Implications of lipid moiety in oligomerization and immunoreactivities of GPI-anchored proteins.

Glycosylphosphatidylinositol (GPI) enriches GPI-anchored proteins (GPI-AP) in lipid rafts by intimate interaction of its lipid moiety with sphingolipids and cholesterol. In addition to such lipid-lipid interactions, it has been reported that GPI may interact with protein moiety linked to GPI and affect protein conformations because GPI delipidation reduced immunoreactivities of protein. Here, we report that GPI-APs that have not undergone fatty acid remodeling exhibit reduced immunoreactivities in Western blotting, similar to delipidated proteins, compared with normal remodeled GPI-APs. In contrast, immunostaining in flow cytometry and immunoprecipitation did not show significant differences between remodeled and unremodeled GPI-APs. Moreover, detection with premixed primary/secondary antibody complexes or Fab fragments eliminated this difference in Western blotting. These results indicate that normally remodeled GPI enhanced oligomerization of GPI-APs and that inefficient oligomerization of unremodeled GPI-APs was responsible for reduced immunoreactivities. Moreover, the reduction in immunoreactivities of delipidated GPI-APs was most likely caused by the same effect. Finally, by chemical cross-linking of surface proteins in living cells and cell killing assay using a pore-forming bacterial toxin, we showed that enhanced oligomerization by GPI-remodeling occurs under a physiological membrane environment. Thus, this study clarifies the significance of GPI fatty acid remodeling in oligomerization of GPI-APs and provides useful information for technical studies of these cell components.

and not freely accessible to the attached protein. Thus, reduced immunoreactivities appeared unexplained by altered protein moiety interactions with cleaved GPI and/or proximate surface circumstances, as was explained in the case of delipidated GPI-APs. Moreover, it is not easy to explain why many GPI-APs with completely different peptides were similarly affected by the same delipidated GPI. Therefore, it was considered that the mechanism causing reduced immunoreactivities remained unclear.
Here, we report that both a defect in remodeling and delipidation of GPI-APs lead to reduced immunoreactivities in Western blotting due to the reduced oligomerization during denaturation/renaturation, and that the nature of remodeled GPI to oligomerize is also conserved under physiological membrane conditions. Thus, through the studies of fatty acid remodeling of GPI-APs, an important characteristic of GPI to enhance oligomerization of GPI-APs is revealed here, which would affect many biological aspects, such as immunoreactivities, signal transduction, and stabilization of rafts.

Cells, plasmids, and antibodies
PGAP3 knock-out mice were created as previously described. Mouse embryonic fi broblasts (MEF) isolated from the mice ( 14 ) were cultured in DMEM high-glucose medium supplemented with 10% fetal calf serum (FCS). Mouse PGAP3 was cloned by PCR using primers (GTTGGTCGAGGTGTGACA-GAATGGCCAAGC and CTTCTCTAGACACTCTCTCCCAGC-CTCTTTAGTC) and subcloned into pLIB2-BSD retroviral vector, which was made from pLIB vector (Clontech Laboratories, Inc., Madison, WI) by modifying the multicloning site and inserting blasticidin resistance gene (BSD) driven by pgk promoter ( 15 ). Human placental alkaline phosphatase N -terminally tagged with signal sequence derived from CD59 and hemagglutinin peptide (HA-PLAP) was described previously ( 16 ). EGFP-Flag-CD59 was generated by tandem fusion of signal peptide derived from human growth hormone, enhanced green fl uorescent protein (EGFP), Flag-tag, and CD59, and subcloned into pRCMV retroviral vector. Wild-type Chinese hamster ovary (CHO) cells (3B2A, GD3S-37) and PGAP2/PGAP3 double mutant (DM) CHO cells that stably expressed human decay accelerating factor (DAF) and CD59 have been previously described ( 6 ). These CHO cells were cultured in Ham's F-12 supplemented with 10% FCS. Monoclonal anti-CD59 (5H8, prepared in this lab), Thy-1 (G7, prepared in this lab), PLAP (8B6, Sigma-Aldrich, Inc., St. Louis, MO), urokinase-type plasminogen activator receptor (uPAR, 5D6, prepared in this lab), and DAF (IA10, prepared in this lab) were used for detection of GPI-APs. Additionally, anti-EGFP (Roche Applied Science, Indianapolis, IN), HA (HA7, Sigma-Aldrich), Flag (M2, Sigma-Aldrich) were used for tag peptide detection. Anti-Ribophorin-1, GAPDH, and Synthaxin-6 were used as quantitative controls. Fab fragments of G7 were prepared using the ImmunoPure Fab preparation kit containing immobilized papain (Pierce Biotechnology, Inc., Rockford, IL). Premixed antibodies were generated by incubating same amounts of primary and secondary antibodies that had been prepared by diluting the stock solutions (1:1000) in 5% skim milk containing TBS (20 mM Tris-Cl, pH 7.4, 150 mM NaCl) at room temperature for 2 h. saturated fatty acid chains, is critical for accumulation of GPI-APs in lipid rafts where (glyco)sphingolipids and cholesterol are enriched in a liquid-ordered manner ( 7 ). Enrichment of GPI-APs in lipid rafts, owing to this lipid composition, is considered to be the most important function of GPI, as this enrichment caused by lipid-lipid interaction affects many biological behaviors of GPI-APs, such as signal transduction, interaction with other molecules, intracellular traffi cking and localization, and secr etion (8)(9)(10).
In addition to the above features of GPI, several reports have indicated another possibly important aspect of GPI, in which it might interact with its protein moiety, based on the unexpected facts that removal of the GPI lipid moieties, namely delipidation, by phospholipases, such as phospholipase C and GPI-specifi c phospholipase D, drastically reduces reactivities to antibodies (11)(12)(13). This phenomenon has been commonly observed in many GPI-APs, such as mammalian Thy-1, CD58, and CD59 and protozoan VSG, EP-, and GPEET-procyclins. The reduction in immunoreactivity is most likely due to conformational changes of proteins caused by altered interactions of the protein moiety with delipidated GPI and/or proximate surface circumstances ( 11 ). Thus, protein-GPI interactions may modulate the functions of these proteins.
Surprisingly, we found that such considerably reduced immunoreactivities were also observed in a wide range of GPI-APs in PGAP3-defi cient cells, which have not undergone fatty acid remodeling and still carry two fatty acid chains. The differences between remodeled and unremodeled GPI-APs exist only in the composition of the sn -2 fatty acyl chain, either saturated or unsaturated, that is completely buried within the membrane. Also, unremodeled GPI is tethered to the surface membrane Processing of GPI-anchor by PGAP3 and PI-PLC. GPI-APs are subjected to fatty acid remodeling in the Golgi. An sn -2 unsaturated fatty acid (for example, C20:4) is cleaved by PGAP3 and substituted with a saturated chain (typically C18:0) by an unknown acyltransferase, remodeling GPI to two saturated long acyl chains. PI-PLC cleaves between phosphate and diacylglycerol in the phosphatidylinositol portion of GPI. The cleaved form becomes a soluble delipidated protein.

Immunoprecipitation of HA-PLAP and EGFP-Flag-CD59
Cells were lysed with O ␤ G buffer. The lysates were incubated with 5H8 or 8B6 antibodies together with protein-G beads (GE Healthcare Bio-Sciences AB, Uppsala, Sweden) in a cold room overnight. After washing the beads with TBS several times, bound proteins were eluted using 2× SDS sample buffer (4% SDS, 0.7M Tris-Cl, pH 6.8, 10% glycerol).

Flow cytometry for GPI-APs
Cells were harvested using trypsin/EDTA mixture (Sigma-Aldrich) or with 5 mM EDTA-containing PBS (for EGFP-Flag-CD59), and the surface expression of GPI-APs was determined by staining with 5H8, IA10, G7, or 8B6 antibodies, followed by fl ow cytometric analyses (CantII; BD Biosciences Co., Franklin Lakes, NJ). Control staining was obtained with isotype-matching antibodies or without primary antibodies, using cells that were PGAP3restored PGAP3 Ϫ / Ϫ MEF cells or cells equivalent to wild-type cells

Extraction of proteins, phosphatidylinositol-specifi c phospholipase C treatment, and immunoblotting
Whole-cell lysates for Western blots were prepared using n -octyl-␤ -D -glucoside (O ␤ G) buffer (20 mM Tris-Cl, pH 7.4, 150 mM NaCl, 1 mM EDTA, and 60 mM O ␤ G) on ice for 30 min, then centrifuged at 4°C for 15 min to remove debris. For phosphatidylinositol-specifi c phospholipase C (PI-PLC) treatment, the lysate was incubated at 20°C for 1 h with 1 unit/ml of Bacillus cereus PI-PLC (Molecular Probes, Life Technologies, Grand Island, NY). For analysis of intracellular GPI-APs, cells were incubated at 10°C for 6 h or at 37°C for 30 min with PI-PLC, washed with PBS, and then lysed with O ␤ G buffer. Samples were prepared with boiling or nonboiling treatment, and then subjected to SDS-PAGE or alkaline phosphatase (ALP) analyses. For blotting, samples were loaded onto SDS-PAGE and then transferred to PVDF or nitrocellulose membranes, and then probed with each antibody.

RESULTS
Considerably weakened Western blotting reactivity of GPI-APs lacking fatty acid remodeling Discrepancies were found in the expressions of unremodeled GPI-APs lacking fatty acid remodeling due to defective PGAP3 between immunoblotting of whole-cell lysate and fl ow cytometry that measured cell surface expression. In Western blot analyses, compared with PGAP3recovered PGAP3-knockout (PGAP3 Ϫ / Ϫ ) MEFs, PGAP3 Ϫ / Ϫ MEFs showed remarkably decreased intensities of Thy-1, CD59, and PLAP bands ( Fig. 2A , left). However, in fl ow cytometry, the surface expression of these GPI-APs on PGAP3 Ϫ / Ϫ MEFs was comparable with PGAP3-recovered MEFs ( Fig. 2A , right). In addition to MEF, CHO cells stably expressing CD59 and DAF were employed to check whether this phenomenon was present in remodeling-defi cient CHO mutant cells. Wild-type 3B2A parent cells and PGAP2/PGAP3 DM cells lacking fatty acid remodeling also showed discrepancies in CD59 and DAF expressions, similar to PGAP3 Ϫ / Ϫ MEF ( Fig. 2B ). These results indicate that the immunoreactivity discrepancies between Western blot analyses and fl ow cytometry were general phenomena in unremodeled GPI-APs.

Intracellular pools of GPI-APs do not account for different estimates by Western blotting
As fl ow cytometry measures only surface GPI-APs, one possible explanation for the above discrepancies was that wild-type cells possessed much larger intracellular pools of GPI-APs than the cell surfaces. To evaluate the amounts of intracellular GPI-APs, MEF cells expressing HA-tagged PLAP (HA-PLAP) were incubated with PI-PLC to remove surface GPI-APs at 37°C or at 10°C, the latter blocking exocytosis and endocytosis. Both cell preparations showed signifi cantly decreased surface expression of PLAP ( Fig. 3A ), indicating effective PI-PLC activity. Subsequently, intracellular PLAP analyses by Western blotting ( Fig. 3B ) showed that PI-PLC treatment strikingly decreased PLAP amounts in whole-cell lysates as well as surface expression. This was considered to indicate that the majority of PLAP existed on cell surfaces. as a representative cell type; control staining obtained from one control cell type was used in all fi gures of the same experiments.

Measurement of ALP activity
ALP activity of HA-PLAP was measured using three methods. In the fi rst method, cell extracts were prepared with O ␤ G lysis buffer, and the lysate ALP activity measured using an SEAP assay kit (Clontech Laboratories). The second method involved a protein denaturation/renaturation experiment , in which cell lysis with Triton lysis buffer [100 mM Tris-Cl, pH 9.5, 100 mM NaCl, 5 mM MgCl 2 , 1% Triton X-100, and 1 Protease inhibitor cocktail without EDTA (Roche Applied Science)] with or without 2.5% SDS was performed, and then ALP activity was measured by addition of 0.25 mM CSPD (Roche Applied Science). And fi nally, ALP activity was measured on transferred membranes after SDS-PAGE using O ␤ G lysis buffer, and ALP activity was measured by adding CDP-star according to manufacturer's instruction (GE Healthcare Bio-Sciences AB), a luminous substrate for ALP.

Detergent-resistant membrane fractionation
Cells were harvested from the plate by using PBS containing 2.5 mM EDTA and 0.5% BSA. After centrifugation, the cell pellet was resuspended in MBS-E [25 mM 2-(N-morpholino)ethanesulfonic acid, pH 6.5, 150 mM NaCl, and 5 mM EDTA] containing protease inhibitors supplemented with 1% TX-100, incubated for 20 min on ice, and homogenized by a potter-type Tefl on homogenizer. The volume of lysis buffer was 25 times the weight of cell pellet (typically 25-30 million cells/ml lysis buffer). One milliliter of lysate was mixed with 1 ml of 80% sucrose in MBS-E, transferred to a centrifuge tube for SW41-Ti, overlaid with 7 ml of 30% and 2 ml of 5% sucrose in MBS-E, ultracentrifuged at 38,000 rpm for 16-18 h at 4°C, and fractionated from the top using Piston Gradient Fractionator (BioComp Systems) with each fraction of 1 ml (total 11 fractions). Aliquots of each fraction were mixed with 6 × sample buffer, without a reducing reagent, and applied to 5-20% gradient SDS-PAGE.

Chemical cross-linking of cell surface proteins
Cells cultured in 6-well plates were treated with 1 mM of cross-linking agent 3,3 ′ -Dithiobis(sulfosuccinimidylpropionate) (DTSSP) dissolved in PBS for 30 min at room temperature. DTSSP contains amine-reactive NHS-ester at each end of a spacer arm 12.0 Å (8-atom). DTSSP-treated and nontreated cells were lysed in O ␤ G lysis buffer. Samples for Western blotting were prepared under nonboiling, nonreducing conditions. The dimers and monomers were visualized using primary (G7 anti-Thy-1)/secondary antibody premix.

Treatment of aerolysin and viability assay
Nontoxic fl uorescein-conjugated proaerolysin (FLAER) was used to analyze binding effi ciency to the cell surface. Cells (1 × 10 6 ) were stained with 10 Ϫ 8 and 10 Ϫ 9 M of FLAER. For viability assay, 5 × 10 3 MEFs per well were cultured in 96-well plates. Prewarmed toxin (0.1 ml) was added to the wells and incubated for 3 h at 37°C. After   Fig. 2 ) were lysed, and whole-cell lysates prepared under nonreducing boiling conditions were applied to SDS-PAGE. EGFP fl uorescence in gel and on transferred membrane (left and middle panels, respectively) were detected using fl uorescence laser scanner; blotting membrane was then probed against 5H8 anti-CD59 antibody (right panel); nitrocellulose membrane was used instead of PVDF membrane due to high fl uorescent background. The same MEF cells were used for fl ow cytometry, Western blotting, and enzyme assay (B-D, respectively). B: Cells were analyzed with anti-PLAP antibody in fl ow cytometry. Gray shadows, staining with 8B6 antibody; dotted lines, staining with isotype-matching control antibody. Numbers indicate the ratio of the geomeans. C: Loading samples were prepared from whole-cell lysates under nonreducing boiling conditions. HA-PLAP were detected with 8B6 anti-PLAP and HA7 anti-HA antibodies in Western blotting. GAPDH, quantitative control. D: ALP activities of HA-PLAP in whole-cell lysates were measured.
In addition, it appeared that intracellular pools were small enough to exclude the possibility that the immunoreactivity discrepancies between Western blot and fl ow cytometric analyses came from differing intracellular pools of remodeled and unremodeled GPI-APs. This was also confi rmed by PGAP2/PGAP3 DM CHO cells. PI-PLC removed more than 90% of DAF, uPAR, and CD59 expressed on the cell surface at 10°C and even more at 37°C, as determined by fl ow cytometry ( Fig. 3C ). The amounts of remaining GPI-APs after PI-PLC treatment in wholecell lysates correlated best with those of surface expression detected by fl ow cytometry ( Fig. 3D ), indicating that CHO cells also had much smaller GPI-AP intracellular pools than on the surface.

Weakened signals of unremodeled GPI-APs in Western blotting were most likely due to their weak reactivities to antibodies
The suspicion was that a technical issue still remained, such that unremodeled GPI-APs might have been lost during the experimental procedure. Thus, the presence of GPI-APs was confi rmed in each step of Western blot analyses. Analysis by fl uorescence laser scanner confi rmed that similar amounts of remodeled and unremodeled EGFP-Flag-CD59 proteins were loaded into polyacrylamide gels and transferred to nitrocellulose membranes, albeit immunodetection of EGFP-Flag-CD59 on the membrane with anti-CD59 antibody exhibited signifi cant differences between the gels and membranes ( Fig. 4A ). Consistent with surface expressions of HA-PLAP in fl ow cytometry ( Fig. 4B ), anti-HA antibody detected similar amounts of HA-PLAPs by Western blotting, contrary to the response of anti-PLAP antibodies that showed big differences between remodeled and unremodeled samples ( Fig. 4C ). Measuring PLAP enzymatic activities in whole-cell lysates from both cells also confi rmed that similar amounts of PLAP were present ( Fig. 4D ). Thus, all results obtained here were comparable, except for Western blot analyses with anti-GPI-AP antibodies, suggesting that the discrepancy with Western blotting was most likely produced by differences in antibody immunoreactivities against remodeled and unremodeled GPI-APs.
indicates that SDS was strong enough to convert the majority of the dimer to monomer even under nonboiling conditions. Furthermore, this indicates that the very minor population of dimeric HA-PLAP ( Fig. 5C , triple asterisks) was very effi ciently detected with anti-PLAP antibody compared with major monomeric HA-PLAP (single asterisk) in the same lanes. The intensities of monomeric HA-PLAP bands produced under nonboiling conditions were weaker and almost invisible compared with those produced under boiling ( Fig. 5C , double asterisks), most likely due to the broadness revealed by anti-HA antibody ( Fig. 5B ). Importantly, no differences in dimeric HA-PLAP band intensities were observed between the remodeled and unremodeled HA-PLAPs ( Fig. 5C , triple asterisks, lower panel), indicating no differences in sensitivity to SDS and consistent with results in Fig. 5A . Thus, remodeled and unremodeled HA-PLAPs showed different signal intensities only after denaturation/renaturation cycles in Western blotting. These results suggested the hypothesis that antibody affi nities against GPI-APs were not very high and, therefore, that dimerization or oligomerization of GPI-APs might be critical for effi cient detection by allowing multivalent binding with antibodies, increasing these avidities. This also suggested that remodeled GPI could have induced oligomerization more effi ciently than unremodeled GPI during renaturation that occurred when SDS was removed from monomeric GPI-APs. Boiling in the presence of SDS might have caused complete monomerization, but during transfer from electrophoresis gels to the blotting membranes, some monomers might have oligomerized due to GPI hydrophobic interactions. In addition, the effi ciency of oligomerization might have been higher in remodeled GPI than in unremodeled GPI, by analogy with the fact that saturated long acyl chains of remodeled GPI associate with each other more densely and tightly than the bending unremodeled GPI. This effect would have led to GPI-AP enrichment in lipid rafts in the cell membrane. This hypothesis was supported by detection of EGFP-Flag-CD59, a chimeric protein in which EGFP, Flag-tag, and CD59 were tandemly fused ( Fig. 5F ). Similar amounts of remodeled and unremodeled EGFP-Flag-CD59, confi rmed by fl uoroscanning of loaded gels, were detected with three different antibodies against EGFP, Flag-tag, and CD59 in Western blot analyses. Antibodies against EGFP and Flag also showed large differences in signal intensities, as did an antibody against CD59, clearly indicating that these differences did not arise from the specifi c protein-GPI interactions, such as between CD59 and GPI, and that the present hypothesis was reasonable. This hypothesis was proven more directly using premixed primary/secondary antibody complexes to increase their avidities, overcoming the supposed low affi nity of the primary antibody by increasing the chances for multivalent binding. In fact, premixed primary/secondary antibody complexes strongly increased the faint intensity of unremodeled HA-PLAP, moderately increased that of remodeled HA-PLAP ( Fig. 6A ), and eventually equalized the intensities of remodeled and unremodeled GPI-APs, such as HA-PLAP, Thy-1, and EGFP-Flag-CD59 ( Fig. 6A-D ).

Ineffi cient oligomerization, not altered conformation of unremodeled GPI-APs, was most likely responsible for weak signals in Western blotting
It has been reported that delipidation of GPI from GPI-APs by phospholipases induces conformational changes of proteins linked to GPI and results in the drastically decreased reactivities to antibodies ( 11,12 ). By analogy to these GPI-cleaved proteins, low immunoreactivities of unremodeled GPI-APs might be caused by altered conformations of the protein moieties. In Western blot analyses, GPI-APs were denatured by SDS plus boiling and then renatured to some extent during transfer onto membranes. It could be interpreted that the remodeled and unremodeled GPIs affected renaturation of proteins to different extents by interacting differently with the protein moiety during the denaturation/renaturation cycle.
Therefore, the question of whether unremodeled PLAP was more sensitive to denaturation by SDS and exhibited more ineffi cient renaturation after removal of SDS than did remodeled PLAP was examined by monitoring PLAP enzymatic activities. Contrary to expectations, PLAP activities from samples treated with various SDS concentrations and denaturation/renaturation did not show any significant differences between remodeled and unremodeled PLAPs ( Fig. 5A ). These results indicated that differences in conformational changes were small under these conditions, consistent with the fact that both PLAPs were similarly recognized by antibodies in fl ow cytometry ( Figs. 2A  and 4B ). Similar results were obtained when urea was used instead of SDS (data not shown).
Next, the effect of boiling was examined using samples prepared from whole-cell lysates, under boiling or nonboiling conditions in the presence of SDS and absence of a reducing reagent, and subjected to Western blotting. It was confi rmed by detection with anti-HA antibody that similar amounts of remodeled and unremodeled HA-PLAPs were loaded to PVDF membrane, with boiled samples reproducing the bands of mass shown in Figs. 2A and 4C ( Fig. 5C , double asterisks), although nonboiling treatments produced proteins that moved more slowly and yielded broader bands than boiling, which were thought to be partially denatured ( Fig. 5B , single and double asterisks, respectively). Antibody against PLAP was then applied to a membrane prepared under the same condition as in Fig. 5B . Surprisingly, nonboiled samples exhibited much stronger intensities than did boiled samples ( Fig. 5C , triple and double asterisks, respectively), and the apparent molecular masses of the intense bands did not coincide with those detected by HA7 antibody ( Fig. 5B, C , single asterisk). Judging from these sizes, the intense bands corresponded to dimeric PLAP. As PLAP is known to be functional only as a homodimer ( 17 ), the enzymatic activities of HA-PLAP blotted on membranes were measured using CDP-star, a luminous substrate for PLAP. As predicted, the intense bands observed completely coincided with bands revealed by CDP-star ( Fig. 5C, D , triple asterisks). This functional dimeric PLAP was barely detectable with HA7 antibody ( Fig. 5B ), although the same antibody normally recognized HA-PLAP in fl ow cytometry ( Fig. 5E ). This clearly indicates that ineffi cient oligomerization of unremodeled GPI, with relatively low antibody affi nities but not altered GPI-AP conformations, was responsible for the remarkably weak signals observed in Western blot analyses.
Moreover, monovalent Fab fragments produced weak but similar intensities in remodeled and unremodeled Thy-1 ( Fig. 6C ). Thus, equalization of band intensities by premixed primary/secondary antibody complexes and Fab fragments

Fig. 5. Conformational change-independent reduction of immunoreactivities in unremodeled GPI-APs. A: Comparison of ALP activities in PGAP3
Ϫ / Ϫ MEF cells restored with PGAP3 or mock (gray and dark gray bars, respectively) after denaturation or renaturation (left and right panels, respectively). Whole-cell lysates were prepared using Triton lysis buffer (100 mM Tris-Cl, pH 9.5, 100 mM NaCl, 5 mM MgCl 2 , and 1% Triton X-100). For denaturation, SDS was added to lysate at indicated SDS percentage; for renaturation, whole-cell lysates were prepared with Triton lysis buffer containing 2.5% of SDS, then diluted to indicated SDS percentage with Triton lysis buffer without SDS.   Loading samples were prepared from whole-cell lysates under nonreducing, nonboiling conditions. ALP activity (left), immunoblotting with anti-PLAP antibody (middle), and immunoblotting of anti-HA antibody (right). Note that the intense bands in the left and middle panels represent dimeric functional HA-PLAPs, whereas the bands in the right panel represent partially denatured monomeric HA-PLAPs, as explained in the text and Fig. 5 legend. E and F: Immunoblotting of whole-cell lysates by sequential and single-step incubations as described in Fig. 6A (left and right, respectively) with anti-PLAP (E) and anti-Thy-1 (F) antibodies. E: Single asterisk, nonspecifi c bands (upper); double asterisks, specifi c bands (lower). F: Long exposure (top) and short exposure (bottom). Note that Thy-1 is an endogenous protein. G: EGFP-Flag-CD59 was immunoprecipitated with 5H8 anti-CD59 antibody plus protein G beads. Loading samples (8% of total lysates for input and 40% of the precipitates for output) were prepared under nonreducing, nonboiling conditions and applied to SDS-PAGE. Fluorescence was measured by fl uorescence laser scanner, and the ratios of band intensities in output to those in input were calculated and are shown at the bottom. H: Immunoreactivity of EGFP-Flag-CD59 after incubation with or without PI-PLC was analyzed as described in Fig. 5F . Strongly reduced immunoreactivity on blotted membranes of PI-PLC-treated GPI-APs was most likely due to inability to oligomerize and not due to the altered conformation of delipidated proteins Finally, the possible causes of severely reduced immunoreactivities of delipidated GPI-AP produced by PI-PLC were examined in terms of protein conformational changes caused by interaction with delipidated GPI, as reported previously ( 11 ), or decreased oligomerization similar to unremodeled GPI-APs, as proposed here. Western blot analyses of HA-PLAP after PI-PLC treatment showed drastically reduced immunoreactivities, even more markedly than unremodeled HA-PLAP ( Fig. 7A ). The molecular weight of delipidated HA-PLAP appeared to be similar to nontreated HA-PLAP, because the molecular mass of the lipid portion eliminated by PI-PLC is less than 1 kDa and much smaller than that of HA-PLAP. Next, the enzymatic activity of HA-PLAP and the immunoprecipitation effi ciencies were examined to determine whether GPI cleavage affected protein conformations. PI-PLC treatment did not signifi cantly affect enzymatic activity ( Fig. 7B ) or the effi ciency of immunoprecipitation by anti-PLAP antibody, among remodeled, unremodeled, and PI-PLC-treated HA-PLAPs, evaluated using anti-HA antibody ( Fig. 7C ). EGFP-Flag-CD59 also did not show signifi cant differences in immunoprecipitation effi ciency using anti-CD59 antibody among remodeled, unremodeled, and PI-PLC-treated ones, which was also evaluated in antibody-irrelevant fl uoroscanning images of EGFP-Flag-CD59 within the polyacrylamide gel ( Fig. 7G , upper panel). These results clearly indicate that immunoreactivities of delipidated and unremodeled GPI-APs were not reduced if they were not denatured. On the other hand, the sensitivity of delipidated HA-PLAP to SDS was greatly increased under nonboiling conditions, based on clear decreases in dimeric HA-PLAP that was detected by both enzymatic activities of PLAP on the blotting membrane ( Fig. 7D , left) and in its immunoreactivities to anti-PLAP antibody ( Fig. 7D , middle), compared with remodeled and unremodeled HA-PLAPs as explained in Fig. 5 . This was most likely due to complete monomerization that was accelerated by the lack of hydrophobic interactions among delipidated GPIs. Nor could premixed primary/secondary antibody complexes restore immunoreactivity in either HA-PLAP or Thy-1, being different from unremodeled forms that were shown in Fig. 6  ( Fig. 7E, F ). This inability of premixed antibody complexes to restore the immunoreactivities of delipidated GPI-APs was due to inability to form oligomers and not due to altered conformations of GPI-cleaved proteins during a denaturation/renaturation cycle. The immunoreactivities of not only anti-CD59 but also anti-EGFP and anti-Flag antibodies were also strongly reduced in delipidated EGFP-Flag-CD59, an effect that could not be explained by conformational changes caused by specifi c interactions of CD59 with delipidated GPI ( Fig. 7H ).

Fatty acid remodeling affects cellular dynamics of GPI-APs through the oligomerization
We previously reported that fatty acid remodeling in CHO cells is critical for raft association of GPI-APs ( 6 ), based on the enrichment in a detergent-resistant membrane (DRM) fraction (cold 1% TritonX-100) representing lipid rafts. Because we used a normal sequential probing procedure in Western blotting, which resulted in the decreased intensities of GPI-AP bands selectively in DRM of PGAP3-defi cient cells, we reevaluated it using anti-HA antibody or premixed primary/secondary antibody complexes. HA-PLAP was fractionated by low-speed centrifugation following solubilization with a cold buffer containing 1% TritonX-100. Western blotting with anti-PLAP antibody showed very weak signal of unremodeled HA-PLAP in DRM fraction (6) . In contrast, blotting with anti-HA drastically strengthened the signal, resulting in intensities comparable with remodeled HA-PLAP ( Fig. 8A ). Not much but a signifi cant amount of unremodeled HA-PLAP was separated into a detergent-soluble fraction, which was confi rmed by fractionation with density-gradient ultracentrifugation ( Fig. 8B ). Thy-1 showed more prominent change in distribution between remodeled and unremodeled Thy-1 ( Fig. 8C ). More than half of unremodeled Thy-1 was solubilized, whereas majority of remodeled Thy-1 was in DRM, which clearly indicated that fatty acid remodeling is critical for enrichment of GPI-APs in rafts. The different ratios of DRM to the detergent-soluble fraction between HA-PLAP and Thy-1 were most likely due to the tendency of the protein moiety to form dimer/oligomer, as PLAP is known to form very stable dimer but Thy-1 is not ( 17 ).
DRM assay is thought to be somewhat artifi cial, although many researchers use this assay to analyze the affi nity of proteins to lipid rafts. To study the signifi cance of GPI-anchor in terms of oligomerization of GPI-APs under physiological conditions, we used a cross-linker reagent that directly proves physical proximity. Crosslinking of surface proteins in living cells by membraneimpermeable cross-linker 3,3 ′ -Dithiobis[sulfosuccinimidyl propionate] (DTSSP), whose spacer length is 12 Å, revealed that signifi cantly more amount of dimeric (and trimeric) Thy-1 is present in PGAP3-restored PGAP3 Ϫ / Ϫ MEF than in PGAP3 Ϫ / Ϫ MEF ( Fig. 8D, E ), indicating the nature of remodeled GPI to oligomerize is seen under physiological membrane environment.
To further ensure the biological signifi cance of oligomerization through the remodeling of GPI, we performed a cell-killing assay using aerolysin, a pore-forming bacterial toxin. It is known that monomeric proaerolysins bind to single GPI-APs and heptamerize to make lethal pores after the activation by protease-mediated cleavage ( 18 ). The binding effi ciency of the toxin to surface GPI-APs was similar between PGAP3-defi cient and PGAP3-suffi cient cells ( Fig. 8F ), whereas cell-killing ability was signifi cantly weaker in PGAP3-defi cient MEF ( Fig. 8G ), most likely due to the slower oligomerization. These results further support our interpretation that discrepancy between different immunoreactivities in Western blotting and fl ow cytometry refl ects different degrees of oligomerization of GPI-APs in the plasma membrane due to fatty acid compositions.

DISCUSSION
What are the merits of employing GPI for anchoring proteins to the cell membrane? Cell surfaces of protozoa are densely covered with GPI-APs, but cell surfaces of mammals are not so densely covered . GPI biosynthesis is essential for the survival of yeast and protozoa but not essential for mammalian cells, albeit it is essential for mammalian embryo development. Thus, GPI might have  different signifi cance in different organisms, but as GPI has been conserved through eukaryotic evolution, some important aspects should be retained. One such conserved fundamental GPI property is based upon the structural characteristic that the GPI moieties of almost all surface GPI-APs contain two saturated long acyl chains, such as ceramide, or two saturated fatty acid chains, which endow GPIs with high affi nity for the liquid-ordered phase. In most organisms, the majority of PI, a source of GPI, has an sn -2 unsaturated fatty acid; therefore, it appears reasonable that lipid remodeling is evolutionarily conserved, although the mechanisms might differ among organisms. Fatty acid remodeling by the PGAP proteins in mammalian cells guarantees this property by replacement of unsaturated chains with saturated ones. Thus, defective PGAP3 activity leads to impaired enrichment of GPI-APs in the detergent-insoluble membrane fraction corresponding to lipid rafts ( 6 ) as well as to altered immune responses in PGAP3-knockout mice ( 14 ). Therefore, study of differences in behaviors of remodeled and unremodeled GPI-APs is very important for understanding the underlying basis of GPI. Kukulansky et al. employed antibodies that recognize various epitopes of Thy-1 to evaluate the reactivities against delipidated Thy-1. In their report, all monoclonal antibodies, 30-H12 and HO-13-4 recognizing the Thy-1.2 allelic determinant and 31-11, 42-21, and G7 recognizing nonpolymorphic determinant of Thy-1, showed reduced immunoreactivities on Western blotting, as well as did polyclonal antibodies against Thy1 ( 13 ). Therefore, they considered that the phenomenon could be explained by conformational change of delipidated Thy-1. Because the delipidation by PI-PLC released GPI-APs from cell surface, they could not investigate whether delipidated GPI-APs on the plasma membrane could be recognized normally by antibodies. Although we have not used various antibodies that recognize different epitopes of one GPI-AP, all antibodies against GPI-APs used here showed similar phenomenon on Western blotting when unremodeled GPI-APs were used as antigens. Nevertheless, these antibodies normally recognized unremodeled GPI-APs in fl ow cytometry, which eventually led to a hypothesis different from theirs, i.e., that reduced oligomerization could be responsible for the reduced immunoreactivities.
The major fi ndings in the present study are: i ) GPI-APs in PGAP3-defi cient cells, which have not undergone fatty acid remodeling but still carry two lipid chains, were found to demonstrate considerable reductions in immunoreactivities in Western blotting; ii ) reduced immunoreactivities of unremodeled GPI-APs and delipidated GPI-APs in Western blotting were most likely not due to altered protein conformations that have been proposed as a mechanism for the effects of delipidated GPI-APs; iii ) remarkable losses in oligomerization during denaturation/renaturation cycles were responsible for immunoreactivity reductions in unremodeled GPI-APs, implying that spontaneous oligomerization is a fundamental property of remodeled GPI; and most importantly, iv ) the nature of spontaneous oligomerization through the remodeled GPI, which was revealed by in vitro experiments, is conserved under physiological membrane environment. In addition to these fi nding, we reconfi rm that fatty acid remodeling is critical for the enrichment of GPI-APs in lipid rafts. Cross-linking of surface proteins gives more solid evidence to prove remodeled GPI-dependent oligomerization of GPI-APs under physiological membrane environment, because of the very short spacer length (DTSSP has a 12 Å spacer) than in the DRM assay, which refl ects accumulation of GPI-APs in liquid-ordered phase by the interaction with other specifi c lipids but not direct protein-protein binding. The cell-killing assay using aerolysin also indicated that remodeled GPI accelerates heptamerization of GPIbound aerolysin, although we cannot eliminate the possibility that remodeled GPI accelerates activation of aerolysin by increasing the chance to meet with the protease. Thus, these results support our interpretation that the discrepancy between different immunoreactivities in Western blotting and fl ow cytometry refl ects different degrees of oligomerization of GPI-APs in the plasma membrane due to fatty acid compositions.
These fi ndings revealed certain signifi cant aspects of GPI, in that remodeled GPI intrinsically accelerated GPI-AP oligomerization, which is considered critical for stabilizing and expanding lipid rafts, activation of signaling pathways, and traffi cking of GPI-APs to the apical side of polarized cells (8)(9)(10)19 ). Suzuki et al. used advanced single-molecule fl uorescent imaging to show that, in resting cells, virtually all GPI-APs are mobile and continually forming transient homodimers through both GPI-GPI and protein-protein interactions of GPI-APs and that the lack of interaction through GPI-GPI shortened the lifetime of homodimers by 2.5 times ( 20 ). These transient rafts containing homodimeric GPI-APs are most likely one of the basic units that form larger and more stable rafts capable of signaling ( 20 ). Sengupta et al. have also reported using a combination of pair-correlation and photoactivated localization microscopy to show that, in the steadystate, GPI-APs are organized into clusters with radii less than 60 nm and containing about two GPI-APs with relatively high local density, indicating that they may form transient dimers ( 21 ). These reports are consistent with the present fi ndings that remodeled GPI effi ciently formed homodimer/oligomers by GPI-GPI interactions, and these reports have clarifi ed this important nature of remodeled GPI. This is in addition to GPI's ability to form liquid-ordered phases by interactions with sphingolipids and cholesterol, as reported previously ( 6 ).
Most GPI-APs are considered to exist as monomers outside of lipid rafts and to move freely on the plasma membrane during staining with antibodies in fl ow cytometry, which allows the divalent conjugation of one antibody with two GPI-APs, even if GPI-APs exist as monomers at the beginning. This results in high avidity binding of antibody with GPI-APs, which is the reason why antibodies against GPI-APs in fl ow cytometry did not show signifi cant differences in immunoreactivities between remodeled and unremodeled GPI. The situation with Western blotting is different, because GPI-APs are immobilized on PVDF