Structure and function of glycosphingolipids and sphingolipids: Recollections and future trends

Abstract

Based on development of various methodologies for isolation and characterization of glycosphingolipids (GSLs), we have identified a number of GSLs with globo-series or lacto-series structure. Many of them are tumor-associated or developmentally-regulated antigens. The major question arose, what are their functions in cells and tissues? Various approaches to answer this question were undertaken. While the method is different for each approach, we have continuously studied GSL or glycosyl epitope interaction with functional membrane components, which include tetraspanins, growth factor receptors, integrins, and signal transducer molecules. Often, GSLs were found to interact with other carbohydrates within a specific membrane microdomain termed "glycosynapse", which mediates cell adhesion with concurrent signal transduction. Future trends in GSL and glycosyl epitope research are considered, including stem cell biology and epithelial-mesenchymal transition (EMT) process‥

I. Introduction

Professor Carl G. Gahmberg, the Editor of this specialized volume on membrane biology and glyco-sphingobiology, asked me to write a review article for this volume. I understand that the review is to be based on the topics which I presented at the beginning of the Glyco-sphingobiology symposium organized by Professors Roger A. Laine, Reiji Kannagi, and Yasuyuki Igarashi, at Tokushima, Japan, Feb 26 - Mar 01, 2007. The context of my talk was a retrospective of studies with many colleagues, mainly on glycosphingolipids (GSLs), during the past 50 years, with some comment on future trends in this area.

I started my research career (late 1950s) with characterization of blood group antigens with "lipoid nature" [1], which were then highly ambiguous, since it was difficult to purify them and characterize their properties. Struggling in purification and characterization of "lipoid antigens", it was fortunate that many ways were opened through development of new technologies, which are briefly mentioned in Sec. II-ii. "Lipoid antigens" are now identified as complex GSLs with many defined carbohydrate sequences [2]. A large number of GSLs with new structures, including developmentally-regulated and tumor-associated antigens, have been characterized (see Sec. II).

A series of studies initiated by Dr. Shoei Iseki (Gunma Univ., Gunma, Japan) [3], and Drs. Salvador Luria, Phil Robbins, and colleagues (MIT, Boston, MA, USA) [4, 5], described dramatic changes in bacterial phenotype (e.g., growth behavior, and colony shape) through structural changes of cell surface lipopolysaccharides by phage-induced lysogenic transformation. It was a great impact for me to learn of a realistic role of cell surface glycans in control of cellular phenotype.

These studies with bacteria provided a new concept that cell surface glycans, even in eukaryotic cells, may define various cellular phenotypes, including oncogenic transformation and development. The first clear evidence for dramatic changes of GSLs in hamster and mouse fibroblasts associated with oncogenic transformation by DNA virus was provided by us [6] and others [7]. These studies, together with various others (see Sec. III), based on dynamic glycosyl response upon cell contact, led to the concept of cell "sociology", as first termed by Herman Kalckar [8]. However, it took close to 40 years to substantiate the functional role of cell surface glycosylation, which causes contact inhibition of cell growth (see Sec. IV-i).

All our studies so far on functional roles of glycosyl epitopes in GSLs and glycoproteins are based on their cis or trans interaction with membrane proteins, to modulate cell adhesion and signal transduction, which are the central theme of this review.

II. Studies on basic structures of glycosphingolipids

i. Structures of GSLs known up to early 1960s

Isolation and characterization of molecular components present in cells and tissues were the major theme in classic biochemical science at the early stage of development. The same situation applied to GSLs. Only cerebroside (GalCer and GlcCer), sulfatide, LacCer, brain gangliosides, and GSLs with tetrasaccharide composition, later termed "globoside", were known by the early 1960s, through pioneering studies by G. Blix, E. Klenk, T. Yamakawa, and R. Kuhn. Readers will find these classic studies described in a few monographs [9–11]. Major events such as discovery of sialic acid, gangliosides, globoside, and their composition and structures, are summarized in Table 1.

Table 1.

Classic studies on glycosphingolipids up to the early 1960s.

Topic		Ref.
Polyhydroxy carboxylic acid isolated from submaxillary mucin, named "sialic acid" (gave characteristic color reaction)		[12, 13]
Presence of lipid containing "sialic acid"		[14, 15]
	(a) "Substance X", a lipid compound showing the same color reaction as sialic acid, accumulated in brain of patient with Tay-Sachs disease	[16]
	(b) Presence of a similar compound in normal brain	[17]
	(c) Isolation of a crystalline "sialic acid-like" compound from Substance X, termed "neuraminic acid"	[18, 19]
Composition and possible structure of globoside		[10]
Composition of purified lipid from brain and spleen, claimed to contain sphingosine, fatty acid, hexose, and neuraminic acid in ratio 1: 1: 3: 1, termed "ganglioside"		[20]
	(a) Galactosamine was identified in aqueous hydrolysate of brain ganglioside	[13]
	(b) Identified galactosamine in brain ganglioside	[21]
Ganglioside containing Glc, GalCer, and sialic acid found as major component in horse erythrocytes, termed "hematoside"		[22]
Complete structural characterization of brain gangliosides, based on gangliotetraosylceramide		[23]
Further confirmation of brain ganglioside structures with sialyl derivatives by methylation analysis		[24]

ii. Basic methods developed for determination of glycosyl structures in GSLs and other glycoconjugates

These methods, developed during the early 1960s to 1970s, as listed in Table 2, greatly facilitated our studies.

Table 2.

Methods for structural analysis of glycosyl residues

♦	Positions of sugar linkages and branching points
	i. Methylation with sodium hydride/ methyl iodide in DMSO [25]
	ii. Hydrolysis of glycosidic linkage of permethylated sugar
	a. For neutral sugars: hydrolytic yield of partially O-methylated hexitol acetate is high; no problem [26]. b. For aminosugars: hydrolysis of glycosidic linkage is hampered if NAc is not protected
	The problem is solved by application of acetolysis, which inhibits de-N-acetylation and promotes glycosidic hydrolysis to get quantitative yield of partially O-methylated hexosaminitol acetate [27–29].
♦	Anomeric linkage and sequence analysis
	Step-by-step application of α or β exoglycosidases with defined specificity gives information on anomeric linkages of each sugar, and their sequences. Examples of enzymes used, and elucidation of structures: see text.
♦	Release of oligosaccharide from GSL for characterization
	Release of oligosaccharide from GSL by oxidation of olefinic double bond of Sph, followed by β-elimination; see text.

Methylation analysis for determination of sugar linkages, furanoside vs. pyranoside distinction, and branching points, has a long history. But the methylation reaction per se is sluggish, taking a long time to complete, and methods for identification of partially O-methylated sugars were difficult. It was a coincidental occasion that an effective methylation method was developed in 1964 [25], and well-designed gas chromatography-mass spectrometry (GC-MS) equipment was commercially available at that time. Thus, identification of partially O-methylated sugars was performed with great success for neutral sugar samples [26]. However, yield of aminosugar derivative after permethylation was unsatisfactory, for the reason described in Table 2. This problem was solved by application of acetolysis, followed by hydrolysis and reduction, which greatly enhanced yield of partially O-methylated hexosaminitol [27–29].

Electroimpact MS of entire permethylated GSL was found to be useful for sequence analysis [30]. This was the first example of total MS of permethylated glycan. The approach is widely used nowadays.

Analysis of anomeric linkages and sequences. Well-qualified exoglycosidases such as jackbean β-N-acetylglucosaminidase [31], jackbean β-galactosidase [32], fig α-galactosidase [33, 34], α-N-acetylgalactosaminidase of pig liver [35], and many other enzymes are useful for sequential analysis of various types of GSLs and glycosyl residues. As an example, complete anomeric structure and sequence of Gb4, Gb3, and Forssman antigen have been identified by use of these enzymes [36, 37].

Release of oligosaccharide from GSL. This was accomplished by oxidation of olefinic double bond of Sph, followed by alkaline β-degradation. Originally, ozonolysis [38] and periodate/ osmium tetroxide [39] were used. Finally, a simpler procedure based on [39] with better yield was described [40]. These methods are useful for identification of oligosaccharides, for not only chemical but also immunological or cell biological purposes. A novel ceramidase capable of releasing oligosaccharide from ceramide is available, e.g., [41, 42]. These enzymes are highly useful to separate oligosaccharides from GSLs, if carbohydrate chain length is relatively short. There is a limitation if carbohydrate chain length is >6–8 monosaccharides.

iii. Characterization of GSLs with ganglio-, lacto-, and globo-series

Ganglio-series

Brain gangliosides are characterized by the presence of common core structure GalNAcβ4Gal and/or ganglio-N-tetraosyl core (Galβ3GalNAcβ4Galβ4Glc) [23], and one or two α2-3NeuAc linked to internal or terminal Gal, or both [24] (see Table 1). Ganglio-series gangliosides are based on such core structures, and are basically different from globo-series or lacto-series structures as described below. Our lab has not been involved in structural analysis of ganglio-series gangliosides.

Globo-series

Presence of ceramide tetrasaccharide as major component of human erythrocytes [43] was confirmed, and the compound was termed "globoside" since it was obtained as a globular, birefringent precipitate [44]. In contrast to brain gangliosides, carbohydrate composition and sequence of globoside (Gb4) were clearly distinct, and the sequence GalNAc→ Gal→ Gal→ Glc→ Cer was proposed [45]. Globotriaosylceramide (Gb3), accumulated in Fabry's Disease, was found to have the structure Galα1-4Galβ1-4Glcβ1-Cer, and this structure is the inner core of all globo-series GSLs [34, 36]. Sequence of sugars, and position of sugar linkages in globoside, were determined previously [45, 46]. Complete globo structure was identified as GalNAcβ1-3Galα1-4Galβ1-4Glcβ1-Cer, by a combination of enzymatic hydrolysis and NMR spectroscopy [36].

Heterogenetic Forssman antigen had been known as GSL [47], and was found to have similar carbohydrate composition as globoside, but with α-GalNAc in contrast to β-GalNAc [48]. In subsequent studies, structure of Forssman antigen was established as GalNAcα3GalNAcβ3Galα4Galβ4Glcβ1-Cer [37]. In some cases, the Forssman epitope (GalNAcα3GalNAcβ3Galα4→R) was found to be carried by other structures [49].

Rat renal cell carcinoma has a novel GSL similar to but claimed to antigenically different from globoside ("cytolipin R") [50]. This GSL was characterized as having position of internal α-Gal different from globoside, i.e., GalNAcβ1-3Galα1-3Galβ1-4Glcβ1-Cer, and was termed "isogloboside" [51, 52].

Extended globo-series antigens, i.e., Galβ1-3, Fucα1-2Galβ1-3, or NeuAcα2-3Galβ1-3 linked to terminal GalNAc of Gb4, were isolated and characterized from human embryonal carcinoma as stage-specific embryonic antigen-3 and/or -4 [53, 54]. These structures were highly expressed in early embryo or embryonal carcinoma of mouse, primates, and humans, and are assumed to be important functional molecules involved in defining stage of stem cell development [55]. Globo-series structures were all identified as listed in Table 3.

Table 3.

Globo-series GSL structures

Galα4Galβ4Glcβ1Cer	GB3[36]
Galα3Galβ4Glcβ1Cer	iso-Gb3 [52,56]
GalNAcβ3Galα4Galβ4Glcβ1Cer	Gb4 [36]
GalNAcβ3Galα3Galβ4Glcβ1Cer	iso-Gb4 [51,21]
GalNAcα3GalNAcβ3Galα4Galβ4Glcβ1Cer	Forssman [37]
Galβ3GalNAcβ3Galα4Galβ4Glcβ1Cer	Gb5 (SSEA-3a) [54]
Fucα2Galβ3GalNAcβ3Galα4Galβ4Glcβ1Cer	globo-H (SSEA-3b) [54]
SAα3Galβ3GalNAcβ3Galα4Galβ4Glcβ1Cer	monosialyl-Gb5 [53] (SSEA-4)
SAα3Galβ3GalNAcβ3Galα4Galβ4Glcβ1Cer	disialyl-Gb5 [57]
$\text{SA}\mathrm{\alpha }\stackrel{6}{2}$

Lacto-series

Gangliosides having glucosamine, and releasing by ozonolysis an oligosaccharide similar to lacto-N-tetraose, were found in bovine erythrocytes and spleen [58]. Ganglioside with NeuAcα2-3Gal terminus was found as major ganglioside in human erythrocytes, and a neutral GSL having the structure Galβ1-4GlcNAcβ1-3Galβ1-4Glcβ-Cer was found as cross-reactive with anti-pneumococcal type XIV antibody [59]. Le^a GSL antigen, Galβ1-3[Fucα1-4]GlcNAcβ1-3Galβ1-4Glcβ-Cer [60] and/or Le^x GSL antigen, Galβ1-4[Fucα1-3]GlcNAcβ1-3Galβ1-4Glcβ-Cer [28] were found to be highly accumulated in colorectal, gastric, and non-small cell lung cancer tissues. Some of these structures were characterized by novel ceramide having phytosphingosine and α-hydroxy fatty acid [28]. Later, blood group ABH antigens present in human erythrocytes were found to be carried by common structure Galβ1-4GlcNAcβ1-3Galβ1-4Glcβ-Cer [2]. These early studies clearly indicated the presence of GSLs with lacto-series type 1 and type 2 chain.

Since then, extensive further studies were directed to isolate a number of branched and unbranched lacto-series GSLs. In total, 15 novel types of lacto-series GSLs were isolated and characterized; some of them are listed in Table 4. Of particular interest is the presence of long, unbranched poly-LacNAc having terminal NeuAc linked to Gal, and internal GlcNAc with α1-3Fuc, termed "myeloglycan" [61, 62], and highly branched poly-LacNAc carrying multiple epitopes [63]. If the epitope is sialyl 2-3Gal at the I-terminus, as found in placenta, it reacts with mAb NUH2. If the epitope is Le^x, the branched glycan could be SSEA-1, presumably having multiple Le^x, although the exact structure of SSEA-1 remains to be elucidated.

Table 4.

Lacto-series GSL structures

$\underset{\text{LacNAc}}{\underbrace{\text{Gal}\mathrm{\beta }4\text{GlcNAc}}}\mathrm{\beta }3\text{Gal}\mathrm{\beta }4\text{GlcCer}$ [59]	$\begin{array}{c}\text{SA}\mathrm{\alpha }3\\ \text{SA}\mathrm{\alpha }6\end{array}\underset{\text{LacNAc}}{\underbrace{\text{Gal}\mathrm{\beta }4\text{GlcNAc}\mathrm{\beta }}}3\text{Gal}\mathrm{\beta }4\text{GlcCer}$ [59]
$\begin{array}{c}\text{Gal}\mathrm{\beta }4\text{GlcNAc}\mathrm{\beta }3\text{Gal}\mathrm{\beta }4\text{GlcCer}*\hfill \\ \underset{{\text{Le}}^{\text{x}}}{\underbrace{\text{Fuc}\stackrel{3}{\mathrm{\alpha }}}}\underset{\mathrm{\alpha }-\text{OH FA}}{*\text{phyto}-\text{Sph}}\hfill \end{array}$ [28]	$\begin{array}{c}\text{SA}\mathrm{\alpha }3\text{Gal}\mathrm{\beta }4\text{GlcNAc}\mathrm{\beta }3\text{Gal}\mathrm{\beta }4\text{GlcCer}\\ \text{Fuc}\stackrel{3}{\mathrm{\alpha }}\text{SA}-{\text{Le}}^{\text{x}}\end{array}$ [64]
$\begin{array}{c}\text{Gal}\mathrm{\beta }3\text{GlcNAc}\mathrm{\beta }3\text{Gal}\mathrm{\beta }4{\text{GlcCer}}^{*}\\ \underset{{\text{Le}}^{\text{a}}}{\underbrace{\text{Fuc}\stackrel{4}{\mathrm{\alpha }}}}\end{array}$ [60]	$\begin{array}{c}\text{SA}\mathrm{\alpha }3\text{Gal}\mathrm{\beta }3\text{GlcNAc}\mathrm{\beta }3\text{Gal}\mathrm{\beta }4\text{GlcCer}\\ \text{Fuc}\stackrel{4}{\mathrm{\alpha }}\text{SA}-{\text{Le}}^{\text{a}}\end{array}$ [65]
Galα3-LacNAc-LacCer [29]
Lex-Lex-LacCer [66]	SA-Lex-Lex-LacCer [67]
Lex-Lex-Lex-LacCer [66]
Lea-Lea-LacCer [68,69]
GSL with unbranched poly-LacNAc	[61,62]
GSL with branched poly-LacNAc	[63]

IV. Characterization of GSL changes associated with oncogenic transformation, or with developmental processes

i. Blocked or incomplete synthesis, with or without accumulation of precursor

Some GSLs are highly expressed in normal cells which are reduced or deleted in transformed cells. Typical examples: (a) GM3 highly expressed in BHK cells or chicken embryonic fibroblasts was reduced or deleted upon transformation by polyoma virus, whereby level of LacCer increased several-fold [6]. A more recent observation along this line is that GM3 highly expressed in mouse and chicken fibroblasts was reduced or deleted upon Jun-induced oncogenic transformation. In this case, transformed phenotype was reverted to that of normal cells by enhanced GM3 synthesis through GM3 synthase gene transfection [70]. (b) Higher gangliosides GD1a, GD1b, and GT1b were deleted, and precursor GM3 was enhanced, by SV40 transformation of 3T3 cells [7]. These changes were interpreted to mean that ganglioside synthase gene is blocked by transformation mechanism ("incomplete synthesis" model).

ii. Specific GSLs or glycosyl epitopes expressed in transformed cells through neosynthesis

Gg3 expressed in normal mouse 3T3 fibroblast is minimal. The level was greatly enhanced by transformation by Kirsten strain murine sarcoma virus (Kimsv). Kimsv tumors grown in Balb/c mice expressed high-level Gg3, but Gg3 expression in various cells and tissues in normal Balb/c mice was highly limited, except for a few sites in spleen and gastrointestinal epithelia [71]. Similarly, Gg3 is highly expressed in murine leukemia L5178Y grown in DBA/2 mice, and treatment of L5178Y leukemic mice with anti-Gg3 IgG3 mAb completely suppressed leukemia growth [72].

The concept of GSLs as tumor-associated antigens, expressed in experimental tumors as above, was further extended to various human tumors, when mAb approach was applied for characterization of human tumor-associated antigens (e.g., [73]). Gb3, a common core structure of globo-series GSLs, was highly expressed in Burkitt lymphoma, defined by IgM mAb 38-12 [74]. GD3 ganglioside is highly expressed in human melanoma, defined by mAb 4.2 [75] or R24 [76]. Extended Le^a (Le^a-on-Le^a) antigen [68] is highly expressed in gastric and colorectal cancer, defined by mAb ST-421 [69]. Disialyl type 1 chain (disialyl-Lc4) is highly expressed in colorectal cancer, defined by mAb FH9 [77]. These antigens are listed in Table 5. It is difficult, however, to assess whether appearance of these tumor-associated GSL antigens is due to precursor accumulation from incomplete synthesis, or to "neosynthesis" through activation of glycosyltransferase genes.

Table 5.

Human tumor-associated GSL antigens

sialyl-Lea	Galβ3GlcNβ3Galβ4GlcCer	[65]	CA19-9
	$\mathbf{\text{SA}}\stackrel{\mathbf{3}}{\mathrm{\alpha }}\mathbf{\text{Fuc}}\stackrel{\mathbf{4}}{\mathrm{\alpha }}$	Gl, pancrease
sial y l-Lex	Galβ4GlcNβ3Galβ4GlcCer	[64]	CSLX
	$\mathbf{\text{SA}}\stackrel{\mathbf{3}}{\mathrm{\alpha }}\mathbf{\text{Fuc}}\stackrel{\mathbf{3}}{\mathrm{\alpha }}$	Gl, breast, lung
SA-Lex-Lex	SAα3Galβ4GlcNβ3Galβ4GlcNβ3Galβ4GlcCer	[67]	FH6
	$\mathbf{\text{Fuc}}\stackrel{\mathbf{3}}{\mathrm{\alpha }}\mathbf{\text{Fuc}}\stackrel{\mathbf{3}}{\mathrm{\alpha }}$	Gl, colorectal
dimeric-Lea (Lea-Lea)	Galβ3GlcNβ3Galβ3GlcNβ3Galβ4GlcCer	[68,69]	ST-421
	$\mathbf{\text{Fuc}}\stackrel{\mathbf{4}}{\mathrm{\alpha }}\mathbf{\text{Fuc}}\stackrel{\mathbf{4}}{\mathrm{\alpha }}$	gastric, colorectal
Gb3	Galα4Galβ4GlcCer	[74,78]	38-13
		Burkitt lymphoma, ovary
globo-H	Fucα2Galβ3GalNβ3Galα4Galβ4GlcCer	[79,80]	MBr1
		breast, prostate
GD3	SAα8SAα3Galβ4GlcCer	[75,76]	R24
		melanoma	4.2
fucosyl-GM1	Galβ3GalNβ4Galβ4GlcCer	discovery and chemical characterization: [81]
	$\mathbf{\text{Fuc}}\stackrel{\mathbf{2}}{\mathrm{\alpha }}\mathbf{\text{SA}}\stackrel{\mathbf{3}}{\mathrm{\alpha }}$	mAb production: [82]	TKH5
		small cell lung cancer
GalNAcβ4 disialyl-Lc3	GalNβ4Galβ3GlcNβ3Galβ1→R	[83,84]	RM2
	$\mathbf{\text{SA}}\stackrel{\mathbf{3}}{\mathrm{\alpha }}\mathbf{\text{SA}}\stackrel{\mathbf{6}}{\mathrm{\alpha }}$	renal cell Ca, prostate Ca

iii. Developmentally-regulated expression of GSLs or glycosyl epitopes

A few types of antigens expressed highly at a defined stage of embryonic development were identified by specific mAbs, and were termed "stage-specific embryonic antigens" (SSEA) [85, 86]. These antigens were clearly identified as GSLs, or as glycosyl epitopes with carrier molecules whose properties are not fully characterized. Their structures and expression patterns at early stage of mouse embryogenesis are shown in Fig. 1.

Fig. 1. Stage-specific transition of glycosyl epitopes from globo- to lacto-, and to ganglio-series, during early mouse embryogenesis.

Le^x epitope carried by SSEA-1 is not expressed at 4-cell stage, but is maximally expressed at 1.5 – 2 days (8–32 cell; morula stage), and induces Le^x-mediated adhesion (compaction). Le^x disappears after compaction, but is expressed at inner cell mass of blastocyst (3.5 days). Extended globo-series SSEA-3 and -4 (yellow background) are expressed at 0.5 – 1 day, before SSEA-1 appears. Ganglio-series is not expressed until 7 days, when neural crest appears. Data for developmental pattern based on [85–87]. Structure for SSEA-1 [87, 88, 216]; structures for SSEA-3 and -4 [53, 54].

SSEA-1, defined by mAb directed to F9 mouse embryonal carcinoma, is not expressed in 2–4 cell stage embryo, but is maximally expressed at 16–32 cell cleavage stage (morula), which induces Le^x-dependent adhesion in the presence of Ca²⁺, i.e., "compaction", and declines in later stages [85, 87]. The SSEA-1 epitope was clearly identified as Le^x glycan [87, 88], and Le^x oligosaccharide, but not Le^a oligosaccharide, inhibited the compaction process [88, 89]. SSEA-3 (Gb5 and globo-H) [54] or SSEA-4 (sialyl-Gb5) [53], defined by respective mAbs which were originally raised against 4- to 8-cell stage mouse embryo, are highly expressed in human embryonal carcinoma and in human embryonal stem cells [86]. In mouse embryo, SSEA-3 and -4 are maximally expressed at 4-cell stage, and decline later [86]. The functional roles of these GSLs remain to be studied. Preliminary studies indicate that SSEA-3 (Gb5 and globo-H) is involved in adhesion and signal transduction of human embryonal carcinoma cells; it is therefore assumed to be involved in adhesion and induced differentiation of human ES cells [55].

IV. Functional role of GSLs and glycosyl epitopes in control of cell growth, adhesion/ motility, and signal transduction

i. Two approaches used for functional analysis of GSLs and glycosyl epitopes

Two approaches as below have been successfully utilized.

a. Cell surface labeling with galactose oxidase, followed by Na[³H]BH₄

depicts the pattern of surface exposed Gal or GalNAc, and their changes associated with oncogenic transformation or other cellular phenotype [90]. Application of this method to distinguish cell surface profile of hamster fibroblast NIL and its polyoma transformant (NILpy) revealed the presence of a dominant peak (peak a, i.e., "galactoprotein a") in NIL which was absent in NILpy [91]. In contrast, NILpy cells had a dominant peak (peak c, i.e., "galactoprotein b") [91] (Fig. 2A-i). Change of glycoproteins corresponding to galactoprotein a and b in cell extract was further confirmed by slab gel electrophoresis with direct fluorography [92] (Fig. 2A-ii). The same protein, with molecular mass ~220 kDa, was detected by cell surface labeling with ¹²⁵I-lactoperoxidase [93]. A similar component was also found as "fibroblast surface antigen" [94]. These ~220 kDa glycoproteins were later collectively termed "fibronectin". Subsequent extensive studies indicated that fibronectin has multiple functional domains with specific binding capability to heparin, collagen, cell surface integrin, fibrin, etc. [95] (Fig. 2A-iii). Galactoprotein b in NIL cells was identified based on its gene cloning as integrin α3 by Tsuji et al [96], who also found that the ligand for α3 is laminin-5 [97], originally termed "epiligrin" by Carter et al [98].

Fig. 2. Two approaches used for functional analysis of cell surface glycosyl epitopes and GSLs.

Panel A. Analysis of Gal oxidase surface-labeled component by SDS-PAGE column (A-i), or by slab gel electrophoresis with fluorography (A-ii). Samples from NIL cells: A-i, left; A-ii, lanes 1 and 2. Samples from NILpy cells: A-i, right; A-ii, lanes 3 and 4. Note that peak "a" (galactoprotein a) in NIL is lost in NILpy, while peak "c" (galactoprotein b) is greatly enhanced in NILpy. Lane 1, total protein of surface-labeled NIL cells. Lane 2, fluorography of lane 1. Lane 3, surface-labeled NILpy cells, extracted with empigin BB (detergent), purified by RCA (Ricinus communis lectin) and slab gel electrophoresis. Lane 4, fluorography of lane 3. Empigin BB extract contained mainly galactoprotein b (~125 kDa), which is present mainly in NILpy cells. A-iii: Functional domain structure of human plasma fibronectin [217]; for review see [218].

Panel B. Exogenous addition of Gb4 to NIL cells induced growth-inhibited and "contact-oriented" appearance (B-i-b), whereas cells without Gb4 addition grew randomly, with partial overlapping (B-i-a).

Synchronized NIL cells were subjected to: determination of cell number increase at G2 phase (B-ii-a), mitotic index (B-ii-b), and thymidine incorporation (B-ii-c). Note that Gb4-added cells did not show cell number increase, and had minimal mitosis and DNA synthesis.

Gal oxidase approach for cell surface labeling was also used for GSLs, and showed distinctive differences between NIL and NILpy cells. nLc4Cer was only present, and labeled strongly, in NILpy cells [99]. Gal oxidase labeled GSLs of NIL cells were greatly affected by Ricinus communis lectin treatment of glycoprotein, indicating GSL crypticity is affected by organizational framework of membrane [100]. In general, cryptic GSLs such as Gb3Cer or Gg3Cer are greatly influenced by glycoprotein organization in membrane, since both Gb3Cer and Gg3Cer were well labeled by Gal oxidase method. Organizational status of these GSLs has been studied, as related to physiological difference of cell phenotypes [101, 102].

b. Exogenous addition of GSLs, inducing phenotypic changes of cells

GSLs have the novel property to interact with cell surface membrane, and incorporate into membrane to affect cellular function and phenotype. The first such experiment was performed by Roger Laine using NIL cells with exogenous addition of globoside (Gb4), although the exact organizational status of GSLs in membrane was unknown at that time [103]. ~8×10⁻⁴ M Gb4 strongly inhibited NIL cell growth, but NILpy cell growth was less inhibited. Growth-inhibited NIL cells showed "contact-oriented" appearance, i.e., cells were aligned in a certain direction, in contrast to randomly-oriented non-growth-inhibited cells. Effect of Gb4 on synchronized cell cycle of NIL cells was determined for cell number, mitotic index, and [³H]thymidine incorporation. Globoside greatly reduced cell number at mitotic stage. Mitosis was not observed in Gb4-added cells. Therefore, peak of thymidine incorporation at N phase was abolished in Gb4-added cells (Fig. 2B).

This experimental design with exogenous addition of GSL to observe phenotypic changes has been often used subsequently in functional analysis of GSLs. Effects of GSLs are more clearly observable when cells are grown in serum-free condition, i.e., chemically-defined media (e.g., GM3 effect on fibroblast growth factor receptor in BHK cells [104]. Specific targets of GSL have been considered after the microdomain concept was introduced. E.g., GM3 inhibited neuroblastoma cell growth by enhancing translocation of Csk into microdomain GEM, where Csk inhibit cSrc and other Src family kinases (Prinetti, Iwabuchi, & Hakomori [105]. A number of studies along this line have been performed since then, e.g., effect of GSLs on contact inhibition of cell growth (see Sec. IV-ii); effect of gangliosides, particularly GM3, on growth factor (EGF, PDGF, FGF) receptors (see Sec. IV-iii); effect of TSP/ ganglioside complex on growth factor receptors (see Sec. IV-iv); etc.

Enhanced expression of specific GSLs can now be tested by transfection of specific glycosyltransferase genes. Effect of exogenous GSL can be confirmed by depletion of specific GSLs expressed in cells by RNAi approach (see Sec. IV-iv). initial experiments along this line were started by phenotypic changes of NIL cells by exogenous addition of Gb4.

ii. Contact inhibition of cell growth and motility is associated with cell surface glycosylation profile

Many studies indicate that oncogenic transformation is closely associated with incomplete synthesis of GM3, Gb4, or Forssman, in hamster fibroblasts BHK or NIL, or mouse or chicken fibroblasts (see Sec. III-i). This indicates that terminal sugar structure in glycosyl residue at the cell surface may function in maintenance of growth control in normal/ contact-inhibited cells. Loss of terminal sugar residue, associated with loss of contact inhibition in transformed cells, indicates a possibility that terminal sugar could be a susceptible site for cell recognition. This hypothesis has not been unambiguously clarified, but is strongly supported by the following observations:

Confluent/ contact-inhibited BHK or NIL cells display enhanced synthesis of specific glycosyltransferase such as α-Gal transferase for synthesis of Gb3/ Gb4 is therefore enhanced [106]. Similarly, synthesis of Gb4, GM3, and Forssman was enhanced in non-transformed BHK or NIL cells when growth of cells was arrested at contact-inhibited state. In contrast, transformed cells displayed greatly reduced level of contact-sensitive GSL species such as Gb4, GM3, and Forssman [107–109]. α-Gal transferase response (for Gb3 synthesis) upon increase of cell population density, reflecting degree of cell contact, in contact-inhibitable BHK and NIL cells, and loss of this response in polyoma virus-transformed cells, is shown in Fig. 3. Loss of Gb3, Gb4, and Forssman in transformed NIL cells was also detected by cell surface labeling with galactose oxidase / NaB³H₄.

Fig. 3. α-Gal transferase response (for Gb3 synthesis) upon cell contact in contact-inhibitable BHK and NIL cells, and loss of this response in transformed cells.

UDP-Gal: LacCer α-Gal transferase (Gb3 synthase) (○-○) increased significantly with increase of cell population density in both non-transformed BHK and NIL cells (left panels). This response was lost in polyoma virus-transformed BHK and NIL cells (right panels). UDP-Gal: GlcCer β-Gal transferase (LacCer synthase) (●-●) did not increase with increase of cell population density in either BHK or NIL cells, or their transformants. The figure is based on data summarized from [106] and [109].

The mechanism by which cell contact induces inhibition of cell growth and motility through change of glycosylation status remains largely unknown. There is a high possibility that GSLs are complexed with TSP at the cell surface, and such complex interacts with growth factor receptors or integrins to induce signal transduction to inhibit cell growth or motility (see Sec. IV-iii).

iii. Tyrosine kinases associated with growth factor receptor (GFR) are inhibited by surrounding gangliosides

Studies described above (Sec. IV-i), which were performed during the early 1970s, suggested that GSLs may interact with unidentified functional membrane components, which in turn may cause changes in cellular phenotype. However, at that time, no realistic information on such functional components was available. It took almost 10 years for development of the current concept of growth factor receptors with tyrosine kinases [110–112], or integrin receptors defining cell adhesion and motility [113, 114]. In addition, for understanding GSL effect on cell growth in culture, basic knowledge on types of growth factors required for culturing specific types of cells was needed. This knowledge was acquired through extensive studies by G. Sato and colleagues, to establish chemically-defined media [115].

a. Fibroblast growth factor receptor

Thus, the first experiment on effect of GSL on BHK cell growth in chemically-defined media was undertaken. BHK cells require fibroblast growth factor (FGF) but not epidermal growth factor (EGF) or platelet-derived growth factor (PDGF). Only cells pre-incubated with GM3 displayed complete refractoriness of growth to FGF stimulation, whereby ¹²⁵I-labeled FGF was accumulated at the cell surface due to failure of internalization of FGF. Labeled FGF added to control BHK cells without GM3 pre-treatment was well internalized. It was assumed that high GM3 level blocked function of FGFR [104]. However, at that time, we had no knowledge on tyrosine kinase associated with FGFR; studies along this line were not performed until 20 years later [116].

b. PDGF receptor

Growth of mouse 3T3 cells depends on PDGF and its receptor. Therefore, effects of gangliosides on PDGF-dependent growth of 3T3 cells, and on tyrosine kinase associated with PDGFR, were studied. GM1 showed stronger inhibition of PDGF-dependent growth and PDGFR tyrosine kinase than GM3 [117]. Similar studies by Yates and colleagues indicated that GD1a, GD1b, and GT1b had much stronger inhibitory effect than other gangliosides on PDGFR tyrosine kinase [118].

c. EGF receptor

This line of studies was further extended to effects of gangliosides on EGF-dependent A431 cell growth, and on tyrosine kinase associated with EGFR. GM3, but no other GSLs, strongly inhibited EGF-dependent cell growth, and EGFR tyrosine kinase [119]. Since EGFR is highly expressed in various epidermal cancers, and its tyrosine kinase activity is closely associated with cancer malignancy, a possibility was investigated whether any ganglioside could have better inhibitory effect than GM3 (see below).

d. Effect of lyso-GM3 and de-N-acetyl-GM3 on EGFR

Hanai et al. found that lyso-GM3 showed much stronger inhibitory effect than GM3 on EGFR tyrosine kinase in vivo as well as in membrane extract in vitro. Furthermore, lyso-GM3 was detected in normal A431 cells [120]. In contrast, exogenously added "de-N-acetyl-GM3" (GM3 having de-N-acetyl sialic acid) strongly promoted EGFR tyrosine kinase and promoted growth and motility of A431 cells [121].

Thus, effect of gangliosides on EGFR tyrosine kinase is more complex than originally considered. I.e., (i) tyrosine kinase is modulated by GM3 when EGFR is activated by EGF under normal conditions; (ii) trace quantity of lyso-GM3 present, which may result from GM3 by de-N-acylation, strongly inhibits receptor function; (iii) de-N-acetylation of GM3 in resting A431 cells may promote cell growth, possibly through a channel different from simple activation of EGFR (see Fig. 4). Exogenous lyso-GM3 is highly cytotoxic, whereas lyso-GM3 dimer is not cytotoxic, but inhibits EGFR tyrosine kinase as strongly as lyso-GM3. Therefore, synthetic lyso-GM3 dimer has been studied for inhibition of EGFR activity and A431 cell growth, for the purpose of developing pharmacologically effective inhibitors of epidermal tumor cell growth [122].

Fig. 4. Effects of GM3, lyso-GM3, and de-N-acetyl-GM3 on EGFR in A431 cells.

When epidermal growth factor (EGF) binds to its receptor (EGFR), cytoplasmic tyrosine kinase is activated [110, 112, 219], possibly through receptor-receptor interaction or some other conformational change of EGFR [220]. EGF-induced receptor tyrosine kinase activity was inhibited by ganglioside GM3 [119], presumably through surrounding EGFR in membrane microdomain. However, lyso-GM3, derived from GM3 by ceramidase, inhibited EGFR tyrosine kinase more strongly than GM3 [120]. De-N-acetyl-GM3, derived from GM3 by de-N-acetylase, strongly promoted receptor kinase [121]. The scheme shown is based on these results.

iv. GSL/ tetraspanin complex inhibits integrin-dependent cell motility, and HGF-induced cell motility and growth

Tetraspanins (TSPs) are ubiquitously present in various normal cell membranes [123], and are known to interact with integrins and modulate cell motility [124]. TSP CD9 was found as a motility-modulating protein [125], and TSP CD82 was identified as cell motility inhibitory gene product KAI-1 [126, 127]. Their expression is down-regulated in various types of human cancer, and survival of cancers patients is correlated with this reduced expression [127, 128]. We now find that these TSPs are complexed with gangliosides, and their function is based on such complex.

a. Loss of GM3/ CD9 complex in Jun-induced oncogenic transformation, and reversion to normal phenotype by restoration of the complex

GM3/ CD9 complex inhibits integrin α3, α5, β1 in mouse or chicken fibroblasts. GM3 synthase gene and GM3 levels were greatly reduced when these cells underwent Jun-induced oncogenic transformation. In this process, CD9 complex with GM3 was lost, leading to activation of integrin-dependent signaling to enhance cell motility, and consequent formation of large colonies in soft agar. When the transformed cells were transfected with GM3 synthase gene, and GM3 level was restored, oncogenic phenotype was reverted to normal cell phenotype, i.e., colony formation in soft agar was inhibited as in non-transformed cells [70]. The mechanism of such reversion remains to be studied. The crucial factor in this process is GM3 level, but not CD9 or integrin level. It is hypothesized that GM3/ CD9 complex regulates integrin activity and integrin-dependent signaling. This possibility is supported by studies of a bladder cancer model.

b. Level of GM3/ CD9 complex in bladder cancer cells defines degree of malignancy

Benign, non-invasive bladder cancer cell line KK47 has high level of GM3/ CD9 complex, and displays low cell motility. In contrast, highly invasive bladder cancer cell line YTS-1 has very low level of GM3, and therefore GM3/ CD9 complex is absent. YTS-1 displays high cell motility, particularly when adhered to laminin-5 (LN5)-coated plate, which leads to α3 integrin-induced signaling. Exogenous GM3 added to YTS-1 causes reversion of highly invasive to non-invasive phenotype similar to that of KK47 [129]. GM3 depletion from KK47 by incubation of cells with GlcCer synthase inhibitor P4 causes conversion to YTS-1-like phenotype with high invasiveness. Low GM3 level activates cSrc, whereas high GM3 level causes Csk translocation into "glycosynaptic microdomain", with subsequent inhibition of cSrc. Inhibition of cSrc by PP2 (the Src inhibitor) reduces motility. Thus, GM3 in microdomain plays a dual role in defining cell motility and invasiveness: (i) modulates interaction of α3 with GM3/ CD9 complex; (ii) activates or inhibits cSrc through translocation of Csk [130].

c. GM2 complexed with CD82 inhibits cell growth and motility

Growth and motility of many epithelial cells, and human cancers, are controlled by "scatter factor" or hepatocyte growth factor (HGF). This process involves HGF-induced activation of Met receptor kinase [131] and its functional interaction with α3β1 integrin receptor [132]. The major difference between normal bladder epithelial HCV29 cells and highly malignant YTS-1 cells is the presence vs. absence of CD82; there is no difference in GM2 or GM3 level. Specific interaction of CD82 with GM2 but not with GM3 was demonstrated based on adsorption of CD82 to GM2-coated polystyrene beads. GM2/ CD82 complex interacts specifically with Met, and inhibits Met kinase. In addition, the complex blocks LN5-induced α3β1 cross-talk with Met [132]. Other GSLs, including GM3, GM1, and Gb3, are incapable of forming a complex with CD82, and therefore have no effect on HGF-induced or α3β1-dependent signaling leading to cell growth/ motility [132].

V. Cell adhesion and signal transduction mediated by carbohydrate-to-carbohydrate interaction

Major processes of cell adhesion/ recognition have been considered to be mediated by: (i) Interaction between the same protein receptor, such as Ig superfamily protein receptors [133] and cadherin receptors [134], or between different protein receptors, such as integrins and extracellular matrix proteins [113, 114], or leukocyte integrins (CD11/ CD18) and intercellular adhesion molecules (ICAM, VCAM) [135]. These mechanisms are basically protein-to-protein interaction (PPI) [133, 136]. (ii) Glycosyl epitopes expressed in glycoprotein or GSL are recognized by carbohydrate-binding proteins through carbohydrate-to-protein interaction (CPI). The carbohydrate-binding proteins involved are galectins [137], selectins [138], and siglecs [139, 140].

On the other hand, we and others have observed a third mechanism mediating cell adhesion, carbohydrate-to-carbohydrate interaction, as described below.

i. The novelty of carbohydrate-to-carbohydrate interaction (CCI), and two lines of well-studied CCI-based processes

PPI and CPI processes as in (i) and (ii) above are well-established. In contrast, a novel process based on carbohydrate-to-carbohydrate interaction (CCI) has been found as a basis of cell adhesion and, more recently, specific signal transduction. Two lines of studies clearly established the occurrence of CCI mediating adhesion of mouse embryonal cells, by us [87, 141, 142], and CCI mediating species-specific autoaggregation of sponge cells, by Max Burger's group [143–145]. The former was based on Le^x-to-Le^x interaction; the latter was based on 3-sulf-GlcNAcβ3Fucα or Galβ4GlcNAcβ3Fucα as a unit oligosaccharide linked to proteoglycan. In both Le^x glycan interaction and sponge cell glycan aggregation, the adhesion process requires Ca²⁺, and is abolished in the presence of EDTA (Fig. 5).

Fig. 5. Two well-established carbohydrate-to-carbohydrate interactions that mediate homotypic cell adhesion.

Top: self-recognition of 3-O-sulfated GlcNAcβ3Fucα-O-Ser/Thr linked to proteoglycan, that mediates species-specific sponge cell adhesion and autoaggregation. Bottom: self-recognition of Galβ4[Fucα3]GlcNAcβ3Gal (Le^x epitope) carried by SSEA-1 glycan, that mediates homotypic adhesion of embryonal stem cells or embryonal carcinoma cells, to induce compaction or autoaggregation. In both cases, presence of Ca²⁺, which promotes interaction of glycan, is essential.

ii. Occurrence of various CCI based on GSL-to-GSL interaction

CCI may occur widely, from primitive marine animals to mammalian cells, although well-established cases are limited, and we assume there are a number of undiscovered cases of CCI. Using purified GSLs isolated from various mammalian tissues and cells, preliminary studies have been conducted regarding possible occurrence of homotypic or heterotypic CCI, besides Le^x-to-Le^x as above. Strong CCI was found between GM3-Gg3 [146, 147], and sulfatide-GalCer [148, 149]. Interactions between H-Le^y, H-H [150], Gb4-Gb5, Gb5-Gg3, and Gb5-nLc4 [55] have been suggested, although unambiguous chemical/ physical basis, and physiological significance of such interaction, remain to be studied. Interactions of various CCI based on GSL-to-GSL interaction are summarized in Fig. 6.

Fig. 6. GSL structures displaying specific CCI.

GSL interactions were determined by binding of liposomes containing specific GSL and labeled with ³H-cholesterol to specific GSL coated on polystyrene plate [141, 146, 147]. Strong, weak, or no interaction are indicated by different lines.

iii. GSL interactions with N-linked glycans

More recently, studies were extended to GSL interactions of N-linked glycans of glycoproteins. We found that GM3 bound strongly to a specific complex-type N-linked glycans with 5–6 GlcNAc termini (e.g., "Os. Fr.B"), to a lesser degree to glycans with 3–4 GlcNAc termini, and essentially did not bind to glycans with 2 GlcNAc termini [151]. GM3 interaction with EGFR, which has been known for years (see Sec. VII-ii), is now shown to be based on interaction of GM3 with N-linked glycans having multiple GlcNAc termini, since: (i) GM3 interacts with EGFR, as demonstrated by binding of EGFR to GM3-coated polystyrene beads, and this interaction is inhibited by Os Fr.B; (ii) Os Fr.B blocks the inhibitory effect of GM3 on activation of EGFR; (iii) swainsonine-treated A431 cells, which express more terminal GlcNAc, show enhanced inhibition of GM3 on EGFR activation [152].

iv. Interactions of specific N-linked glycans among themselves

Recently, complex-type N-linked glycans similar to Os Fr.B as above were found to interact among themselves (self-recognition). This interaction was revealed by strong binding of Os Fr.B conjugated to aminoceramide, to plates coated with the same Os Fr.B/ aminoceramide conjugate [153].

VI. Microdomains displaying GSL- or glycosylation-dependent adhesion and signal transduction

The essential function of GSLs in eukaryotic and animal cells is their strong ability to interact with specific functional proteins (see Sec. III and IV). In addition, GSL clusters in membrane are capable of binding among themselves, or to N-linked glycans (see Sec. V). Such property of GSLs is based on their clustering ability (cis-CCI), to form microdomains in which TSPs, integrins, growth factor receptors, and signal transducers (such as cSrc, Src family kinases, RhoA, Ras) are also present, and which are involved in GSL- or glycosylation-dependent adhesion, signal transduction, and changes in cellular phenotypes such as cell growth, motility, and differentiation. Such microdomains are distinct from "lipid raft" [154] in the following properties: (i) they are cholesterol-independent, and are often resistant to cholesterol-binding reagents such as nystatin, filipin, and β-cyclodextrin; (ii) they often depend on TSPs, and are solubilized together with TSPs in organic solvents, i.e., lipophilic property depends on TSP rather than cholesterol. These distinctive properties of such microdomains vs. "lipid raft" are summarized in Table 6. Such microdomains showing a close association with glycosylation function were termed "glycosynapse" [155, 156], in analogy to "immunological synapse" [157], the microdomain of immunocytes involved in adhesion of T-cells or B-cells to antigen-presenting cells.

Table 6.

Contrasting properties of "lipid raft" vs. "glycosynapse" or TCR synapse (immunological synapse)

Lipid raft		Glycosyn./ Immunosyn.
♦	1% Triton X-100 insoluble	♦	1% Triton X-100 soluble 0.5% Brij95 insoluble
♦	cholesterol-dependent (disrupted by chol.-binding reagent)	♦	cholesterol-independent (resistant to chol.-binding reagent)
♦	tetraspanin (TSP)-independent	♦	tetraspanin-dependent
♦	diameter 10 nm - <100 nm	♦	diameter >100 nm; usually 500–1000 nm
♦	highly mobile	♦	less mobile, or non-mobile
♦	not involved in cell adhesion	♦	involved in cell adhesion with concurrent signaling

VII. Studies on sphingolipids controlling cellular phenotype

Following studies by Bell, Hannun, and Merrill, indicating that sphingosine (Sph) inhibits protein kinase C (i.e., PKC α and β) [158, 159], we found that D-erythro-N,N-dimethyl-Sph (DMS) displayed much stronger inhibitory effect than Sph on conventional PKC [160]. Sph and DMS also displayed enzyme-specific and substrate-specific inhibition of c- and v-Src kinases determined with α- and β-casein, respectively. Sph, DMS, GM3, and psychosine also inhibited Src kinases [161].

The effect of Sph is unstable, since Sph is converted to ceramide (Cer) or Sph-1-phosphate (Sph-1-P). Therefore, the real Sph effect is often difficult to interpret in terms of biological significance. In contrast, the effects of DMS, and its analogue N,N,N-trimethyl-Sph (TMS), are assumed to be stable, since they are not converted to Cer or Sph-1-P. Sph, DMS, and TMS inhibit tumor cell metastasis [162–164], as well as inflammatory processes [165, 166], because of their strong inhibitory effect on activation of endothelial cells or platelets. Tumor cells or myelocytes secrete lymphokines (e.g., TNF-α), which stimulate endothelial cells or platelets to induce de novo synthesis of E-selectin, or to translocate P-selectin to the cell surface. This process leads to adhesion of tumor cells or myelocytes to endothelial cells or platelets through E- or P-selectin, to initiate tumor cell metastasis or inflammatory processes. Since PKC plays an essential role in signal transduction through lymphokines, to activate endothelial cells or platelets, DMS/ TMS inhibit tumor cell metastasis and inflammation. DMS/ TMS were more effective inhibitors of PKC than H-7 or calphostin, the well-established PKC inhibitors [165].

Effects of sphingolipids as modulators of signal transduction were greatly progressed by a series of studies focused on Cer or Sph-1-P, rather than Sph, during the early 1990s. It is generally accepted that Cer inhibits cell growth and induces apoptosis [167–172], whereas Sph-1-P, the first Sph catabolite, promotes cell growth (e.g., for 3T3 fibroblasts), induces differentiation (for endothelial cells), or inhibits motility (for B16 melanoma cells), depending on the type of Sph-1-P receptor [173–178].

Separately from this major trend of sphingolipid studies focused on Cer and Sph-1-P, we performed studies on functional roles of Sph and DMS in activation of various protein kinases. Protein kinases activated only by Sph or DMS, but not by Cer, Sph-1-P, or twelve other sphingolipids, phospholipids, and glycerolipids, were termed "Sph-dependent kinases" (SDKs) [179, 180]. Three groups of SDKs have been separated, based on difference of substrate and their specificity.

i. SDK-1. Sph or DMS activates caspase-3 through activation of a "cascade" of caspases. Caspase-3 hydrolyzes full-length PKC δ into regulatory domain (N-terminal half) and kinase domain (C-terminal half). The kinase domain of PKC δ per se is SDK-1 [181], which was known to phosphorylate 14-3-3 at interfacing Ser (Ser60 of β, Ser59 of η, and Ser58 of ζ isoform) to induce 14-3-3 dimer formation. 14-3-3 dimer inhibits anti-apoptotic Bcl-2 at mitochondrial membrane (i.e., promotes apoptosis) [180]. SDK-1 effect is summarized in Table 7. Apoptotic stimuli (e.g., exposure of cells to serum-free condition) induce a rapid Sph increase within 30 min, which induces multiple mechanisms, such as translocation of PKC δ to mitochondrial membrane, activation of caspase-3, release of SDK-1 through hydrolysis of PKC δ, 14-3-3 dimer formation through phosphorylation of interfacing 14-3-3, its interaction with mitochondrial membrane, and with cell surface integrin, etc. [181, 182]. These mechanisms are summarized schematically in Fig. 7.

Table 7.

Sphingosine-dependent kinases (SDKs)

♦	Protein kinases activated only by Sph or N,N-dimethyl-Sph (DMS), but not by Cer, Sph-1-P, or 12 other sphingolipids, phospholipids, and glycerolipids
♦	The activation is stereospecific by D-erythro, but not L-threo or D-threo derivative

Fig. 7. Multiple Sph-induced mechanisms leading to apoptosis.

Apoptotic stimulation causes (i) higher Cer level, followed by release of Sph by ceramidase; and (ii) an unknown mechanism causing translocation of PKCδ and its possible accumulation in mitochondrial membrane. These two events are considered to trigger multiple Sph-induced channels for apoptosis processes as below.

a. Sph activates a "cascade of caspases" leading to caspase-3 activation (process 1).

b. Activated caspase-3 cleaves PKCδ to produce PKCδ KD (SDK1) (process 2).

c. Sph activates SDK1 to phosphorylate 14-3-3 (→14-3-3-P) (process 3).

d. 14-3-3-P inhibits 14-3-3 dimer formation (process 4), which in turn inhibits binding of 14-3-3 to pro-apoptotic BAD/BAX, promoting their pro-apoptotic effect (process 5).

e. Sph inhibits integrin-dependent survival signal, e.g., Akt (process 6), leading to apoptosis (process 7).

f. Sph inhibits BCL-2 gene expression (process 8), thereby inhibiting anti-apoptotic BCL-2/BCL-X (process 9). Processes 5, 7, and 9 cause release of cytochrome c (process 10), which contributes to caspase-3 activation (process 11).

g. Activated caspase-3 releases PKCδ KD (SDK1), promoting a "vicious cycle" of Sph-induced apoptosis (process 12).

The overall process is based on enhanced Sph level, and translocation of PKCδ. The higher the Sph level, the greater the effect of the cycle through released caspase-3 and released PKCδ KD. Susceptibility of adherent tumor cells to Sph-induced apoptosis is less than that of non-adherent cells, because adherent cells require inhibition of integrin-dependent survival signal by Sph (process 6). Which Sph-induced channel is dominant may vary depending on type of cell. From [182].

ii. SDK-2 phosphorylates calreticulin (50 kDa) and protein disulfide isomerase (PDI) (55 and 58 kDa) [183].

iii. SDK-3 is associated with glucose-regulated proteins (GRPs) and heat shock proteins (HSPs), and is co-purified with them during separation. Purified SDK3 separated on Q-Sepharose column specifically phosphorylates GRP94, GRP105, and HSP86 [183].

Both SDK-2 and SDK-3 are chaperone proteins associated with endoplasmic reticulum. The cell biological significance of SDK-2 and SDK-3 requires extensive further study.

VIII. Future trends

Based on our current knowledge of functional roles of glycosyl epitopes in GSLs and glycoproteins, future trends of studies can be predicted. A few such trends, described below, are simply presented for my own "memorandum", while I doubt that such predictions are well justified.

♦ Trends of studies on cell recognition through glycosyl epitopes

The basic function of glycosyl epitopes in GSLs and glycoproteins is cell recognition, which consists of weak cell-cell tethering to strong cell-cell binding or cell-substratum binding, and subsequent signal transduction to alter cell phenotype. The mechanisms of these processes are summarized in Table 8. Among these, item 1, carbohydrate binding by protein (CPI), and item 2, carbohydrate modulating protein receptor function, which is essentially PPI, have been well studied. In contrast, item 3, carbohydrate-to-carbohydrate interaction (CCI) (see Sec. V) has been studied to some extent, but what is known is only a small fragment. A number of unknown processes based on CCI remain to be studied. Regarding Item 4, specific oligosaccharides are known to bind DNA with distinct target sequence, and such oligosaccharides have the ability to interfere with transcription factor function [184]. On the other hand, specific DNA sequences termed "aptamers" bind with high selectivity and affinity to cellobiose, but not to lactose, maltose, or gentiobiose [185]. These are the only studies I know so far, but the results indicate a tremendous future for functional role of carbohydrates in regulation of transcription or other processes.

Table 8.

Trends in cell recognition through glycosyl epitopes

1.	Chydr recognition by protein (chydr-protein interaction; CPI)
	• selectins, galectins, siglecs, and their ligands
	• many other chydr-binding proteins, their ligands
2.	Chydrs that modulate protein receptor function
3.	Chydr-chydr interaction (CCI)
	• Homotypic: e.g., self-recognition of Lex or SO3-GlcNAcβ4Fucα (see Fig. 5)
	Heterotypic: e.g., H-to-Ley, Os Fr.B-to-GM3 (see Fig. 6)
	• faster reaction, higher specificity than CPI or PPI
4.	Chydr-nucleic acid interaction (CNI)

♦ Molecular assembly and organization of glycosyl epitopes in membrane microdomains

All the processes listed in Table 8 take place in membrane microdomains where glycosyl epitopes as GSLs or as glycoproteins are organized with specific membrane receptors, auxiliary protein TSP, or signal transducers. So far, a few microdomains are distinguished, such as (i) "lipid rafts" as signaling platforms which are strongly affected by cholesterol; (ii) "immunosynapse" for immunocyte adhesion and signaling, typically observed for T-cell receptors in T lymphocytes and antigen-presenting cells; (iii) "glycosynapse" as functional units displaying glycosylation-dependent adhesion and signal transduction. Since the organizational framework of glycosyl epitopes in GSLs and glycoproteins, with various types of membrane receptors and signal transducers, is increasingly important, another major future trend will be detailed studies on membrane microdomains.

Our knowledge of microdomains has not been progressed for the past 15 years, since available methodology to separate different microdomains is highly limited. What we really need is to create new methods to distinguish subtle differences of microdomains. This is possible only through advancement of biophysical approaches.

♦ Genetic and epigenetic regulation of glycosylation

Glycosylation is controlled by (a) well-established glycosyltransferase genes, and their transcription factors involved in regulation of glycosyltransferase expression; and (b) less-studied epigenetic mechanisms through (i) methylation or acylation of coding nucleosome DNA, which controls phenotypic expression of mRNA, and (ii) non-coding DNA, which is outside the genomic DNA encoding small non-coding RNA such as "riboswitch" and "microRNA" [186, 187].

The area (a) has been extensively studied, and most of the mechanisms of transcription factors are known. Genes for glycosyltransferases, sugar nucleotide synthases and transporters, and sulfotransferases have been extensively studied; the total number of such "glyco-genes" is ~180 so far. In contrast, genes for glycosylhydrolases are also present in large numbers, and many of them remain to be studied.

The area (b-i) is in the early stage of study; further extensive studies are required for epigenetic control mechanisms of each type of glycosyltransferase gene. The area (b-ii), through small non-coding RNA, may affect expression of "glyco-genes". This area is totally unknown, and remains entirely for future study.

♦ Genetically-defined glycosylation differences that may cause disease susceptibility

A few cases have been known:

i. Possible expression of incompatible blood group P¹PP_k antigen in gastric cancer of a patient with blood group pp

A blood group pp patient with gastric cancer had anti-P¹PP_k antibodies (then termed Tj^a) in serum; mistaken blood transfusion caused disappearance of cancer; the cancer tissue had P¹-like antigen [188].

ii. Nonsecretor has much higher urogenital E. coli infection than secretor

Nonsecretor patients expressed sialyl-Gb5 epitope (NeuAc2-3Galβ3GalNAcβ3Galα4Galβ→R) in urogenital epithelia, to which infectious E. coli could bind, since it had "adhesin" to recognize this epitope. In contrast, secretor patients expressed globo-H and globo-ABO epitope (Fuc1-2Galβ3GalNAcβ3Galα4Galβ→R), to which infectious E. coli did not bind. Thus, nonsecretors have 4–5x higher E. coli infection in urogenital epithelia, particularly in women [189, 190].

iii. Blood group A or B expression in tumors suppresses tumor cell malignancy

Expression of blood group A or B determinants in various tumors is often reduced or deleted, which promotes tumor cell malignancy [191, 192]. A group of lung cancer patients with blood group A showed much higher survival rate when A expression in tumor was continued, than those in whom A expression in tumor was deleted [192]. Transfection of A or B gene in colonic cancer cell lines greatly reduced integrin-dependent tumor cell motility and cell growth [193, 194].

iv. Infertility in women is caused by antibodies directed to glycosyl epitope of sperm cells carried by male-specific CD52

Based on long-standing studies, a significant number of infertility cases were found to be associated with presence of anti-sperm antibodies in patient sera, which inhibit sperm cell motility in the presence of complement [195]. Subsequently, human mAb H63C4 was established based on fusion of lymphocytes of a patient (having high anti-sperm titer) with mouse myeloma NS1. The antigen epitope of H63C4 was identified as i, sialyl-i, and sialyl-I [196]. Previously-established anti-sialyl-I mAb NUH2 [197] also inhibited sperm cell motility. The antigen present in sperm cells was recently identified as male-specific CD52, the GPI-anchored glycan through a short peptide to which N-linked glycans, sialyl-i and sialyl-I, are linked [198, 199] (Fig. 8).

Fig. 8. Infertility in women is caused by the presence of antibodies directed to sperm antigens sialyl-i (SA-i) and sialyl-I (SA-I), carried by male-specific CD52.

Analysis of gangliosides present in human sperm, and thin-layer chromatography (TLC) immunostaining data with mAb H6-3C4, indicate that the male-specific epitope is sialyl-i, i, or sialyl-I, carried by GPI-anchored CD52 [198, 199] which consists of a short peptide (12 a.a.) with bulky N-linked glycan as shown (sialyl-i, top; i, middle; sialyl-I, bottom). Since H6-3C4 showed stronger reactivity with sialyl 2–6 lactonorhexaosylceramide than with sialyl 2–3 lactonorhexaosylceramide, the sialyl epitope could be a 2–6, 2–3 mixture. Data are explained in the text. From [196].

The line of these studies as exemplified in items i–iv above will be increasingly important. More effort should be focused to find differences in glycosylation correlated with development of specific diseases, such as diabetes, autoimmune diseases, and diseases caused by degenerative processes with unclear etiology (e.g., arthritis, Alzheimer's, Parkinson's).

♦ Possible role of glycosylation in stem cell biology

Stem cells with multi-differentiation ability are found in early embryo, and display two distinct processes: (i) differentiate into various tissues, organs, and eventually individuals; or (ii) lose differentiation ability to form embryonal carcinoma cells, and grow in vivo as malignant cancer to kill the host. Process (i) vs. (ii) is determined by difference in microenvironment of stem cells. The process (i) is provided by contact/ adhesion of stem cells with surrounding "niche", which promotes differentiation with normal cell cycle and growth. Process (ii) fails to provide the same microenvironment as in process (i), presumably due to absence of similar adhesion with "niche" as in process (i), which induces de-differentiation and uncontrolled cell growth.

Cells similar to embryonal stem cells are found in various adult tissues and organs. Properties of such "adult stem cells" are maintained by adhesion with surrounding niche cells, to display restricted, regulated growth as in normal cells in normal tissues. "Niche", a group of cells that maintains stem cell function through their adhesion, activates various signaling molecules such as Notch, TBGF, Wnt, and BMP (bone morphogenetic protein), although the molecular mechanism of adhesion remains unclear (for review see [200, 201].

A major question at this time is whether glycosylation plays a key role in defining stem cell function, and their adhesion to induce activation of signaling molecules as above. Mouse embryonal stem cells express glycosyl epitopes SSEA-1 [87], -3, and -4 [53, 54], which may mediate adhesion of stem cells to induce differentiation. Recent studies indicate that a group of cells in neuronal tissues expressing SSEA-1 is correlated with stem cells [202], and glycosyl expression changes during the 12th through 28th day of differentiation, as defined by many antibodies and lectins [203, 204]. However, critical studies on functional role of glycosylation which may mediate adult stem cell adhesion, with subsequent differentiation, have not been performed.

Studies of stem cell processes and their molecular mechanism will provide answers to two basic questions: (i) how cells initiate multiple directions of differentiation, to form specific types of cells with specific functions; (ii) whether cancer cells are derived from adult stem cells when the microenvironment changes, according to the "Cohnheim hypothesis" [205]. Question (ii) needs some explanation. Under this hypothesis, oncogenic phenotype of stem cells can be inhibited when stem cells are "strayed" in tissues and placed in a certain microenvironment. Oncogenic phenotype is "awaken" when the stem cells are exposed to a different environment; see also [201]. According to this hypothesis, genetic mutation, activation of oncogenes, or inhibition of anti-oncogenes are not necessary. Rather, epigenetic changes caused by the microenvironment are sufficient to induce oncogenic phenotype. Cohnheim's view, which was forgotten for over 130 years, is now revived by increasing evidence for epigenetic conversion and reversion of oncogenic phenotype (e.g., [70, 130, 206–208]. It should be noted that malignancy of mouse embryonal carcinoma cells was completely lost, and cells differentiated into "perfect mice", when embryonal carcinoma cells were placed in blastocyst. This surprising observation, shown by Beatrice Mintz 30 years ago [209, 210], is now clearly explained.

♦ Epithelial-mesenchymal transition (EMT): Is there any glycosylation effect?

The EMT, originally discovered by Elizabeth Hay [211], is the major target mechanism of oncogenic transformation and developmental process [212, 213]. The majority of tumors are derived from epithelial cells, and the process of oncogenic transformation is regarded as EMT. A number of current studies suggest that cancer cells display properties similar to those of "myofibroblasts" that underlie epithelial cells. "Cancer-activated fibroblasts" result from EMT, and are characterized by down-regulation of E-cadherin, up-regulation of N-cadherin, vimentin, and oncofetal fibronectin, and different susceptibility to various growth factors, particularly TGF-β, chemokines, and ECM-degrading proteases; in many cases, EMT is induced by TGF-β [213, 214]. Interestingly, expression of oncofetal fibronectin with change of III-CD is based on site-specific O-glycosylation change [215].

So far, the functional role of glycosylation in EMT processes, according to the EMT definition, has not been systematically studied. Morphology and HGF-dependent growth of normal bladder epithelial cells were converted to HGF-independent, highly malignant cellular phenotype by deletion of TSP CD82 by RNAi, whereby ganglioside GM2 complexed with CD82 was lost, and epithelial morphology was converted to fibroblastic morphology [132]. This oncogenic conversion could be EMT. On the other hand, reversion of tumor cells to normal cells can be observed as discussed in Sec. IV-iii, in which TSP CD9 or CD82 complexed with ganglioside GM3 or GM2 inhibits GFR or integrin receptor. The process is therefore ganglioside-induced inhibition of GFR and integrin.

Abbreviations

CCI

carbohydrate-to-carbohydrate interaction

Csk

C-terminal Src kinase

DMS

D-erythro-N,N-dimethyl-Sph

EGF

epidermal growth factor

EGFR

epidermal growth factor receptor

EMT

epithelial-mesenchymal transition

FGF

fibroblast growth factor

FGFR

fibroblast growth factor receptor

FN

fibronectin

GEM

glycolipid-enriched microdomain

GSL

glycosphingolipid

LN

laminin

MS

mass spectrometry

PDGF

platelet-derived growth factor

PDGFR

platelet-derived growth factor receptor

PKC

protein kinase C

Sph

sphingosine

Sph-1-P

sphingosine-1-phosphate

SSEA

stage-specific embryonic antigen

TMS

D-erythro-N,N,N-trimethyl-Sph

TSP

tetraspanin