Carbohydrates, their structures and the enzymes responsible for their synthesis and degradation, offer numerous possibilities for the design and application of probes with which to study and treat disease. The intracellular dynamic O-GlcNAc (O-linked β-N-acetylglucosamine) modification is one such glycosylation with considerable medical interest, reflecting its implication in diseases such as Type 2 diabetes and neurodegeneration. In the present paper, we review recent structural and mechanistic studies into the enzymes responsible for this modification, highlighting how mechanism-inspired small-molecule probes may be applied to study potential disease processes. Such studies have questioned a causal link between O-GlcNAc and Type 2 diabetes, but do offer potential for the study, and perhaps the treatment, of tauopathies.
Carbohydrate-active enzymes, in particular glycosyltransferases and glycoside hydrolases, are fascinating and diverse catalysts. Approx. 1–3% of most genomes are dedicated to the synthesis and breakdown of the glycosidic bond . Much of the work in our group focuses on these carbohydrate-active enzymes, in particular the study of their three-dimensional structure and reaction mechanism. Structural techniques have been used to map the reaction co-ordinate in conformational terms providing insight into the distortions of the pyranoside rings that pertain to catalysis (examples include [2–8]). Of particular interest is how conformational analyses (reviewed in [9,10]) may lead to the design and exploitation of enzyme inhibitors both to inform mechanistic understanding and to act as probes of cellular function (for example [11–14]). In the present paper, we focus on the mechanistic enzymology of one particular glycosylation/deglycosylation: the O-GlcNAc (O-linked β-N-acetylglucosamine) modification. We review how mechanistic and structural insight has aided in the development of small-molecule probes of cellular function that are beginning to have an impact on disease models, notably for Type 2 diabetes and neurodegeneration.
The O-GlcNAc modification
The attachment of a single O-GlcNAc (reported by Torres and Hart in 1984 ) unit to the side chains of serine and threonine residues of nuclear and cytoplasmic proteins (Figure 1) has emerged as an important post-translational modification comparable with phosphorylation . Unlike other types of protein glycosylation, the O-GlcNAc modification is not extended further and, perhaps more significantly, this modification is cycled several times in the lifetime of proteins, strongly supporting its implication in cell signalling events . Hundreds of proteins such as transcriptional regulators, cytoskeletal network proteins and ubiquitin–proteasome components are now known to be O-GlcNAc-modified . In many cases, the presence of the O-GlcNAc residue is correlated with the phosphorylation status of the protein, and although the phosphorylation/O-GlcNAcylation balance is not simply reciprocal, it is evident that some competition exists between these two modifications (see  and references therein).
The O-GlcNAc modification, dissection and intervention
In contrast with the abundance of kinases and phosphatases sustaining the phosphorylation signalling network , O-GlcNAc cycling (in humans) is performed by only two proteins, a glycosyltransferase that attaches the O-GlcNAc residue using UDP-GlcNAc as substrate [OGT (O-GlcNAc transferase)] and a glycoside hydrolase that removes it from proteins [OGA (O-GlcNAcase)] (Figures 1 and 2). The fact that only two enzymes regulate the O-GlcNAc activity makes the regulatory role of this modification appear deceptively simple. In reality, the O-GlcNAc modification is regulated at several levels, such as the processing of mRNA encoding the O-GlcNAc enzymes [20,21], the availability of UDP-GlcNAc substrate supplied at the end of the hexosamine biosynthetic pathway, the post-translational modifications of the OGT and OGA enzymes by phosphorylation or self O-GlcNAcylation and the specific association of these enzymes with various other partners in transient protein–protein complexes .
Much of the impetus for the study of O-GlcNAc is derived from its proposed link to Type 2 diabetes through the hexosamine biosynthetic pathway  with O-GlcNAc acting as a nutrient sensor . Hyperglycaemia certainly increases cellular O-GlcNAc levels, and, in at least one model system, diabetic rats appear to have elevated O-GlcNAc levels . The numerous potential protein targets and sites for O-GlcNAc modification (where they could thus interfere with phosphorylation-dependent signalling cascades) coupled to the nutrient-sensor role offers a seductive model for the significance of the O-GlcNAc modification in insulin resistance. Strong support for this was given by Vosseller et al.  through the application of an O-GlcNAc hydrolase inhibitor, ‘PUGNAc’ (Figure 2d), in a model cell line 3T3-L1 adipocytes. In these foundation studies, the adipocyte cells treated with PUGNAc showed elevated O-GlcNAc levels, decreased phosphorylation of the key signalling kinase Akt and also developed insulin resistance (reviewed in ). Subsequent studies have supported these proposals, including observations that OGT overexpression leads to insulin resistance and hyperglycaemia in mouse studies [27,28]. As with any area of scientific research, these different approaches all have benefits and all suffer from drawbacks. The loss, or gain, of crucial protein–protein interactions is clearly a drawback of knockout/overexpression approaches, whereas the promiscuity of potential ‘chemical genetic’ tools results in complex phenotypes (as is well-documented in the kinase field, e.g. ). Given the known drawbacks with many of the chemical agents used to probe the O-GlcNAc modification, we applied structural and mechanistic insights into the O-GlcNAc enzymes in order to develop less-ambiguous chemical probes.
The O-GlcNAc enzymes and their exploitation
The drive for more mechanistically based and specific inhibitors led us to collaborate extensively with the Vocadlo group at Simon Fraser University. In the present review, we consider how the resultant ‘retrobiologic’ approach (Figure 1b), whose thought processes – splitting a target experiment down into necessary precursor experiments – are somewhat analogous to retrosynthesis in organic chemistry, is beginning to shed light on the three-dimensional aspects of the O-GlcNAc-modifying enzymes and how this work dovetails with the design and application of small-molecule enzyme inhibitors as probes of the linkage of O-GlcNAc to disease. Aspects of this work are outlined below.
The three-dimensional structure of human TPR (tetratricopeptide repeat) domains and an OGT homologue
The human OGT enzyme is a large (>1000 residues) protein composed of two main regions: an N-terminal TPR domain and a C-terminal GT (glycosyltransferase) domain. At the N-terminal side, OGT displays several repetitions of TPR domains whose length varies according to the OGT localization . In humans, the longest OGT isoform (110 kDa) is found in the nucleus and cytoplasm, whereas a shorter version (103 kDa) localizes to the mitochondria. The number of TPRs correlates with the ability of OGT to glycosylate protein substrates [30,31]. The removal of 2.5 TPRs at the N-terminus of OGT reduces the processing of protein substrates, whereas a further removal of up to five TPRs completely abolishes the activity on proteins, leaving the OGT able to process only short peptide substrates. OGT-deletion mutants without any TPR domains still modify short peptides . The crystal structure of the first 11.5 TPRs is available for the human OGT . The TPR domains are arranged in a superhelical architecture that displays a conserved asparagine ladder that, by similarity with importin α  and β-catenin , suggest how peptide substrates could be recognized without restrictions on their amino acid sequences. At the time of writing, there are no structures of these TPR domains in complex with protein or peptide substrates.
The C-terminal domain of OGT contains the GT domains. Currently, there are no structural data of full-length eukaryotic OGTs, but the availability of the structure of a bacterial homologue from Xanthomonas campestris (XcOGT) solved by two groups [32,36] provides insights into the architecture of OGT (Figure 2a). The first TPRs at the N-terminus of XcOGT follow the same arrangement seen in the human TPR superhelix forming a 260° superhelical turn. The TPR helices in the interface with the GT domain depart from the canonical TPR structure to form a large contact area of approx. 1500 Å2 (1 Å=0.1 nm) with the GT domains. This interaction anchors the TPR superhelix to the entrance of the GT catalytic site in a rigid way and is in this respect different from other TPR-containing enzymes that show a flexible linkage between the TPR and the catalytic subunit . The structure of XcOGT together with the human OGT TPR structure allows us to compose a model of a full-length OGT in which the TPR superhelix protrudes 130 Å from the GT domains and its central axis points directly to the interface of the GT domains where the glycosyltransfer reaction occurs. The crevice leading to the active site is approx. 17 Å wide and 25 Å deep, and these dimensions may accommodate protein substrates in an extended conformation or with limited secondary-structure elements. The overall length of the TPR superhelix provides extensive contact area for the formation of complexes between OGT and various other proteins that may impart additional specificity to the hundreds of known OGT substrates.
The GT domains of OGT are composed of two β/α/β ‘Rossmann’ folds typical of the nucleotide-sugar-dependent glycosyltransferases of the GT-B class with the OGT enzymes classified in the family GT-41 [38,39]. Catalysis in this glycosyltransferase architecture occurs at the interface between the GT domains with the N-terminal GT domain in charge of binding the glycosyl acceptor, while the C-terminal domain binds the nucleotide-sugar donor. The structure of XcOGT was obtained in the presence of UDP allowing the location of the substrate-binding site. The substrate-binding site regions of XcOGT and human OGT share 36% sequence identity and, with the caveats that the donor and acceptor for XcOGT are currently unknown, the XcOGT structure combined with sequence alignment information allowed us to probe the relevance of various human OGT residues for catalysis in the glycosyltransferase-inverting mechanism . In contrast with the progress seen in the design of enzymatic inhibitors for OGA (described below), work on inhibitors of OGT is still scarce (with some notable exceptions, including ) a trend that partially reflects the traditional lag of the structural work on glycosyltransferases with respect to that on glycoside hydrolases .
The three-dimensional structures of O-GlcNAc hydrolase homologues
In mammalian systems, only one hydrolytic enzyme, the O-GlcNAc hydrolase OGA (also known as hexosaminidase C) is responsible for the removal of O-GlcNAc linked to serine and threonine residues. The human OGA enzyme is a large (>900 residues) protein that contains in its N-terminal portion the catalytic glycoside hydrolase domain and in its C-terminus a putative histone acetyltransferase domain whose functional characterization remains elusive [41,42]. The OGA glycoside hydrolase domain is found in family GH84 of the classical (indeed the original papers are now Biochemical Journal ‘citation classics’ ) Henrissat ‘CAZy’ classification  of carbohydrate-active enzymes. Similar to the OGT situation and despite several years of study, the three-dimensional structure of a mammalian OGA has not yet been reported. Conveniently for structural studies, family GH84 is also populated by bacterial enzymes, many from human symbionts and pathogens that show high sequence similarity to the human enzyme. Thus, in 2006, two groups were able to report the three-dimensional structures of OGA homologues: NagJ from Clostridium perfringens  and BtGH84 from (Bacteroides thetaiotaomicron GH84)  (Figure 2b).
Both bacterial OGA homologues display a multi-domain structure in which the first two domains are homologous with the human enzyme, but in which the bacterial C-terminal modules are both different from each other and also not present in the human OGA. The two conserved modules consist of an N-terminal predominantly β-sheet domain in which a six-stranded antiparallel β-sheet that also possesses two α-helices that abut the subsequent catalytic domain. This β-sheet domain is similar to the equivalent domain in both human and bacterial hexosaminidases from family GH20, exemplified by HexB (hexosaminidase B) and chitobiase respectively. The next domain, a β/α-barrel domain, contains the catalytic apparatus and is, again, highly similar to hexosaminidases from family GH20. Furthermore, the catalytic domain shows active centre similarity to a ‘superfamily’ of enzymes including GH18, GH20, GH25, GH56, GH84 and GH85 (recently reviewed in ) that together harness an unusual ‘neighbouring-group participation’ catalytic mechanism  (described below). The key players in this mechanism, throughout the superfamily, are two carboxylates (GH85 enzymes are an exception) and in the case of BtGH84 these residues are Asp242 and Asp243.
Reaction mechanism: probing the reaction co-ordinate
When considering enzyme inhibition strategies, an understanding of chemical mechanism is fundamental. Ever since Pauling's concept  of enzyme inhibition through mimicry of the reaction transition-state (or as Pauling termed it the “strained…activated complex” ) the concept of studying enzyme mechanism with a view to designing enzyme inhibitors has always been a main thread in enzymology. In the context of the O-GlcNAc hydrolase, the key stereochemical feature is that the enzyme acts with net retention of anomeric configuration. Although most glycoside hydrolases that perform catalysis with net retention of anomeric configuration do so through a mechanism involving the formation and subsequent breakdown of a covalent glycosyl-enzyme intermediate (reviewed in ) an alternative mechanism occasionally pertains for enzymes acting on 2-N-acetyl sugars (such as N-acetylglucosamine here). Where there is a 2-N-acetyl substituent, it is possible to envisage a neighbouring-group mechanism in which catalysis is still performed via the formation of a covalent intermediate, but this intermediate is not covalently bound to the enzyme, but, through the nucleophilic attack of the N-acetyl carbonyl group, a bicyclic oxazoline (or oxazolinium-ion) non-covalently bound intermediate is formed (Figure 3a). A key experiment, and one which preceded the three-dimensional structures described above, was to determine whether catalysis by OGA involved the formation of a covalent glycosyl-enzyme intermediate or was instead performed via an oxazoline. Such mechanistic insight is essential if one is to consider the Pauling approach of designing high-affinity and specific enzyme inhibitors based upon reaction mechanism. In 2005, it was shown, through classical physical organic enzymology including a ‘Taft’ analysis in which fluoroacetyl substrates, with increasing numbers of fluorine substituents are used to probe the participation of the adjacent carbonyl group in catalysis, that the OGA mechanism involved the formation of the oxazoline intermediate through nucleophilic attack of the N-acetyl carbonyl group to the anomeric centre (Figure 3a) . Such a mechanism demands two catalytic orchestrators. The first cardinal residue is a catalytic acid/base which first protonates the poor leaving group to facilitate departure and then subsequently acts as a base to activate an incoming water molecule for breakdown of the intermediate ; in the case of BtGH84, this residue is Asp243. The second residue aids formation of the intermediate probably through deprotonation of the incipient oxazolinium ion, through close interaction with the N-acetyl amide hydrogen. In the case of BtGH84, this residue is Asp242. Armed with knowledge of a neighbouring-group reaction mechanism through an oxazoline/oxazolinum ion intermediate, and knowledge of the identity and roles of the catalytic constellation, it became possible both to conceive experiments with which to analyse the conformational aspects of catalysis along the reaction co-ordinate and also to design and apply inhibitors based upon the mechanism. Examples of both are given below.
Reaction mechanism and its conformational analysis
In order to study the ‘Michaelis’ enzyme–substrate (ES) complex of BtGH84, we took advantage of the insight gleaned from the initial Taft analyses, above, and more recent multi-dimensional linear free energy relationships . Thus, by using the combination of a difluoroacetyl substrate with a comparatively poor aryl leaving group, we were able to trap a Michaelis ES complex for BtGH84 . The structure (Figure 3b) showed distortion of the pyranoside ring to an approximate 1S3/1,4B conformation (as observed for the mechanistically analogous GH20 chitobiase and GH18 chitinases previously; reviewed in ). Such a conformation allows in-line nucleophilic attack of the N-acetyl carbonyl group with an axial leaving group conformation (although these days, we might consider this more of an ‘electrophilic migration’ of the anomeric carbon rather than a classical nucleophilic attack at this centre). Through such studies, an ES complex is conformationally defined. The study of the oxazoline intermediate demands other approaches, not least because with natural substrates on wild-type enzyme, the formation of the intermediate is very slow compared with its breakdown; its fleeting existence proving too challenging to study.
The oxazoline intermediate was therefore accessed with two different strategies, both of which were dependent upon manipulating the rates of formation and breakdown of the intermediate such that the enzyme-bound intermediate (EI) species is sufficiently long-lived to facilitate its structural observation. In the first approach, a 5-fluoro-GlcNAc fluoride substrate was used. The rationale (based on years of successful work in the Withers laboratory; see  for a review) was that the 5-fluoro substituent would inductively destabilize the transition states for the formation and the breakdown of the intermediate, but that the effect on the formation would be (partially) mitigated through the use of a good chemical leaving group (here F−). Thus the intermediate forms rapidly, but is only slowly broken down. The rate of breakdown could be further decreased through the use of a general acid/base variant (here D243N). This combined approach indeed allowed formation and observation of a stable 5-fluoro-oxazoline intermediate .
The second strategy employed by He et al.  allowed formation of an unmodified oxazoline intermediate. Here, inspiration was provided by the 2009 structure determinations of several GH85 enzymes. GH85 is a family of N-glycanases that remove N-linked glycans through cleavage of the chitobiosyl core. Two facets of these enzymes are pertinent: they do not possess the classical twin carboxylate signature; instead the oxazoline-interacting residue is an asparagine as opposed to aspartate or glutamate and also these enzymes have found considerably usage in organic synthesis using oxazoline donors which appear to be stable enough to facilitate transglycosylation as opposed to hydrolysis. We thus made the equivalent D242N enzyme variant and were gratified to observe rapid formation of the oxazoline (Figure 3c) using an aryl glycoside substrate coupled to a slow breakdown whose ‘burst-phase’ kinetics informed crystal freezing and X-ray data collection. Together, the Michaelis complex in 1S3 conformation coupled to an oxazoline intermediate in 4C1 chair strongly points to a classical ‘electrophilic migration’ of the anomeric carbon along the reaction co-ordinate, consistent with past physical organic approaches to transition-state poise .
Historically, much of the work on the O-GlcNAc modification has used small-molecule chemical probes to inhibit the hydrolase in vivo and perturb O-GlcNAc levels. Two compounds have received much attention: PUGNAc (initially developed as a generic hexosaminidase inhibitor by Vasella and colleagues ) and STZ (streptozotocin) (itself a relatively weak OGA inhibitor ). The problems associated with these compounds are now well-documented (recently reviewed in ): STZ causes nitrous oxide release, DNA alkylation and cell death, whereas PUGNAc, the ‘workhorse’ of O-GlcNAc study, is a known promiscuous inhibitor that is not even specific for α- compared with β-glycosidases , let alone specific for the OGA, and which therefore probably causes complex phenotypes in which the off-target effects might be difficult to dissect from the OGA inhibition. The unusual neighbouring-group mechanism of the OGA, discussed above, does at least open up new avenues for the selective inhibition of this enzyme.
Dating back to Spencer Knapp's work in the 1990s , there has been a classical approach to the specific inhibition of those enzymes that use an oxazoline intermediate; the thiazoline. Thus Vocadlo and colleagues were able to synthesize the GlcNAc-thiazoline (Figure 2d) and show that it was a tight-binding competitive inhibitor (Ki ~70 nM) of the human OGA . The beauty of the thiazoline approach is that by virtue of its resemblance to the unusual intermediate (and the transition state for its formation ), the compound immediately provides considerable specificity for those enzymes that use the neighbouring-group reaction, thus immediately evading some of the problems associated with more mechanistically agnostic and hence promiscuous inhibitors. Indeed, we were subsequently able to solve the three-dimensional structure of BtGH84 in complex with ‘GlcNAc-thiazoline’ where it binds, as expected, in the active centre with the two catalytic aspartate residues, 242 and 243, positioned exactly as would be expected for their respective roles in catalysis (see above and Figures 2 and 3) with Asp243 acting as the general acid/base and Asp242 acting through its interaction with the N-acetyl nitrogen.
What is immediately also obvious from this BtGH84 complex is that there is a large cavity below the N-acetyl pocket (Figure 2c) whose volume is not exploited by a simple acetyl moiety. This pocket offered the potential for the introduction of extended acyl chains to bring even more specificity for OGA over other, structurally related, GH20 enzymes. This may prove important for human HexA (hexosaminidase A) and HexB in the lysosome, enzymes found in CAZy family GH20, are also hexosaminidases that exploit the neighbouring-group mechanism. Thus, if one wants to also improve inhibitor specificity for OGA over HexA/HexB (whose inhibition certainly changes ganglioside levels and may have other effects), exploitation of this OGA-specific cavity (the acyl group is much more tightly packed in HexA and B [59,60]) is an exciting prospect.
Indeed, before the three-dimensional structural work, Macauley et al.  had synthesized a panel of inhibitors, both on the PUGNAc and thiazoline scaffolds, and shown that chain-extended thiazolines, such as N-butylthiazoline (Figure 2d), retained tight binding, but gained considerable specificity for OGA over HexA/B. Three-dimensional structural analysis of the N-butylthiazoline complex confirmed that the N-butyl group indeed protrudes down into the activity revealed by the original structure determinations . Thus, equipped with a potent and specific inhibitor, the Vocadlo group set out to probe some of the proposed links of O-GlcNAc to disease, notably Type 2 diabetes and neurodegeneration, discussed below. At this point, it is pertinent to stress that although the present review outlines the work performed in the authors' and collaborators' laboratories, synergistic work on the exploitation of the O-GlcNAcase cavity, by the van Aalten group in Dundee, has led to the development of a series of compounds, termed GlcNAcstatins [61,62]. These GlcNAcstatins are also a group of compounds which couple elements of transition-state mimicry to modified N-acetyl groups and which may prove useful in vivo as probes to study O-GlcNAc in future.
Applications of thiazoline-derived inhibitors in neurodegeneration and Type 2 diabetes research
Macauley et al.  applied the N-butylthiazoline to study the proposed link of elevated O-GlcNAc to insulin resistance. If one recalls the original experiments of Vosseler et al. , 3T3-L1 adipocyte cells, the standard model cell line for diabetes research with respect to insulin resistance of peripheral tissues, became partially insulin resistant after treatment with the inhibitor PUGNAc. Given that PUGNAc increases O-GlcNAc levels through inhibition of the OGA, it was a logical conceptual leap to suggest that elevated O-GlcNAc was causally linked to insulin resistance in this cell line. Decreased insulin-dependent activation of Akt, as measured by its phosphorylation, in these PUGNAc-treated cells was also indicative of an impaired insulin signalling pathway. To our surprise, although the effects of PUGNAc are recapitulated, N-butylthiazoline-treated 3T3-L1 adipocytes show no signs of insulin resistance (over a range of insulin concentrations), yet their O-GlcNAc levels are increased to the same levels as with PUGNAc-treated cells and the kinetics of the change in O-GlcNAc levels is likewise indistinguishable between PUGNAc and N-butylthiazoline-treated cells. Furthermore, Akt phosphorylation is unimpaired in the N-butylthiazoline-treated cells. This discrepancy between the effects of PUGNAc and N-butylthiazoline is significant. The data show that elevated O-GlcNAc is not correlated with insulin resistance in these experiments. The simplest explanation for this inconsistency is that one, or more, of the off-target effects of PUGNAc are responsible for the insulin-resistant phenotype and that experiments that use PUGNAc to elevate O-GlcNAc levels should be interpreted with this possibility in mind. Alternatively, one may consider that N-butylthiazoline itself has an as yet unknown target which both coincidentally and completely serves to ‘cure’ cells of an insulin-resistant phenotype; arguing against this position is the observation that (in the context of O-GlcNAc/phosphorylation ‘cross-talk’ at least) cells exposed to both PUGNac and GlcNAc-thiazoline show a PUGNAc phenotype . Furthermore, were N-butylthiazoline able to reverse the diabetic phenotype, this would have to occur upstream of Akt activation as this process was unimpaired in the N-butylthiazoline-treated 3T3-L1 cells . Clearly more studies will be required to address this conundrum.
During the evolution of this work, it had become apparent that better, potentially more specific, inhibitors could be designed and exploited in vivo. One such compound, termed ‘Thiamet-G’ (Figure 2d), harnesses an isothiourea moiety, coupled to an extended acyl chain, to yield a ~20 nM inhibitor with approx. 40000-fold specificity for human OGA over HexB . The rationale behind this compound is that by virtue of its net positive charge, it should be better able to interact with Asp242 in the active centre of the GH84 enzymes than a regular thiazoline. Yuzwa et al.  have therefore been able to use Thiamet-G in animal studies to probe an additional aspect of O-GlcNAc cell biology; that of limiting tau phosphorylation in vivo with a view to evaluating the potential beneficial effect in neurodegeneration.
Tau is a microtubule-associated protein that exhibits several different post-translational modifications including phosphorylation and O-GlcNAcylation. Hyperphosphorylation of tau is widely viewed as being a requisite for the formation of neurofibrillary tangles in vivo, which is a signature feature of a family of neurodegenerative diseases collectively termed ‘tauopathies’ (reviewed in ). Past work in vitro had shown that the levels of phosphorylation on tau were indeed reciprocal to levels of O-GlcNAcylation, suggesting that the modification with O-GlcNAc may influence tau phosphorylation (reviewed recently in [67,68]). Additionally, glucose metabolism is frequently impaired in Alzheimer's disease patients , providing an additional link between decreased O-GlcNAc levels and hyperphophorylation. Furthermore, the OGT enzyme resides at a chromosomal locus in humans that is linked to increased susceptibility to Alzheimer's disease , again suggesting there might be a causal link between decreased O-GlcNAc and neurodegeneration. Together, these observations suggest that modification of O-GlcNAc levels may be a way to influence tau phosphorylation and, in 2004, Gong and colleagues showed that PUGNAc could be used to increase tau O-GlcNAcylation and hence reduce tau phosphorylaton in tissue sections .
Unfortunately, in addition to the potential problems with PUGNAc described above, it is also difficult to synthesize in quantities required for extensive live-animal studies and it is also believed not to cross the blood–brain barrier. Yuzwa et al.  used Thiamet-G both in PC-12 cells (a neuronal cell model) and in rat studies to perturb the O-GlcNAcylation, and thus phosphorylation, of tau. In the PC-12 cell line, Thiamet-G-treated cells show significant increases (approx. 7-fold) in O-GlcNAc levels with an EC50 of approx. 30 nm. Furthermore, phosphorylation of tau, at physiologically relevant sites (Thr231 and Ser396) implicated in pathological hyperphosphorylation is reduced 2.7- and 2.3-fold. Armed with these encouraging data, Thiamet-G was also used in a rat model, where it was found to be orally bioavailable and also to perturb tau O-GlcNAcylation of tau in the living brain, indicative of an ability to cross the blood–brain barrier. Both experiments on total brain homogenates and also immunohistochemistry using antibodies against pSer396, pThr231 and O-GlcNAc clearly shows a reduction in tau phosphorylation at these sites and a corresponding increase in O-GlcNAc levels. Thiamet-G and related compounds are thus proving to be exciting chemical probes for neurochemistry and may ultimately inspire therapeutic avenues for the treatment of Alzheimer's disease.
The diverse roles for the O-GlcNAc modification in mammalian (indeed also in plant; see  for review) systems only continue to increase. Applications of small-molecule inhibitors  for the modification of tau phosphorylation is clearly a major area, and, in the diabetes context, it will be very interesting to see whether other O-GlcNAcase inhibitors will continue to question the link between elevated O-GlcNAc levels and insulin resistance and whether these effects manifest themselves in animal models. Other areas of future research in the O-GlcNAc field will probably be directed to clarify the role of this modification in the regulation of gene transcription. Currently, many of the known OGT targets are master regulator transcription factors that control essential gene-expression programmes in turn in charge of orchestrating, for example, the onset of immune responses  or stem cell differentiation . The recent discovery that OGT is part of the polycomb group of proteins in Drosophila adds to the significance of this modification on the regulation of the cell fate [74,75]. At present, it is not unreasonable to think of the O-GlcNAc modification as an important link that informs or at least might modulate the transcriptional response according to the metabolic status of the cell. The potential for the application of small-molecule probes, based upon a foundation of three-dimensional structure and chemical mechanism in synergy with other techniques, is clearly set to rise.
GlaxoSmithKline Award Lecture Delivered at Royal Holloway, University of London, on 30 March 2010 as part of the Structural Glycobiology and Human Health meeting Gideon Davies
The work highlighted in the present review is the result of a fruitful collaboration with Professor David Vocadlo and his group at Simon Fraser University, Burnaby, British Columbia, Canada. Professor Vocadlo is thanked for stimulating scientific and philosophical discussions.
We thank the Biotechnology and Biological Sciences Research Council for funding. G.J.D. is a Royal Society/Wolfson Research Merit award recipient.