Catalytic flexibility of rice glycosyltransferase OsUGT91C1 for the production of palatable steviol glycosides

The study addresses the need for low-calorie sweeteners with favorable taste profiles to reduce the health risks associated with excessive consumption of high-calorie sugars. Steviol glycosides from Stevia rebaudiana are intensely sweet but vary in organoleptic properties depending on glycosylation patterns at the R1 (C13-hydroxyl) and R2 (C19-carboxylate) ends of the steviol aglycone. While stevioside (ST) and rebaudioside A (Reb A) are abundant, they can have bitter aftertastes. Rebaudioside D (Reb D) and M (Reb M) have superior taste but occur only in trace amounts, largely due to inefficient β(1→2) glucosylation at the R2 end by the native UGT91D2 enzyme in Stevia. The research aims to characterize and engineer a rice glycosyltransferase, OsUGT91C1, hypothesized to efficiently catalyze the critical β(1→2) glucosylation at the R2 end, thus enabling conversion of abundant precursors (e.g., Rubusoside, Reb A) into desirable products (Reb E, Reb D, Reb M).

Steviol glycoside biosynthesis in S. rebaudiana involves several UDP-glucose-dependent GT1 family glycosyltransferases: UGT85C2 adds the first glucose at R1; UGT74G1 adds the first glucose at R2; UGT91D2 typically performs β(1→2) addition at R1 but shows only trace activity at R2; UGT76G1 catalyzes β(1→3) additions at both ends, with efficient R2 addition requiring prior β(1→2) at R2. Prior work showed UGT76G1 can bind substrates in two orientations and dictates β(1→3) specificity via active site positioning. A rice enzyme (EUGT11/OsUGT91C1) sharing 40% identity to UGT91D2 was identified as a potential β(1→2) transferase at R2, enabling production of Reb D/M from Reb A via sequential β(1→2) then β(1→3) additions. Organoleptic studies indicate Reb D/M have superior taste vs ST/Reb A, motivating biotechnological routes to increase their yields.

- Enzyme cloning and expression: OsUGT91C1 coding sequence was synthesized and cloned into pET28b, expressed in E. coli BL21 (DE3). Native protein induced with 0.5 mM IPTG at 37 °C for 4 h; SeMet-labeled protein produced in M9 medium with SeMet, induced and expressed 18 h at 16 °C. - Purification: Ni2+-affinity (HisTrap) in 20 mM Tris-HCl pH 7.8, 0.5 M NaCl, 250 mM imidazole; gel filtration in 10 mM HEPES-NaOH pH 7.2, 150 mM NaCl, 2 mM DTT; concentrated to 20 mg/ml, stored at −80 °C. Site-directed mutants generated by QuickChange. - Crystallography: Sitting-drop vapor diffusion with protein (20 mg/ml) ± 1 mM UDP and 5 mM steviol glycoside ligands; crystals in 100 mM HEPES-NaOH pH 6.5, 20% PEG 4000; cryoprotected with 25% glycerol; data collected at SSRF BL18U1/BL19U1 (λ=0.97853 Å) on Pilatus detectors. Data processed with DIALS (CCP4i2). SeMet SAD at 3.45 Å used to solve apo by CRANK2; molecular replacement with Phaser for higher-resolution datasets. Model building/refinement with Coot and REFMAC5; ligand restraints via ProDRG/AceDRG; carbohydrate validation with Privateer. Structures determined: apo (7ERY), complexes with UDP+Reb E (7ES0), UDP+ST (7ES1), UDP+STB (7ERX), H27A+UDP+Reb D (7ES2). - Biochemical LC-MS assays: Reactions (200 µl) with 0.3 mM steviol substrate (Rubu, Reb A, STB, Reb E, Reb D, ST), 1 mM UDP-glucose, 20 mM Tris-HCl pH 7.2; enzyme at defined concentrations (1× or 5×) at 25 °C; aliquots at 0, 2, 18 h quenched by 1-butanol extraction; analyzed by UFLC coupled to AB SCIEX QTRAP 5500 (negative ion mode), Waters ACQUITY UPLC BEH C18 (2.1×100 mm, 1.7 µm); gradient 5–95% acetonitrile over 10 min, 0.5 ml/min; detection at 210 nm; MS parameters: −4500 V, DP −90 V, 500 °C; Q1 200–1200 Da; MS/MS on selected parent ions with 20–60 eV CE. - Steady-state kinetics: UDP-Glo Glycosyltransferase Assay (Promega) quantified UDP production; luminescence linear <25 µM UDP. Substrates: Reb A (R2 β(1→2)), S13G (R1 β(1→2)), Rubu (combined R1+R2 β(1→2)); UDPG at 200 µM (~10× Km). Time-course aliquots every 1 min; ensured <10% substrate and donor consumption. Global nonlinear Michaelis–Menten fitting (n=3 replicates) produced kcat, Km, kcat/Km. - Mutagenesis/engineering: Rational mutations targeting aglycone tunnel flexibility (F208M) and cryptic glucose pocket adjacent to active site (H93A, H93W, F379A) and double mutants (H93W/F208M, F379A/F208M) to modulate β(1→2) vs β(1→6) activities.

- OsUGT91C1 catalyzes β(1→2) glucosylation at both ends of steviol glycosides: • With Rubusoside (Rubu), LC-MS detected formation of ST (via 2-1-R1) and a second isomer Rub-X (via 2-1-R2), both m/z 803 (addition of 162 Da). With higher enzyme, Reb E (m/z 965) formed, confirming sequential β(1→2) additions at both ends. • Reb A is converted to Reb D by adding glucose 2-1-R2. Reb I (with 3-1-R2) is not a substrate, indicating that 3-1-R2 blocks R2 β(1→2) addition and enforcing reaction order (β(1→2) before β(1→3)). - Structural basis of promiscuity and catalysis: • OsUGT91C1 is a monomeric GT-B fold enzyme with two Rossmann-like domains; UDP binds identically across complexes. • Active site catalytic dyad: His27 (general base) hydrogen-bonded to Asp128; H27A mutation abolishes β(1→2) activity at both ends. • Substrates bind in two opposite orientations (R1-in or R2-in), enabling β(1→2) at either end without protein rearrangement (r.m.s.d. ~0.4 Å among complexes; apo vs complex r.m.s.d. ~2.3 Å due to UDP binding). • Key recognition: glucose 1-R2 3- and 4-OH form bidentate H-bonds to Glu283; E283Q/A reduce activity. Trp22 H-bonds to 3-OH of 1-R2, rationalizing inability to accept 3-1-R2 (Reb I). • Aglycone sits in a loose, shallow hydrophobic tunnel (Val129, Phe130, Leu149, Met155, Ile159, Arg162, Ala205, Phe208) allowing 180° rotational swap. - Additional catalytic activity discovered: β(1→6) glucosylation at R1 end. • Structures and LC-MS show OsUGT91C1 adds a 6-linked glucose to glucose 1-R1 on STB and Reb E (products STB-X and Reb E-X). β(1→6) occurs only at R1, and is blocked by presence of 3-1-R1. • Structural mechanism: after β(1→2) at R1, glucose 1-R1 flips 180°, placing 6-OH near His27 and moving glucose 2-1-R1 into an adjacent cryptic pocket (contacts His93, Phe379), clearing the active site for UDP-glucose and enabling β(1→6). - Reversibility observed: With UDP present, OsUGT91C1 catalyzes deglycosylation of β(1→2) linkages (e.g., Reb E to ST and Rub-X, then to Rubu), consistent with reverse transfer back to UDP (H27A inactive). - Enzyme engineering outcomes (kcat/Km, s−1 M−1; fold vs WT): • WT: Rubu 673; S13G 1941; Reb A 445; STB β(1→6) 148. • F208M (more flexible aglycone tunnel): Rubu 1914 (2.8×), S13G 3476 (1.8×), Reb A 1918 (4.3×); preference shift toward R2 addition in Rubu assays. • H93W (blocks cryptic pocket): eliminates β(1→6) on STB (ND); reduces β(1→2) (Rubu 381; S13G 1796; Reb A 172). • F379A (disrupts cryptic pocket): eliminates β(1→6); improves β(1→2): Rubu 2727 (4.1×), S13G 5088 (2.6×), Reb A 523 (1.2×). • Double mutant H93W/F208M: suppresses β(1→6); β(1→2) similar to WT (Rubu 688; S13G 2005; Reb A 535). • Double mutant F379A/F208M: suppresses β(1→6); enhances β(1→2): Rubu 3989 (5.9×), S13G 6014 (3.1×), Reb A 1876 (4.2×); shows preferred R2 addition on Rubu. - Comparison to UGT76G1: Different aglycone positioning leads to β(1→2) (OsUGT91C1, 2-OH to His27) versus β(1→3) (UGT76G1, 3-OH to His25) regioselectivity.

The findings address the core bottleneck in producing superior-tasting steviol glycosides by providing an enzyme, OsUGT91C1, that efficiently installs the critical β(1→2) glucose at the R2 end. Structural analyses reveal that a loose hydrophobic aglycone tunnel enables dual binding orientations (R1-in/R2-in), and a catalytic His27–Asp128 dyad activates the acceptor 2-OH, explaining efficient β(1→2) transfer at both ends. A cryptic adjacent glucose pocket and a flip of glucose 1-R1 rationalize the unexpected β(1→6) activity at R1. Guided mutations can tune this promiscuity: disrupting the cryptic pocket (F379A and/or H93W) suppresses β(1→6), while increasing tunnel flexibility (F208M) enhances desired β(1→2), especially at R2. The best engineered variant (F379A/F208M) combines elimination of β(1→6) with up to ~6-fold gains in catalytic efficiency on relevant substrates and preferred R2 addition, making it a strong candidate catalyst for converting abundant substrates (Rubu, Reb A) into Reb E/D and ultimately Reb M (in combination with UGT76G1). The work clarifies how GT-B fold enzymes can achieve multiple linkages from one active site and provides a structural basis to control and optimize regioselectivity for industrial biocatalysis.

This work biochemically and structurally characterizes rice glycosyltransferase OsUGT91C1 as a single-active-site enzyme capable of three reactions: β(1→2) at R1 and R2 and β(1→6) at R1 on steviol glycosides. Crystal structures in multiple substrate poses uncover the structural origins of promiscuity (dual aglycone orientations and a cryptic glucose pocket) and identify the catalytic dyad (His27–Asp128). Structure-guided engineering yields variants that suppress undesired β(1→6) glucosylation and enhance the desired β(1→2), particularly at the R2 end. The optimized double mutant F379A/F208M shows strong improvements in catalytic efficiency and preferred R2 modification, offering a promising biocatalyst to relieve the bottleneck in producing Reb D and Reb M from abundant precursors. Potential future directions include further optimization for substrate specificity and stability, process-scale validation for industrial production, and integration with β(1→3) transferase UGT76G1 in cascade or whole-cell systems to maximize yields of Reb D/M.

- Organoleptic properties and sweetness potency of newly formed β(1→6) products (STB-X, Reb E-X) were not tested; their desirability remains unknown. - The reverse reaction (deglycosylation) was inferred from product profiles; the production of UDP-glucose during reversal was not directly measured. - In one complex (Reb E), density for glucose 2-1-R2 was absent, possibly due to cleavage or disorder during crystallization. - Reb I (bearing 3-1-R2) is not a substrate for OsUGT91C1, imposing a strict order of operations for R2 end elaboration toward Reb M. - Kinetic analysis with Rubu reports combined R1 and R2 β(1→2) activities and does not deconvolute site-specific rates.