Glycosphingolipids (GSLs) comprise a heterogeneous group of membrane lipids formed by a ceramide backbone covalently linked to a glycan moiety. Hundreds of different glycans can be linked to tens of different ceramide molecules, giving rise to an astonishing variety of structurally different compounds, each of which has the potential for a specific biological function. GSLs have been suggested to modulate membrane-protein function and to contribute to cell–cell communication. Although GSLs are dispensable for cellular life, they are indeed collectively required for the development of multicellular organisms, and are thus considered to be key molecules in ‘cell sociology’. Consequently, the GSL make-up of individual cells is highly dynamic and is strictly linked to the cellular developmental and environmental state. In the present review, we discuss some of the available knowledge, open questions and future perspectives relating to the study of GSL biology.
In 1873, the British archaeologist Alexander Cunningham published a report of a seal with a short string of symbols belonging to what was later referred to as the Indus script . Subsequently, thousands of objects bearing similar symbols have been discovered, produced by a civilization that flourished 4500 years ago in northern India and Pakistan . As soon as the repertoire of such inscriptions increased, the scientific community faced the question as to whether they are the graphic representation of a ‘language’ ; in other words, they asked whether these symbols are arranged in strings according to any recognizable set of syntactic rules, and if they might ‘code for’ something meaningful.
Biologists faced very similar problems in the biopolymer era [3, 4], when attempting to understand the meaning of ordered sequences of residues in a given DNA, RNA or protein . Such efforts led to the discovery of the genetic code and to the birth of molecular biology . Nevertheless, among the biopolymers, the polymers of monosaccharide residues known as the glycans have proven to be more difficult to decipher in terms of any possible coding [5, 6]. Although the idea of a glycan code has indeed been proposed (i.e. the glycocode) [5, 6], it is not clear to what extent glycans might serve as an information media in eukaryotic organisms. Indeed, there are hundreds of different complex glycan polymers that are a part of glycoproteins and glycolipids .
In vertebrates, the most represented class of glycolipids is that of glycosphingolipids (GSLs) . These are a heterogeneous class of amphipathic compounds that are characterized by complex glycan structures linked to a ceramide backbone by a β-glycosidic bond. Many aspects of GSL structure, synthesis and disposal have been defined over the years , which has led to an accurate molecular description of the GSL metabolic pathways and to an understanding of the pathogenic mechanisms of a number of genetic diseases that are caused by defects in GSL-catabolizing enzymes .
Manipulation of GSL synthesis in animal models has led to the concept that, although cells can survive, grow and divide in the complete absence of GSLs , metazoa and especially vertebrates, need GSLs (collectively) to correctly complete their development . Furthermore, ablation of significant subparts of the GSL synthetic pathways yields much milder, or even elusive, phenotypes in animal models, which has been either demonstrated to depend on the compensatory potential of the remaining GSLs  or was suggested to be the consequence of the involvement of given GSLs in very specific physiological functions  (Fig. 1B). For a number of GSLs, modulatory effects on specific plasma membrane receptors have indeed been demonstrated , thus providing the proof of principle for their involvement in environment ‘sensing’ and in cell identity establishment/maintenance . Nevertheless, the biological functions of the vast majority of GSLs, as well as the regulatory layers controlling their ‘expression’, remain to be defined.
Keeping with the linguistic metaphor, to date, GSLs are for biologists what the Indus script is for archaeologists, namely a collection of strings/structures composed of symbols/residues, the possible grammar and deep meaning of which remain largely unknown. Despite the little information that we have available about these compounds, their undoubted importance in higher eukaryote biology invites further investigation, which promises to provide new clues to the physiology of multicellular organisms.
The GSL metabolic pathways are embedded in the endomembrane system [14, 15]. GSL synthesis is initiated in the endoplasmic reticulum (ER) with the condensation of a sphingoid base (as either sphinganine in de novo synthesis, or sphingosine in the salvage pathway) with an acyl-CoA, in a reaction that is catalyzed by a group of six enzymes: the ceramide synthases (CerS)1–6 . The product of this reaction is either ceramide (Cer) or dihydroceramide, depending on the substrate (sphingosine or sphinganine, respectively) . Different CerS catalyze the same reaction but differ in their substrate preference with respect to the acyl-CoA involved . Thus, Cer and dihydroceramide with different acyl chain lengths are produced by different CerS .
Cer can be galactosylated in the ER, to produce galactosylceramide (GalCer)  or it can be transported to the Golgi complex via two alternative pathways . Cer can be picked up from ER membranes by the ceramide-transfer protein CERT, and delivered to the trans-Golgi network (TGN) where, after translocation across the membrane bilayer, it is primarily used for the synthesis of sphingomyelin . Alternatively, Cer reaches the cis-Golgi, most probably through vesicular transport , where it is glucosylated to produce glucosylceramide (GlcCer) [18, 20].
GlcCer is produced on the cytosolic leaflet of early Golgi membranes. It can then be transported across the Golgi complex via vesicular trafficking, translocated to the lumenal membrane leaflet of the Golgi by an uncharacterized transporter , and subsequently glycosylated by specific Golgi-resident glycosylating enzymes, to produce complex GSLs . Alternatively GlcCer can be picked up from the cis-Golgi membranes by the lipid-transfer protein FAPP2 [22, 23]. FAPP2 then delivers GlcCer to the TGN [22, 24], where it is translocated to the lumenal membrane leaflet. Here, this is most probably mediated by the multidrug resistant P-glycoprotein . GlcCer is then used for the production of complex GSLs through the action of TGN-specific glycosyltransferases  (Fig. 1A).
GlcCer and GalCer are the common precursors of all of the GSLs that are synthesized in the Golgi complex by the Golgi glycosyltransferases [7, 26]. These enzymes transfer a specific carbohydrate from the appropriate sugar nucleotide (e.g. UDP-Glc, UDP-Gal, CMP-sialic acid) to a specific position on a particular type of acceptor (Cer, or the nonreducing end of a growing carbohydrate chain attached to Cer). After GlcCer is translocated to the luminal leaflet of the Golgi and TGN membranes, it is galactosylated to produce lactosylceramide (LacCer; Galβ1-4Glcβ1Cer) by β4-galactosyltransferases V and VI [27-29]. Once produced, LacCer cannot be translocated back to the cytosolic leaflet of cell membranes [17, 21]. Instead, LacCer is the metabolic branch point for the formation of the different classes of complex GSLs . Indeed, LacCer is the substrate of: (a) the β1,4-N-acetylgalactosylaminyltransferase B4GALNT1 to produce GA2 (GalNAcβ1-4Galβ1-4Glcβ1Cer) [30, 31]; (b) the α-2,3-sialyltransferase ST3GAL5 to produce GM3 (NeuAcα2-3Galβ1-4Glcβ1Cer) ; (c) the α1-4-galactosyltransferase A4GALT to produce Gb3 (Galα1-4Galβ1-4Glcβ1-Cer) ; and (d) the β-1,3-N-acetylglucosaminyltransferase B3GNT5 to produce Lc3 (GlcNAcβ1-3Galβ1-4Glcβ1Cer) . GA2, GM3, Gb3 and Lc3 are then the precursors for the synthesis of GSLs belonging to the asialo, ganglio, globo/iso-globo and lacto/neo-lacto series, respectively . GalCer on its side is transported to the Golgi complex where it can be sialylated to produce GM4 ganglioside, or sulfated to produce sulfogalactolipids  (Fig. 1B).
From the TGN, GSLs and sphingomyelin are transported to the plasma membrane (PM) in membrane-bound transport carriers . At the PM, GSLs can undergo partial remodelling through the action of specific glycosidases . Alternatively, GSLs are transported along the endocytic routes from the PM to lysosomes to be degraded [8, 15]. GSL degradation in lysosomes is operated by a set of specific glycohydrolases and accessory proteins , which assist in the stepwise dismantling of the glycan moieties, all the way down to Cer . The Cer in lysosomes is then catabolized by acid ceramidase to produce a fatty acid and sphingosine, which can be transported to the ER and used for the synthesis of GSLs (salvage pathway)  (Fig. 1A).
According to the LIPID MAPS Structure Database , more than 400 different glycan structures have been found attached to Cer in vertebrates, the composition of which ranges from one to 20 sugar residues (Fig. 2A). Twelve different sugars make up the GSL ‘alphabet’, with sulfation representing a further modification (Fig. 2B). The occurrence of each single residue in the GSL glycan repertoire is not evenly balanced, with galactose (Gal) by far the most frequent sugar (~ 40%), followed by N-acetylglucosamine (GlcNAc; ~ 20%), glucose (Glc; ~ 14%), fucose (Fuc; ~ 10%), N-acetylgalactosamine (GalNAc; ~ 8%) and N-acetylneuraminic acid (NeuAc; ~ 5%), with other sugar residues being found more rarely (Fig. 2C).
Although, in principle, any two residues can be linked in a GSL glycan structure, only a subset (n = 29) of the theoretical (n = 144) disaccharide combinations are involved in vertebrates (Fig. 2D). These include, for example, Fuc as an obligatory glycan chain terminator, Glc always linked to Gal, sialic acid species [NeuAc, N-glycolylneuraminic acid (NeuGc)] elongated at their nonreducing ends only by other sialic acid residues (NeuAc, NeuGc) and the very frequent Gal-GlcNAc (polylactosamine) repetitions.
According to these considerations of the available GSL structure database, it is possible to construct a descriptive hierarchical tree (Fig. 2E) and to define the tentative generative ‘grammar’ for GSL glycan chain assembly (Fig. 3). The syntactic rules of this ‘grammar’ derive from the existence of the given GSL metabolizing enzymes that can catalyze a subset of all of the possible synthetic reactions. When the occurrence of different residues is considered along the reducing/nonreducing synthetic axis, a strong regularity is seen (Fig. 4A), with Glc present only at position 1, Gal as the most represented residue at even positions (i.e. 2, 4, 6, 8, etc.) and GlcNAc, GalNAc, NeuAc and Fuc frequently at odd positions (i.e. 3, 5, 7, etc.). This particular behaviour is not shared by other directionally arranged biopolymers, such as mature human miRNAs (2080 sequences obtained from the miRBase) (Fig. 4B), where no preference for any specific base is observed along the 5′- to 3′ synthetic axis. The consequence of this phenomenon can be seen through computation as the Shannon conditional entropy, which measures the amount of information associated with the elongation of a string of characters . This demonstrates that the theoretical information content oscillates along GSL chains, with odd-positioned residues being extremely more information-rich than even-positioned ones (Fig. 4C). Thus, when analyzed as linear entities, these GSL glycan chains show: (a) adherence to cogent syntactic rules for their assembly; (b) regularity in their structure, with very frequent Gal-X repetitions; and (c) high theoretical information content.
Further layers of complexity are associated with GSL glycan moieties that are not shared by other information-bearing biopolymers, including variability of the linkage point, heterogeneity in the anomeric nature of the glycosidic bond (α or β) and ‘branching’. Thus, in GSLs, two residues can be linked by either α or β bonds between different atoms in their structures. Moreover, the glycan chain can be branched, with one residue accepting more than one residue at its nonreducing end during elongation . These peculiar features mean that the GSLs have the possibility to engage multiple interactions with one or more sugar-binding partners (a property known as multivalency) by adopting particular three-dimensional conformations . Interestingly, when the occurrence of these branching events is considered along the reducing/nonreducing synthetic axis of GSL glycans, they appear mainly at even positions that are frequently occupied by Gal residues (Fig. 4D).
The second main source of complexity in GSL structures is linked to heterogeneity in their ceramide backbone. Ceramide is a family of molecules distinguished by specific components and modifications . These include the presence of double bonds and hydroxylations at particular positions, as well as the presence of fatty acids with different acyl chain lengths both as part of the sphingoid base or of the amide-linked acyl part of the molecule . Taking in account the combination of these modifications, mammalian cells have the potential to synthesize more than 200 different ceramide species, with dozens of them having been found in cells . Nevertheless, ceramide heterogeneity is not a consequence of a ‘lack of fidelity’ in the synthetic enzymes because it derives from the action of dedicated enzymatic machineries controlling the production of specific species [43-45], each of which has the potential to generate its own ‘GSL world’ [41, 42].
Altogether, although not complete, the collection of GSL structures currently available already allows us to formulate some hypotheses about the properties of GSLs in terms of their possible ‘syntax’ (Fig. 3) and even their ‘coding’. However, the understanding of the biological significance of such a vast repertoire of molecules remains bound to the determination of two key aspects: (a) how individual cells determine their GSL make-up and (b) the cellular factors (e.g., proteins, glycans, lipids) that interact with and are regulated by specific GSLs.
Control of GSL expression
By contrast to other biopolymers, the assembly of glycans into the GSLs is not template driven but rather depends on the combined actions of the enzymes involved in their synthesis [26, 46]. Unexpectedly, despite the lack of an inheritable template, individual cell types usually show high fidelity in their expression of a few selected GSL species at detectable levels. The mechanisms by which this specificity is accomplished primarily involve the expression of the relevant enzymes , the control of the subcellular localization of these enzymes  and the formation of multi-enzyme complexes that can convey precursors to specific metabolic fates  (Fig. 5). Thus, for example, the expression of GSLs is strictly controlled during development  in such a way that GSLs can be used as lineage-specific differentiation markers . Similar changes occur during in vitro differentiation of embryonic stem cells, with the globo and lacto series of GSLs expressed at the stem-cell stage, and gangliosides being more abundant in embryonic bodies , especially for neuronal lineages . These changes are secondary to the transcriptional reorganization of the GSL synthetic system [47, 48], which involves the coordinated down-regulation and up-regulation of the enzymes involved in the synthesis of the globo/lacto and ganglio series of GSLs, respectively [47, 48]. Similar changes in the GSL content of non-embryonic cells under differentiation or other stimuli have been frequently observed, whereas their dependence on coordinated transcriptional programmes has not been approached systematically [49-56]. Moreover, to our knowledge, the structure of the gene networks that might coordinate the expression of GSL synthesizing enzymes is not known. Nevertheless, concomitant expression of enzymes devoted to different branches of GSL metabolism is commonly observed in a number of cell types and under various conditions, thus leaving open the question as to whether the final membrane composition in terms of GSLs is determined by stochastic/probabilistic means, or whether more active mechanisms exist. An active means by which cells can control their GSL make-up involves the localization and quaternary structure of the Golgi-resident glycosyltransferases and sugar transporters . Individual glycosyltransferases are localized to specific Golgi subcompartments , where they interact with each other [57, 58] and with the sugar transporters . These can thus form dedicated metabolic machines to catalyze a number of sequential reactions, and thus to convert a substrate GSL into a much more complex product with high efficiency . An example of how this can influence the GSL synthetic outcome is provided by the finding that LacCer synthase physically interacts with both GM3 synthase and Gb3 synthase [60-62]; these interactions are mutually exclusive and relocate LacCer synthase to different Golgi subregions . Consequently, the delivery of GlcCer to one or the other multi-enzymatic complex will determine a different metabolic outcome . Interestingly in this respect, GlcCer can be transported by independent mechanisms to different Golgi subregions . Indeed, if GlcCer is progressively transported through the Golgi stack via vesicular trafficking, it is preferentially used by the LacCer synthase/GM3 synthase complex to produce GM3 and the consequential downstream gangliosides. If, instead, GlcCer is transported directly to the TGN by the action of the transfer protein FAPP2, it is used by the TGN-localized LacCer synthase/Gb3 synthase complex for the production of Gb3 and its consequential downstream globosides . Again, although some mechanistic aspects of GlcCer transfer by FAPP2 have been described , whether and how FAPP2 is regulated during the conditions that lead to cell GSL remodelling remains unknown.
Along similar lines, it has been shown that GlcCer translocation to the luminal leaflet of Golgi membranes discriminates between two GlcCer pools . The best-described of these is the MDR1 glycoprotein that acts as a GlcCer ‘flippase’ at the late Golgi (the Rab6-positive compartment) to participate in the synthesis of neutral (globo series) and not acidic (ganglio series) GSLs in a number of cell lines . Alternatively, there is the uncharacterized ‘flippase’ in the more proximal Golgi compartment that provides substrate for ganglioside synthesis. In addition to this evidence, according to recent studies , glycosylation in the Golgi is subjected to extensive control through many signalling pathways. Although the details of the intricate relationships between individual signalling modules and the given glycosylation patterns are far from being understood, what emerges from these studies is that environmental cues can have profound impact on the glycan expression patterns of glycoproteins, and possibly of glycolipids. In agreement with this, single-cell GSL expression has been found to vary remarkably within an isogenic cell population, which depends on the cell population context and, ultimately, on the micro-environment . The control circuits that integrate such environmental/signalling components with the assembly of the glycosylating machineries, with GSL trafficking and with dedicated transcriptional programmes responsible for GSL expression have not been characterized to date, and thus await further investigation.
The physiological role of GSLs has been studied using genetic, biochemical, biophysical and cell biology approaches. Mouse genetics has provided a general framework for our understanding of the roles of GSLs in mammals. According to these studies, ablation of the gene for GlcCer synthase in mice leads to embryonic lethality during gastrulation as a result of massive apoptosis . Similarly, ablation of the B4GALT-V gene that is responsible for LacCer synthesis expression  leads to embryonic lethality by embryonic day 10.5, possibly as a result of growth inhibition , which suggests that GSLs synthesized downstream of GlcCer are cumulatively required for correct embryo development. On the other hand, ablation of the ST3GAL5 gene, which is responsible for GM3 synthesis expression, does not lead to any major abnormalities, although this is associated with enhanced insulin sensitivity , impaired neuropsychological behaviour  and hearing loss . Ablation of the downstream GA2/GM2/GD2 synthase  leads to male infertility , axonal degeneration, myelination defects , motor deficit  and Parkinsonism , whereas ablation of GD3 synthase leads to thermal hyperalgesia, mechanical allodynia  and reduced neuroregeneration , all of which strongly involve gangliosides in neuronal function. Along the same lines, combined ablation of the GM3 and GA2/GM2/GD2 synthases leads to severe neurodegeneration , and combined ablation of GD3 and GA2/GM2/GD2 synthases induces lethal audiogenic seizures  and peripheral nerve degeneration, leading to reduced sensory function and skin lesions as a result of over-scratching  in mice. Ablation of GalCer synthase, which results in the loss of all of the gala series GSLs, also induces profound neuronal phenotypes that appear to be secondary to defects in myelination, in line with the extreme enrichment of these lipids in myelin . For the lacto/neolacto series, ablation of the B3GNT5 gene that is responsible for Lc3 synthesis expression leads to either preimplantation lethality  or multiple postnatal defects, including early death, growth inhibition, loss of fur, obesity, reproductive problems and B-cell functional defects . By contrast, A4GALT and FAPP2 knockout mice show either absent or reduced globoside synthesis, respectively, with no overt phenotypes [12, 24] (Fig. 1B).
GSLs are therefore associated with several diseases in humans, including cancers. Indeed, GSLs actively modulate various aspects of the biology of the cell, including apoptosis, cell proliferation, endocytosis, intracellular transport, cell migration and senescence, and inflammation. These all represent crucial aspects relating to tumourigenesis and cancer progression, as well as the responses to anti-cancer therapies . Moreover, a large number of tumour-associated antigens have been identified as GSLs . Altered cell surface GSL-expression patterns are associated with tumour-relevant phenotypes in different cancer cells . Thus, the gangliosides Gt1b, GD1A, GM3 and GM1 inhibit cell proliferation and epidermal growth factor (EGF) receptor tyrosine phosphorylation , whereas the globosides Gb4 and Gb5 strongly enhance cell proliferation and motility . On the same line, disialyl GSLs GD2 and GD3 have been demonstrated to enhance tumour phenotypes  such as cell growth and invasiveness in malignant melanoma (GD3) [84-87], in small cell lung and breast cancers (GD2)  and in osteosarcoma cells (GD3/GD2)  by modulating Src family kinases and Focal adhesion kinase activation. On the other hand, the ganglioside GD1a inhibits cell migration in highly metastatic osteosarcoma cells  by suppressing matrix metalloproteinase-9 , tumour necrosis factor α , nitric oxide synthase 2  and hepatocyte growth factor (HGF) expression , thus impacting on HGF induced c-Met phosphorylation . Moreover, Gt1b has been shown to inhibit integrin dependent keratinocyte adhesion to fibronectin , whereas GM3 and GM2 inhibit integrin dependent cancer cell motility, promoting the formation of a ganglioside/tetraspanin/integrin complex and negatively regulating Src or Met [97, 98].
Cumulatively, these data assign specific roles in mammalian physiology and pathology to different classes of GSLs, and the molecular mechanisms through which they exert these functions involve interactions of GSLs with proteins and glycans . Indeed, over the last 30 years, numerous GSL interactors have been characterized and, for some of these, the functional outcomes of their interactions have also been defined . A set of GSLs (mostly gangliosides) have been found to interact with a number of PM-located signalling receptors and to modulate their activation [81, 99-116] (Table 1). Probably the best-characterized example is that of the interaction between the EGF receptor and GM3. The original finding that exogenously added GM3 inhibits cell growth in different cell lines [117, 118] through its modulation of EGF receptor phosphorylation has been approached at the synthetic biology level . According to Coskun et al. , inhibition by GM3 of EGF receptor auto-phosphorylation in liposomes depends on the presence of the NeuAc residue in GM3 and of lysine 642 in the EGF receptor. The binding of GM3 to the EGF receptor inhibits EGF receptor dimerization in the absence of its ligand and, accordingly, keeps the EGF receptor in an inactive state in the absence of a productive stimulus . A similar interaction has been reported between the insulin receptor and GM3, which again involves a key lysine positioned in the proximity of the transmembrane portion of the insulin receptor .
Table 1. Known interactions between GSLs and proteins.
GSLs can also interact with a number of nonreceptor PM proteins, including tetraspanins, integrins and caveolin-1 [96, 119-125], as well as with galectins [126, 127]. Some of these interactions involve the binding of GSL glycans to protein modules or specific amino acids [99, 114, 126, 127], whereas a further option is for GSLs to interact with the glycan moieties of glycoproteins or other GSLs through so-called carbohydrate–carbohydrate interactions (CCIs) . CCIs can be established both in cis (i.e. with the glycomolecules on the same membrane) and in trans (i.e. with the glycomolecules on the limiting membrane of an adjacent cell) . The first class of CCIs (i.e. those in cis) include the reported interactions between GM3 and the terminal GlcNAc moieties of the EGF receptor  and HGF receptor  glycans, which have been suggested to contribute to the modulation properties of receptor activation by GM3 . Similarly, the interaction between Gt1b/GD3 and mannose residues in integrin α5 linked glycans has been shown to modulate integrin α5-β1 function . For the second class of CCIs (i.e. those in trans), there are the GM3–Gg3 and GM3–LacCer interactions that have been reported to contribute to the adhesion of tumour cells to endothelial cells [129, 130].
In addition to the ability of GSLs to interact with specific partners, they have the unique feature of forming molecular clusters by acting as both hydrogen bond donors and acceptors [7, 13, 15, 131]. This promotes their self-aggregation, which can create membrane heterogeneity through reduced mixing with the other membrane lipids . These ‘platforms’ have been referred to in different ways, such as lipid rafts, GSL-enriched membranes and glycosynapses , with these different terms relating to their operational definition, and probably referring to different entities . Although the very existence of lipid-driven membrane heterogeneities has been matter of intense debate in the field of membrane biology [133, 134], as a result of difficulties with respect to visualizing them in living cells, recent advances in imaging techniques  and new experimental strategies  have provided direct evidence for sphingolipid enriched membrane domains in living cells. These lipid platforms on the PM appear to serve to cluster signalling and adhesion molecules, to regulate their functions  and to participate in cargo sorting at the different traffic stations along the secretory and endocytic pathways . The composition of membrane lipid domains in terms of GSLs can vary , with distinct GSLs being enriched in different domains in the same cell [138-141], which influences the function and architecture of these domains [83, 142].
Two lines of evidence suggest that also ceramide heterogeneity can have a biological relevance in GSL functions: (a) the GSL composition in terms of ceramide backbone influences their unmixing properties in membrane bilayers, thus impacting on their partitioning in lipid domains  and (b) sphingolipids can interact with proteins through their ceramide moiety in a fashion where interaction specificity is conferred by ceramide composition . Although ceramide heterogeneity has not been systematically approached as a factor influencing GSL function, these data, along with the observation that GSL ceramide composition is tightly controlled during GSL remodelling [55, 145], open new perspectives with respect to the determination of GSL functions.
Altogether, the available knowledge on GSLs indicates that they have membrane-organizing functions [13, 17], whereas specific GSL species are involved in interactions with specific proteins and/or lipids . These two properties concur with the role that GSLs have in ‘environment sensing’, both in terms of modulation of cell responsiveness to hormonal stimuli and of cell–cell adhesion/recognition. These concepts position GSLs as important modulators of multicellularity, and more generally relate to ‘cell sociology’ . Thus, GSLs have emerged to be key controllers in processes that imply cell differentiation and tissue patterning, whereas their deregulation plays a driving role in ‘cell sociology’ related diseases such as cancers. It should also be noted, however, that, although these concepts are supported by a number of studies, only a modest number of GSLs have been studied in any real detail, thus leaving the understanding of the specific roles of most GSLs to future research.
Open questions and possible future perspectives
In conclusion, the specific state of the art at present indicates that: (a) GSLs comprise a vast group of biological polymers that show remarkable heterogeneity in their structures; (b) given GSLs are specifically expressed by mammalian cells under particular developmental and/or physio/pathological conditions; and (c) specific GSLs modulate cell functions by influencing signal transduction pathways and, by doing so, they ultimately affect gene expression . These properties position GSLs not only among the cellular factors that impact on cell phenotype at a nongenetic level, but also among the ‘epigenetic’ mechanisms that shape the cell and organism phenotype. However, in contrast to other classes of molecules that share similar properties (i.e. micro-RNAs), little is known about the targets of GSL regulation, the control of GSL expression, and the possible metabolic and genetic circuits that integrate GSLs with other cell-regulatory elements. The main reasons for this gap are technological. Indeed, although nucleic acid and protein sequencing have been possible for many years, such that these techniques are in routine use in most cell biology and biochemistry laboratories, the determination of GSL structures is still a labour-intensive and specialist-restricted task. Also, GSL–protein interaction studies have suffered as a result of the lack of high-throughput methods to systematically approach this issue. Moreover, although imaging of proteins and nucleic acids is well developed both in cells and in tissues, GSL imaging has proven to be extremely difficult to standardize, with very few reagents managing to faithfully describe GSL distributions in cells and tissues. Also, along the same lines, although extremely valuable as an option, the ‘tagging’ of GSLs with fluorescent dyes [146, 147] is accompanied by serious doubts about the ability of these tagged compounds to fully and correctly reproduce the behaviour of endogenous GSLs.
However, in recent years, some of these technological barriers have been overcome. The development of MS for GSLs has provided a fast and reliable method for determination of GSL levels and structures in biological samples  and, more recently, also for GSL imaging in biological sections . The use of fluorescently-labelled bacterial toxins binding with high specificity to GSLs on cell membranes [150, 151] has provided a fast and reliable approach for visualizing GSLs in cells and tissues [24, 64, 152]. Also, optochemical strategies for cross-linking of GSLs with their interacting proteins have been developed with success [144, 153]. These new techniques allow those interested in GSL biology to approach the study of these compounds in a more systematic way and at a more quantitative level, with significant steps forward already being accomplished. A further aspect to be considered is the tremendous advances made in recent years in the field of tissue engineering and stem cell biology . The development of in vitro models for organogenesis provides a unique opportunity to study the function of GSLs in experimental systems where they have a greater chance of being relevant, thus dissecting their involvement in cell sociological processes.
In the light of these advances, it is conceivable to imagine that GSL research will experience an acceleration in the coming years, which will surely uncover new and unexpected roles for this class of molecules, ultimately allowing us to decipher the so-far-enigmatic GSL ‘language’.
We thank Dr S. Parashuraman for discussions and for critically reading the manuscript, as well as Dr C. P. Berrie for editorial assistance. G.D.’A. acknowledges the support of AIRC (MFAG 10585).