New Globoseries Glycosphingolipids in Human Teratocarcinoma Reactive with the Monoclonal Antibody Directed to a Developmentally Regulated Antigen, Stage-specific Embryonic Antigen 3*

Glycolipids in a cultured human teratocarcinoma cell line (2102Ep) were investigated. The major glycolipids in these cells are globoseries glycolipids having the following structures: Synthesis of these structures by serial addition of gal- actose, fucose, and N-acetylneuraminic acid to globoside (Gb,) in this teratocarcinoma is obvious, although further elongation of Gb4 in human cells and tissues has not been previously found with the exception of the presence of a small quantity of Forssman glycolipid in some tissues in the human population (Fs’

The abbreviations used are: SSEA, stage-specific embryonic an-carried by lipids and/or by protein molecules. Precise understanding of the glycoconjugates at the surface of embryos and teratocarcinoma cells is of increasing interest as these molecules may be important to the developmental potential of the cells. This paper describes characterization of the glycolipids in a cultured human teratocarcinoma, 2102Ep, cell line. It has been previously reported that human teratocarcinoma cell lines express an embryonic antigen, SSEA-3, detected by a monoclonal antibody raised against 4-to 8-cell stage mouse embryos (10). The antigen is expressed in a stage-specific manner during early mouse embryogenesis and a change in the expression of SSEA-3 is also detected during the course of differentiation of human teratocarcinoma cells (11). The presence of an embryonic antigen common to mouse oocytes, mouse embryos, and human teratocarcinoma cell lines is of interest because of the possibility of conserved expression of such an antigenic determinant on functionally related cells from different species. The SSEA-3 antigenic determinant appears to be carbohydrate in nature and carried by both membrane glycolipids and glycoproteins (10). We have now purified glycolipids from human teratocarcinoma cells and studied their structure and reactivity with this monoclonal antibody.

Glycolipids 8935
Extraction and Purification of Glycolipids-Packed cells were homogenized and extracted with 20 volumes of chloroform/methanol (2:1, 1:1, and 1:2, v/v). After Folch's partition (14), the lower layer glycolipids were freed from phospholipid contamination by acetylation (15). Upper layer and lower layer glycolipids were pooled and subjected to DEAE-Sephadex column chromatography to separate neutral and acidic glycolipids (16). Further purification of the glycolipids was performed by high performance liquid chromatography with a Varian HPLC system (model 5000, Varian Associates Inc., Walnut Creek, CA), using a column (1 X 50 cm) of Iatrobeads (IRS 8010, 10-p diameter, Iatron, Tokyo) and eluted with a gradient of isopropyl alcohol/hexane/water (17). The solvent composition for gradient elution is shown in the legend to Fig. 1.
Purified fig n-galactosidase was a gift from Drs. Y.-T. Li and S.-C. Li, Tulane University, New Orleans, LA. &Galactosidases from jack bean, Escherichia coli and Aspergillus niger, and jack bean 8-Nacetylhexosaminidase were obtained from Sigma; Charonia lampas 8galactosidase was purchased from Miles Biochemicals (Elkhart, IN). Defucosylation of glycolipids was done with 0.1 N trichloroacetic acid at 100 "C for 1 h, and desialylation was performed in 1% acetic acid a t 100 "C for 1 h.
Immunological Reactivities of Glycolipids-The reactivity of the glycolipid with antibodies was ascertained by three different methods. TLC immunostaining was performed as described previously (8, 21). Briefly, glycolipid was chromatographed on HPTLC plates (Si-HPF plates, J. T. Baker Chemical Co.) and reacted successively with 1:lOOO diluted monoclonal SSEA-3 antibody or Pk antibody (both rat IgM), 1:lOOO rabbit anti-rat IgM (p-chain specific), and '251-labeled protein A solution. After washing, TLC plates were subjected to autoradiography. Solid phase radioimmunoassay was performed with vinyl assay strips (Costar, Cambridge, MA). Glycolipids (-50 ng/well) were dissolved in ethanol with phosphatidylcholine and cholesterol (250 ng and 125 ng/well), dried a t 37 'C to achieve adsorption to the bottom of each well. After treatment with PBS containing 5% bovine serum albumin, the well was reacted successively with 1:500 diluted monoclonal SSEA-3 antibody (IgM), 1:lOOO diluted rabbit anti-rat IgM, and "'I-protein A solution. After washing, the radioactivity of each well was measured by a y-scintillation counter. The cell-binding inhibition test was performed as follows: 5 X lo4 2102Ep cells were incubated for 1 h a t 37 "C with a 1:lOOO dilution of SSEA-3 antibody (50 plltube) in the presence or absence of liposome suspensions (50 pl/tube) containing various amounts of glycolipids. Liposomes were made from 50 pg of glycolipid, 200 pg of phosphatidylcholine, 150 pg of cholesterol, and 7.5 pg of dicetylphosphate, suspended by sonication in 0.25 ml of PBS. Serial dilution was made by diluting this liposome suspension with PBS. After incubation a t room temperature for 1 h, liposomes and unreacted antibody were removed by washing the cells three times with PBS. The cells were then incubated with 50 plltube of a 1:lOOO dilution of second antibody (rabbit anti-rat IgM) at room temperature for 1 h, washed, and reacted with 9labeled protein A solution. Radioactivity absorbed to the cells was determined with a y-scintillation counter.
Chemical Characterization of GL-1 to 4-GL-1 and 2 showed the same TLC mobilities as HexCer and LacCer standards prepared from human erythrocytes. Based on this finding and the following results on GL-3 to 7, the structure of these two glycolipids must be Glc@l+lCer and Galfi14Glcfil"rlCer.
The anomeric structure of the terminal galactose residue in GL-5a and c was difficult to determine. Neither glycolipid was cleaved with jack bean @-galactosidase, although a nLc4 standard bearing the Gal@l4GlcNAc-terminus and a Gg, standard having a Gal@1-3GalNAc-terminus were completely cleaved under the same conditions. Purified n-galactosidase from fig and @-galactosidase from E. coli or from A.
niger also failed to cleave GL-5a and c. The only enzyme that cleaved GL-5a and c was &galactosidase from C. lampas, as shown in Fig. 6. The complete hydrolysis of 20 pg of GL-5a or c was obtained with 20 pl of enzyme (1 unit/ml) after overnight incubation, yielding glycolipids having the same TLC mobility as the upper and lower spots of GL-4. Since this enzyme has so far not been utilized to determine the anomeric structure of glycolipid carbohydrates, a strict control study was performed. Under the same conditions, the enzyme preparation degraded nLc4 into LC?, Gal@l-3nLc4 into Lcn, and Gg, to Ggn. Gbn prepared from human erythrocytes having a G a l n l 4 G a l terminus, and iCb:, having a Galnl"r3Gal terminus, prepared by enzymatic degradation of iGb, (cytolipin-R) obtained from rat kidney, were not cleaved by the enzyme preparation. These findings confirm the specificity of the enzyme toward the @-galactose terminus. Based on these findings, we conclude that the terminal galactose in GL-5a and c is the @-anomer. The finding that GL-Sa and c react with PNA lectin can be taken as additional evidence for the @-Gal terminus (data not shown). Thus, the entire structure of GL-5a and c must be Gal@1+3GalNAc@1+3Galcul+ 4Gal@1+iGlcBl+1Cer.
This structure was further confirmed by the results of 'H-NMR study for the anomeric reasonances of underivatized GL-5a ( Fig. 7a and Table I) G a l a 1 4  G a l B 1 4 Glc@l-lCer Glycolipid 6 ( J ) 6 ( J ) 8 ( J ) 6 ( J ) 6 ( J ) 6 ( J ) tose H-5 resonance, which is poorly resolved, but apparently very close to its position in Gb4 (-4.16 uersus 4.14 ppm) (25). The resonance at 4.61 ppm of GL-5a was assigned to the penultimate GalNAcPl-3 H-1, shifted downfield (A6 = 0.09 ppm) from its position in Gb4, as would be expected for glycosylation of this residue (25). One additional signal in the region for H-1 of &galactose or glucose units, at 4.20 ppm, is assigned to the terminal GalPl-3 of GL-5. Its coupling constant (J1,* = 7.3 Hz) confirms its assignment to a P anomer.
In view of the fact that this resonance appears to be shifted upon further elongations of the saccharide chain (see later discussion), we consider this assignment to be of fairly high reliability. Chemical Characterization of GL-6-GL-6 had almost the same TLC mobility as neolactonorhexaosylceramide, suggesting it to be a ceramide hexasaccharide. Direct probe mass spectrometry of GL-6 showed the characteristic ions for deoxyHex- Methylation analysis showed the presence of 2,3,4-0-Me3-Fuc (terminal Fuc), 4,6-O-Me2-GalNAc (-3GalNAcl4) and 2,3,6-0-Mes-Glc (-4Glcl+) (Fig. 3d). As to the derivatives of the galactose residues, GL-6 contained 2,3,6-0-Me3-Gal and twice the amount of 2,4,6-0-Me3-Gal and/or 3,4,6-0-Me3-Gal as shown in Fig. 3d-1. Since the peaks of 2,4,6-0-Me3-Gal and 3,4,6-0-Me3-Gal overlapped on the DB-5 column, an OV-225 column was utilized for the separation of those compounds. With this column, 2,4,6-0-Me3-Gal elutes significantly earlier than with DB-5, and 3,4,6-0-Me3-Gal overlaps with 2,3,6-0-Me3-Gal, confirming the presence of 2,4,6-0-Me:3-Gal, as shown in Fig. 3d-2. From these findings, it is obvious that GL-6 contains equimolar amounts of 2,4,6-0-Me:]-Gal, 2,3,6-0-Me3-Gal, and 3,4,6-0-Me3-Gal (-+3Gall+, +4Gall+, and +2Gall+), respectively. Thus, GL-6 yielded the same partially methylated alditol acetates as those from GL-5a or 5c; the only differences are the presence of terminal fucose and 2-substituted galactose residues, with the concomitant disappearance of the terminal galactose residue which was present in GL-5a and 5c. After defucosylation, GL-6 yielded glycolipids having the same mobility as GL-Sa and 5c. The conversion of GL-6 to GL-5 by defucosylation in 0.1 N trichloroacetic acid was quantitative. Based on these findings, the structure of GL-6 is proposed to be a fucosylated GL-5, Fucal~2Gal~l-+3GalNAc~1+3Galal+4Gal~1+ 4Glc/31+lCer. This structure was further confirmed with the 'H-NMR study. As shown in Fig. 7b, the spectrum of GL-6 contained six anomeric resonances, two of which coincide at 4.46 ppm. Three resonances from internal aGal, @Gal, and PGlc were unchanged from their positions in the spectrum of GL-5 and were assigned similarly. The additional signal at 4.95 pprn had the extreme downfield position and vicinal coupling constant (J1,* = 2.4 Hz), which is compatible with a terminal Fucal-t2 residue. The proton signals at 4.46 ppm, therefore, correspond to the GalNAcP1+3 H-1 which has shifted upfield (A6 = -0.15 ppm) and the Galpl-3 H-1 which has shifted downfield (A6 = 0.26 ppm) from their positions in the GL-5a spectrum ( Table I). The exact reason for the large changes in chemical shifts for these residues is not clear at present, but an analogous effect of terminal fucosylation on the anomeric resonances of internal sugar residues is reported with a type 1 chain H-active glycolipid and has been ascribed to the effect of steric crowding upon fucosylation at the terminus (26). The type 1 chain H-terminal trisaccharide differs from that of GL-6 only at the C-4 configuration of the internal HexNAc. This difference should not alter the gross relative stereochemistry of the substitutents. The analogy is supported by the presence of a quartet at 4.07 ppm, the position assigned for H-5 of Fucal-2 of the type 1 chain H-terminal. For the type 2 chain H-terminal, this resonance was found at 4.00 ppm (26). The other signal in this region (Fig. 7b), a triplet centered at 4.10 ppm, is most probably the H-5 of a-galactose resonance, shifted upfield (A6 = -0.06) from its position in GL-5. This shift can be taken as evidence for a very long range effect of fucosylation on the steric alignment of the glycosyl chain.
The structure was further confirmed by the results of 'H-NMR study, which showed the presence of similar anomeric  proton resonances as found in GL-5 ( Fig. 7c and Table I).
The Gal@1+3 H-1 moved downfield by 0.04 ppm, as would be expected upon sialylation of this residue; the GalNAcB1-3 H-1 moved upfield -0.04 ppm. The characteristic doublet of doublets for the H-3, of sialic acid was found at 2.77 ppm (not shown), which is within the region between 2.77-2.75 ppm. This signal is found for all terminal a2-3 sialylated glycolipids we have tested.' The position of this resonance also has been used extensively in studies of oligosaccharides and glycopeptides for determination of sialic acid linkage with a high degree of reliability (see for example, Vliegenthart et al. (41) and references cited therein). The triplet a t 4.13 ppm is again assigned to the H-5 of a-galactose which is shifted upfield (A6 = -0.03) less than for GL-6.
Immunological Reactivity of Teratocarcinoma Glycolipids to SSEA-3Antibody-As shown in Table 11, GL-4 to 7 all reacted with the antibody to SSEA-3 in the solid state radioimmunoassay; the Forssman antigen was also weakly reactive. GL-5 showed the highest reactivity. Significant reactivity of GL-4 and GL-5 with the antibody was detected by TLC immunostaining using a 1:250 dilution of the antibody (Fig. 4 4 . However, when a 1:lOOO dilution was used, only GL-5 was significantly reactive (data not shown). Results of binding inhibition test with 2102Ep cells also showed that GL-5 had a higher affinity for the antibody than GL-4 (Fig. 8). A weak cross-reaction was observed with iGb4 (cytolipin-R) purified from rat kidney, the isomer of Gb,.
A 50% inhibition of binding was obtained with 180 ng of G L -~c , 375 ng of GL-4, and 8.6 pg of cytolipin R per tube, respectively. S. B. Levery, R. Kannagi, and S. Hakomori, unpublished data.
The antibody does not seem to react with the terminal structure of GL-5, which is Galpl+3GalNAc@l+R, since Gg4 having the same terminus did not cross-react with the antibody. In addition, GL-4, GL-6, GL-7, and Forssman antigen, which have entirely different terminal structures than that of GL-5, clearly cross-reacted with the antibody. The antibody seems to recognize the internal structures of these glycolipids, most probably the R+3GalNAcpl+3Gala14Gal~l+R', the common internal sequence of these glycolipids. That GL-5 exhibited higher reactivity than the other glycolipids indicates that a favorable comformation of the internal determinant, R+3GalNAc~l+3Galal+4Gal/3l+R', may be obtained by the Galpl-3 substitution at the GalNAc residue in the determinant. The finding that the antibody did not react with Gg3 which has the GalNAcpl4Galp-R terminal or IV3BGalNAcnLc4 (X, glycolipid, Ref. 22) which has a GalNAcp1+3Galp+R terminal indicates the importance of the 1-3 linkage between IV-pGalNAc and 111-aGal and the a-anomeric structure of the 111-Gal. The 1 4 linkage between 111-aGal and 11-@Gal also seems important, since the reactivity of iGb, was significantly weaker than that of Gb,.

DISCUSSION
This study was initiated to elucidate the structure of the glycolipid antigen reactive with the monoclonal antibody directed to SSEA-3 in human teratocarcinoma. Glycolipids in the teratocarcinoma 2102Ep cells were thoroughly analyzed by direct probe mass spectrometry, methylation analysis, enzymatic digestion, and nuclear magnetic resonance spectrometry after extensive purification by HPLC. Almost all glycolipids which were visible on TLC by orcinol reaction were characterized, with the exception of one minor glycolipid comigrating with GL-5a and 5c. The proposed carbohydrate structures of these glycolipids are summarized in Table 111. The human teratocarcinoma cell showed a characteristic glycolipid composition. All of the glycolipids characterized belonged exclusively to the globoseries glycolipids; no appreciable amounts of ganglio-or lactoseries glycolipids were detected. The synthetic pathway of these glycolipids in this cell line appears obvious from the carbohydrate structure of these glycolipids, the sequential conversion of each precursor glycolipid to a higher glycolipid by the stepwise addition of one terminal sugar residue.
The major terminal product of the synthetic pathway of globoseries glycolipids in human tissue was thought to be Gb, (globoside). The presence of these new structures revises the concept of the globoseries glycolipids in humans and raises the possibility that "extended globoseries" glycolipids, such as GL-5, 6, and 7, could be expressed in undifferentiated human tissues or embryos.
The presence of a large quantity of "extended globoseries" glycolipids detected in this cell line, including the novel structures GL-5, 6, and 7, may be unique for human teratocarcinoma and embryo; their chemical concentration in adult human cells and tissues must be very low or undetectableP Previously we described a ganglioside (G5) (39) which has very similar properties to teratocarcinoma GL-7 presented in this study. The TLC mobility of erythrocyte G5 was the same as a standard IV3aNeuAcGg4 and desialylated G5 had a TLC mobility identical with Gg,, similar to the teratocarcinoma GL-7 and GL-5 described in this study. A t that time, erythrocyte G5 was tentatively identified as IV3aNeuAcGg4, since the desialylated G5 reacted with a conventional anti-Gg4 (asialo GM1) antibody. The only difference between erythrocyte G5 and IV3aNeuAcGg4 (GMlb) was that the desialylated G5 (supposed to be Gg, at that time) was not cleaved with any exoglycosidases tested, including the p-galactosidase from jack bean, which readily degraded a standard Gg4 prepared from bovine brain under the same condition (see footnote of Ref.   (27) and "para-Forssman antigen" as a very minor component of human erythrocytes (28). GL-5 carries the terminal sugar sequence, GalP1-3GalNAcP-, which is identical with the terminal sequence of Gg, (asialo GM,). Some of the asialo GM1-reactive antibodies may cross-react with GL-5. Because of this terminal sugar sequence, Gg, can react with PNA lectin and has been regarded as the glycolipid receptor for PNA lectin. GL-5 can be another PNA receptor glycolipid, which is carried by the globoseries core structure. GL-6 carries an H-active terminus. It is known that the H-active terminus in erythrocytes and/ or intestinal tissue is carried by lactoseries and/or neolactoseries core structures (29); the H-terminus carried by ganglioseries glycolipids has also been reported (30). GL-6 is the first example of a globoseries glycolipid which carries the Hterminal structure. The terminal structure of GL-7 is identical with that of IV3aNeuAcGg, (GMlb). It would be of interest to test if the glycosyltransferases involved in the synthesis of the terminal structures carried by the globoseries glycolipids are identical with those active in the synthesis of the same terminal structures which are ordinarily found in ganglio-or lactoseries glycolipids in other human cells and tissues. A glycolipid having the same sugar sequence as GL-5 has been suggested to be present in cultured green monkey cells (31). However, the anomeric structure and/or linkage of sugar residues have not been fully elucidated. The anomeric structure of the terminal Gal in pentaglycosylceramide isolated from green monkey kidney cells was tentatively assigned as P because the glycolipid did not have any blood group B or PI activity. The assignment of the anomeric structure by NMR was difficult because of a shortage of material (31). The chemical basis of the structure of a similar glycolipid to GL-7 detected in chick muscle has not been described so far (32,33).
Since the antibody defining SSEA-3 seems to react with the sequence R+3GalNAc/31+3Galal+R', the terminal structure of GL-4 (globoside) and the internal structure of GL-5,6, and 7, it is a useful reagent to detect the globoseries glycolipids. Most carbohydrate-reactive antibodies are directed to the terminal sugar structures; however, antibodies reacting to an internal sequence are known; a monoclonal IgM antibody reactive with both globoside and Forssman (34) and various types of Ii-reactive antibodies (29) are good examples. Even though the antibody is directed to the internal structure, its reactivity is indirectly affected by the terminal lated erythrocyte G5 toward various exoglycosidases is very similar to that of the teratocarcinoma GL-5 in this study. Based on these findings, we predict that the structure of erythrocyte G5 is most probably identical with that of teratocarcinoma GL-7 described in this study. structure, probably due to changes in the tertiary structure of the internal sugar chain, as suggested by the NMR study.
The presence of SSEA-3 antigens in human teratocarcinoma cells raises the possibility that the antigen is also present in human embryos and plays a role as a stage-specific antigen. The presence of P and Pk antigens on the mouse embryo has been detected using polyclonal antisera (9). It is noteworthy that the structure of SSEA-3 active human glycolipids described in this paper includes the P-blood group antigen and its further metabolites. Since all the globoseries glycolipids so far characterized play a role as alloantigens in the P-blood group system (35), it could be predicted that some of the new structures found in the human teratocarcinoma cells may display previously uncharacterized antigens in Pblood group system. It is well known that individuals of rare pp-phenotype have a high incidence of abortion, and it is suggested to be due to the reaction of anti-PPIPk antibody in the maternal serum with corresponding antigens in the fetus (36,37). Frequency of the abortions is particularly high at the early stages of pregnancy. It is possible that P-antigen and/ or other antigens in P-blood group system play a role as stagespecific developmental antigens not only in mouse but also in human embryogenesis, and the frequency of abortion depends upon the variable degree of surface expression of these antigens on the fetal cells and tissues during the course of embryogenesis.
Recently we have also elucidated the complete structures of the SSEA-1-containing glycolipids (8). This antigenic determinant, like SSEA-3, is found on the surface of murine embryo, but it is expressed at a later stage of preimplantation development (7, 10). The finding that both of these antigenic determinants are carbohydrates which can be borne by glycolipid molecules indicates the importance of changes in the cell surface glycolipids in the developing embryo.
The antigenic transformation from SSEA-3+/SSEA-1-to SSEA-3-/SSEA-l+ status has been also detected during the course of in vitro differentiation system of human teratocarcinoma cells (11). SSEA-1 antigens are carried by a set of lactoseries glycolipids (S), and SSEA-3 antigens are carried  FIG. 9. Synthetic pathways of SSEA-1 and SSEA-3 active glycosphingolipids. a, the synthetic pathway of globoseries glycolipids which leads to the synthesis of a set of SSEA-3 active glycolipids; b, the synthetic pathway of neolactoseries glycolipids, fucosylation at internal GalNAc residues which leads to the synthesis of a set of SSEA-1 active glycolipids (8). A switching or shift of glycolipid synthesis from pathway a to pathway b is suggested to occur in mouse early embryogenesis. by a set of globoseries glycolipids. Thus, in terms of glycolipid antigens, SSEA-1 and SSEA-3 antigens belong to entirely different species of glycolipids, and the synthetic pathways for the two antigens are also entirely different (Fig. 9). Therefore, the transition in expression of these antigens observed in mouse embryogenesis and differentiation of human teratocarcinoma cells is not due to the simple addition of one or a few sugar residues to pre-existing carbohydrate chains, but involves dynamic changes covering multiple synthetic pathways of cellular glycolipids, i.e. synthesis of lactoseries and globoseries glycolipids. Thus an extensive change in the synthesis of cell surface carbohydrates might occur between the 4-8 cell and morula stages. This type of drastic change in glycolipid synthesis involving more than one synthetic pathway is not necessarily unusual; a similar type of alteration of cellular antigenicity carried by lacto-and globoseries glycolipids was reported to occur during the course of differentiation of a mouse leukemia cell line (38).