b molecule (Sharrow et aL, 1989) and a polyclonai antibody (B6.Kl anti-B6 antiserum, Flaherty, 1976) (data not shown). Quantitation of the results obtained with different antibodies revealed that approximately 80% of the cell-associated protein was lost in 8 h ( Fig.  2C and data not shown), but only 1-10% was recovered in the culture medium ( Fig. 2 0 and data not shown). This indicated that, in L cells, most of the newly synthesized protein disappeared without being released into the culture medium. These observations were consistent with retention of the precursor protein in a pre-Golgi compartment, where a small amount u n d e~e n t cleavage into a form that could be secreted from the cells and the rest was progressively lost intracellularly, The intracellular retention and lack of processing of Q7b were not due to a general defect in transport of MHC molecules in L cells since the class I MHC molecule, H2-Kk, was completely processed into an Endo 33-resistant form by 1 h (Fig. 3).
The hydrophobic carboxyl terminus of the Q7b precursor protein, which acts as a signal for GPI anchor addition, is remarkably similar to the transmembrane domain of other Class I MHC molecules, such as H2-Kk (Fig. 1). Comparison of the hydrophobic domains of Q7b and H2-Kk reveals that the only non-conservative change is Asp316 in Q7b to Val in H2-Kk (Fig. 1). In addition, H2-Kk has a 37-amino acid c~o p~s m i c tail compared to a potential 3-amino acid cyto- n n n.n n n plasmic tail for Q7b. Replacing Asp316 with valine in the hydrophobic domain of Q7b results in the mutant protein being expressed at the cell surface in a phosphatidylinositolspecific phospholipase C-resistant form, presumably as a transmembrane protein (Waneck et al., 1988a). When stably expressed in L cells, the Q7b D316V mutant protein was found to be rapidly processed into an M, 38,000 Endo H-resistant form (Fig. 4). &Microglobulin was associated with the processed form, and little protein was shed in the medium (14%). Thus, removal of the charged amino acid residue from a potential transmembrane sequence resulted in release from pre-Golgi retention. The localization of Q7b at an ultrastructural level was examined by pre-embedding immunoperoxidase staining of fixed, permeabilized cells and electron microscopy. In CHO cells, Q7b was found predominantly at the plasma membrane, as expected for a GPI-linked protein (Fig. 5A, small arrowheads). In contrast, no surface staining was detected in L cells (Fig. 5B, small arrowheads). Instead, the staining was confined to the cisternae of the ER and the nuclear envelope (Fig. 5B, large arrowheads). Immunoreactive material was often seen in dilated cisternae connected to the ER system ( Fig. 5B, asterisk). No obvious staining of the Golgi system or lysosomes was apparent in these cells (Fig. 5B). Mutation of the aspartic acid residue in the GPI anchoring signal to a valine residue (D316V) caused a dramatic change in the distribution of the Q7b precursor that was now almost exclusively detected at the cell surface (Fig. 5C). The globular structures indicated as LD in the different panels of Fig. 5 correspond to lipid droplets; the deposition of osmium in these structures was also observed in cells incubated without antibodies and probably represents nonspecific staining unrelated to the peroxidase reaction product. These observations identified the cisternae of the ER as the pre-Golgi compartment where Endo H-sensitive forms of Q7b are retained.

Decreased Stability of QP Association with Membranes in L
Cells-We next examined whether the different fates of the various Q7b species correlated with changes in the physicochemical properties of the protein. To confirm the presence or absence of a GPI anchor in the different Q7b forms, cells were labeled with [3H]ethanolamine and proteins isolated by immunoprecipitation. Q7b was found to be labeled with [3H] ethanolamine in CHO cells, consistent with normal cleavage of the polypeptide signal and addition of GPI in these cells (Fig. 6A). In CHO cells, Q7b appeared as a doublet (Fig. 6A,  lane 2), which probably corresponded to the high mannose and terminally processed forms of the protein. Neither Q7b nor Q7b D316V were labeled with [3H]ethanolamine in L cells (Fig. 6A), thus confirming the absence of a GPI anchor in these proteins.
To examine the nature of the association of different Q7b forms with membranes in greater detail, microsomes were prepared from labeled cells and extracted with sodium carbonate, pH 11.3. Treatment with sodium carbonate, pH 11.3, extracts soluble and peripheral membrane proteins but not integral membrane proteins (Fujiki et al., 1982;Howell and Palade, 1982). The GPI-anchored form of Q7b in CHO cells was found to be predominantly associated with the membrane fraction when extracted by this method (Fig. 6B), whereas Q7b in L cells was found mainly in the soluble fraction. This association with the membrane. This lack of stable association was observed in spite of the fact that the Q7b precursor in L cells had an uncleaved hydrophobic sequence at its carboxyl terminus, as inferred from the molecular weight of the non-GPI-anchored species. Mutation of Asp316 to Val in the GPI anchoring signal resulted in a protein that was strongly associated with the membrane fraction, consistent with the conversion into a transmembrane protein by a single amino acid change (Fig. 623). P2-rnicroglobulin, which is a peripheral membrane protein, followed Q7b in L cells in its partition between the membrane and soluble fractions, even after extraction at this high pH, further indicating that there is a strong association between the Q7b precursor and &microglobulin. Persistent Aggregation of Q7b Species Retained in the ER-AS an additional property that could explain the different fates of Q7b species, we examined the size of the Q7b forms in sedimentation velocity experiments using sucrose gradients.
Metabolically labeled cells were solubilized with Triton X-100, and proteins separated by centrifugation on a 2-15% sucrose gradient. The gradient was fractionated, and Q7b species immunoprecipitated, then analyzed by SDS-PAGE. Unexpectedly, most of the newly synthesized Q7b protein in CHO cells ran near the human transferrin receptor ( M , 180,000) in the pulse, and near the ovalbumin marker ( M , 45,000) after a 4-h chase (Fig. 7 A ) . Since no other proteins were found to be specifically associated with Q7b under these conditions of solubilization, even after long-term labeling with [35S]methionine (data not shown), this result would suggest a maturation from a multimeric to a monomeric state. In L cells, the Q7b precursor migrated as a heterogeneous species starting close to the transferrin receptor ( M , 180,000) and extending to the bottom of the gradient for both the pulse and the chase samples (Fig. 7B). This observation suggested that in L cells maturation of newly synthesized Q7b was arrested at a multimeric state. Mutation of Asp316 in Q7b to valine resulted in a protein that ran as a homogeneous species near the ovalbumin marker (Mr 45,000) for both the pulse and the chase (Fig. 7C), similar to the class I MHC molecule, H2-Kk (Fig. 70). Thus, retention in the ER of the uncleaved Q7b precursor correlated with persistence of a multimeric complex, the existence of which was dependent upon the presence of the negatively charged residue within the hydro- phobic sequence of the Q7b GPI anchoring signal. Mutation of the Cleavage/Attachment Site Results in ER Retention in GPI-competent Cells-To further test whether retention in the ER was solely due to the presence of the uncleaved signal for GPI anchoring in the Q7b precursor protein, independently of the GPI anchoring competency of the cells, we constructed a mutant Q7b protein that could not be cleaved by the GPI anchoring enzyme(s). The mutant replaced 2 residues at or near the predicted cleavage/attachment site from A~n~' ' -M e t~~ to Tyr-Phe. The choice of amino acids was based on the observation that only a certain subset of amino acids is found at the cleavage/attachment site (Ala, Asn, Asp, Gly, Cys, Ser, Met), and bulky amino acids are never found at the GPI addition site (reviewed by Ferguson and Williams (1988) and Cross (1990)). Expression of the mutant protein in CHO cells, which are otherwise competent for GPI anchor addition, resulted in a protein that remained in an Endo H-sensitive form and was gradually lost (tip = 2-3 h) without significant secretion into the culture medium (Fig. 8). By immunofluorescence microscopy, Q7b NM-YF in CHO cells gave an ER staining pattern similar to that seen for Q7b in L cells (data not shown). Thus, the fate of proteins with uncleaved signals for GPI anchoring is the same in GPIcompetent cells (CHO) as in GPI-deficient cells (L), and is therefore not dependent on metabolic defects other than the inability to cleave the carboxyl-terminal polypeptide signal.
Information Leading to ER Retention Is Transferable-The above results suggested that information leading to retention of the uncleaved precursor protein was contained in the carboxyl-terminal GPI anchoring signal. If this were true, then it would be possible to transfer the retention phenotype onto another protein. To test this hypothesis, we constructed a chimeric protein with the extracellular domain of the interleukin-2 receptor a chain (Tac), which is normally expressed at the cell surface as a transmembrane protein (Leonard et al., 1984), fused to the GPI anchoring signal at the carboxyl terminus of Q7b (amino acids 299-333) (Fig. 1). The stable transfection of this chimeric protein, T~c -Q~~, into L cells resulted in a protein that remained Endo H-sensitive at all times after synthesis ( Fig. 9) and had a half-life of 1-2 h. Little (51%) secretion of T~c -Q~~ into the culture medium was found (Fig. 9). A similar loss of newly synthesized protein was observed where cells were solubilized with 1% SDS and proteins immunoprecipitated with a polyclonal anti-Tac antibody (data not shown). Mutation of Asp316 to Val in the context of the T~c -Q~~ chimera ( T~c -Q~~ D316V) also resulted in processing of the chimeric protein into an Endo Hresistant form (Fig. 9). The ability to transfer the ER retention phenotype onto another protein confirmed that the information leading to ER retention was entirely located in the uncleaved GPI anchoring signal and was dependent on the presence of an Asp residue within the context of a hydrophobic sequence.
To ascertain whether the ability to cause retention and subsequent protein loss was a property common to other GPI signals, we examined the effect of adding the GPI anchor signal of a CD16 species (Simmons and Seed, 1988;Scallon et al., 1989) to the Tac extracellular domain. Although the hydrophobic carboxyl terminus of GPI-linked CD16 has a sequence different from that of Q7b, they both have similar general properties and serve as GPI signal sequences (Fig. 1). Like Q7b, there is an aspartic acid residue in the hydrophobic domain of CD16, but this charged group is positioned further downstream of the attachment site. In CHO cells, the Tac-CD16 chimera was found to be processed into an Endo Hresistant form (Fig. lo), in contrast to L cells where the protein remained Endo H-sensitive and rapidly disappeared without being secreted into the medium (Fig. 10). These results demonstrate that another GPI signal is also capable of causing retention and loss of precursor protein in cells defective in GPI anchor addition. Similar observations were made when cells were solubilized in 1% SDS and the rates of disappearance of both the Tac-CD16 and the T~c -Q~~ chimeras did not change in the presence of NH4Cl, which inhibits lysosomal degradation (data not shown). synthesis. The Q7b precursor protein was found to remain sensitive to Endo H and was localized to the ER cisternae by immunoelectron microscopy. The ER-retained protein disappeared with a half-time of 2-4 h in these cells. Within 8 h, 80% of the precursor protein was lost and only 1-10% was secreted into the culture medium as a proteolytically cleaved form. The small amount of Q7b secreted into the medium was of lower molecular weight than the precursor protein, and may result from cleavage of the GPI anchoring signal by the putative transamidase that normally adds GPI anchors or by an ER resident proteolytic activity. The net loss of newly synthesized Q7b precursor protein, observed even when cells were solubilized under denaturing conditions, suggests that the protein was degraded. The localization of the precursor protein to the ER, as demonstrated by biochemical and morphological analyses, and the inability to prevent losses of the precursor protein with NH&1 are indicative of degradation occurring by a pre-Golgi, nonlysosomal pathway described previously (Klausner and Sitia, 1990;Bonifacino and Lippincott-Schwartz, 1991). These results demonstrating localization of the uncleaved Q7b precursor to the ER cisternae are in line with suggestions of a pre-medial Golgi localization for unprocessed precursor proteins in GPI-deficient cells (Conzelmann et al., 1988). Recent studies using pharmacologic inhibitors of GPI addition have also suggested that the unprocessed precursor proteins are incompetent for transport out of the ER and undergo variable cleavage into forms that are secreted into the culture medium (Lisanti et al., 1991;Takami et al., 1992). Other precursors of GPI-anchored proteins have been localized to post-ER compartments (Jost et al., 1991;Singh et al., 1991;Moran and Caras, 1992), suggesting that molecules that escape the ER can still be retained in other compartments of the secretory pathway. ER Retention-What causes the precursor of GPI-anchored proteins to be retained in the ER? For many proteins, retention in the ER is due to aberrant folding of the polypeptide chains (Hurtley and Helenius, 1989). Global misfolding of the Q7b precursor protein in L cells, however, seems unlikely since &microglobulin is tightly associated with the precursor protein and even partitions with Q7b upon sodium carbonate extraction of microsomes. Retention in the ER is more likely related to local conformational changes or to the intrinsic tendency of GPI anchoring signals to engage in protein interactions. One possibility is that the uncleaved GPI anchoring signal causes sustained association with ER resident proteins, such as components of the GPI anchoring system or the protein, p88, that has been shown to associate transiently with newly synthesized MHC class I molecules (Degen and Williams, 1991;Wada et al., 1991;Hochstenbach et al., 1992). Even after long term (24 h) labeling with [35S]methionine, however, we observed that no other proteins consistently coprecipitated with either Q7b or the Tac chimeras under the conditions used in our experiments? An alternative possibility is that the presence of the uncleaved signal for GPI anchoring causes self-association of the precursor proteins leading to the formation of transportincompetent aggregates. Analysis by sedimentation on sucrose gradients revealed that the Q7b precursor protein indeed * M. D. Delahunty, F. J. Stafford, L. C. Yuan, D. Shaz, and J. S.

Aggregation of the Precursor Proteins as
Bonifacino, unpublished observations. formed heterogeneous multimeric species, under conditions in which no other associated proteins were detected. An additional correlation that suggests a role for multimer formation in ER retention is the observation that proteins lacking charged residues in their hydrophobic sequences, such as H2-Kk and the Q7b D316V mutant, migrate as monomers on sucrose gradients and are normally transported out of the ER. In GPI-competent cells (CHO), there is also an initial multimer formed that is not found after the 4-h chase period, when the protein is presumably GPI-anchored. It is thus possible that the process of GPI anchor addition may reverse aggregation of the precursor proteins, thereby allowing transport of the protein from the ER to the Golgi system.
Similarity of GPI Anchoring Signals and Heterophilic Sequences-The polypeptide signals for GPI anchoring in the carboxyl terminus of the precursor proteins were found to be responsible for targeting these proteins for retention in the ER. This was concluded from the correlation between the presence of an uncleaved GPI anchoring signal and the manifestation of the retentionldegradation phenotype, and from the demonstrated ability to transfer this phenotype by simply fusing the GPI anchoring signals of either Q7b or CD16 to the Tac extracellular domain. In addition, small changes in the sequence of the GPI anchoring signal resulted in a dramatic alteration in the fate of the precursor proteins. For instance, the acidic residue in the carboxyl-terminal GPI signal sequence of Q7b has been shown previously to be critical for GPI addition and surface expression of the molecule (Waneck et al., 1988a). Mutation of this charged residue to a hydrophobic residue allowed efficient egress of the mutated Q7b precursor from the ER. The Asp residue was also shown in our study to be responsible for the ER retention of a chimeric protein containing the Q7b GPI anchoring signal. To what extent these observations can be extended to other GPIanchored proteins remains to be established. Considerable variation exists in amino acid sequence among GPI signals from different proteins. Instead of acidic residues, the hydrophobic sequence of other GPI signals is interrupted by basic residues (Lys, Arg), strongly polar but uncharged residues (Gln, Asn), groups of weakly polar residues (Thr, Ser) or helix-breaking residues such as Pro (reviewed by Ferguson and Williams (1988) and Cross (1990)). The presence of any of these residues has the effect of reducing the overall hydrophobic character of the carboxyl-terminal sequences and may thus confer similar functional properties to polypeptides with different primary structures. On this basis, it is reasonable to expect that other precursors of GPI-anchored proteins would have similar fates (i.e. ER retention and variable proteolytic cleavage). GPI anchoring signals are in some ways reminiscent of the transmembrane domains of subunits of certain multimeric complexes, such as the T and B cell antigen receptors, and immunoglobulin Fc receptors. The transmembrane domains of these proteins are characterized by the presence of poten-tially charged or strongly polar residues within otherwise hydrophobic transmembrane sequences (Clevers et Weissman et al., 1988;Williams et al., 1990;Hermanson et al., 1988;Sakaguchi et al., 1988;Hombach et al., 1988;Blank et al., 1989). Because of their predominantly hydrophobic character with foci of strong hydrophilicity, we refer to this type of sequences as "heterophilic sequences." Like the uncleaved signals for GPI anchoring, heterophilic sequences can cause ER retention and, in some cases, degradation of unassembled subunits in the ER (Bonifacino et al., 1990;Wileman et al., 1990;Williams et al., 1990). The ER retention phenotype is also transferable to other proteins and depends on the presence of the potentially charged or strongly polar residues Wileman et dl., 1990;Williams et al., 1990).
An additional similarity between some GPI anchoring signals and heterophilic sequences is their potential for mediating subunit interactions. For example, the transmembrane domains of some T cell receptor chains have been shown to promote subunit assembly, in a process that requires the presence of the potentially charged transmembrane residues (Manolios et al., 1990;Cosson et al., 1991). Strikingly, some GPI anchoring signals can also behave like the transmembrane domains of receptor subunits. Recent studies have demonstrated that carboxyl-terminal sequences from mem-.brane-bound IgD and the precursor of a GPI-anchored form of CD16 can serve dual functions as transmembrane domains or GPI anchoring signals, depending on the expression of other components of the B-cell antigen receptor and Fc receptor, respectively (Hibbs et al., 1989;Wienands and Reth, 1992). In cases where GPI anchoring signals and heterophilic sequences are not directly interchangeable, small changes can convert one into another. For instance, deletion of 3 amino acid residues from the cytoplasmic tail of membrane-bound IgG results in conversion of the protein from a transmembrane to a GPI-linked form that is expressed at the cell surface independently of other subunits of the B cell antigen receptor (Mitchell et al., 1991).
The structural and functional similarity of some GPI anchoring signals with heterophilic sequences suggests a common mechanism underlying the various processes in which they are involved. In essence, both GPI anchoring signals and heterophilic sequences can be seen as protein interaction motifs, whose primary function is to mediate binding to either components of the GPI anchoring system or to other protein subunits. In either case, the result of the interaction is the production of species capable of egressing from the ER. In the event that the functionally relevant interactions fail to occur or are unproductive, an additional function of these sequences is manifested, which is the ability to cause retention in the ER.