Unraveling A4GALT Mechanism and Its Modulation With Adamantyl‐Galactosylceramide Analogues: Advancing Fabry Disease Therapeutic Strategies

ABSTRACT

Fabry disease (FD), one of the most prevalent lysosomal storage disorders in Europe, is caused by mutations in the GLA gene leading to deficient α‐galactosidase A activity with lysosomal accumulation of globotriaosylceramide (Gb3). Enzyme replacement therapy (ERT) and pharmacological chaperone therapy (PCT) are used in the clinic to treat FD but are limited in efficacy, underscoring the need for alternative therapeutic strategies. Inhibiting α‐1,4‐galactosyltransferase (A4GALT), the glycosyltransferase responsible for Gb3 biosynthesis, represents an attractive strategy. Here, we reveal the molecular mechanism of human A4GALT at atomic detail using QM/MM simulations. We reveal a conformational rearrangement involving a 3₁₀‐helix that stabilizes the donor substrate and promotes a front‐face S_Ni‐like catalytic mechanism, in which a short‐lived oxocarbenium‐ion intermediate forms. The simulations informed the synthesis of a panel of glycosylceramide substrate analogues. Among these, AdaGalCer (Ada = adamantyl) proved able to reduce Gb3 production in fibroblasts while simultaneously being converted by A4GALT into the galactosylated product AdaGb2. These results provide a clear path towards inhibiting A4GALT, paving the way for potential new and effective FD therapeutics.

A 3₁₀‐helix‐mediated conformational switch promotes a front‐face S_Ni‐like catalysis by human A4GALT. Mechanism‐guided design identifies AdaGalCer as a selective modulator of globotriaosylceramide (Gb3) biosynthesis, opening a clear route toward new Fabry disease therapeutics.

1. Introduction

Fabry disease (FD, OMIM 301500) is one of the most common lysosomal storage disorders (LSDs), with a worldwide estimated incidence of approximately 1 in 10 000 based on genetic screenings [1, 2]. This X‐linked disease arises from mutations in the GLA gene [3, 4], resulting in reduced α‐galactosidase A (α‐Gal A) activity, typically to less than 30% of normal levels—a threshold generally considered diagnostic for Fabry disease [5, 6, 7]. As a result, patients accumulate glycosphingolipids such as globotriaosylceramide (Gb3, also known as P^k) and galactobiosylceramide (Gb2) in the lysosome and ultimately in plasma and urine [8, 9, 10, 11]. The deacylated form of Gb3, globotriaosylsphingosine (LysoGb3), has proven to be a reliable indicator of disease burden and treatment effectiveness [12]. Blood group P¹, NOR1 and NOR2 glycosphingolipids—all presenting a terminal α‐galactosyl moiety—are also known products of α‐1,4‐galactosyltransferase (A4GALT), with NOR1 and NOR2 produced by the Q211E variant (Figure 1A). The progressive accumulation of these metabolites underlies the multisystemic pathology of FD, which includes neuropathic pain [13], renal failure [14, 15], cardiomyopathy [16, 17], and cerebrovascular complications [18, 19].

FIGURE 1.

(A) Glycosphingolipids generated by A4GALT (in red): Gb2, Gb3, P¹, NOR1 and NOR2, starting from ceramide (in green), and enzymes involved in their synthesis (CGT: ceramide galactosyltransferase, GCS: glucosylceramide synthase, LCS: lactosylceramide synthase, and B3GALNT1: β‐1,3‐N‐acetylgalactosaminyltransferase 1) and hydrolysis (GALC: galactocerebrosidase, GLA: α‐galactosidase A, GBA1: glucocerebrosidase, GLB1: β‐galactosidase, NAGA: α‐N‐acetylgalactosaminidase, and ASAH1: acid ceramidase). (B) Chemical reaction catalyzed by A4GALT using LacCer and GalCer as substrates to generate Gb3 and Gb2, respectively. (C) Double‐displacement mechanism suggested for retaining GTs with a putative nucleophilic residue on the β‐face of the donor sugar (left) versus the front‐face or S_Ni‐like mechanism proposed for retaining GTs lacking a putative active site nucleophile (right).

Current therapeutic strategies for FD include enzyme replacement therapy (ERT) [20], pharmacological chaperone therapy (PCT) [21, 22], and, more recently, substrate reduction therapy (SRT) that targets the biosynthesis of the accumulating metabolites [2, 23]. While ERT with recombinant α‐Gal A can slow disease progression in Fabry patients, efficacy is limited by poor biodistribution and cellular uptake [24, 25]. PCT with migalastat benefits only patients with amenable mutations [26], and glucosylceramide synthase (GCS) inhibitors [27, 28], evaluated as SRT in clinical trials, act only indirectly on Gb3/LysoGb3 levels and perturb other lipids such as lactosylceramide (LacCer).

These limitations highlight the need for more specific strategies [23, 25, 29, 30]. Targeting human A4GALT—the enzyme responsible for Gb3 synthesis by transferring galactose from UDP‐Gal to lactosylceramide (LacCer)—represents a direct and rational approach for substrate reduction. Importantly, genetic ablation of A4GALT prevents Gb3 accumulation and disease‐related nephropathy in patient‐derived organoids [31], and naturally occurring A4GALT deficiency in humans is well tolerated [32, 33], underscoring its promise as a therapeutic target. Yet, despite its biomedical relevance, no bona fide A4GALT inhibitors exist. The absence of structural and mechanistic insight into A4GALT hampers our ability to understand substrate recognition and catalysis, thereby precluding a rational starting point for inhibitor development.

α‐1,4‐Galactosyltransferase A4GALT, also known as Gb3/CD77 synthase or P1/P^k synthase, is a type II transmembrane retaining glycosyltransferase belonging to the CAZy glycosyltransferase family 32 (Carbohydrate Active Enzymes Database, CAZy, http://www.cazy.org) [34]. The enzyme catalyzes the transfer of a galactose from UDP‐galactose to GalCer and LacCer (Figure 1B), and requires Mn²⁺ as a divalent cofactor for activity. The catalytic mechanism of retaining glycosyltransferases such as A4GALT remains a matter of debate [35, 36]. A type of double‐displacement mechanism similar to that of glycosidases, involving a covalent glycosyl‐enzyme intermediate in which the donor sugar is bound to a nucleophilic Asp or Glu residue, was initially proposed [37]. This mechanism has been demonstrated only for transferases acting on rare sugars (β‐Kdo transferases) [38, 39] and remains controversial for α‐1,3‐galactosyltransferase (α3GalT) [40]. However, most retaining GTs lack an appropriately positioned nucleophile, and increasing evidence supports instead a front‐face (S_Ni‐like) mechanism, in most cases proceeding via a short‐lived oxocarbenium ion intermediate (Figure 1C). Such mechanism has been established for several retaining GTs, including trehalose‐6‐phosphate synthase (OtsA) [41, 42], polypeptide GalNAc‐transferase 2 (GalNAc‐T2) [43], lipopolysaccharide α‐galactosyltransferase (LgtC) [44], α‐1,4‑N‑acetylhexosaminyltransferase (EXTL2) [45] and mannosylglycerate synthase [46].

Here, we integrate mechanistic analysis of A4GALT with natural and adamantyl‐functionalized glycosphingolipid substrates, providing new opportunities for Fabry disease inhibitor discovery. Using structural modeling, molecular dynamics, mutagenesis, and QM/MM simulations, we obtained a Michaelis complex of A4GALT in complex with UDP‐Gal and LacCer and uncovered how the enzyme undergoes a conformational rearrangement upon substrate binding that enforces a front‐face S_Ni‐type reaction pathway. Guided by these insights, we designed and synthesized a panel of glycosphingolipid analogues to probe substrate recognition and exploit mechanistic features for inhibition. Evaluation of these substrates in vitro and in cellulo identifies the factors that govern inhibition and reduction of Gb3 levels, thereby providing proof‐of‐principle for A4GALT‐targeted Fabry disease therapy.

2. Results and Discussion

2.1. A4GALT Ternary Complex Sheds Light on Substrate Accommodation by a Conformational Switch

Among the three known GT fold types (A, B, and C) [47, 48], A4GALT adopts the GT‐A fold [49], characterized by two closely abutting β/α/β Rossmann‐like domains. In GT‐A enzymes, the N‐terminal domain recognizes the nucleotide sugar donor, whereas the C‐terminal domain forms the acceptor binding site. Another conserved feature of this fold is the D‐X‐D motif, whose carboxylate side chains coordinate the divalent metal cation (Mn²⁺ in A4GALT). Very often, there is also a positively charged residue in the active site (Lys or Arg) [47].

We obtained the structure of A4GALT using AlphaFold2 (AF) (Figure S1) and subsequently added the donor substrate (UDP‐Gal) and the Mn²⁺ ion by structural superposition with the homologous PaToxG enzyme (Figure S3), which belongs to the same GT family [50]. The resulting enzyme complex with UDP‐Gal and Mn²⁺ shows the key features of GT‐A enzymes: a D‐X‐D motif (D192‐T193‐D194) that coordinates Mn²⁺, and a positively charged residue (R179) in the active site. A C‐terminal histidine (H315) completes the six‐fold metal coordination (Figure S2). UDP‐Gal is positioned within a tunnel‐shaped active site, where it is partially shielded by a loop comprising residues 318 to 321 that adopts a 3₁₀‐helical conformation. The full Michaelis complex was built by docking the LacCer acceptor into the active site, and the full model was embedded in a mammalian membrane mimic (Figure 2A) [51].

FIGURE 2.

Structural overview and conformational dynamics of the A4GALT‐LacCer‐(UDP‐Gal)‐Mn²⁺ complex, obtained by AF, docking and molecular dynamics simulations. (A) Membrane‐embedded view of the A4GALT complex. A close‐up shows the spatial arrangement of the sugar moieties and the ceramide tail within the membrane bilayer (gray spheres). (B) Rotated view and zoom in of the catalytic site with the 3₁₀‐helix (residues 318–321) in pink, the LacCer acceptor in yellow and the UDP‐Gal in blue. (C) Structural superposition of the unliganded enzyme (green) and the enzyme in complex (pink) reveals a conformational rearrangement of the 3₁₀‐helix upon binding of the substrates UDP‐Gal and LacCer. (D) Distribution of RMSD values for the 3₁₀‐helix during MD simulations in unliganded and complex states.

To investigate how substrate binding influences the protein conformation, we performed extensive molecular dynamics (MD) simulations (1.5 µs) of both the ternary complex (A4GALT + UDP‐Gal/Mn²⁺ + LacCer) and the unliganded enzyme. Analysis of the resulting trajectories revealed a conformational change, with a backbone RMSD displacement ≈ 2 Å (Figure S4), upon substrate binding. This rearrangement primarily involves the 3₁₀‐helix (Figure 2C,D), which curves over the UDP‐Gal donor, positioning the backbone of K319 near the phosphate groups and thereby providing electrostatic stabilization. This large motion reduces solvent accessibility of the active site, as confirmed by solvent accessible surface area (SASA) calculations (Figure S4). These results suggest that the interaction between the UDP‐Gal phosphates and the 3₁₀‐helix, induced upon substrate binding, acts as a conformational switch that prevents donor hydrolysis while promoting the formation of catalytically competent configurations.

2.2. A4GALT Interactions with Gal‐UDP (Donor) and LacCer (Acceptor)

We next analyzed the main substrate‐enzyme interactions within the active site in the ternary complex (Figure 3A). As illustrated in a 2D representation (Figure 3D), the D‐X‐D motif and the C‐terminal His residue (H315) coordinate to the Mn²⁺ metal ion (Figure S5). In addition, residue E83 interacts with the uridine group of the UDP moiety, while two residues (D176 and N217) interact with the Gal donor. Residue D176 forms two hydrogen bonds with 4‐OH and 6‐OH groups of Gal, whereas N217 mainly interacts with the 3‐OH. The terminal Gal unit of the LacCer acceptor is also engaged in hydrogen bond interactions, particularly through its 2′‐OH and 3′‐OH groups with Q211. Furthermore, two tryptophan residues located at the entrance of the active site interact with the acceptor substrate: W246 forms a hydrogen bond with the Gal 4′‐OH, whereas W289 establishes a π‐stacking interaction with the Gal ring (Figure S6). These interactions ensure a correct orientation of the acceptor, thereby enhancing specificity toward Gal‐terminated substrates. Several hydrogen bond interactions between the donor and the acceptor further stabilize their position in the active site. For instance, the 4′‐OH and 6′‐OH groups of the Gal acceptor interact with the donor β‐phosphate, while the 2‐OH of the donor interacts with the 3′‐OH of the acceptor (Figures 3B and S7).

FIGURE 3.

(A) Close‐up of the A4GALT active site, highlighting the coordination of Mn²⁺, the D‐X‐D motif (purple), the C‐terminal histidine (khaki), the 3₁₀‐helix (pink), the donor UDP‐Gal (blue), and the LacCer acceptor (yellow). (B) Details of the interactions (black dashed lines) between the donor and the acceptor. (C) 2D histogram of the main catalytic distance (C1‐O4′) and the angle of nucleophilic attack (O1_P‐C1‐O4′) measured during classical MD simulations. (D) 2D schematic representation of the interaction network (dashed lines) between the active site residues, UDP‐Gal, Mn²⁺, and LacCer. Hydrogen bond interactions are in blue, coordination interactions in purple, and π‐stacking in green. (E) Results of site‐directed mutagenesis (alanine screening), showing the activity of each A4GALT‐V5‐His₆ mutant compared to the wildtype enzyme (WT). Activity is expressed as the amount of Gb3 lipid in pmol produced per mg of protein, corrected for the total protein count (BCA) and expression level (V5‐signal), normalized against the activity in pmol/mg of the wildtype. The experiment was performed in triplicate as a biological n = 2, with the ± S.D. calculated over the mean of the two replicates. D: donor binding residue, M: metal binding residue, G: glycosylation site, A: acceptor binding residue. Statistical significance was calculated by performing a one‐way ANOVA with Dunnett's test using the wildtype as the control group. Ns = not significant, *p <0.05, ** p <0.001, *** p <0.001 and **** p <0.0001.

The MD simulations showed that the anomeric carbon (C1) of the Gal donor remains near the oxygen atom of the acceptor 4′‐OH group, with a C1‐O4′ distance of 3.33 ± 0.18 Å (Figure 3C). No nucleophilic residue (Asp or Glu) was observed on the β‐face of the donor Gal, ruling out the possibility of a double‐displacement mechanism. Notably, the acceptor sugar is positioned towards the α‐face of the sugar donor (the O1_P‐C1‐O4′ angle amounts to 103.6 ± 6.8⁰), and there is a hydrogen bond between the acceptor and the β‐phosphate. This is a configuration that is well‐suited for a front‐face S_Ni‐like reaction [41]. Therefore, the MD simulations reveal a complex and intricate network of substrate‐enzyme interactions in the active site, with the relative orientation of donor and acceptor consistent with an S_Ni‐like reaction mechanism.

2.3. Functional Mutagenesis Confirms Critical Roles of Conserved Residues in A4GALT Catalytic Activity and Substrate Recognition

To better understand A4GALT binding, we generated mutants of residues observed to interact with the substrates and assessed their catalytic activity by measuring Gb3 synthesis. We designed expression plasmids encoding 22 different structural or active site mutants, in which the targeted residues were substituted with alanine, and expressed them in WT HEK293T cells (Figure S8). The residual activity of these mutants was quantified by measuring Gb3 lipid production 48 h post‐transfection using LC‐MS/MS [52]. Lipid levels were corrected for cell density variability and transfection efficiency within each experiment. To achieve this, we produced wild‐type A4GALT and mutant analogues, all functionalized with an N‐terminal V5‐His₆‐tag, allowing for quantitative comparison of signals by Western blot analysis. The catalytic activity of A4GALT remains unaffected by the V5‐His₆ tag, as demonstrated by our in vitro assays using the purified A4GALT‐V5‐His₆ construct.

The activity levels of the A4GALT‐V5‐His₆ mutants align with the results of the in silico analysis (Figure 3E). The lowest activity levels (<25%) were observed for mutations of residues that are expected to be involved in catalysis and/or binding of the substrates. Lower activity was observed for mutants E83A and N217A (two residues involved in donor binding), Q211A (residue involved in acceptor binding, most likely responsible for acceptor specificity), as well as D192A, D194A, and H315A (metal‐binding residues). Mutations in the loops that interact with residues N217 and Q211—Y285A and T255A, respectively—also significantly reduced enzyme activity.

In line with previous data, the aspartic acids D192 and D194, which are involved in the formation of the D‐X‐D metal‐binding motif, are essential for activity [53, 54, 55]. In our study, activity was reduced to 10–15%, supporting the hypothesis that this motif plays an important role in catalysis [56]. Interestingly, activity is also reduced when N203 is mutated, which is one of two reported N‐glycosylation sites [49]. Western Blot analysis indeed revealed the production of an N203A A4GALT variant with a lower molecular weight (Supporting Information, Section 4.1). Although this mutant retained some activity, previous findings indicate that the loss of glycosylation at this site may lead to cellular mislocalization, thereby reducing Gb3 synthesis in the Golgi apparatus. Finally, while D176 was expected to contribute significantly to catalysis by orienting the Gal donor through hydrogen‐bond interactions, it proved not essential for enzymatic activity (∼75% activity retained). This is likely because the active‐site architecture is preserved and N217 plays a dominant role in positioning the donor substrate.

In all, with few exceptions, mutations of amino acid residues involved in substrate binding result in a significant reduction of catalytic activity. Our combined in silico and in vitro results suggest that the most important residues in substrate recognition are E83 and N217 (involved in donor binding), the D‐X‐D motif D192‐T193‐D194 (which coordinates Mn²⁺), and Q211 (involved in acceptor binding). Notably, substitution of the last residue by a glutamate is known to be the cause of a rare inherited polyagglutination syndrome by also accommodating a GalNAc instead of a galactose in A4GALT acceptors [57], whereas the alanine variant (Q211A) abrogates Gb3 synthase activity.

2.4. A4GALT Catalysis Proceeds via an S_Ni‐Like Reaction Involving a Short‐Lived Oxocarbenium Ion Intermediate

The catalytic mechanism of A4GALT was modeled by QM/MM MD simulations, starting from a representative snapshot of the MD simulations of the ternary complex. We used the On‐the‐fly Probability Enhanced Sampling Method (OPES) [58] to activate the glycosyltransferase reaction. OPES is a recently developed method that, like metadynamics, enables an efficient exploration of free energy landscapes (FELs) of complex systems [59]. A path collective variable (PathCV), integrating the key bond distances expected to form or break during the reaction, was used (Figure S9B). The PathCV included the distance of nucleophilic attack (C1‐O4′), the distance measuring departure of the phosphate leaving group (C1‐O1_P) and the distance describing the proton transfer between the Gal acceptor and the β‐phosphate (H4‐O1_P). During the simulation, the UDP‐bound galactose detached from the UDP and transferred to the LacCer acceptor, forming the product Gb3 with retention of the anomeric stereochemistry (Figure S9C). The resulting free energy landscape (FEL) (Figure 4A) reveals two deep minima, corresponding to the reactants (Michaelis complex, MC) and the product (P) states. The highest energy point along the reaction coordinate is at ≈ 14.0 kcal/mol from MC, in reasonable agreement with the free energy barriers estimated for other retaining GTs (14‐21 kcal/mol) [44, 46, 60, 61, 62]. An interesting feature of the FEL is the presence of a small minimum (an intermediate state) along the pathway (I in Figure 4A), separated by energy barriers of ≈ 2.5 kcal/mol from both MC and P. This intermediate state corresponds to a short‐lived oxocarbenium ion (Figures S10 and S11), being a signature of an S_Ni‐like reaction [41, 63].

FIGURE 4.

A4GALT mechanism obtained by Path‐CV QM/MM OPES simulations. (A) Reaction free energy landscape showing the most relevant stated: Michaelis complex (MC), the same complex after rotation of the β‐phosphate (MC’), first transition state (TS1), oxocarbenium ion intermediate (I), second transition state (TS2), and product (P). (B) Evolution of the main catalytic distances along the reaction coordinate from MC → MC’ → I → P, and the conformational itinerary of the Gal donor ring. (C) Representative snapshots of the main states along the reaction coordinate. (D) Schematic representation of the overall A4GALT reaction mechanism.

Analysis of the atomic motion along the reaction coordinate provides further insight into the reaction mechanism (Figure 4B,C). In the MC, the Gal donor forms hydrogen bonds with both the β‐phosphate of UDP and the galactose of the acceptor. However, direct nucleophilic attack by the 4‐OH of the acceptor on C1 is hampered by the specific position of the β‐phosphate, which precludes the approach of the acceptor Gal. At this point, a conformational rotation of the β‐phosphate around the P–O linkage that connects with the α‐phosphate (Figure S12) takes place. This exposes one β‐phosphate oxygen atom (O1_P) towards the 4‐OH of the acceptor (MC’ in Figure 4C), facilitating the formation of the right hydrogen bond interaction between donor and acceptor. This interaction has been observed in several retaining GTs that follow a S_Ni‐like mechanism, and it is expected to be an essential feature of the Michaelis complex. Subsequently, the sugar‐phosphate bond breaks, the Gal of the donor adopts a planar conformation around the anomeric carbon (an envelope ⁴ E), and the C1‐O5 bond shortens (from 1.43 ± 0.02 Å to 1.28 ± 0.01 Å), evidencing the formation of an oxocarbenium ion‐like species. According to the reaction FEL, this species is a short‐lived intermediate. The β‐phosphate is fully detached from the donor sugar at I (O1_P‐C1> 3 Å), and the negative charge on O2_P redistributes across O1_P, allowing both atoms to interact with the 4‐OH and 6‐OH groups of the Gal acceptor, respectively. The reaction continues with nucleophilic attack of the acceptor O4′ on the donor anomeric carbon, which occurs concomitantly with proton transfer from O4′H to O1_P, ultimately forming the new α‐1,4 glycosidic bond of the Gb3 product. Overall, the simulations are consistent with an S_Ni‐like reaction, highlighting the critical role of hydrogen bonding between the donor and acceptor molecules in stabilizing the short‐lived reaction intermediate.

Our mechanistic analysis reveals how A4GALT recognizes its donor and acceptor substrates and orchestrates catalysis by phosphate rotation to enable a front‐face S_Ni‐like mechanism. Key residues (D176, N217, Q211, W246, and W289) shape the catalytic pocket and enforce productive positioning of galactose‐terminated ceramides. In addition, productive catalysis requires specific donor‐acceptor interactions, including a short C1‐O4′ distance, the formation of a hydrogen bond between the donor and the β‐phosphate, and a favorable angle of attack (<O_P‐C1‐O4′ ≈ 90⁰). These insights offer guidance for the rational design of substrate analogues that exploit the substrate recognition machinery of A4GALT to achieve high‐affinity binding yet resist further turnover.

2.5. In Silico Evaluation of Galactosylceramide Analogues

More than a decade ago, Lingwood et al. reported adamantyl galactosylceramide (AdaGalCer, 1) and adamantyl glucosylceramide (AdaGlcCer, 2) as amphipathic glycosphingolipid (GSL) mimics with enhanced water solubility able to reduce Gb3 levels in cells [64]. Although these compounds effectively lower Gb3 accumulation, they are also metabolized into their corresponding galactosylated products, whose biological consequences remain unclear. Specifically, AdaGlcCer 2 was suggested to be galactosylated by LacCer synthase (LCS) in cells, and the product (AdaLacCer) was proposed to act as LCS inhibitor, thereby decreasing the concentration of LacCer, and thus depleting complex GSLs that arise from it. AdaGalCer 1 selectively inhibits Gb3 and Gb4 biosynthesis through the generation of a galactosylated metabolite AdaGb2 in FD lymphoblasts. While both 1 and 2 reduce Gb3 levels, the exact mechanism by which these compounds interact with A4GALT remains unclear.

To gain a deeper understanding of A4GALT substrate binding specificity, we generated a focused library of adamantyl galactosylceramide compounds (Figure 5A), including AdaGalCer (1), AdaGlcCer (2), AdaLacCer (3), and their galactosylated derivatives AdaGb2 (4), AdaGal‐α‐1,4‐GlcCer (5), and AdaGb3 (6). We investigated whether these glycosylated ceramide derivatives could retain binding while avoiding reactivity, by superposing each compound onto the natural substrate bound to A4GALT and running MD simulations to characterize their binding modes. Furthermore, we developed a scoring function to rank their predicted reactivity.

FIGURE 5.

(A) Chemical structures of potential A4GALT inhibitors: AdaGalCer (1) and AdaGlcCer (2) (previously reported to lower Gb3 levels in cells), new potential substrate AdaLacCer (3), and their galactosylated products Ada(Gal)₂Cer or AdaGb2 (4), AdaGal‐α1,4‐GlcCer (5), and AdaGb3 (6). Note that 4, 5, and 6 are potential products of the reaction of 1, 2, and 3, respectively, catalyzed by A4GALT. (B) Binding poses of the six compounds in the A4GALT active site, obtained from MD simulations. (C) Distribution of reactivity scores obtained from MD simulations.

The MD simulations showed that all compounds 1–6 bind to A4GALT (Figure 5B). In all cases, the polar head of the inhibitor (Gal or Glc) resides within the enzyme active site, whereas the adamantyl group remains embedded in the hydrophobic intermembrane region (Figure S14), without interacting with active site residues. To assess the potential reactivity of these compounds, we considered the key structural parameters relevant to catalysis, as defined by the previously determined reaction coordinate (Figure 4): (i) the distance between the nucleophile and the anomeric carbon (C1‐O4′), (ii) the hydrogen bond between the acceptor O4′‐H and the β‐phosphate (O4′H···O2_P), and (iii) the acceptor nucleophilic attack angle (<O_P‐C1‐O4′). We also quantified the deviation of the acceptor sugar from a chair conformation (Figure 5C), since large distortions would preclude reactivity. We then developed a scoring function that integrates these four descriptors (see Methods) and evaluated it for each frame of the MD trajectories. The compounds were then scored according to the similarity of their active site configurations to the ideal catalytic arrangement (a perfect configuration scores 1, whereas less favorable geometries receive exponentially lower values). This approach provides a quantitative measure of the system's ability to adopt a reactive configuration.

Among the six compounds, AdaGalCer (1) and AdaLacCer (3), which most closely resemble the natural A4GALT acceptors, displayed the highest average reactivity scores (median ∼0.92) (Figure 5C). AdaGb2 (4) and AdaGal‐a1,4‐GlcCer (5) also adopted reactive poses, despite featuring an α1→4 glycosidic linkage (in contrast with the β‐linkages of GalCer and LacCer), and thus behaved similarly to compounds 1 and 3. In contrast, AdaGlcCer (2) scored the lowest (median ∼0.62), indicating that it binds but rarely achieves a productive catalytic configuration. This reduced reactivity arises from the equatorial orientation of the 4‐OH group in glucose, which prevents interaction with W246, leading to an unfavorable positioning of the terminal acceptor sugar in the active site. AdaGb3 (6), despite its mean score of 0.79, samples many non‐catalytic poses, with a quarter of its frames scoring below 0.15. This is due to the pronounced deviation of the terminal Gal from the chair conformation. In summary, the simulations and reactivity scoring demonstrate that all compounds bind to A4GALT, but only compounds 1, 3, 4, and 5, which contain a terminal galactose, can adopt catalytically competent configurations.

2.6. Synthetic AdaLacCer Analogues 1–5 Inhibit A4GALT In Vitro but Only 1 and 2 Reduce Gb3 Levels in Fibroblasts

To evaluate their A4GALT inhibition activity, we developed an in vitro activity assay quantifying the synthesis of Gb3 using the fluorescently labelled substrate C12‐NBD‐LacCer (2.5 µM), UDP‐Gal (500 µM) and a truncated soluble recombinant human A4GALT (rhA4GALT, ∼250 nM) carrying a C‐terminal V5‐His₆‐tag in MOPS buffer at pH 6.6 supplemented with BSA (1 mg/mL) and sodium taurocholate (3.7 mM) (Figure S13). Enzymatic activity was strictly dependent on the presence of Mn²⁺. Both AdaGalCer (1) and AdaGlcCer (2) inhibited A4GALT in vitro in the low micromolar range, with 2 exhibiting stronger inhibitory potency. Although GlcCer has not been reported to inhibit A4GALT activity, in silico analysis suggests that AdaGlcCer (2) can bind to A4GALT without being catalytically processed. AdaLacCer (3), which closely resembles the natural substrate LacCer, showed the highest inhibition potency (apparent IC₅₀ = 4.7 ± 0.3 µM). Interestingly, the galactosylated products of 1 and 2, namely AdaGb2 (4) and AdaGal‐α1,4‐GlcCer (5), respectively, retained comparable inhibition activity as their parent compounds, whereas AdaGb3 (6) showed only high micromolar inhibition. All tested compounds, except for AdaGlcCer (2), exhibited selectivity for A4GALT over a panel of related glycosidases and glycosyltransferases involved in the same biochemical pathway, including glucosylceramide synthase (GCS), β‐glucosidases (GBA1 and GBA2), β‐galactosidases (GALC and GLB1), α‐glucosidases (GAA and GANAB), as well as α‐galactosidase A (GLA).

The susceptibility of compounds 1–6 to act as A4GALT acceptors was experimentally investigated by LC‐MS/MS analysis, demonstrating that 1, 3 and 5 (and, to a lesser extent, compound 4) are galactosylated in vitro when incubated at 1 mM concentration with rhA4GALT for 6 h (Figures 6A and S17). In contrast, these products were only observed in human fibroblasts when incubated with AdaGal 1 and AdaLac 3, although significantly lower levels of AdaGb3 (6) originating from 3 were observed when compared to AdaGal 1 (Figures 6B and S18). To investigate whether this decrease was due to impaired cell or Golgi membrane permeability, cells were incubated in situ with compounds 1–6, thoroughly washed, and then homogenized to release all compartmentalized enzymes before a second incubation of 6 h. As a result, the levels of AdaGb3 derived from AdaLacCer 3 increased significantly, whereas the formation of AdaGb2 from AdaGalCer 1 increased only modestly upon lysis (Figures 6C and S19). This suggests that cellular delivery of AdaGalCer 1 is highly efficient, while AdaLacCer 3 has a reduced membrane permeability or intracellular trafficking. In contrast, galactosylation of 4 and 5 was not detected in cells, even after the lysing step, and only degalactosylation products were observed, most probably resulting from the activity of endogenous galactosidases. Interestingly, incubation of fibroblasts with AdaGlcCer 2 led to the detection of both AdaLacCer and AdaGb3, likely through the consecutive action of LacCer synthases (LCS: B4GALT5 and B4GALT6) and A4GALT, although AdaLacCer levels were minimal and AdaGb3 levels remained comparatively low (Figure S19).

FIGURE 6.

Galactosylation studies by A4GALT of adamantyl analogues 1–6. (A) Incubation of rhA4GALT with compounds 1, 2, 4 and 5 at 1 mM concentration for 6 h. (B) Incubation of human dermal fibroblasts for 6 h with compounds 1 and 3 at 50 µM concentration, with or without an additional incubation of 6 h after cell lysis, and analysis by HILIC‐MS/MS. The experiment was performed as a biological triplicate. (C) Incubation of fibroblasts with increasing concentrations of compounds 1–6, followed by feeding with ¹³C₅‐Sph. Activity is expressed as the levels of ¹³C₅‐GlcCer, ¹³C₅‐LacCer and ¹³C₅‐Gb3 measured by LC‐MS/MS normalized to a DMSO control (100%). The experiment was performed in biological triplicate, with ± S.D. calculated from the mean of two separate experiments (n = 2).

We then investigated the effect of the adamantyl analogues on glycosphingolipid metabolism in live cells, including the levels of Cer, GlcCer, LacCer, Gb3 and their corresponding sphingoid bases. Normal human dermal fibroblasts were pre‐incubated with compounds 1–6, followed by treatment with neutron‐encoded metabolic precursor ¹³C₅‐sphingosine (¹³C₅‐Sph). The levels of ¹³C₅‐labelled neutral glycosphingolipids (nGSLs) and lysolipids were then quantified by LC‐MS/MS. A clear concentration‐dependent effect on ¹³C₅‐Gb3 metabolism was observed for compounds 1 and 2 (Figures 5C and S20–S27), with half maximal effective concentration (EC₅₀) values for Gb3 reduction of 7.6 ± 1.0 and 2.1 ± 0.1 µM, respectively. Compound 1 selectively lowered ¹³C₅‐Gb3 levels, indicating direct action on A4GALT. AdaGlcCer 2 reduced both ¹³C₅‐LacCer levels and ¹³C₅‐Gb3 levels. The low micromolar in situ activities of compounds 1 and 2, consistent with their in vitro potencies, suggest good cell permeability and effective trafficking to the Golgi apparatus. In contrast, but in line with the above galactosylation studies in cells, compounds 3–6 showed no inhibitory or stimulatory effects on ¹³C₅‐Gb3 metabolism at concentrations up to 50 µM. Given that compounds 3–5 were among the most potent A4GALT inhibitors in vitro, their lack of cellular activity again suggests that their additional glycoside moiety—combined with the adamantyl modification—may significantly impair cell and/or Golgi membrane permeability. Notably, concentrations above 50 µM of compounds 3–6 were cytotoxic to fibroblasts.

Additional effects on other endogenous and ¹³C₅‐labelled lipids were also observed. Notably, compound 2 induced a marked increase of (¹³C₅‐)HexSph, most likely corresponding to (¹³C₅‐)GlcSph, along with a rise in endogenous Cer levels. Previous studies have shown that compound 2 inhibits lysosomal rhGBA1 and also functions as a substrate for rhGBA1 in vitro and for endogenous GBA1 in cells [64]. Consistent with this, compound 2 competes with 4‐MU‐glucoside at micromolar concentrations, with 50% displacement observed at 77 µM using Cerezyme (Table 1). It is plausible that any accumulating ¹³C₅‐GlcCer undergoes rapid turnover to ¹³C₅‐GlcSph via endogenous ceramidase. Similarly, compound 1 also elevated Cer levels, albeit to a lesser extent. In this case, galactosylceramidase (GALC) could be responsible for its hydrolysis, although no inhibitory activity was observed in our in vitro assay up to 100 µM (Table 1).

TABLE 1.

A4GALT inhibition of compounds 1–6 and selectivity against related glycosyltransferases and glycosidases. Half‐maximal inhibitory concentration (IC₅₀) values (µM) for in vitro inhibition of recombinant human A4GALT (rhA4GALT), GCS inhibition in vitro (HEK293T lysate) and in situ cell (RAW264.7 macrophages); in vitro inhibition of recombinant human GBA1 (rhGBA1, Cerezyme), in vitro inhibition of GBA2 (lysate of HEK293T cells overexpressing GBA2), in vitro inhibition of GLB1 (fibroblast lysate), in vitro inhibition of recombinant murine GALC (culture medium from HEK293T cells overexpressing rmGALC), in vitro inhibition of recombinant human GAA (rhGAA, Myozyme), in vitro inhibition of GANAB (lysate of fibroblasts of infantile Pompe patients who are GAA deficient), and in vitro inhibition of recombinant human α‐Gal A (rhGLA, Fabrazyme). Reported values are the mean ± standard deviation from three technical replicates.

		IC50 (µM)
Enzyme/inhibitor		1	2	3	4	5	6
Galactosyltransferase	rhA4GALT	38 ± 4.0	12 ± 2.3	4.7 ± 0.3	9.5 ± 0.2	9.9 ± 1.1	625 ± 16.0
Glucosyltransferases	GCS (in vitro)	> 100	> 100	> 100	> 100	> 100	n.d.
	mGCS (in cellulo)	> 100	> 100	> 100	> 100	> 100	n.d.
β‐Glucosidases	rhGBA1	> 100	77 ± 7	> 100	> 100	> 100	n.d.
	rhGBA2	> 100	> 100	> 100	> 100	> 100	n.d.
β‐Galactosidases	GLB1	> 100	> 100	> 100	> 100	> 100	n.d.
	rmGALC	> 100	> 100	> 100	> 100	> 100	n.d.
α‐Glucosidases	rhGAA	> 100	> 100	> 100	> 100	> 100	n.d.
	GANAB	> 100	> 100	> 100	> 100	> 100	n.d.
α‐Galactosidases	rhGLA	> 100	> 100	> 100	> 100	> 100	n.d.

3. Conclusion

A4GALT is an attractive pharmacological target not only for Fabry Disease but also for a range of other pathological conditions associated with elevated Gb3 levels, including cancer and anti‐cancer drug resistance³², immunodeficiency (AIDS)³³, and pathogen infection and virulence (Shiga toxin and verotoxin B). In addition, targeting bacterial A4GALT homologues, such as Neisseria meningitidis’ lipopolysaccharyl‐α‐1,4‐galactosyltransferase C (LgtC), could also lead to the discovery of new antibacterials³⁴. In this study, we combined structural modeling, molecular dynamics, mutagenesis, and QM/MM simulations to elucidate how A4GALT recognizes its donor and acceptor substrates and catalyzes galactosyl transfer. Our analyses revealed that A4GALT undergoes a conformational rearrangement of a 3₁₀‐helix upon substrate binding, which stabilizes the donor substrate and reorganizes the active site for catalysis. QM/MM simulations demonstrated a front‐face S_Ni‐like mechanism involving a short‐lived oxocarbenium ion intermediate, consistent with observations in other retaining glycosyltransferases that lack a nucleophilic residue on the  β‐face of the donor sugar [35, 62]. Notably, rotation of the β‐phosphate primes the reaction, so that the reactive configuration of the active site differs from that present immediately after substrate binding. Mutagenesis of key active‐site residues confirmed their critical roles in donor and acceptor binding, metal coordination, and catalysis, providing experimental validation of the computational model. These mechanistic insights not only resolve a long‐standing question in A4GALT enzymology but also define structural features that are exploitable for inhibitor design.

Guided by these findings, we synthesized a panel of adamantyl glycosylceramide analogues to probe how A4GALT accommodates acceptor‐like ligands and to evaluate their potential as inhibitors. MD simulations indicated that they all bind to A4GALT, and biochemical assays confirmed that they indeed inhibit the enzyme in vitro with low micromolar potencies. MD simulations further revealed that compounds bearing terminal galactose moieties may warrant caution as they engage A4GALT in catalytically competent configurations. This suggests that these inhibitors could be processed by the enzyme to form undesired side products of unknown physiological consequences. Galactosylation studies using LC‐MS/MS showed that AdaGal (1), AdaLac (3), AdaGb2 (4) and AdaGal‐α1,4‐GlcCer (5) are indeed galactosylated in vitro when incubated at high concentrations (1 mM) with rhA4GALT. In contrast, AdaGlcCer (2), the glucose‐terminated analogue, is unreactive and thus represents a bona fide A4GALT inhibitor. Cellular studies, however, demonstrated that AdaGlcCer (2) also reduces LacCer levels and undergoes galactosylation to AdaLacCer and AdaGb3, indicating that it is not selective for A4GALT.

The inhibitory behavior of compounds 1–5 cannot be rationalized by a classical inhibition model. Although several analogues can act as acceptor substrates at high concentrations (1 mM) in vitro (Figure 6A), the low‐micromolar apparent IC₅₀ values obtained in the in vitro competition assay (Table 1) may primarily reflect their ability to bind at the acceptor site and compete with LacCer. Product formation only becomes apparent in cells with compound 1, indicating that, in this case, glycosylation and compound 4 formation might contribute to inhibition. Among the series, only AdaGalCer (1) penetrates cells efficiently and selectively reduces Gb3 levels in fibroblasts, whereas more extended analogues lacked activity in situ, likely due to impaired membrane permeability or Golgi trafficking. The 4‐phosphate adaptor protein‐2 (FAPP2) is a glycolipid transfer protein that transports GlcCer from the cis‐Golgi or cytosol to the trans‐Golgi, where it is converted into complex GSLs [65]. FAPP2 shows a preference for simple GSLs, such as GlcCer and GalCer, and to a lesser extent LacCer, but does not accommodate certain modified congeners, such as the sulfated derivative of GalCer (sulfo‐GalCer) [66]. This may explain limited cell membrane permeability or inefficient Golgi trafficking of disaccharide ligands 3–5. Structural modelling highlights how FAPP2 selects glycosphingolipids via a specificity loop that excludes bulky analogues through steric hindrance (Figure S28). Collectively, these results emphasize the importance of coupling mechanistic insights of substrate recognition with an understanding of lipid transport pathways when developing A4GALT inhibitors.

Together, our work provides the first comprehensive study of A4GALT substrate recognition, reaction mechanism, and inhibition, establishing a framework for further rational development of competitive inhibitors. Future research may build on these findings by designing non‐reactive acceptor analogues that retain strong binding affinity, optimizing physicochemical properties for efficient transport to the Golgi apparatus, and minimizing off‐target effects on related glycosyltransferases and glycosidases.

Author Contributions

Nicky de Koster: investigation, writing – original draft, methodology, formal analysis, visualization, data curation, writing – review and editing. Òscar Vidal‐Gironès: investigation, writing – original draft, methodology, visualization, writing – review and editing, formal analysis, data curation. Rowan de Graaf: data curation, formal analysis, methodology, investigation. Ken Kok: investigation, methodology, formal analysis. Lindsey T. Lelieveld: investigation, methodology. Jackelien Hoogervorst: investigation, methodology. Maria J. Ferraz: investigation, methodology, formal analysis, data curation. José Pablo Rivas‐fernández: investigation. Rob F. Lammers: investigation, formal analysis, data curation, methodology. Herman S. Overkleeft: writing – review and editing, supervision. Johannes M. F. G. Aerts: writing – review and editing, supervision. Rolf G. Boot: methodology, writing – review and editing, supervision. Carme Rovira: conceptualization, funding acquisition, writing – original draft, writing – review and editing, resources, supervision. Marta Artola: conceptualization, funding acquisition, writing – original draft, writing – review and editing, visualization, project administration, formal analysis, supervision, resources, data curation, methodology.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Supporting File: anie73810‐sup‐0001‐SuppMat.docx.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.