The catalytic mechanism of retaining glycosyltransferases (ret-GTs) remains a controversial issue in glycobiology. By analogy to the well-established mechanism of retaining glycosidases, it was first suggested that ret-GTs follow a double-displacement mechanism. However, only family 6 GTs exhibit a putative nucleophile protein residue properly located in the active site to participate in catalysis, prompting some authors to suggest an unusual single-displacement mechanism [named as front-face or SNi (substitution nucleophilic internal)-like]. This mechanism has now received strong support, from both experiment and theory, for several GT families except family 6, for which a double-displacement reaction is predicted. In the last few years, we have uncovered the molecular mechanisms of several retaining GTs by means of quantum mechanics/molecular mechanics (QM/MM) metadynamics simulations, which we overview in the present work.
Glycosyltransferase mechanisms: enzymatic formation of glycosidic bonds
The large amount of complex carbohydrates that govern many cellular functions requires the action of a diverse range of selective enzymes for their biosynthesis . This is the role of glycosyltransferases (GTs) and transglycosylases (TGs), which are ubiquitous in every living organism. They are responsible for protein glycosylation, the most recurrent post-translational modification occurring in nature for cell recognition and signalling. GTs catalyse the formation of glycosidic linkages by the transfer of a saccharide, typically a monosaccharide, from an activated donor substrate to an acceptor substrate . GT alterations cause several diseases such as infection, inflammation and either normal or abnormal cellular developments [3,4]. One of the major challenges for the rational design of specific and potent drugs/inhibitors for this class of enzymes [2,5,6] is unravelling their detailed reaction mechanisms, which is still not well understood.
The transfer of the glycosyl group can take place either with inversion or retention of the anomeric carbon stereochemistry with respect to the donor sugar (Figure 1) . The mechanism of inverting GTs follows a single-displacement SN2 reaction (Figure 1a) in which a general base catalyst increases the nucleophility of the attacking group. This mechanism is analogous to the well-characterized mechanistic strategies used by inverting glycoside hydrolases (GHs) to catalyse the cleavage of glycosidic bonds . However, the mechanism of retaining GTs (ret-GTs) has been controversial for some years, with several mechanisms being proposed .
Possible reaction mechanisms for enzymatic glycosyl transfer proposed in the literature
It was first suggested that the enzyme actively participates in the reaction by covalently binding to the donor molecule while it is being transferred to the acceptor (Figure 1b) . This is in analogy with the double-displacement mechanism of retaining GHs (ret-GHs), which form a covalent intermediate, with clear evidences not only from experiments  but also theory [11–13]. Despite attempts to study trapped glycosyl–enzyme intermediates by X-ray crystallography , there are evidences of the formation of covalent adducts in experimental probes (MS  and chemical rescue ) using GT6 enzyme mutants (cysteine or alanine mutants of the possible nucleophile). However, it is not clear whether these adducts belong to a catalytically competent reaction path . On the other hand, successful chemical rescue is a necessary but not sufficient proof for the formation of a glycosyl–enzyme intermediate in the WT enzyme . Therefore, although encouraging, these experiments also raise the question whether the mutation could affect the molecular mechanism. More importantly, only family GT6 enzymes have a nucleophilic protein residue (glutamate or aspartate) well located and well oriented to act as nucleophile in a double displacement reaction. The non-conserved architecture in the region of the active site where the catalytic nucleophile would have to be positioned remains puzzling.
The lack of clear experimental evidence for the double-displacement mechanism prompted some authors to suggest an alternative mechanism in which the reaction proceeds via a front-side single-displacement, taking place on a ‘single face’ of the sugar (Figures 1c and 1d) . This unusual reaction, termed as front-face or front-side, would hold for GTs lacking a putative nucleophile residue, such as glycogen phosphorylase, a ret-GT that does not use nucleotide sugars (non-Leloir GTs) , lipopolysaccharyl α-galactosyltransferase (LgtC) and trehalose-6-phosphate synthase (OtsA).
Antecedents of the front-face reaction
The front-face reaction is uncommon in comparison with other nucleophilic substitutions, but it has important chemical precedents. This type of reaction was early described to explain the net retention of configuration in the decomposition of alkyl chlorosulfites . The reaction was termed SNi (substitution nucleophilic internal), as the nucleophile comes from decomposition of the leaving group and it is held on the same face as the leaving group as an ion-pair (Figure 2).
The SNi reaction for the decomposition of alkyl chlorosulfites
The front-face reaction has also been described for glycosyl transfer in solution . Sinnott and Jencks  proposed this reaction for the solvolysis of glycosyl fluorides in a mixture of ethanol and chloroethanol and their predictions have been previously confirmed by measurements of the kinetic isotope effect . Several structural studies have also described glycosyl transfer in ret-GTs in terms of the front-face reaction. In fact, structures of ternary complexes obtained by superposition of binary complexes show that the donor and acceptor are properly oriented for the front-face reaction to occur . This is the case of the galactosyltransferase LgtC from Neisseria meningitides , retaining phosphorylases (non-Leloir GTs) and ret-GTs lacking a putative nucleophile on the β-side of the donor sugar . Information on the fine details of the molecular mechanism is not yet available from experimental probes, but most authors have argued that the reaction needs to take place via a short-lived oxocarbenium species  (Figure 1d), with a lifetime longer than a bond vibration (a condition often invoked to be considered as an intermediate ). Nevertheless, the front-face mechanism, in which two covalent bonds are formed and broken, respectively, in the same region of space, continued generating debate. Is the front-face mechanism feasible? Does it take place on a single step or is it a stepwise reaction (i.e. are the two bonds formed and broken simultaneously?)? What is the base that deprotonates the phosphate leaving group? All these questions have now been solved to a large extent by theory and simulation.
Early theoretical analyses of enzymatic glycosyl transfer with retention of configuration
The first theoretical analysis of an enzymatic glycosyl transfer reaction with retention of configuration was performed in 2004 by I. Tvaroska , considering small models (without the protein environment) of the active site. The electronic structure was described with density functional theory (DFT) using a small-size basis set . The front-face reaction was found to be feasible (i.e. involving a low energy barrier) when the nucleophile and the leaving group were constrained on the same face of the sugar. Surprisingly, the reaction was found to be one-step, i.e. a concerted reaction with only one transition state (TS). A subsequent study with similar models predicted that the double-displacement would also be feasible when a nucleophile is placed near the anomeric carbon of the donor sugar . Even though these results were promising, the simplicity of the models (lack of enzyme environment, heavy constraints on atomic motion, small basis sets used, large corrections applied on energy barriers and neglect of the room temperature dynamics of the atoms) precluded establishing firm mechanistic conclusions and the glycosyl transfer reaction continued being controversial over the years .
Recent theoretical analyses of the front-face reaction by quantum chemical/molecular mechanics methods
Complementing the previous studies and going one step forward in accuracy and predictive power, we analysed the glycosyl transfer reaction using a full model of the enzyme, embedded in water, by applying an efficient quantum mechanics/molecular mechanics (QM/MM) method  (i.e. the active site is treated quantum mechanically and the rest of the system by molecular mechanics). The electronic structure was described by DFT with the Perdew–Burke–Ernzerhof (PBE) functional , a choice that gave good results in the study of GH mechanisms . In addition, we took into account the dynamics of the enzyme , which is known to be crucial in atomistic modelling of enzymes [28–30]. In particular, we used ab initio MD  to move the atoms at room temperature and metadynamics  to drive the chemical reaction.
Metadynamics  is a novel approach (based on MD) that allows exploration of the free-energy landscape (FEL) associated with slow motions of the system or rare events (e.g. going from reactants to products in a chemical reaction) as a function of a limited set of collective variables (CVs). The CVs are functions of the coordinates of the atoms, such as distances and angles and need to be chosen a priori . In a nutshell, the method is based in adding a Gaussian potential to the real energy landscape of the system at given time intervals to surmount the energy barrier of the process of interest. In metadynamics, the system escapes the free energy minimum through the lowest energy pathway and information of the final state (e.g. the products of the chemical reaction) is not required. Ultimately, the method can be exploited not only for accelerating rare events but also for mapping the FEL, which can be estimated, after a sufficient time, as the negative of the sum of the added biasing potential [33,34] (i.e. we obtain the FEL and reaction mechanism). It should be noted that metadynamics, as well as other free-energy-based approaches , is different (and computationally more costly) than the standard (static) approach based on geometry optimizations along a pre-defined reaction co-ordinate, in which temperature effects on atomic motion and entropy are neglected and potential energy instead of free energy is obtained. The two different approaches have been described as dynamic versus static QM/MM approaches in the literature [24,26]. We summarize below our findings for three different ret-GTs: trehalose-6-phosphate synthase (OtsA) , polypeptide GalNA c-transferase 2 (GalNAc-T2)  and α-1,3-galactosyltransferase (α3GalT)  using metadynamics.
The first ret-GT modelled with the QM/MM approach was OtsA, a family 20 GT that synthesizes trehalose-6P from UDP-Glc and Glc-6P (Figure 3), being responsible for the penultimate step in the synthesis of trehalose. In the absence of crystal structure of the ternary complex for OtsA, we used the structure of a complex with a natural product derivative obtained in the group of G. J. Davies , which we manually converted into the reaction products (trehalose-6-phosphate+UDP). Subsequently, the Michaelis complex (UDP-Glc and Glc-6P) was modelled by breaking the glycosidic bond and forming the phosphate-sugar bond in two consecutive metadynamics simulations . The ternary complex obtained (Figure 3a) was fully stable under MD (classical or QM/MM) and in agreement with the X-ray structures of the binary complexes of the enzyme with the donor and the acceptor .
Reaction mechanism for glycosyl transfer catalyzed by OtsA
As OtsA does not exhibit a putative nucleophile on the donor sugar β face, our investigation was focused on the front-face reaction for glycosyl transfer reaction. To model the reaction, we used metadynamics with two collective variables that take into account all covalent bonds that are being formed and broken during the reaction. We expected that the a priori unknown reaction co-ordinate would be a combination of these distances. One collective variable measures the degree of glycoside transfer (i.e. the breaking of the phosphate-sugar and the formation of the glycosidic bond), whereas the second collective variable measures the degree of proton transfer. As defined, these collective variables do not force any order of the events or reaction mechanism (proton transfer or leaving group departure). They do not impose either a concerted mechanism (with a unique TS) or a mechanism in several steps. During the simulation, the whole enzyme moves at room temperature so it accommodates to the changes along the reaction co-ordinate.
The FEL obtained in the simulation (Figure 3b), with respect to the two collective variables, shows two deep energy minima (reactants and the products), with a free energy barrier [23 kcal/mol (1 kcal ≡ 4184 J). This value is compatible with the experimental rate constants for glycosyl transfer [15,16,41,42], indicating that the reaction is feasible. Importantly, the FEL contains a region in which the FEL is rather flat, except for a small free energy minimum (∼3 kcal/mol more stable than the surrounding states along the pathway), indicating that the reaction is stepwise.
Reaction mechanism for glycosyl transfer catalyzed by GalNAc-T2
Figure 3(c) illustrates the OtsA reaction pathway. Most of the reaction energy is being invested to cleave the phosphate-sugar bond, leading to the formation of an intermediate in which there is a clear ion-pair separation (a negative phosphate and an oxocarbenium ion). At the same time, the OH of the acceptor changes hydrogen bond partner, from the negative oxygen atom of the phosphate leaving group to the oxygen of the covalent bond being broken. Therefore, the O1′-H moves to stabilize the negative charge developed on this leaving oxygen, driving the acceptor molecule closer to the donor. This switch of the O1′-H…O interaction favours the covalent bond cleavage, being probably the driving force of the reaction. Noteworthy, there is a part of the reaction pathway (shaded in grey in the distance analysis of Figure 3d) in which both the phosphate oxygen and the acceptor oxygen are quite separated from the anomeric carbon. Any structure in this region represents an oxocarbenium ion-like species. The lifetime of this intermediate is very short (indeed, separate ab initio QM/MM MD simulations starting from this species show that it relaxes to reactants or products in a few picoseconds) and, as such, it would be very difficult to trap experimentally. In addition, these structures show the maximum development of positive charge at the anomeric carbon (Figure 3e). The lifetime of the reaction intermediate is short, but long enough to enable active site re-organization: the phosphate group leaves and the acceptor molecule approaches. Eventually collapsing with the oxocarbenium ion and forming the new glycosidic bond. Finally, the phosphate leaving group withdraws the proton to reach the products. These simulations also demonstrate that the phosphate group acts as a base enhancing the nucleophilicity of the incoming acceptor molecule. Noteworthy, proton transfer can occur before or after glycosidic bond formation (in the last case, an oxonium ion, corresponding to the local minima on the upper-right side of the FEL of Figure 3b, is formed). The resulting mechanism is in very good agreement with the data obtained in a simultaneous experimental study of the OtsA mechanism in which kinetic isotope effects were measured . What the QM/MM metadynamics study added to the experimental description were the fine details of the molecular mechanism and the demonstration that that the reaction is stepwise, with the formation of a short-lived oxocarbenium ion-like intermediate.
Subsequent theoretical studies of glycosyl transfer were also consistent with the front-face reaction mechanism, with different outcomes concerning the details of the reaction. A stepwise reaction was found for glycosyl transfer in solution, involving an ion-pair intermediate. Similar results were also obtained for α-1-2-mannosyltransferase Kre2p/Mnt1p, the reaction was predicted as a stepwise SNi-like nucleophilic substitution via oxocarbenium ion intermediate . A recent structural, chemical and theoretical study found a timid ion-pair intermediate for glucosyl-3-phosphoglycerate synthase (GpgS) . In contrast, a concerted reaction (one single step and TS) was found for LgtC  using the static QM/MM approach.
A recent analysis of a number of ret-GT X-ray structures concludes that the front-face one-step reaction, (Figure 1c), is the only mechanism in ret-GTs . However, it seems unlikely that the fine details of the TS or intermediates of the glycosyl transfer reaction could be inferred from structural data of Michaelis complexes alone.
Whether the concerted one-step reaction is possible or not in ret-GTs has not been fully clarified, but it has been argued to be either orbitally forbidden  or sterically and entropically unfavourable . As mentioned above, most authors support that the front-face reaction is stepwise in ret-GTs, as evidenced in the most sophisticated QM/MM studies.
Polypeptide GalNAc-transferase 2: protein O-glycosylation
O-GalNAc glycosylation  is by far the most complex and differentially regulated type of protein glycosylation and it may also be the most abundant, with over 80% of all proteins passing through the secretory pathway predicted to be O-glycosylated . One of the key enzyme isoforms controlling human protein O-glycosylation is polypeptide GalNAc-T2. This enzyme is part of a large family of retaining isoenzymes that transfer a GalNAc residue from UDP-GalNAc to serine/threonine side chains and thereby initiate mucin-type or GalNAc-type protein. How GalNAc-Ts target specific sites on proteins and glycoproteins and how they catalyse O-GalNAc transfer is not understood.
We recently uncovered the mechanism of O-glycosylation in GalNAc-T2 by a combination of X-ray crystallography and QM/MM metadynamics . The simulations were initiated from a ternary complex structure obtained with a thio-derivative donor analogue. To reconstruct the natural ternary complex, we replaced the donor sugar S by O and equilibrated the complex by classical MD (Figure 4a). Subsequent QM/MM metadynamics simulations clearly indicated the discrete formation of an oxocarbenium–phosphate ion-pair (2 in Figure 4b) that is further stabilized by the formation of OThr-H…OUDP and NGalNAc-H…OUDP hydrogen bonds (2 in Figure 4c). The formation of such intimate ion-pair is emerging as the common feature of the mechanism of ret-GTs lacking a nucleophile in the sugar β face. Two recent QM/MM studies using the static approach on different starting structures reached similar conclusions [50,51], but in one of them the ion-pair intermediate could not be characterized , thus the reaction was described as a single-step SNi reaction. It is likely that capturing the fine details of the front-face reaction, such as the short-lived oxocarbenium ion-like intermediate requires the explicit treatment of protein dynamics.
α-1,3-glycosyltransferase: a nucleophile-containing ret-GT
Family 6 GTs deserve particular attention as is the only known GT family in which there is a nucleophilic residue in the β-face of the donor sugar that is well oriented for attack on the donor sugar anomeric carbon. To date, only two ret-GTs, α3GalT and blood-group A and B α3GalT (GTA/GTB), both belonging to family 6, have been found to exhibit a nucleophile within the active-site residues (aspartate or glutamate, as in ret-GHs) [2,7]. Therefore, the natural question is whether these GTs would operate via a front-face or a double-displacement mechanism.
To answer the above question, we recently modelled the glycosyl transfer reaction in bovine α3GalT, i.e. the transfer of a galactose (Gal) moiety from UDP-Gal to a lactose acceptor. To build the Michaelis complex, we aligned the structures of binary complexes of the enzyme with either the nucleotide sugar donor (UDP-2F-Gal, PDB code 2VFZ)  or both the nucleotide and the acceptor (UDP+lactose, PDB code 1GWV) . As in the previous cases, we equilibrated the system by classical MD simulation, followed by QM/MM MD, to accommodate all the actors (donor, Glu317 and acceptor) optimally in the binding site.
The ternary complex structure obtained from the simulation (Figure 5a) shows that the side chain of Glu317 (the putative nucleophile) bridges both the donor and the acceptor (Figure 5a). As a consequence, it comes closer to the anomeric carbon [3.45 Å (1 Å=0.1 nm) than it was in the model obtained purely by structural superposition (4.09 Å). Therefore, the ternary complex (the Michaelis complex) is not the direct superposition of binary complexes, but there are small adjustments that might be relevant for the reaction. In view of the inherent dynamics of GTs , these small changes are probably not limited to α3GalT.
Reaction mechanism for glycosyl transfer catalyzed by α3GalT
As shown in Figure 5(a), Glu317 is well positioned for nucleophilic attack on the donor anomeric carbon and, therefore, a SN2-type of reaction (Figure 1b) seems to be feasible. However, a front-face type of reaction (Figures 1c and 1d), cannot be ruled out, as the O3′H of the acceptor interacts with the UDP-Gal donor (specifically, the O3′H group interacts either with the phosphate O atom or the O5 of the Gal, oscillating among the two on a timescale of a few femtoseconds). In fact, the relative position between the donor and the acceptor is similar to what we previously found for OtsA . Therefore, the structure of the ternary complex alone does not give any clue on the type of mechanistic transformation.
To inspect the glycosyl transfer reaction in α3galT, we used QM/MM metadynamics with two CVs corresponding to the main bonds undergoing breaking and formation. In order not to bias the reaction towards the formation of a covalent glycosyl–enzyme intermediate, the nucleophile was not included in the CV definition. Nevertheless, analysis of the FEL (Figure 5b) shows that, once the sugar-phosphate bond starts to break (Figure 5c), the anomeric carbon approaches one of the negative oxygens of the Glu317 carboxylate and bounds to it. Simultaneously, the O3′H bond rotates to establish a strong interaction with the phosphate leaving group. Formation of the glycosyl–enzyme complex (R → 1 → 2) is a dissociative process, in which the cleavage of the C1-OP bond precedes the formation of the C1-OGlu317 bond and involves an energy barrier of 23 kcal/mol, similar to the one obtained for OtsA. As shown in Figure 5(c), the conformation of the donor sugar changes from being a chair (4C1) to a boat (1,4B) on the glycosyl–enzyme intermediate (2). This conformational change, triggered by the nucleophilic substitution, facilitates the subsequent step of the reaction (deglycosylation) by placing the O3′ atom in a suitable position to attack the anomeric carbon. The reaction thus follows a conformational itinerary that favours catalysis, as found in glycosyl hydrolases . To complete the reaction, the C1-OGlu317 bond breaks as the C1-O3′ glycosidic bond forms, in concert with the transfer of the O3′H proton to the leaving phosphate group. This step involves a lower energy barrier (13 kcal/mol) than the cleavage of the C1-OP bond, supporting previous experimental predictions .
To quantify the role of Glu317 in the reaction mechanism, additional simulations were performed by using a model that attenuates the reactivity of this residue, either by preventing its bonding to the sugar donor or eliminating its electrostatic interaction. In both cases, the energy barrier of the reaction raises by more than 8 kcal/mol, highlighting the crucial role of Glu317 in the reactive process. These results agree with the loss of enzyme activity for Glu317 mutants . Altogether, and in contrast with the conclusions of analyses static QM/MM [56,57], our results are only consistent with the double-displacement mechanism for α3GalT.
By means of ab initio QM/MM metadynamics simulations, we have modelled the long controversial front-face reaction in ret-GTs, showing that the reaction is feasible and takes place via formation of a short-lived ion-pair intermediate, for most ret-GT families, i.e. those do not exhibiting a putative nucleophile residue on the β face of donor sugar. In contrast, our simulations are consistent with the formation of a glycosyl–enzyme intermediate for family 6 GTs. An intriguing question rises though, which is why nature has placed a carboxylate residue in the active site of a few GTs so that they undergo a different type of reaction (double-displacement) than most GTs which, presumably follow a front-face mechanism? In our opinion, the two types of mechanisms for GTs are not opposed. Their main difference is the way the enzyme stabilizes the oxocarbenium ion-like species that forms upon cleavage of the donor sugar-phosphate bond. In OtsA (and related GTs), the electrostatic potential of the active site is such that it can stabilize the oxocarbenium ion-like intermediate for a very short time (picoseconds), but long enough for the active site to reorganize and the oxocarbenium-ion species and the acceptor to move one towards the other . In the case of family 6 GTs (α3GalT and blood group galactosyltransferases), the oxocarbenium ion-like species stabilizes by forming a covalent bond with an optimally-located carboxylate side chain. The particular type of donor and acceptor substrates and the regioselectivity of the glycosyl transfer reaction are probably at the basis of these mechanistic variations. In this respect, both modes of operation can be considered as variations of a common mechanism, a two-step reaction via oxocarbenium ion-like TSs that flank an intermediate, either an oxocarbenium ion or covalent glycosyl–enzyme  depending of the fine structure of the active site.
We acknowledge the computer support, technical expertise and assistance provided by the Barcelona Supercomputing Center-Centro Nacional de Supercomputación (BSC-CNS).
This work was supported by the Generalitat de Catalunya [grant number 2014SGR-987]; and the Ministerio de Economía y Competitividad [grant number CTQ2014-55174-P].
density functional theory
substitution nucleophilic internal
Carbohydrate Active Enzymes in Medicine and Biotechnology: Held at University of St Andrews, Fife, Scotland, U.K., 19–21 August 2015.
Present address: Max-Planck Institut für Biophysik, Max-von-Laue-Straße 3 60438 Frankfurt am Main, Germany
Present address: Department of Chemistry, King's College London, Britannia House, 7 Trinity Street, London SE1 1DB, U.K.