Diversification of 4ʹ-Methylated Nucleosides by Nucleoside Phosphorylases

The growing demand for 4'-modified nucleoside analogs in medicinal and biological chemistry is contrasted by the challenging synthetic access to these molecules and the lack of efficient diversification strategies. Herein, we report the development of a biocatalytic diversification approach based on nucleoside phosphorylases, which allows the straightforward installation of a variety of pyrimidine and purine nucleobases on a 4'-alkylated sugar scaffold. Following the identification of a suitable biocatalyst as well as its characterization with kinetic experiments and docking studies, we systematically explored the equilibrium thermodynamics of this reaction system to enable rational yield prediction in transglycosylation reactions via principles of thermodynamic control.

The growing demand for 4'-modified nucleoside analogs in medicinal and biological chemistry is contrasted by the challenging synthetic access to these molecules and the lack of efficient diversification strategies. Herein, we report the development of a biocatalytic diversification approach based on nucleoside phosphorylases, which allows the straightforward installation of a variety of pyrimidine and purine nucleobases on a 4'-alkylated sugar scaffold. Following the identification of a suitable biocatalyst as well as its characterization with kinetic experiments and docking studies, we systematically explored the equilibrium thermodynamics of this reaction system to enable rational yield prediction in transglycosylation reactions via principles of thermodynamic control.
File list (2) download file view on ChemRxiv Manuscript_Me-nucleosides_ChemRxiv.pdf (3.87 MiB) download file view on ChemRxiv SI_Me-nucleosides_ChemRxiv.pdf (11.41 MiB) The growing demand for 4ʹ-modified nucleoside analogs in medicinal and biological chemistry is contrasted by the challenging synthetic access to these molecules and the lack of efficient diversification strategies. Herein, we report the development of a biocatalytic diversification approach based on nucleoside phosphorylases, which allows the straightforward installation of a variety of pyrimidine and purine nucleobases on a 4ʹ-alkylated sugar scaffold. Following the identification of a suitable biocatalyst as well as its characterization with kinetic experiments and docking studies, we systematically explored the equilibrium thermodynamics of this reaction system to enable rational yield prediction in transglycosylation reactions via principles of thermodynamic control.
Nucleosides are central biomolecules that play key roles in a variety of cellular processes by serving as enzymatic cofactors, building blocks of DNA and RNA and energy transport systems. As such, modified nucleosides mimicking their natural counterparts have a long history in medicinal and biological chemistry. [1][2][3][4] Today, modified nucleosides are indispensable pharmaceuticals for the treatment of various types of cancer and viral infections and further represent important tools in chemical biology for a spectrum of imaging applications. [5,6] Despite the great demand for these molecules, the synthesis of nucleosides is still regarded as challenging and inefficient. [7] While nucleosides with ribosyl or 2ʹ-desoxyribosyl moieties can be accessed from naturally occurring nucleosides or carbohydrates, [7][8][9][10] the preparation of sugar-modified nucleosides typically suffers from lengthy reaction sequences and low total yields. [11][12][13][14][15][16][17][18] Furthermore, a heavy reliance on protecting groups entails low overall efficiencies [7] and several sugar modifications at the 2ʹ or 4ʹ positions are known to limit diastereoselectivity in glycosylation approaches, [19,20] severely complicating the synthetic access to many target compounds. More importantly, established routes typically exhibit a lack of divergence as they tend to be specific to one nucleoside. As such, the introduction of desired substitutions at the nucleobase often requires complete or partial re-synthesis of the target molecule since a general strategy for the efficient diversification of modified nucleosides has not been reported to date (Scheme 1, top). With the advent of scalable routes for the de novo synthesis of selected 4ʹ-modified nucleoside analogs, as reported recently by Britton, [21] such a diversification strategy would readily provide access to a variety of sought-after nucleosides.
We envisioned that nucleoside phosphorylases could provide a biocatalytic platform for late-stage diversification of 4ʹ-modified nucleosides. These enzymes catalyze the reversible phosphorolysis of nucleosides to the corresponding nucleobases and pentose-1-phosphates via an SN2-like mechanism. [22,23] The reaction sequence involving phosphorolysis of one nucleoside and in situ reverse phosphorolysis to the target nucleoside is generally known as a transglycosylation, and effectively transfers the sugar moiety from one nucleobase to another. [24] While this reactivity is well-established for ribosyl and 2ʹdesoxyribosyl nucleosides [9] and a few 2ʹ-modified nucleosides (Scheme 1, center), there are no examples in the literature of the enzymatic synthesis of 4ʹ-modified nucleosides, except for Merck's recent report of a 5-step enzymatic cascade for the synthesis of the 4ʹ-alkynylated nucleoside drug Islatravir. [25] Therefore, the feasibility of transglycosylation reactions with 4ʹ-modified nucleosides as well as the thermodynamics of such a cascade process are notably underexplored. Herein, we address this gap by reporting on the phosphorolysis and transglycosylation of the simplest 4ʹ-alkylated pyrimidine nucleoside, 4ʹmethyluridine (1a). Following the identification of a suitable biocatalyst, and a characterization of its reactivity with kinetic experiments and docking studies, we explored the thermodynamics of the phosphorolysis of 1a and leveraged this information in transglycosylation experiments to access a range of 4ʹ-methylated pyrimidine and purine nucleosides.
In the absence of obvious pyrimidine nucleoside phosphorylase (PyNP) candidates for the phosphorolysis of 1a, we began our investigation by screening a small panel of PyNPs with known broad substrate spectra. To our surprise, only the PyNP from Thermus thermophilus (TtPyNP) [26,27] showed measurable conversion of 1a under screening conditions ( Figure S1). Other broad-spectrum PyNPs, such as those from Geobacillus thermoglucosidasius (GtPyNP) [28] or Bacillus subtilis, [23] displayed no activity with 1a ( Figure 1A). To substantiate the observed conversion of 1a by TtPyNP, we performed a series of control experiments. Reactions either without Scheme 1. Synthesis and biocatalytic diversification of nucleosides with modified sugars. NB = Nucleobase, NP = nucleoside phosphorylase. phosphate, without enzyme or with denatured enzyme gave no conversion. Similarly, no conversion was observed under reaction conditions outside of the working space of TtPyNP (pH 3 or pH 12, Figure 1A). [26] NMR analysis of a reaction mixture with TtPyNP and 1a corroborated the proposed reactivity and creation of the pentose-1-phosphate 3, as evident from the rise of an additional 1 H NMR signal at 5.57 ppm showing a strong H,P-HMQC signal ( Figure 1B). Consistent with the native reactivity of PyNPs, inversion at the anomeric position was evident by this signal lacking NOE contacts to the 4ʹ-methyl group of 3, while the corresponding anomeric proton in 1a showed clear correlation to the methyl substituent.
Having established the activity of TtPyNP with 1a, we conducted kinetic experiments to provide further insights into this enzymatic transformation. Although TtPyNP is inhibited by pyrimidine nucleobases such as uracil (2a), [26] we could observe Michaelis-Menten behavior of the enzyme with 1a ( Figure 1C). Interestingly, the apparent Michalis-Menten constant KMʹ of the phosphorolysis of 1a (KMʹ = 3.37 mM) indicated that TtPyNP has a much lower affinity for 1a compared to natural nucleosides like uridine or thymidine (KM < 1 mM), [27] suggesting that productive binding of the modified substrate 1a might present a challenge due to the increased steric bulk. In addition to a lower affinity for 1a, TtPyNP also displayed a lower rate constant compared to uridine (0.59 vs 5.05 s -1 for 1 mM substrate at 60 °C and pH 9) [26] which showed a similar temperature-dependence as indicated by phosphorolysis experiments at different temperatures monitored by UV spectroscopy ( Figure 1D). [29,30] Collectively, these results demonstrate that, unlike other nucleoside phosphorylases, TtPyNP selectively converts the 4ʹ-methylated nucleoside 1a to the corresponding sugar phosphate 3, albeit with a lower rate constant and substrate affinity compared to the native substrates.
Next, we performed preliminary in silico docking studies to rationalize why 1a is only converted by TtPyNP and not by other closely related and highly promiscuous enzymes such as GtPyNP. We hypothesized that conversion of this substrate would primarily be limited by steric hindrance during substrate binding, since i) uridine and 1a only differ by a single methyl group distant from the anomeric position and ii) TtPyNP displays significantly lower affinity for 1a than for uridine. PyNPs generally exhibit marked flexibility during their catalytic cycle with a transition from an open conformation to a closed state requiring a domain movement of approximately 8 Å. [31] Since all first sphere residues in the closed state are highly conserved and identical between the tested PyNPs, we anticipated that initial binding in the open conformation would be a limiting factor, as TtPyNP offers slightly more space than GtPyNP due to a threonine-serine substitution at the back of the active site, as evident from sequence alignments. [28] To examine this hypothesis, we obtained an X-ray crystal structure of GtPyNP at 1.9 Å resolution (see Supporting Information for details; PDB ID 7m7k) and used AutoDock Vina implemented in YASARA to dock uridine and 1a into the open conformations of this structure and the known Xray crystal structure of TtPyNP (PDB ID 2dsj). [32] Docking of uridine and 1a into TtPyNP yielded structures in good agreement with the native mode of substrate binding via Hbonding to the nucleobase and positioning of the anomeric carbon near the phosphate binding pocket (Figures 2A and  2B). Likewise, uridine could be docked into GtPyNP in a similar position to the cocrystallized substrate (Figures 2C  and S8), where the 4ʹ-position of uridine is located in proximity to Thr84 (Ser83 in TtPyNP). However, we were unable to obtain sensible docking results for 1a with GtPyNP as the increased steric bulk at the 4ʹ-position consistently led to a rotation of the sugar scaffold into an unproductive pose ( Figure 2D). This suggested that the subtle space-creating mutation to a serine in TtPyNP might be a key factor for conversion of 1a. Consistent with this conclusion, the slightly more sterically congested TtPyNP-S83T mutant significantly lost activity compared to the parent enzyme (kobs = 0.25 s -1 vs kobs = 0.59 s -1 , Table S2), while the reverse substitution in GtPyNP installed a low but measurable level of activity in this enzyme (kobs = 0.02 s -1 for GtPyNP-T84S). Moreover, all other enzymes we screened initially, and which were inactive with 1a, also possess a threonine at this position, which likely impedes their ability to bind this substrate productively. Although such subtle but crucial space-creating mutations are rare, there is precedent from other enzymes in the literature. [34] Together, these results indicate that sufficient space in the open conformation of PyNPs is a prerequisite for conversion of sterically more demanding substrates such as 1a. Clearly, there are other factors influencing the rate constant of this Figure 1. Phosphorolysis of 4ʹ-methyluridine (1a). The data for uridine in D were taken from ref. 26. Please see the Supporting Information for details and the externally hosted supplementary information for raw data. [33] transformation, as evident from the order of magnitude difference between the rate constants of the active enzymes, but these must arise from mutations far from the active site, as all other residues in possible contact with the substrate are identical between the tested enzymes.
Since the phosphorolysis of ribosyl and 2ʹ-desoxyribosyl nucleosides is under tight thermodynamic control, [23] we were then interested in the thermodynamics and reversibility of the phosphorolysis of 1a to enable a diversification of the scaffold via transglycosylation. Time-course experiments with 1a and varying excesses of phosphate revealed incomplete conversion of the substrate, with the equilibrium positions being consistent with an equilibrium constant K of 0.16 (at 60 °C and pH 9, Figure 1D). Further experiments to monitor the equilibrium at 75 °C and 90 °C revealed that the phosphorolysis of 1a has an apparent reaction enthalpy Δ ′ of 8.9 kJ mol -1 and an apparent reaction entropy Δ ′ of 11.7 J mol -1 K -1 ( Figure S2). Interestingly, these values closely resemble the equilibrium constants and thermodynamic parameters of the phosphorolysis of uridine, [23] indicating that substitutions distant from the anomeric center have little influence on the equilibrium thermodynamics of nucleoside phosphorolysis. These results also pointed to the reversibility of this transformation, opening the door for transglycosylation reactions from the sugar phosphate 3 to yield other nucleosides.
With a solid understanding of the thermodynamics and kinetics of the phosphorolysis of 1a by TtPyNP, we proceeded to diversify this scaffold by subjecting the sugar phosphate 3 to subsequent enzymatic catalysis with different nucleobases in situ. Using this transglycosylation approach ( Figure 3A and Scheme 1, center), we aimed to access a variety of 4ʹ-methylated nucleosides from 1a in a one-pot manner. After confirming the stability of 3 through equilibrium shift experiments ( Figure S6), [35] we subjected 1a to phosphorolysis using only minimal phosphate in the presence of different pyrimidine nucleobases 2b−2e belonging to a panel of 5-substituted uracil analogs ( Figures  3A and 3B). Analysis of the reaction mixtures by HPLC revealed consumption of 1a and the respective uracil analog with concurrent formation of new products ( Figure 3B), which HRMS analysis identified as the nucleoside products arising from glycosylation of 2b−2e with 3. Equilibrium state thermodynamic calculations [24] based on transglycosylation experiments with different sugar donor concentrations revealed apparent equilibrium constants of phosphorolysis of 0.12−0.73 for these products 1b−1e ( Figure 3B and S3). The trifluoromethylated pyrimidine 2f could also be converted, although the instability of the starting material and product in aqueous solution [36]  [a] 2f is converted, but 1f and 2f hydrolyse to the corresponding carboxylates under the reaction conditions. [b] Reaction mixtures with purines additionally contained the purine nucleoside phosphorylase from Geobacillus thermoglucosidasius (PNP). Please see the externally hosted supplementary information for raw data and calculations. [33] precluded us from obtaining equilibrium data ( Figure S4). A similar elaboration of in situ generated 3 with purine nucleobases proceeded smoothly using the promiscuous purine nucleoside phosphorylase from Geobacillus thermoglucosidasius. [37] Notably, the adenosine analogs 1g−1i were generated in much higher conversions, corresponding to equilibrium constants of phosphorolysis of 0.01−0.02, reflecting the more favorable thermodynamics typically observed for 6-aminopurines. [35,[38][39][40] The guanosine and inosine analogs 1j and 1k could also be accessed, although with lower conversions indicative of higher equilibrium constants ( Figure 3C). These experiments not only confirmed that nucleoside transglycosylations with the methylated precursor 1a can deliver a range of modified nucleosides in a one-pot manner, but also that the equilibrium thermodynamics of this system largely resemble those of the well-described ribosyl nucleosides. These findings further indicated that these transglycosylations would offer themselves to rational reaction engineering using established principles of thermodynamic reaction control to predict and maximize conversions in these reactions. [24] Indeed, thermodynamic calculations based on the obtained equilibrium constants suggested that 1b could, for instance, be obtained in 84% conversion from 1a using 4 equivalents of nucleobase, which we confirmed experimentally ( Figures 3D and S5). Similarly, 1i could be obtained in quantitative conversion with 4 equivalents of 2i, in agreement with our predictions. As a proof of synthetic utility, we subjected 1a to transglycosylation with 5 equivalents of 2e and obtained the iodinated 1e in 68% conversion (61% predicted) and ca. 40% isolated yield.
In conclusion, we identified and characterized TtPyNP as a biocatalyst for the diversification of 4ʹmethylated nucleosides. Reversible phosphorolysis of a methylated precursor 1a yields stable the pentose-1phosphate 3 which can be employed as a sugar synthon to access a range of modified nucleosides in one pot. Our investigations revealed that sufficient space near the active site in the open conformation of PyNPs appears crucial for binding and conversion of 1a. Furthermore, the equilibrium thermodynamics of the phosphorolysis of 4ʹ-methylated nucleosides largely resemble those of ribosyl nucleosides, indicating that substitutions distant from the anomeric position have only minor effects on the conversions in these systems. Leveraging principles of thermodynamic reaction control enabled us to access a spectrum of 4ʹ-methylated nucleosides bearing different pyrimidine and purine bases in transglycosylation reactions. Lastly, we expect that other 4ʹ-modified nucleoside analogs can be obtained with such biocatalytic systems in a similar fashion (probably with comparable equilibrium thermodynamics), although bulkier 4ʹ-substitutions will likely require some extent of protein engineering to improve activity. Author contributions (with definitions as recommended by Brand et al. [1] )

Data availability
All data depicted visually in the items in the main text (Figures 1−3

) as well as in the Supplementary
Information (Figures S1−S16, see below) are available as tabulated data from the text below and from the externally hosted Supplementary Information at zenodo.org. [2] The data and model of GtPyNP with bound uridine were deposited to the Protein Data Bank (PDB) under accession code 7m7k.

General remarks
All chemicals used in this study were of analytical grade or higher and purchased from Sigma Aldrich (Steinheim, Germany), Carbosynth (Berkshire, UK), Carl Roth (Karlsruhe, Germany), TCI Deutschland (Eschborn, Germany) or VWR (Darmstadt, Germany) and used without prior purification. 4′-Methyluridine (1a) was synthesized as described recently. [3] Water deionized to 18.2 MΩ•cm with a Werner water purification system was used for the preparation of all enzymatic reactions as well as purification and storage buffers. For the preparation of NaOH solutions for quenching, deionized water was used. Analytical HPLC analyses were carried out with an Agilent 1200 series system equipped with LibreOffice, spectral unmixing with data_toolbox, [5,6] modelling and docking in YASARA and protein viewing in ChimeraX. [7] Crystallographic software is described below. 4

Experimental details and supplementary items
Wild-type enzymes were cloned as described in previous reports [5,8] and the available glycerol stocks of the enzymes from previous projects [5,9,10] were used directly for this work. BsPyNP was obtained as a freeze-dried enzyme from Sigma Aldrich and dissolved to 1 g L -1 in 2 mM phosphate buffer (pH 7).
Cloning of the mutant enzymes was carried out via BamHI/HindIII sites in the plasmid pGW3 (gift by Matthias Gimpel, unpublished). Codon-optimized genes were obtained (GeneArt Invitrogen/Thermo Fisher Scientific, Massachusetts, USA) and cloned into pGW3 using the recipient strain Escherichia coli DH5α. The correct sequence was confirmed with Sanger Sequencing (LGC Genomics, Berlin, Germany).
pGW3 is a 2 nd generation derivative of pCTUT7, which was optimized with respect to tightness of the LacO in comparison to the 1 st generation derivate used in a previous work. [8] Protein expression and purification was performed as described recently [5,8] in Escherichia coli BL21 using the EnPresso protocol for 50mL (Enpresso, Berlin, Germany). Briefly, all enzymes were heterologously expressed in E. coli as His6-tagged proteins through IPTG-induced overexpression.
Purification was achieved through cell disruption, heat treatment of the crude extract (80 °C for . Samples were withdrawn at timely intervals after reaction initiation, as detailed in the metadata files freely available online. [2] 6 Reaction monitoring of phosphorolysis reactions was achieved via spectral unmixing. From live reactions, samples were withdrawn and quenched in 100 mM aqueous NaOH as described previously. [5,6] Sample dilution factor was adjusted to reach final concentrations of 100−200 µM UVactive reaction components (please note that the exact concentration is not relevant here since spectral unmixing only takes spectral shape and not absolute intensity into account). For instance, from reactions with 1 mM 1a 50 µL of the reaction mixture were pipetted into 250 µL 100 mM NaOH for quenching and dilution. Of the diluted alkaline sample, 200 µL were transferred to UV/Vis-transparent 96-well plates (UV star, GreinerBioOne, Kremsmünster, Austria) for analysis. UV absorption spectra were recorded from 250−350 nm with a BioTek PowerWave HT platereader and subjected to spectral unmixing using analogously obtained reference spectra of 1a and 2a. [2] Reference spectra used in this study are freely available from the externally hosted Supplementary Information. [2] The degree of conversion was determined directly from the spectral fit which considers the UV-active substrate and product in relation to one another. [5] For activity determination, only sampling points showing 3−10% conversion of the nucleoside substrate were considered. This lower bound was set due to the inherent inaccuracy of the UV-based method employed (roughly ±0.3 percentage points, due to the inherent error in spectral acquisition, as described in the original publication) [5] and the upper bound was applied as recommended by Cornish-Bowden [11] for equilibrium reactions. All datapoints outside this window were not included for calculation of activity and marked accordingly in the datasets available in the Supplementary Information. [2] Datapoints that displayed baseline shifts or other spectral anomalies were also excluded from consideration. Background correction was performed as described recently. [6] Experimental spectra were fitted either across the entire spectrum or over one of the information-rich shoulder regions of pyrimidine nucleosides/nucleobases, as appropriate for the analysis. All background corrections and the corresponding datafiles are detailed in the metadata files in the externally hosted supplementary information. [2] Enzymatic activity was determined by linear approximation of the conversion over time with a forced intercept at the origin. All raw data and the datapoints considered for calculation are freely available online with outliers and excluded datapoints clearly marked. [2] The observed rate constant was obtained by considering the degree of conversion (mol per second) per mol enzyme applied, using the molar extinction coefficient of TtPyNP of 26,930 cm -1 M -1 as predicted by Protparam [12] (i.e. the stock solution of 1 g L -1 had a concentration of 37.1 µM).
Enzyme screening ( Figure 1A) was performed using reaction mixtures of 1 mM 1a, 20 mM potassium phosphate and 30 µg mL -1 enzyme (TtPyNP, GtPyNP, BsPyNP or EcTP) at 50 °C in 50 mM MOPS buffer pH 7 in a final volume of 50 µL. These reactions were carried out at a neutral pH in this buffer system 7 to accommodate for the working space of the enzymes used. Later reactions were performed at pH 9 since TtPyNP retains excellent activity and stability under alkaline conditions [9] and pentose-1phosphates (such as 3) are much more resistant to hydrolysis under alkaline conditions. [13,14] The reactions were quenched by addition of 250 µL 100 mM NaOH to the reaction mixtures after 30 min.
The resulting samples were analyzed by UV spectroscopy as described above. For each protein, control reactions with uridine were performed under identical conditions, all of which gave conversion of the nucleoside to or near equilibrium ( Figure S1). For both uridine and 1a, control reactions without protein were carried out, which resulted in no conversion of the starting material. only submaximal activity at 50 °C. [9] Only TtPyNP displayed appreciable activity with 1a (B). The raw data are available online. [2] For illustrative purposes, all UV spectra shown here were background corrected and normalized to the isosbestic point of base cleavage (271 nm for this substrate). [6] The control reactions for TtPyNP ( Figure 1A) [9] prior to addition of 1a. The reaction at pH 3 was carried out in a buffer mix consisting of 5 mM citrate, 10 mM MOPS and 20 mM glycine (all final concentration) adjusted to pH 3.
The reaction at pH 12 was carried out with 25 mM NaOH instead of MOPS buffer. All reactions were quenched through addition of 200 µL 100 mM NaOH and analyzed as described above.
The NMR spectra of the sugar phosphate 3 ( Figure 1B) were recorded directly from a reaction mixture.     Table S1.
Raw data and calculations for this experiment are freely available online. [  The temperature-dependence of the activity of TtPyNP with 1a ( Figure 1D) was determined using reaction mixtures of 1 mM 1a and 50 mM potassium phosphate in 50 mM glycine/NaOH buffer at pH 9 and the indicated temperature in a total volume of 150 µL. Depending on the temperature (and, therefore, on rate of phosphorolysis), 6−24 µg mL -1 TtPyNP were used (6 µg mL -1 for 70 °C, 12 µg mL -1 for 60 °C and 24 µg mL -1 for 50 °C), to permit sampling of all reactions within the same time domain.
The thermodynamic control of the phosphorolysis of 1a ( Figure 1E) was probed with reaction mixtures of 1 mM 1a and 100 µg mL -1 TtPyNP (3.71 µM, 0.37 mol%) in 50 mM glycine/NaOH buffer at pH 9 and 60 °C in a total volume of 160 µL with either 2, 5, 10 or 20 mM potassium phosphate (equivalent to 2, 5, 10 or 20 equivalents of phosphate over the nucleoside 1a). The reaction mixtures were incubated in a PCR cycler with lid heating (70 °C). Samples of 25 µL were quenched in 225 µL 100 mM NaOH after 2, 8, 30, 60, 111 and 165 min and analyzed by spectral unmixing as described above and in the metadata files available online. [2] Likewise, the raw data for this experiment are freely available in the externally hosted supplementary information. [2] The obtained data were fit to equation (S4) which was derived as detailed below.

Derivation of equation (S4) for the determination of phosphorolysis equilibrium constants from phosphorolysis equilibria with different phosphate excesses:
Nucleoside phosphorolysis is a thermodynamically controlled reaction which closely adheres to the law of mass action. [ This quadratic equation can be rewritten to equation (S14) and solved via equations (S15) and (S16). Information. [2] Figure S2. Temperature-dependence of <= . Data for 60, 75 and 90 °C were used for calculation. (Table S2,  and 60 min, quenched in 300 µL 100 mM NaOH and analyzed by spectral unmixing. Kinetic constants were calculated as described above, using the extinction coefficient of 21,890 cm -1 M -1 of GtPyNP. The raw data and calculations for this experiment are available online. [2]   (13,000 rpm, 10 min) and analyzed by HPLC. Conversion was calculated according to equation (S18), which assumes that the molar extinction coefficients of the nucleobase 2 and the corresponding nucleoside 1 are equal.

The activity of the mutant PyNPs
where $ 9 is the conversion in the transglycosylation reaction (i.e. conversion of the nucleobase to the target nucleoside), . < is the peak area of the target nucleoside (1) and . > is the peak area of the nucleobase (2). Typical retention times of the compounds used herein are given in Table S3. The identity of all target compounds was confirmed by high-resolution mass spectrometry (HRMS) as detailed below in Table S4. The raw data for all HPLC runs used for calculation of equilibrium constants are freely available online. [2] Since opening and processing of the files requires Agilent software, all chromatograms are depicted below ( Figure S3) and integration results are listed in Table S3. The equilibrium constants of phosphorolysis of the target nucleoside were determined by fitting the conversion of the nucleobase to the corresponding nucleoside as a function of the excess of 1a according to equation (S19), which is derived below.

Derivation of equation (S19) for the determination of phosphorolysis equilibrium constants from transglycosylation equilibria with different excesses of the donor nucleoside:
The basis for this equation is given by an expression reported and derived in our previous report (equation (4) in the original paper). [16]  nucleoside ( ) ; as described in a previous work). [16] Since this deviation is well within the inherent error of HPLC, equation (S19) provides an accurate output for realistic experimental data.  Figure S3.    labile to the alkaline conditions needed for stability of 3. Therefore, neither remaining starting material (as expected for a thermodynamically controlled reaction), nor product can be observed at pH 9 after 4 h (B). At pH 7 and with 4 equivalents of 2f, clear product formation is visible, which was confirmed by HRMS. However, significant hydrolysis is also apparent under these conditions, as is obvious by the large peak at the solvent front, corresponding to the hydrolysis product 2f*. Formation of 2f* as well as 1f* was also confirmed by HRMS analysis (Table S4, see below). were also weakly detected.  Table S3. The stability of the sugar phosphate 3 was assessed through an equilibrium shift experiment which provided stability information for this compound without having to isolate it or detect it directly (for details on the approach and equations, please see our method paper). [14] To this end, reaction mixtures for this experiment are freely available online [2] and the fit results are shown in Figure S6. Considering the data reported by Bunton [13] and us, [14]  NMR experiments with a reaction mixture provided further insights into the stability of 3 under relevant conditions. In phosphate buffer at pH 7, we observed no loss of 3 from a reaction mixture incubated for 2 months at room temperature, indicating that the sugar phosphate is quite stable under moderate conditions. However, 3 is, like other sugar phosphates, labile to acidic conditions. At pH ≈1

Reactions with excess of nucleobase
(achieved via addition of HCl to a reaction mixture in equilibrium), full hydrolysis of 3 was apparent by disappearance of the signal corresponding to the anomeric proton after 1 month of incubation at room temperature ( Figure S16).
4′-Methyl-5-iodouridine (1e) was prepared by TtPyNP-catalyzed transglycosylation. To this end, 5iodouracil (2e, 13.9 mg, 0.059 mmol, 5 equivalents) and 4'-methyluridine (1a, 3 mg, 0.012 mmol, 1 equivalent) were dissolved in 40 mL 10 mM glycine buffer (pH 9) with 0.09 mM potassium phosphate (0.3 equivalents) and 4 µg mL -1 TtPyNP (0.15 µM, 0.05 mol%). The reaction mixture was intentionally kept very diluted since TtPyNP is inhibited by nucleobases such as 2e and more concentrated mixtures severely compromise the productivity of the enzyme. The reaction mixture was heated to 60 °C in a water bath. After 3 d, HPLC analysis revealed 68% conversion of 1a to the iodinated analogue 1e (please see Figure S7 and the externally hosted Supplementary Material for HPLC trace). [2] The mixture was then concentrated to ≈7 mL in vacuo, filtered to remove precipitated protein and injected into preparative HPLC. An HPLC method consisting of 10 min isocratic elution with 1% MeCN in water, followed by linear gradient to 10% MeCN over 40 min, cleanly afforded 1e after 35 min retention time.
The fraction containing 1e was concentrated in vacuo. Quantification of recovered 1e proved surprisingly difficult and inaccurate since the compound is quasi-intractable and practically insoluble in all solvents we tried. HRMS data were collected directly from the dilute eluate from the preparative HPLC and 1 H-NMR analysis was performed with a saturated solution of 1e in D2O (ca. 0.5 mM, Figure   S14). We estimate the isolated yield to be around 1.5−2 mg, corresponding to around 40% from 1a.  of 1a were converted. Please note that 1a/2a and 1e/2e have significantly different extinction coefficients at 260 nm. [5,6] Docking of uridine and 1a was performed by using the crystal structure of TtPyNP (PDB ID 2dsj) as a receptor structure. Dockings were performed in AutoDock VINA [17] implemented in YASARA (Yet Another Scientific Artificial Reality Application). All water molecules were removed from the structure prior to the docking calculation. The receptor was treated as a rigid structure and the substrate was treated as a flexible molecule. Point charges on 2dsj were initially assigned according to the AMBER99 [18] force field and point charges on the nucleosides were generated with AM1-BCC. [19] Docking results obtained for each ligand with the receptor were analyzed based on docking energy (kcal mol -1 Figure S8. Superposition of the proteins with docked (orange sticks) and cocrystallized uridine (white sticks) in the GtPyNP active site. Only the original crystal structure is shown. Residues interacting with the nucleobase (R168, S183, K187) and the relevant threonine (T84) are shown as grey sticks.

Crystallographic methods
The crystal structure of GtPyNP bound to uridine at a resolution of 1.9 Å was determined to enable a comparison of TtPyNP and GtPyNP via YARASA-docking of the uridine and 1a ligands (Table S5). The structure of GtPyNP revealed the typical two-domain architecture of a NP-II family PyNP enzyme, [20] composed of an α-helical (α) domain and a mixed α-helical and β-sheet (α/β) domain ( Figure S9A).
Inspection of the active site after molecular replacement revealed positive density in the unrefined F0-Fc density map, suggesting presence of the uridine ligand within the catalytic pocket ( Figure S9B).
Modelling of the uridine revealed that the substrate is recruited to the active site by a set of specific interactions, mediated by polar and aliphatic side chains extending from the cleft in between the αand α/β-domains ( Figure S9C). Notably, the positive electron density in the unrefined F0-Fc density map is more pronounced for the uracil moiety, as compared to the ribose residue ( Figure S9B), suggesting flexibility of the ribose, or degradation of the substrate. It is possible that sulfate, which was present in the crystallization solution, might have allowed for partial turnover of the substrate in crystallo. This phenomenon has been observed for the structurally only distantly related uridine phosphorylases, [21] but is, to the best of our knowledge, unprecedented for pyrimidine nucleoside phosphorylases. Thus, we ascribe the lower electron density observed for the sugar moiety to the flexibility of this moiety in the open confirmation. Inspection of crystal packing contacts further revealed that GtPyNP homodimerizes via the α-domains, as suggested from the interaction with a symmetry mate ( Figure   S9D) and as expected for NP-II family PyNP enzymes. [20] The cloning, expression, and purification of GtPyNP were adjusted from the methods described above since the original construct did not crystallize. The gene encoding GtPyNP was amplified by PCR using oligonucleotides A&B (see below) and cloned into pET-24d (Novagen) using the restriction enzymes Diffraction data were collected at Beamline ID29 of the European Synchrotron Radiation Facility (Greanoble, France). [22] Data were processed with the XDS program package for data reduction, [23] and merging and scaling was performed using the AIMLESS program as implemented in the CCP4 package. [24] The data set was solved by molecular replacement using the crystal structure of Bacillus stearothermophilus PyNP (PDB ID 1brw, chain A) [25] via the CCP4 implemented program Phaser. [26] Coot [27] in combination with phenix.refine, as implemented in the Phenix software suite, [28] were used for iterative model building and refinement.  NMR spectra (the full raw data are freely available online [2] and tabulated data are listed above) Figure S10. 1