Enzymatic Generation of Double‐Modified AdoMet Analogues and Their Application in Cascade Reactions with Different Methyltransferases

Abstract

Methyltransferases (MTases) have become an important tool for site‐specific alkylation and biomolecular labelling. In biocatalytic cascades with methionine adenosyltransferases (MATs), transfer of functional moieties has been realized starting from methionine analogues and ATP. However, the widespread use of S‐adenosyl‐l‐methionine (AdoMet) and the abundance of MTases accepting sulfonium centre modifications limit selective modification in mixtures. AdoMet analogues with additional modifications at the nucleoside moiety bear potential for acceptance by specific MTases. Here, we explored the generation of double‐modified AdoMets by an engineered Methanocaldococcus jannaschii MAT (PC‐MjMAT), using 19 ATP analogues in combination with two methionine analogues. This substrate screening was extended to cascade reactions and to MTase competition assays. Our results show that MTase targeting selectivity can be improved by using bulky substituents at the N⁶ of adenine. The facile access to >10 new AdoMet analogues provides the groundwork for developing MAT‐MTase cascades for orthogonal biomolecular labelling.

An engineered MAT (methionine adenosyltransferase) is capable of generating double‐modified AdoMet analogues (DM‐AdoMets) when provided with non‐natural methionine analogues and nucleoside‐modified ATP analogues. DM‐AdoMets have unique properties and not all MTases (methyltransferases) can easily convert them. This increases selectivity for MTase‐based biomolecular labelling applications, enabling differential MTase targeting wherein multiple MTases are present (i. e., in cell).

Introduction

Methyltransferases for labelling

Exploiting nature's bioalkylation systems has become common for labelling biomolecules via transfer of non‐natural groups. ^[1] Methyltransferases (MTases) utilize S‐adenosyl‐l‐methionine (AdoMet) as co‐substrate to methylate a diverse range of DNA, RNA, protein, or small molecule substrates. Non‐natural AdoMet analogues with modifications at the sulfonium centre have proven useful for the transfer of various functional groups, including photocleavable or clickable ones. ^[2] If a promiscuous MTase is available, this approach proves powerful for site‐specific labelling of a substrate with a group of interest, enabling downstream applications, such as enrichment or modulations of the biological function. ^[3]

Cascade reactions and metabolic labelling approaches

AdoMet analogues can be synthesized chemically or generated enzymatically by halide methyltransferases or methionine adenosyltransferases. Halide methyltransferases use S‐adenosyl‐l‐homocysteine and alkyl iodides to generate the respective AdoMet analogues. ^[4] Methionine adenosyltransferases use ATP and methionine analogues to generate the respective AdoMet analogues. In the present study, we focused on the latter approach, that is, AdoMet analogue generation by methionine adenosyltransferase (MAT).

The MAT reaction can be performed in one pot with the MTase as an enzymatic cascade for chemoenzymatic synthesis and site‐specific transfer[ ² , ⁵ ] (Figure 1A). Methionine analogues as starting point provide two key advantages, namely (i) their high stability (compared to AdoMet analogues that have limited half‐lives in aqueous solution) and (ii) their cellular uptake (whereas AdoMet is not cell‐permeable). These features are exploited in metabolic labelling approaches, in which cells are treated with methionine analogues, AdoMet analogues are formed inside the cells and used by promiscuous MTases. ^[6]

Figure 1.

Cascade reaction and substrate analogues investigated in this study. (A) Methionine adenosyltransferase (MAT) for enzymatic generation of AdoMet and AdoMet analogues. The modification at different sites alters how well the AdoMet analogues are converted by different MTases, providing a way to achieve specific modification with non‐natural residues. (B) Schematic illustration of AdoMet binding pocket of MTases. R¹ (blue), R² (green), and R³ (red) indicate residues at the nucleobase, the ribose, or the sulfonium (S) / selenium (Se) centre (X), respectively. A complete list of NTPs and methionine analogue substrates used in this study is shown in Figure S2.

Lack of selectivity in complex systems

While these approaches have proven powerful to analyse methylation target sites in select biomolecular species (notably RNA) that can be readily separated from other components of the cell, the complexity of the living systems brings unique challenges for molecular labelling in more complex mixtures. The main concerns would be selectivity and efficiency issues if two or more MTases are present at the same time. The challenge to realize selective MTase‐based modifications in complex mixtures or in cells is thus enormous, as there are currently >400 unique MTase entries for Homo sapiens and >60 entries for Escherichia coli in UniProt. ^[7]

In many MTases, the sulfonium centre of AdoMet is solvent‐exposed in the AdoMet binding pocket (Figure 1B, Figure S1), enabling many MTases to accept sulfonium centre‐modified AdoMet analogues. ^[8] Consequently, it can be anticipated that metabolic labelling via methionine analogues would lack selectivity. On the other hand, multiple studies have previously focused on the nucleobase, ^[9] the nucleoside, ^[10] or the terminal carboxylate/amine moiety ^[11] modifications. However, the transferable groups used in those cases were methyl or ethyl, which is not suitable for molecular labelling applications. We therefore focused on both the sulfonium centre and the nucleoside moiety to generate double‐modified AdoMet analogues (DM‐AdoMets) that offer selectivity and labelling applications at the same time.

In previous work, we tested MATs from Cryptosporidium hominis (I122A‐I330A) and Homo sapiens (I117 A) on seven nucleobase‐modified ATP analogues. ^[12] We observed efficient formation of N⁶‐benzylated 5′‐methylthioadenosine (MTA) analogue after eight hours when the natural amino acid substrate l‐methionine was used. However, the formation of DM‐AdoMets was very low, yielding 15 % N7‐propargylation of the eukaryotic mRNA 5′ cap in Ecm1 cascade. As MATs show product inhibition, the MAT reaction could not be assessed individually and only the final product of cascade reactions could be analysed, which made it difficult to discern whether MAT or MTase was limiting.

Here, we tested the substrate scope of a recently reported MAT variant (L147A‐I351A) from Methanocaldococcus jannaschii (PC‐MjMAT). PC‐MjMAT has a number of attractive features: it is (i) thermostable (T_m >90 °C), but also active at lower temperature (37 °C), (ii) it can be recombinantly produced in E. coli with satisfactory yields (∼160 mg enzyme per litre culture), (iii) it exhibits less product inhibition (∼100‐fold) compared to most α‐type MATs. ^[10b] Furthermore, with the L147A‐I351A double mutation, the enzyme shows >300‐fold increased activity on methionine analogue with photo‐cleavable (PC) group (ortho‐nitrobenzyl) ^[2d] , hence the “PC‐” prefix of the variant's name.

As a second step, we assessed the versatility of PC‐MjMAT in one‐pot cascade reactions with three different MTases: Streptomyces niveus 8‐demethylnovobiocic acid C8‐methyltransferase (NovO), ^[13] Rattus norvegicus catechol O‐methyltransferase M40A‐Y200L variant (RnCOMT‐var), ^[14] and Giardia lamblia trimethylguanosine synthase 2 V34A variant (GlaTgs2‐var). ^[15]

The broad substrate spectrum of PC‐MjMAT and its implementation in cascades unveiled a set of nucleoside modifications leading to distinguished conversion by certain MTases. However, there are still major challenges (notably competition with cellular ATP, l‐Met, and AdoMet) to be overcome for the ultimate aim of orthogonal metabolic labelling.

Results and Discussion

PC‐MjMAT efficiently generates N⁶ adenine‐modified AdoMet analogues

We tested the activity of PC‐MjMAT on a panel of 19 ATP analogues and two methionine analogues (Figure 1)(Figure 1B, Figure S2). First, all of the 19 ATP analogues were tested together with the co‐substrate ortho‐nitrobenzyl‐dl‐homocysteine (y‐Hcy). The 10 ATP analogues that led to a reasonable conversion were then additionally tested together with the co‐substrate propargyl‐l‐selenohomocysteine (z‐SeHcy). Consequently, a total of 29 different substrate/co‐substrate combinations were tested on PC‐MjMAT. The set of ATP analogues contains 15 unique base moieties (R¹=a–o) and additional variations at the ribose (R²=1‐3) (Figure 1). Specifically, we tested a set of natural NTPs (N1‐methyl‐ΨTP (a1), ΨTP (b1), UTP (c1), CTP (e1), ITP (f1), GTP (g1), ATP (i1), and 2‐amino‐ATP (j1) ^[16] ) and dNTPs (dTTP (d2), dCTP (e2), dGTP (g2), and dATP (i2)), as well as the non‐natural ribose‐modified nucleotide 2′‐azido‐2′dATP (i3) and purine‐modified nucleotides (2‐aminopurine‐riboside‐5′‐TP (h1), N⁶‐propargyl‐ATP (k1), N⁶‐cyclopentyl‐ATP (l1), N⁶‐phenyl‐ATP (m1), N⁶‐benzyl‐ATP (n1), and 2‐(phenylethynyl)‐ATP (o1)) (Figure 1B). As non‐natural amino acid substrates, the methionine analogues ortho‐nitrobenzyl‐dl‐homocysteine (y‐Hcy) and propargyl‐l‐selenohomocysteine (z‐SeHcy) were tested (R³, Figure 1B). They show unique functional properties and can be used for different downstream applications related to the fields of photochemistry (y‐Hcy) ^[2d] and click chemistry (z‐SeHcy). ^[6b]

First, PC‐MjMAT activity alone was tested on the 19 NTPs initially with y‐Hcy as co‐substrate (Figure S4) by monitoring methionine analogue consumption over time in an HPLC‐based assay. The known product inhibition of MAT is circumvented by decomposition of AdoMet (or analogues) at high reaction temperature (65 °C) into methylthioadenosine (or analogues), which can be detected by HPLC as previously described. ^[2d] Out of these 19 NTPs, eight (h1–n1, i2) led to very good or almost quantitative conversions in combination with y‐Hcy (Figure 2A–C, Figure S4). On the other hand, 9 NTPs (a1–f1, i3, g2) resulted in no conversion at all (Figure S4). Two NTPs (g1, o1) gave mediocre conversions (Figure 2A–C, Figure S4). The results of the eight well‐accepted and the two mediocre NTPs are shown in more detail in Figure 2B, C.

Figure 2.

PC‐MjMAT activity was tested on a set of nucleoside‐modified ATP analogues (g1‐, i2‐, h1‐, i1‐, j1‐, k1‐, l1‐, m1‐, n1‐, and o1‐triphosphates). The extended reaction illustration and the detailed results including the inactive NTPs are given in the Figures S3 and S4, respectively. (A) MAT reaction and the AdoMet modifications at positions R¹, R², and R³ are summarized. (B) For each NTP, y‐Hcy conversion is given for end (6 h) time point. (C) Time course of activity assay for y‐Hcy. Measurements at 0, 0.25, 0.5, 1, 2, 4, and 6 hours. (D) For each NTP, z‐SeHcy conversion is given for end (6 h) time point. Data and error bars show mean values and SEM of three independent experiments. The HPLC traces are given in Figures S15‐S34.

As expected, the NTPs yielding near quantitative conversion (h1–n1, Figure 1B, 2B, C) share a high structural similarity with the natural substrate ATP (i1). The lack of the 2′‐hydroxy group in dNTPs (d2, e2, g2, i2) drastically reduces conversion. Only dATP (i2) could be converted (Figure 2B, C), whereas activity was completely abrogated in the three other dNTPs (Figure S4).

Substitution of purines with pyrimidines (1 a–e) is not tolerated, which can be attributed to weaker π‐stacking and fewer H‐bond donors or acceptors. In the case of purines, the modest activity for GTP (g1) suggests that the exact location of H‐bonding atoms on the ring is also crucial. In guanine, the hydrogen donor N⁶ in adenine is replaced by a hydrogen acceptor oxygen (O⁶) and there is an additional exocyclic amino group (N²). The 2‐amino‐ATP (j1) is very well accepted by PC‐MjMAT, suggesting that N² does not present a steric clash. The 2‐aminopurine‐riboside‐5′‐TP (h1) is also very well accepted, suggesting that the N⁶ H‐bond donor is not essential for binding or that the loss of N⁶ can be rescued by N². This means that mediocre activity on g1 is not the result of a possible N² steric clash, nor it is the result of lacking H‐donor at position 6. Taken together, the presence of H‐acceptor O⁶ itself seems to be detrimental for the activity on g1.

A subset of NTPs tested represented ATP‐analogues with substituents at the important N⁶‐position (k1–n1). The conversion of these substrates was consistently high (Figure 2B), indicating that even bulky substituents at the N⁶ position of adenine are well tolerated. This is further substantiated by time‐dependent measurements, revealing at least 90 % conversion of y‐Hcy within the first hour of the reaction for h1–n1 (Figure 2C). Taken together, these data show that PC‐MjMAT efficiently converts y‐Hcy in combination with most of the purine‐based ATP‐analogues (Figure 2B, C, Figure S4). Using these 10 purine NTPs on which PC‐MjMAT is active, we then tested conversion of a second methionine analogue, z‐SeHcy, as clickable amino acid substrate (Figure 2D). In all cases, we observed a markedly lower conversion for z‐SeHcy compared to y‐Hcy (Figure 2D). Given that propargyl is less sterically demanding than ortho‐nitrobenzyl, the lower activity for z‐SeHcy cannot be explained by steric hindrance. z‐SeHcy features a selenium instead of a sulfonium centre. Despite previous reports on the positive impact of a selenium centre on enzymatic transfer by MTases, ^[17] a positive effect is not apparent for MATs.

MAT‐MTase cascades for enzymatic formation and assessment of double‐modified AdoMet analogues

Based on the finding that PC‐MjMAT can efficiently produce a number of nucleoside‐modified AdoMet analogues with ONB or propargyl as transferable moiety, we next wanted to assess how these double‐modified AdoMet analogues (DM‐AdoMet) perform in cascade reactions with different MTases.

To this end, we chose 10 ATP analogues (g1‐, i2‐, h1‐, i1‐, j1‐, k1‐, l1‐, m1‐, n1‐, and o1‐triphosphates) on which PC‐MjMAT is active in the presence of y‐Hcy or z‐SeHcy, and performed cascade reactions in combination with NovO, RnCOMT‐var, or GlaTgs‐var. We performed a comprehensive screening of substrate/co‐substrate pairs for these three different MTases. Figure 3 shows the cascades reactions for ATP analogues, which are most relevant for potential application for selective MTase targeting. In total, we performed 55 biotransformations (out of a total of 60 potential biotransformations, Figure S6).

Figure 3.

MAT/MTase cascade reactions and time courses generating and using a select set of nucleoside‐modified AdoONB (upper panels) and SeAdoYn analogues (lower panels) are given. (A) PC‐MjMAT/NovO cascade reaction. The enzyme activity is given as THPC conversion (%). The HPLC traces are given in Figures S35‐S54. (B) PC‐MjMAT/RnCOMT‐var cascade reaction. The enzyme activity is given as DHBAL conversion (%). The HPLC traces are given in Figures S55‐S74. (C) PC‐MjMAT/GlaTgs2‐var cascade reaction. The enzyme activity is given as m⁷GpppA conversion (%). The HPLC traces are given in Figures S75‐S94. Datasets for all ligands including duplicates for end time points are given in Figure S6.

NovO is an 8‐demethylnovobiocic acid C8‐methyltransferase from Streptomyces niveus and catalyses the methylation of THPC (4,5,7‐trihydroxy‐3‐phenylcoumarin) at the C8 (Figure 3A). It was previously reported to transfer extended alkyl groups with unsaturated bonds in β‐position as well as benzylic residues. ^[18] In cascade reactions with PC‐MjMAT forming a number of AdoMet analogues, NovO transferred the ONB group more efficiently than the propargyl group in all cases, i. e., from AdoONB and the respective DM‐AdoMets (Figure 3A, Figure S6). This partially reflects the efficiency of DM‐AdoMet formation by PC‐MjMAT, which was also higher for y‐Hcy than for z‐SeHcy. In addition to ATP (i1), 2‐amino‐ATP (j1) led to quantitative transfer in the case of y‐Hcy and mediocre transfer in the case of z‐SeHcy (Figure S6). The ATP analogues k1 and l1 with aliphatic substituents at the N⁶ position also led to mediocre conversions in combination with y‐Hcy, whereas m1, n1 with aromatic substituents were barely converted, suggesting a negative impact of the aromatic ring in these cases. In the case of z‐SeHcy most ATP analogues led to no notable conversion (Figure 3A).

RnCOMT, the Rattus norvegicus catechol O‐methyltransferase catalyses the methylation of hydroxyl groups in DHBAL (3,4‐dihydroxybenzaldehyde). The double variant M40A−Y200 L was reported to preferentially methylate the 3‐hydroxy group and to transfer also ethyl, allyl, and benzyl groups. ^[14]

In cascade reactions with PC‐MjMAT forming a number of AdoMet analogues, RnCOMT‐var transferred ONB and propargyl residues and gave good conversions in many cases (Figure 3B, Figure S6). Conversion was almost quantitative for z‐SeHcy in combination with h1–n1, which comprises all adenine‐derivatives except for the bulky 2‐phenylethynyl substituted o1. More than 50 % conversion was obtained with i1–n1 for both y‐Hcy and z‐SeHcy (Figure S6). With y‐Hcy as substrate, conversion was generally lower, but still almost quantitative in 4 cases, i. e., ATP (i1) itself and the analogues j1, k1, and n1 (Figure 3B, Figure S6). These data indicate that RnCOMT‐var is not only promiscuous regarding the substituent at the selenium/sulfonium centre but also the adenine ring, in particular regarding substituents at the N⁶ position. Interestingly, in several cases, RnCOMT‐var transferred the propargyl group more efficiently than the ONB group (Figure 3b, Figures S6). This was particularly pronounced for h1, but also observed for l1–n1. Considering that the DM‐AdoMets derived from z‐SeHcy are made with lower efficiency by PC‐MjMAT than their counterparts with the ONB‐substituent (Figure S4), this data suggests that they are strongly preferred by RnCOMT‐var. This trend is clearly different compared to the substrate preference observed with NovO. A second marked difference in substrate preference compared to NovO is the good acceptance of DM‐AdoMets with k1–n1, out of which m1–n1 is particularly noteworthy as they were not converted by NovO.

GlaTgs2, the Giardia lamblia trimethylguanosine synthase 2 catalyses the N² methylation of m⁷GpppA, a minimal RNA 5′ cap (Figure 3C). We used variant V34 A reported to accept AdoMet analogues with larger substituents at the sulfonium centre. ^[15] In cascade reactions with PC‐MjMAT forming a number of AdoMet‐analogues, the transfer of the ONB‐group was low when ATP (i1) was used and negligible in all other cases (Figure 3C, Figure S6). Using z‐SeHcy, the transfer of the propargyl group was very good (almost 90 %) for ATP (i1), h1, and k1, and good (>60 %) in the case of n1 (Figure 3C, Figure S6). The observation that the propargyl residue was transferred more efficiently than the ONB residue confirms the limited substrate promiscuity of GlaTgs2‐var regarding the substituent at the sulfonium/selenium centre.

The good conversions (>60 %) with z‐SeHcy in combination with ATP analogues h1, k1, n1 suggest steric constraints in the active site of GlaTgs2 also regarding substituents at the N⁶‐position of adenine. The methylene group in k1 and n1 seems to help accommodation of the substituent, most likely by its flexibility, whereas direct attachment of a ring system (as in l1 or m1) is disfavoured.

Based on this direct comparison of the activities of NovO, RnCOMT, and GlaTgs2 for 10 co‐substrates in a time‐dependent manner, we chose l1, m1, and n1‐triphosphates and both methionine analogues as promising combinations to generate DM‐AdoMets more suitable for differential targeting of the indicated MTases.

NovO and RnCOMT‐var are capable of quantitative conversion of their respective substrates THPC and DHBAL with co‐substrate i1y (AdoONB) (Figure 3A, B). However, with n1y (N⁶‐benzyl‐AdoONB), NovO‐dependent modification is reduced to ∼15 % of y‐THPC (Figure 3A), whereas RnCOMT‐var can still quantitatively label its substrate DHBAL (Figure 3B). This means that the DM‐AdoMet n1y favours RnCOMT‐var over NovO, improving selectivity in MTase targeting.

The MTase GlaTgs2‐var is capable of quantitatively converting its substrate m⁷GpppA in the presence of i1z (SeAdoYn) (Figure 3C), as previously reported. ^[15] However, only 13 % and 7 % of m⁷GpppA are converted in the presence of l1z (N⁶‐cyclopentyl‐SeAdoYn) or m1z (N⁶‐phenyl‐SeAdoYn), respectively (Figure 3C). On the other hand, the activity of RnCOMT‐var remains high for either of these two DM‐AdoMets, resulting in at least 95 % conversion of DHBAL (Figure 3B). This means that the DM‐AdoMets l1z and m1z favour RnCOMT‐var over GlaTgs2‐var, improving selectivity in MTase targeting.

These results show that we identified DM‐AdoMets that are converted by different MTases with very different efficiencies and that we are able to generate them in situ from their amino acid and NTP precursors.

Bulky substituents at N⁶ of adenine improve selectivity in dual MTase cascades

Previously, Substrate Multiplexed Screening (SUMS) has been demonstrated to assess enzyme promiscuity, wherein multiple ligands competed for an enzyme. ^[19] With a similar approach, we wanted to assess whether the preferences of MTases for DM‐AdoMets would allow selective modification of a target molecule in a mixture of substrates and MTases. In our case, two enzymes were competed for a common substrate. To this end, we set up a one‐pot competition assay, in which two MTases have to compete for the same co‐substrate formed in situ by PC‐MjMAT (Figure 4). It should be mentioned that the outcome of such a competition assay will heavily depend on the co‐substrate concentrations and the kinetic parameters of the enzymes involved. We therefore had the following considerations: To ensure competition of the two MTases, the concentration of AdoMet‐analogues was limited. Thus, equimolar amounts of MAT substrate (NTP) and MTase substrates (THPC, DHBAL, or m⁷GpppA) were used. We anticipated that in this setup, a highly active MTase would consume the majority of finite DM‐AdoMet resulting in high conversion, whereas the MTase with lower activity on that particular DM‐AdoMet should lead to lower conversion.

Figure 4.

MTase competition assay. Two different MTases are combined with PC‐MjMAT in a one‐pot reaction. In these reactions, equimolar MAT substrate (NTP) and MTase substrate (THPC, DHBAL, or m⁷GpppA) are provided. (A) PC‐MjMAT/NovO/RnCOMT‐var cascade with y‐Hcy (ONB‐Hcy) and i1‐ or n1‐triphosphates. (B) PC‐MjMAT/GlaTgs2‐var/RnCOMT‐var cascade with z‐SeHcy (propargyl‐l‐SeHcy) and i1‐, l1‐, or m1‐triphosphates. Data and error bars show mean values and SEM of n≥3 independent experiments.

We chose MTase and co‐substrate combinations based on the results for the individual cascade reactions (Figure 3). Specifically, both MTases should give quantitative conversion with the ATP‐derived AdoMet analogue, yet their activity should be dramatically different with the DM‐AdoMet analogue. Based on these considerations, we tested two different combinations of dual cascade reactions. NovO/RnCOMT‐var dual cascade was tested for ortho‐nitrobenzyl (y) transfer and GlaTgs2‐var/RnCOMT‐var dual cascade was tested for propargyl (z) transfer.

The combination of NovO and RnCOMT‐var was tested with in situ formed i1y and n1y (Figure 4A). In the presence of i1y, NovO and RnCOMT gave similar conversions of their respective substrate. With n1y, RnCOMT‐var achieved ∼3 fold conversion compared to NovO. This is in line with the observations from the individual cascade reactions (Figure 3A, B) showing that N⁶‐benzylation of the adenosine drastically reduces the activity of NovO, but not RnCOMT.

The combination of GlaTgs2‐var and RnCOMT‐var was tested with in situ produced i1z, l1z, and m1z (Figure 4B). For the propargyl transfer, RnCOMT‐var gives higher conversion than GlaTgs2‐var in the competition assay in all cases. This result is consistent with our single‐MTase cascade assay and the previously reported kinetic parameters of RnCOMT (Y200L) ^[14] and GlaTgs2‐var. ^[15]

These two distinct case studies on dual MTase cascades show that bulky N⁶‐adenine modifications could either improve selectivity (in case of NovO/RnCOMT‐var) or indiscriminately hinder the activity of both MTases (in case of GlaTgs2‐var/RnCOMT‐var). Taken together, the ligand modifications should be designed on a case‐by‐case basis to achieve optimal activity of a target MTase of interest.

Enhanced hydrophobic interactions with bulky base attachments suggests product inhibition

Our findings indicate that bulky N⁶ modifications of adenine in AdoMet analogues provide a promising way to make the co‐substrate selective for certain MTases.

However, if the cascade is to be utilized in a more complex system of cellular origin, one may need to increase the selectivity by using an AdoMet analogue that has even higher steric demands. Ideally, the non‐natural AdoMet analogue should not be processed by undesirable MTases, thus conferring selectivity. However, its precursors should be processed by a MAT of interest, enabling production of the AdoMet analogue in situ. In that sense, o1y (2‐(phenylethynyl)‐AdoONB) is a promising ligand. While all of the MTases tested in this study are practically inactive on o1y, its precursors could be converted at reasonable levels by PC‐MjMAT. However, this conversion is mediocre (45 %), therefore, we found it necessary to have a structural insight for protein‐ligand interactions of PC‐MjMAT in order to support future research.

To this end, we aimed to analyse the binding energetics of PC‐MjMAT with different base‐modified AdoONB analogues. We have previously reported the crystal structure of PC‐MjMAT/AdoONB complex. ^[2d] We now followed a computational approach and performed structural refinement on AdoONB (i1y) and its analogues l1y, m1y, n1y, and o1y in complex with PC‐MjMAT. Here, base‐modified AdoONB analogues were built in PyMOL ^[20] and then subjected to energy minimization using Avogadro v1.2. ^[21] The energy‐minimized ligand and the protein crystal structure together were subjected to structural refinement in explicit water solvent, using HADDOCK2.4 refinement interface. ^[22] The interaction energy scores were calculated throughout the water‐refinement step. These scores were then used for comparative analysis of the binding characteristics in different complexes (see Experimental section and Supporting Information).

Expectedly, the van der Waals energy contributions of the AdoONB analogues to the overall interaction energetics are significantly improved (lower energies) owing to their bulkier base moieties, especially in the case of o1y (Figure 5A). We structurally compared the best (lowest) energy‐state of PC‐MjMAT/AdoONB complex and the best energy‐state of analogue's complexes (Figure 5B). This pairwise comparison shows that the interacting residue side chains efficiently adapt to different ligands. For N⁶ modifications at the adenine moiety, D146 stands out as a potential gatekeeper, presenting a steric block. In case of the C2 modification of adenine, T263 acts as a potential gatekeeper residue. In both cases, when there is a bulky ATP analogue extending towards these residues, the side chains cooperate by changing conformation, accommodating the ligand. These side chain conformational changes are depicted as a transition from the clashing (grey spheres) to the adapted (coloured sticks) conformer (Figure 5B). Among the analogues, l1y is the smallest ligand and thus the original binding cleft does not present a steric clash. However, in the case of m1y, the phenyl ring clashes with the β‐carbon of D146 (grey). The n1y benzyl ring is in a higher degree of steric clash, again with the β‐carbon of D146 (grey). The o1y phenylethynyl ring clashes with the T263 side chain hydroxyl. Upon the structural refinement of these complexes, all of these clashes were resolved, leading to the coloured conformations as presented in Figure 5B.

Figure 5.

Analysis of the water‐refined PC‐MjMAT/AdoONB (analogues) complexes. PC‐MjMAT/ligand crude complexes were subjected to water refinement by using HADDOCK 2.4 web server. HADDOCK generated 20 new interaction states from the starting conformation. (A) The van der Waals (kcal/mol) energy contribution of the ligands are given in the boxplot, with each box representing the different energetic contributions of the 20 different ligand states. (B) The water‐refined structures with the best (lowest) score are shown. All AdoONB analogue‐containing complexes (pink, red, yellow, and blue for l1y, m1y, n1y, and o1y, respectively) are separately superposed to the AdoONB‐containing reference complex (grey) in the respective panels. The potential gatekeeper residue side chains and the base moiety of the ligands are represented in stick form. The remaining groups in the ligands are represented in line form. The gatekeeper residue side chains in the reference structure and the unique bulky ligand groups in the superposed structures are represented in spheres in order to highlight steric clashes. The black circles indicate the extreme steric clashes. (C) The van der Waals (kcal/mol) energies of the potential gatekeeper residues are given in the boxplot, with each box representing all 20 states of the van der Waals energy contribution of that particular residue in the presence of respective ligands. The detailed interfacial van der Waals and electrostatics energy analyses are given in the Figures S7 and S8, respectively.[ ²⁰ , ²¹ , ²³ ]

Importantly, when the interfacial energy contributions of D146 and T263 are considered, it is seen that these conformational adaptations of the side chains do not bring an energy cost that would hinder the ligand binding, as indicated by the negative van der Waals energies (Figure 5C). In fact, there is approximately a 50 % improvement in the van der Waals energy contribution of T263 (from ∼−4 kcal/mol to ∼−6 kcal/mol) in the presence of o1y. The stronger van der Waals interactions and lower enzymatic activity for the AdoMet analogue o1y suggest that o1y poses a product inhibition. In the present study, due to commercial unavailability, the C2‐adenine modifications were not as thoroughly explored as N⁶‐adenine modifications. In future studies, by using different C2‐adenine modifications that have unique steric properties (especially smaller groups) and higher flexibility, it may be possible to tackle PC‐MjMAT product inhibition by weakening the protein‐product van der Waals interactions.

Conclusion

Our results show that PC‐MjMAT is highly active on ATP analogues with bulky N⁶ adenine modifications, quantitatively converting o‐nitrobenzyl‐dl‐homocysteine in the presence of such ATP analogues. PC‐MjMAT is also capable of achieving acceptable conversion of propargyl‐l‐selenohomocysteine in most cases. With the advantages of its high expression yields, ease of purification, and promiscuity, PC‐MjMAT is a favourable AdoMet analogue generator for downstream applications. Moreover, despite its high optimum temperature, the enzyme is remarkably active at 37 °C, enabling us to combine PC‐MjMAT with mesophilic MTases in cascades. These features together make PC‐MjMAT a strong candidate for MTase‐based in vivo metabolic labelling applications.

Following this, PC‐MjMAT is applied in MTase cascades. Compared to using purified DM‐AdoMets, cascades present additional complexity. MAT efficiency varies on different analogues, which directly affects the co‐substrate supply rate of the MTase reaction. Therefore, it is challenging to make a fair comparison of MTase activity on different DM‐AdoMets (in situ). On the other hand, the cascade approach is satisfactory to quickly assess whether an MTase is active on a particular DM‐AdoMet. This provides preliminary data that reveal the viable DM‐AdoMet options for in‐depth kinetic experiments, saving time and resources. Of particular note, RnCOMT‐var is observed to be active on a wider range of DM‐AdoMets compared to NovO and GlaTgs2‐var.

While the cascade approach has the aforementioned disadvantage when comparing the MTase activity on different DM‐AdoMets, it is suitable for comparison of different MTases on a single DM‐AdoMet. Based on this, we demonstrated 3‐in‐1 pot cascades by using PC‐MjMAT together with two different MTases. These MTase competition assays provided a measure of how DM‐AdoMets suit differential targeting of MTases. Of particular note, DM‐AdoMets with benzylic nucleoside attachments were suitable for targeting RnCOMT‐var.

In this study, we explored a plethora of modifications of AdoMet analogues and tested their effect on MTase‐catalysed conversions. Previously, many studies in the literature have focused on the nucleobase, ^[9] the nucleoside, ^[10] or the terminal carboxylate/amine moiety ^[11] when testing modifications of AdoMet. The transferable group was limited to methyl or ethyl in most reports. In our work, nucleoside modifications of AdoMet provided minor improvements in selectivity when sterically more demanding transferable groups were used. We conclude that ‐ if a molecular labelling application is in mind – the preliminary assays should already be performed using the double‐modified AdoMet analogues that carry the transferable groups of interest. Protein homology and ligand similarity across different systems in different studies may aid the design of novel molecular labelling systems. However, our model systems clearly demonstrate that this cannot be a substitute for a preliminary study that tests the actual possibilities.

In the present study, we unveiled a set of sulfonium/selenium centre and base modification combinations that offer increased selectivity in MTase targeting in vitro. However, there are still major challenges to overcome on the way to orthogonal metabolic labelling in vivo. PC‐MjMAT is capable of using the natural substrates ATP and l‐methionine. In a cellular environment, these natural substrates would be abundant, interfering with the in vivo generation of DM‐AdoMet, as well as leading to background labelling. To overcome this issue, in future studies, ligand engineering efforts should be combined with protein engineering to achieve negative selectivity against the natural substrates. To this end, an example approach would be mutating Y293 in PC‐MjMAT binding cleft to abolish π‐stacking of the ATP adenine moiety. The lost stacking interaction can be rescued for 2‐(phenylethynyl)‐ATP by substituting the residues surrounding the phenylethynyl moiety for aromatic residues. In the cascade, the MTase of interest will also likely require a similar protein engineering to be active on 2‐(phenylethynyl)‐AdoONB (o1y). Our results will support future research on developing MAT‐MTase toolboxes that suit orthogonal biomolecular labelling applications.

Experimental Section

Detailed materials and methods section and supplementary data are provided in the Supporting Information file.

Materials: All chemicals are of analytical grade. HPLC grade acetonitrile was purchased from Honeywell. 2′‐azido‐2′‐dATP was purchased from TriLink Biotechnologies. 2‐(Phenyltethynyl)‐ATP was purchased from Tocris Bioscience. ITP (inosine‐5’‐tiphosphate) was purchased from Merck (Sigma‐Aldrich). The standard dNTPs (dATP, dCTP, dGTP, and dTTP) and NTPs (ATP, CTP, GTP, and TTP) were purchased from Solis BioDyne and Jena Bioscience. All other NTPs were purchased from Jena Bioscience.

Software: The LUBLICATE OpenChrom ^[24] and the Agilent ChemStation softwares were used for HPLC data analysis. The Bruker Compass DataAnalysis software (v4.4) was used for analysis of the LC/MS data. The HPLC chromatogram line plots were plotted by using the python (v3.10) ^[25] packages NumPy, ^[26] pandas,[ ²⁷ , ²⁸ ] and Matplotlib ^[29] in Jupyter Notebook (v6.3.0) ^[30] in Anaconda (v3) ^[31] environment. All data presented in the manuscript were visualized in R (v3.6.1), ^[32] invoked by Rstudio (v1.2.5001), ^[33] using the ggplot2 ^[34] R package. The heatmaps were plotted by using the pheatmap ^[35] and RcolorBrewer ^[36] R packages. Pdb‐tools python scripts ^[23] were used for clean‐up of the coordinate files prior to water refinement in HADDOCK2.4. The open source PyMOL ^[20] molecular visualization program was used for visual analysis of the molecules and for preparation of the crude protein‐ligand complexes. Ligand energy minimization was performed in Avogadro (v1.2). ^[21] The 2D chemical structures were drawn in ChemSketch software. The monoisotopic mass calculations were performed in ChemSketch software. ^[37] ApE (A Plasmid Editor) ^[38] and CLC Main Workbench 7.9.1 software were used for biological sequence data handling. LabIcons (https://www.labicons.net/index.html) were used for illustration.

Synthesis of the methionine analogues: The ortho‐nitrobenzyl‐dl‐homocysteine ^[2a] (ONB‐Hcy ; y‐Hcy) and propargyl‐l‐selenohomocysteine ^[12] (propargyl‐l‐SeHcy ; z‐SeHcy) were synthesized as previously reported.

Plasmids and site‐directed mutagenesis: The pET28a‐NovO and pET28a‐RnCOMT expression vectors were kindly provided by Prof. Dr. Jennifer Andexer (Freiburg, Germany). Site‐directed mutagenesis was performed on RnCOMT cDNA by using Phusion high‐fidelity DNA polymerase (Thermo Fisher Scientific). pET28a‐PC‐MjMAT[ ^2a , ^2d ] and pProEx‐MTAN ^[39] expression vectors were previously reported. pET28a‐GlaTgs‐var was prepared by subcloning of the cDNA from pRSET backbone used previously. ^[40]

Recombinant protein expression and purification: Recombinant expression and purification conditions for PC‐MjMAT, ^[2a] RnCOMT‐var, ^[14] GlaTgs2‐var, ^[40] and MTAN were previously described. NovO expression and purification conditions were adapted from Pacholec et al. ^[13] Briefly, the same conditions were used as in the case of PC‐MjMAT, without the heat treatment of the crude extract. The GlaTgs2 V34 A and RnCOMT M40 A substitutions enable the enzymes to transfer bulkier groups. In addition, the RnCOMT Y200 L substitution confers regioselectivity, increasing the selectivity for the meta‐hydroxyl group of DHBAL. ^[14]

Activity assays: The PC‐MjMAT activity assays were performed at 65 °C by incubating 50 μM PC‐MjMAT, 2.5 mM NTP, and 5 mM ONB‐Hcy or 5 mM propargyl‐l‐SeHcy in pH 8 reaction buffer containing 5 mM KCl, 10 mM MgCl₂, and 50 mM HEPES for up to 6 hours.

Activity assays for NovO, RnCOMT‐var (M40 A/Y200 L), and GlaTgs2‐var (V34 A) were performed as MAT/MTase cascade reactions. The reaction mixture content for the first step (MAT reaction) of the cascade is identical for the three MTases. All MTase reactions contain 10 μM MTAN to prevent product inhibition via enzymatic depurination of the MTase product AdoHcy. The NovO cascade mixture additionally contains 50 μM NovO and 300 μM THPC. The RnCOMT‐var cascade mixture additionally contains 50 μM RnCOMT‐var and 300 μM DHBAL. The GlaTgs2‐var cascade mixture additionally contains 200 μM GlaTgs2‐var and 400 μM m⁷GpppA. The activity assays were performed in pH 7.4 reaction buffer containing 5 mM KCl, 10 mM MgCl₂, and 50 mM HEPES. Reactions were incubated up to 6 hours at 37 °C for NovO and RnCOMT‐var, and up to 8 hours at 30 °C for GlaTgs2‐var. All reactions were terminated by adding HClO₄ to a final concentration of ∼100 mM. The quenched reaction mixtures were centrifuged at 4 °C at 21,130×g for 10 minutes. The supernatants were analysed by HPLC. In the case of PC‐MjMAT/GlaTgs2‐var cascade reaction with dATP (i2), reaction was treated with alkaline phosphatase to prevent co‐elution of dATP and m⁷GpppA. To do so, MTase reaction was quenched by incubation at 95 ^oC for 5 minutes. Then, 1 μL (1 unit) of alkaline phosphatase (Thermo Fisher Scientific, FastAP REF #EF0651) was added for 10 μL of reaction. Dephosphorylation was performed at 23 ^oC for 0.5 hours. Dephosphorylation reaction was quenched by acidification as described for other reactions.

Measurements were performed in HPLC (Agilent Technologies 1260 Infinity) using a 5 μm C18 pyramid column (MN Nucleodur REF #760201.40) as the stationary phase. 100 mM pH 6.5 phosphate buffer (buffer A) and 100 mM pH 6.5 phosphate buffer / acetonitrile 1 : 1 (v/v) buffer (buffer B) were used as the mobile phases. The detailed HPLC methods and substrate conversion calculations are described in the Supporting Information file.

LC/MS analysis: Mass spectrometry analyses were performed according to the previous report. ^[12]

Structural refinement and energy analysis of the PC‐MjMAT/ligand complexes: The crystal structure of PC‐MjMAT/AdoONB complex (PDB entry 7P84) was previously reported. ^[2d] Initially, the crystal structure coordinate file was cleaned up for downstream steps by using the “pdb_selaltloc.py”, “pdb_occ.py”, “pdb_reres.py”, and “pdb_reatom.py” python scripts of the pdb‐tools package, in the given order. The coordinates of AdoONB were isolated from the “clean” complex and exported as “.pdb” file. Using AdoONB as the scaffold, base‐modified AdoONB analogues (N⁶‐cyclopentyl‐AdoONB, N⁶‐phenyl‐AdoONB, N⁶‐benzyl‐AdoONB, and 2‐(phenylethynyl)‐AdoONB) were built in PyMOL and exported as “.pdb” files. Then the ligands were subjected to energy minimization in Avogadro (v1.2) software, ^[21] using Ghemical force field and Conjugate Gradient Minimization algorithm.[ ⁴¹ , ⁴² ] The energy minimization was performed for approximately 200 seconds for each ligand. It should be noted that when the other force fields (GAFF, MMFF94, MMFF94 s, and UFF) were tested, the ribose ring became planar (losing its unique 3D conformation), or C−C bonds lost their tetrahedral geometry, or the N7 bond angle became abnormally large, essentially collapsing the 5‐membered ring of base moiety and ultimately causing problems in the downstream computational experiments.

In Avogadro, the energy‐minimized ligands were exported as “.mol2” files. These optimized ligands were then fit to the original crystal structure coordinates of AdoONB. The complete set of AdoONB scaffold heavy atoms were used for the fitting. On the other hand, all of the coordinates in the crystal structure other than the ligand (protein chains, cofactors, and triphosphates) were isolated as the “apo” form. Then, this apo form coordinates and the optimized ligand coordinates were loaded together in PyMOL, and isolated as a single molecule for each AdoONB analogue, yielding the crude complexes. The crude complexes were subjected to structural refinement in HADDOCK 2.4 web server refinement interface[ ^22a , ^22b ] (https://wenmr.science.uu.nl/haddock2.4/refinement/1). Briefly, a very short (femtosecond‐time‐scale) molecular dynamics simulation in explicit water solvent were performed on the complexes. Throughout this water refinement, 20 new models were generated for each complex. For each model, HADDOCK also outputs the interface energy contributions of the interacting residues/ligands. These interfacial residue energies are provided in “ene‐residue.disp”, located under the “/structures/it1/water/analysis/” subdirectory of the result folder. The interfacial residue energies were extracted from this file and filtered for the residues that are within 7 Å shell of AdoONB analogues. The codes and computational methods used in the present study were adapted from our previous study. ^[43]

Conflict of interest

The authors declare no conflict of interest.

Supporting information

As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials are peer reviewed and may be re‐organized for online delivery, but are not copy‐edited or typeset. Technical support issues arising from supporting information (other than missing files) should be addressed to the authors.

Supporting Information

Erguven M., Cornelissen N. V., Peters A., Karaca E., Rentmeister A., ChemBioChem 2022, 23, e202200511.

Data Availability Statement

The data that support the findings of this study are available in the supplementary material of this article.