Characterisation of RNA guanine-7 methyltransferase (RNMT) using a small molecule approach

RNA polymerase II transcripts are modified at the 5ʹ end by the co-transcriptional addition of the RNA cap [1,2]. In mammals, this structure is 7-methylguanosine linked by a 5ʹ-to-5ʹ triphosphate to the first-transcribed nucleotide. The first- and second-transcribed nucleotides can be further methylated by other methyltransferases [3,4]. The RNA cap is formed during the early stages of transcription as the nascent RNA emerges from the polymerase complex [1,4]. RNA guanylyltransferase and 5ʹ phosphatase binds directly to the RNA pol II large subunit and catalyses the addition of the guanosine cap to the nascent transcript, denoted as G(5ʹ)ppp(5ʹ)X, where X is the first-transcribed nucleotide. RNA guanine-7 methyltransferase (RNMT) catalyses the transfer of a methyl group from S-adenosylmethionine (SAM) to the cap intermediate to form the cap m7G(5′)ppp(X) and the by-product S-adenosylhomocysteine (SAH) (Figure 1A).

RNMT was identified based on homology to the yeast and viral homologues and in vitro activity [5,6]. This basic cap structure (cap 0, m7G(5ʹ)ppp(X)) protects RNA from degradation by nucleases during synthesis and recruits factors involved in RNA processing and translation initiation [1,2]. RNMT has an activating co-factor RNMT-activating miniprotein (RAM) clamped at the interface of the catalytic domain and the modular lobe 416–456 [7]. This stabilisation promotes SAM recruitment and induces a significant increase in the methyltransferase activity [8,9] through the formation of a series of interactions affecting the α-helix hinge, α-helix A and multiple active site residues [10]. RAM has also an RNA-binding domain that recruits RNA to RNMT [9,11]. RNMT-RAM is critical for gene expression [12,13]. In addition to its role of catalysing RNA cap methylation during transcription and later, RNMT-RAM has also non-catalytic roles in transcription and potentially other processes [11,14]. Deletion of RNMT results in reduced proliferation and increased apoptosis, and deregulated expression of RNMT promotes cell transformation [15-18].

Several cellular signalling pathways regulate RNMT by modulating the expression of RNMT itself or activating subunit RAM, by altering its activity through post-translational modifications or by controlling metabolism of SAH, the inhibitory by-product of methylation reactions [1,19-21]. The regulation of RNMT accompanies or drives major cellular transitions and directs enhanced expression of specific sets of genes. For example, RNMT-RAM is up-regulated during T-cell activation when rapid cell growth and proliferation requires increased mRNA synthesis for adaptive immune responses [12]. Of particular importance is the enhanced production of TOP-RNAs that encode ribosomal proteins and other proteins involved in ribosome production. Conversely, during embryonic stem cell differentiation, RNMT-RAM is down-regulated, which is required for the repression of pluripotency-associated genes and morphological features of differentiation [19].

Consequently, RNMT is a target of interest in cancers and immune disorders. These diseases exhibit deregulated transcription, which requires a cap, and RNMT inhibition has been demonstrated to reduce proliferation and increase apoptosis in many cell lineages [8,16,21]. Certain oncogenic mutations increase dependency on the cap. Breast cancer cell lines with oncogenic PIK3CA mutations have enhanced sensitivity to RNMT inhibition [15]. Myc-dependent transcription and proliferation is dependent on RNA cap formation [20,22,23]. Recent analyses have linked RNMT expression to cancer outcomes [24,25].

Here, we use biophysical, biochemical and structural biology approaches to build understanding of the mode of enzymatic action of RNMT. Further, by carrying out a diversity screening campaign, we describe two novel, small molecule inhibitors of RNMT that may be developable into useful tools for interrogating the role of RNMT in different cell contexts.

In this manuscript, cap is GpppG (Figure 1B), except in crystallographic studies where a cap surrogate guanosine 5′-[β,γ-imido]triphosphate (GMP-PnP) (Figure 1C) was used. Sinefungin (Figure 1D) is used as a SAM mimetic [26]. This molecule has been shown to bind in the active site of a wide range of methyltransferases with substrates as diverse as DNA [27], RNA [28], glycopeptides [29] and catecholamine synthesis enzymes [30]. Methylated cap and SAH are the final products of the enzymatic process. Two RNMT biochemical assays were developed with different readouts (Figure 2).

The first utilises the MTase Glo reagent (Promega) to convert the product of the methyltransferase assay into a bioluminescent signal [31] through coupling enzymes. We found that this assay had some susceptibility to false positives, presumably through inhibition of the coupling enzymes rather than RNMT. In view of this, we developed a mass spectrometry readout based on the accumulation of the product, SAH, using the RapidFire system. Both assays share the same basic biochemical setup with enzyme and substrate concentrations, and time-courses being similar. However, the mass spectrometry assay is stopped with formic acid and the plates sealed. The RapidFire processing system allows the sample to be aspirated from the well and loaded on to a solid-phase extraction column where the buffer salts and proteins are removed by washing with a TFA solution. The SAH was eluted using an acetonitrile:water:TFA solution and directed to a triple quadrupole mass spectrometer (Agilent 6740). Full details of the SAH RapidFire assay protocol have been published using a different methyltransferase [32].

To elucidate the cap-binding mode, full-length HsRNMT in complex with activating subunit RAM (Avi-HsRNMT-RAM) was investigated by surface plasmon resonance (SPR). We first measured the kinetic and equilibrium binding constants (K_D) for SAM (Figure 3A), SAH (Figure 3B), and Sinefungin (Figure 3C). The kinetics of binding for these compounds appear very similar with a rapid onset of binding reaching a steady state within a few seconds. The loss of binding on washing with buffer was very similar.

No interaction of the cap with Avi-HsRNMT-RAM was detected in the absence of co-substrate or in the presence of SAH. The interaction of the cap with Avi-HsRNMT-RAM was only detected after addition of a saturating concentration (1 µM) of Sinefungin to the assay buffer (Figure 3D). Kinetic and steady-state calculations of affinity constants were consistent. These biophysical results are summarised in Table 1.

Altogether, these results suggest that HsRNMT-RAM has an ordered binding mechanism with SAM binding first, followed by the cap. Furthermore, biophysical characterisations also suggest that the interaction of the cap with HsRNMT-RAM was driven by the 5-amino group on Sinefungin (or the SAM methyl group in a physiological condition) as SAH does not carry a comparable group at position 5 and no cap binding was detected. We hypothesise that in the absence of a 5-amino/methyl group, the cap guanine moiety is rotated by 180° in a clockwise direction, a consequence of the intra-molecular attractive force between the exocyclic -NH₂ of the guanine moiety and the phosphates. This conformation would be detrimental to the cap binding since most of the interactions with HsRNMT-RAM would be lost.

Michaelis–Menten parameters were determined for HsRNMT-RAM using the RapidFire Mass Spec assay for cap (K_M^cap = 3.2 µM [2.0–5.3 µM]), SAM (K_M^SAM = 1.6 µM [1.3–2.1 µM]), and a mean Vmax 0.04 µM (0.04–0.05) SAH/min for both substrates (Figure 4). These data provide assay conditions suitable for characterisation of inhibitors.

Following the establishment of appropriate ‘balanced’ assay [33], a screening campaign was carried out on 48,806 diverse drug-like compounds [34] from our internal compound library at a single concentration (10 µM) using the MTAse Glo assay to measure enzymatic activity. Analysis of the single-point screen revealed a normal distribution ( Supplementary Figure S1) with a mean inhibition of 4.6% and standard deviation of 15.4%. Using a cut-off of 50.8% (mean + 3 × standard deviation; 99.73% confidence assuming a normal hit distribution), the HsRNMT-RAM assay resulted in a hit rate of 1.6%. The screen was robust with a mean Z′ [35] of 0.8 (±0.1) and mean signal:background of 7.7 (±3.6). Two compounds with similar structures (Figure 5) were profiled in the orthogonal RapidFire assay.

Mode of inhibition experiments were performed for Sinefungin, DDD1060606 and DDD1870799 using the RapidFire assay. While Sinefungin showed competitive inhibition with SAM (Figure 6C) and non-competitive inhibition with cap (Figure 6B) as was anticipated, both small molecule inhibitors demonstrated uncompetitive inhibition profiles against both cap and SAM (Figure 6E, F, H, I). The data from these experiments are summarised in Table 2.

We used SPR to investigate the co-factor requirements to permit the binding of the inhibitors to HsRNMT-RAM. No binding was observed in the absence of co-ligand or in the presence of SAM, whereas high-affinity binding (K_D = 1.6 µM and 2.5 µM (Figure 7) was measured for DDD1060606 and DDD1870799, respectively, when the buffer was supplemented with SAH. This suggests that the ligands bind preferentially after the reaction has completed, and SAM has been metabolised into SAH. The data from these experiments are summarised in Table 3.

Previous work [7] demonstrated that the lobe 416–455 becomes highly agitated in the absence of RAM. While we were able to crystallise RNMT catalytic domain in the presence of RAM, the crystallisation process was insufficiently robust for drug discovery efforts. To improve the robustness of the crystallisation process, the lobe 416–455 of HsRNMT catalytic domain was replaced by its homologous sequence region from the Encephalitozoon cuniculi (GLGC) guanine-7 cap methyltransferase (Q8SR66) identified from a pairwise sequence alignment. The sequences used are in Supplementary Figure S2. The two chains of the asymmetric unit for all structures are almost identical and can be well superposed (RMSD < 0.5 Å) to the structure of reference (5e8j) solved in complex with RAM [7] ( Supplementary Figure S3). This validates our hypothesis that the effect of this lobe substitution would be negligible on the RNMT-RAM structure.

The understanding of the cap binding mechanism is currently limited by the absence of a published HsRNMT-cap structure and the lack of knowledge of the key residues involved in cap binding. This deficit hinders the development of potent and/or selective drugs targeting the cap pocket. Since attempts to crystallise RNMT with the G(5ʹ)ppp(5ʹ)G (cap) had been unsuccessful, we co-crystallised RNMT with the cap surrogate, GMP-PnP. GMP-PnP was confirmed as a cap competitive inhibitor with a Ki of 2.3 µM (0.9–5.5 µM) (Supplementary Figure S5). Crystallisation of RNMT with GMP-PnP provides the first structural characterisation of a Class I methyltransferase in a complex with a cap analogue and Sinefungin. The Sinefungin molecule is well resolved in the electron density map (Supplementary Figure S6) for both chains of the asymmetric unit and shares a common network of interaction to that observed for SAH binding to HsRNMT (Supplementary Figure S4). GMP-PnP was successfully built in just one of the chains in the asymmetric unit and is located between the α-helix A and the β-strands 8/9. When co-crystallised with HsRNMT and Sinefungin (Figure 8A), the GMP-PnP guanine moiety is stabilised within the cap pocket by hydrogen bonds to residues H288, Y289, E370, Y467 and the Sinefungin 5-amino group, as well as through a Pi-stacking interaction with F285.

The GMP-PnP ribose makes two hydrogen bonds with N176, whereas the beta-gamma pyrophosphate group interacts with the polypeptide chain via two salt bridges to K208. Finally, the distance measured between the Sinefungin 5-amino group and the guanine N7 (2.4 Å) alongside the angle formed by C5/5-NH2/N7 (140°) confirms that GMP-PnP is observed in an ideal orientation to permit the methyl transfer, similar to the predicted model published [10].

HsRNMT–Sinefungin–GMP-PnP, HsRNMT–SAH–DDD1060606 and HsRNMT–SAH–DDD1870799 co-complexes were solved, respectively, at 2.5, 2.0 and 2.4 Å in space group P2₁ ( Supplementary Table S1).

In all the three structures, SAH or Sinefungin is observed in a hydrophobic pocket made by the α-helices 3 and 4 and the loops connecting the β-strand 1 to the α-helix B and the β-strands 3 and 4. SAH and Sinefungin are stabilised in the active site through an identical network of hydrogen bonds, involving residues K180, G205, D227, D261, S262 and Q284, as well as 2 salt bridges with K180 ( Supplementary Figure S3). These results support the hypothesis that Sinefungin binds in the SAM-binding site in the presence or absence of the cap.

The two inhibitors (DDD1060606 and DDD1870799) identified in the screening campaign were successfully co-crystallised with HsRNMT in the presence of SAH. DDD1060606 (Figure 8B) is observed at the cap pocket and displays a mainly hydrophobic binding mode. Using the HsRNMT–Sinefungin–GMP–PnP as a structure of reference, the pyridine is rotated 180° and sits on top of the SAH homocysteine moiety, making direct and water-mediated hydrogen bonds with Y289 and H288 side chains, as well as F285 main chain.

The piperidine is located nearby the GMP-PnP ribose and is mainly stabilised by a hydrogen bond with N176. Finally, the thiophene near to the GMP-PnP α-phosphoryl group is largely solvent exposed and does not interact with the peptide chain.

DDD1870799 (Figure 8C) is also observed in the cap pocket of RNMT. In the presented structure, the pyrazolo-pyridine interacts with Y289 via two aromatic H-bonds, while the fluoro-benzene is buried in a hydrophobic pocket made by the residues including the main chain R173, as well as the side chains N176 and Y467.

For both DDD1060606 and DDD1870799, we determined the enthalpic and entropic components of the free energy of binding to HsRNMT-RAM to guide both compound selection and the optimisation process. Note that all of the isothermal titration calorimetry (ITC) experiments were performed with protein pretreated with SAH or Sinefungin. We used cap as the reference ligand as it is widely accepted that ligands with more negative enthalpies of binding provide better starting points for lead optimisation [36,37]. The interaction of cap with HsRNMT-RAM is enthalpically driven with an ΔH of −33.5 kcal/mol (Figure 9A) and net free energy change, ΔG of −8.4 kcal/mol (Figure 9B). Furthermore, we determined the binding affinities for DDD1060606 and DDD1870799 to HsRNMT-RAM.

The resulting affinities (both approximately 3 μM) are comparable to the K_D measured by SPR in the previous experiments. A favourable entropic contribution Δ(TΔS) of −1.8 kcal/mol was observed for DDD1060606, and the total Gibbs free energies were similar for both ligands (ΔG = −7.5 kcal/mol). However, a significant gain in net enthalpic change ΔΔH of −1.9 kcal/mol was measured for DDD1870799 in comparison with DDD1060606, suggesting that the interaction of DDD1870799 with HsRNMT-RAM is more enthalpy driven and might result in more selective leads with fewer absorption, distribution, metabolism, excretion and toxicity issues than DDD1060606 [38,39] (Figure 9). The data from this analysis are summarised in Table 4.

We then calculated the molecular interaction fields (MIFs) by FLAP [40] to identify the attributes that lead to the biological activity of the cap analogue and the 2 new inhibitors.

Figure 10A and B shows the calculated FLAP maps for the cap analogue (GMP-PnP) binding to RNMT. The observed polar back pocket requires a hydrogen bond donor (guanine: position 2) to interact with E370 that is bordered by hydrophobic regions (pocket S1), while the position of the 3′O (ribose) coordinates well with the requirement for a hydrogen bond acceptor to generate the hydrogen bond with N176. The DDD1060606 piperidine (Figure 10C), meanwhile, matches the hydrophobic and the hydrogen bond acceptor requirements introduced by F285 and N176, respectively (pocket S2). It is worth noting that neither the pyridine nor the thiophene can form favourable interactions with the protein, while the former sits above the thioether group of SAH. In comparison, the MIFs calculated for DDD1870799 (Figure 8D) suggest that the fluoro-phenyl moiety is responsible for the ligand binding via the interaction with a small pocket (S3) of high hydrophobicity. This interaction, by displacement of disorganised water molecules [41], is likely to be responsible for the significant gain in net enthalpic change ΔΔH measured for DDD1870799. Finally, since the orientations observed for the inhibitor pyridine groups do not appear to provide any advantage in comparison with guanine from the cap, the design of ligands with an improved affinity profile will have to take into consideration substituents that may be capable of stabilising the pyridine in the S1 pocket.

The biophysical characterisation of cap binding to Avi-HsRNMT-RAM described earlier (cap will only bind in the presence of either SAM or Sinefungin but not SAH) suggests that the S-methyl or amine group plays a key role. As the binding poses for cap are very similar when either SAM or Sinefungin are bound, it is unlikely that hydrogen-bond formation between the amine group and the guanine base is the driver. Instead, it is likely that the common steric effects of the amine and S-methyl groups as seen in the crystal structures are important factors for stabilisation. It is only when the Sinefungin or SAM is in place that the interactions observed in the cap structure (polar with H288, Y289, E370, Y467 and Pi stacking with F285) become important. Thus, in the presence of SAH, any new inhibitor ligand should ideally carry features that consider the steric effect of the methyl on SAM to permit the guanine-mimicking group access to the S1 pocket.

Despite a growing interest in HsRNMT-RAM as a valuable target to either inhibit the proliferation and increase the apoptosis of cancer cells [14,16,20,21] or to prevent eukaryotic virus proliferation [42], the lack of understanding of the enzymatic mechanism limits the development of selective and efficient drugs. Here, we present a comprehensive overview of the enzymatic mechanism of HsRNMT monomer and in complex with its co-factor RAM. This characterisation is used as a basis to discuss HsRNMT-RAM druggability through the description of two inhibitors targeting the cap pocket.

In our manuscript, we describe a label-free RapidFire mass spectrometry assay for the enzymatic characterisation of HsRNMT-RAM. Although our data are robust, we observed significant differences compared with the recently published data [43]. Specifically, there were 10- and 40-fold differences in the K_m values for SAM and RNA, respectively. There are many potential sources for these differences. These include detail variation in the expression systems for the protein, the presence or absence of RAM and divergence in the presence or absence of a small RNA attached to the 3ʹ end of the cap substrate. Despite these differences, we report comparable IC₅₀ values for Sinefungin, which differ by less than two-fold.

Biophysical characterisation of substrate binding shows an ordered reaction with SAM required to bind first to enable subsequent cap binding. While an ordered reaction for RNMT has been made [10], the observed order of substrate addition is reversed to the prediction.

We report the structure of two small molecule inhibitors of RNMT as result of a diversity-driven hit discovery campaign. These molecules behave biochemically as uncompetitive inhibitors with respect to both cap and SAM despite the structural biology data leading to an expectation of competitive inhibition with cap. The biophysical data resolve this puzzle by demonstrating that the compounds require the presence of the enzymatic product, SAH, in RNMT to bind efficiently. This stabilisation of enzyme–product complex explains the competitive binding resulting in uncompetitive inhibition. The model we propose for the behaviour of RNMT in binding its substrates and the compounds is presented (Figure 11). This is similar to the mode of action of mycophenolic acid against the inosine 5′ monophosphate dehydrogenase [44,45].

Finally, we used molecular modelling approaches to map the physicochemical properties of the cap pocket of HsRNMT. This, in combination with the crystal co-complexes of the enzyme with the small molecule hits, allows us the prediction of chemical modifications that will generate better tool molecules and potentially new medicines.

Reagents were purchased from Sigma Aldrich (Gillingham, Dorset, U.K) unless otherwise stated.

For biochemical assays, full-length HsRNMT (1-476) was cloned in the pOPC bi-cistronic plasmid as a TEV-cleavable His-MBP fusion with HsRAM (1-90) and expressed in Escherichia coli BL21 (DE3) using LB medium. This protein is referred to as HsRNMT-RAM.

For SPR, full-length HsRNMT was cloned into an LIC adapted pNIC28a with an N-terminal TEV cleavable hexa-Histag and C-terminal Avitag and HsRAM (1-90) into a pACYC vector and is described as Avi-HsRNMT-RAM in this manuscript. Plasmids were co-transformed into E. coli BL21 (DE3) in the presence of a pCDFduet-BirA vector for in vivo biotinylation and grown in TB autoinduction media. For crystallography, the HsRNMT catalytic domain (165–476, Δ416–455 GLGC) was cloned into pET15bTEV and expressed as a TEV-cleavable His tag fusion in E. coli BL21 DE3 using ZY autoinduction medium.

Both HsRNMT full-length proteins (1-476) in complex with HsRAM (1-90) (HsRNMT-RAM and Avi-HsRNMT-RAM) were purified using the same procedure. Harvested cells were re-suspended in a lysis buffer comprising 20 mM HEPES, 175 mM NaCl, 25 mM KCl, 10% glycerol, 0.5 mM TCEP, pH 7.5 supplemented with DNAse and EDTA-free protease inhibitors (Sigma) before lysis at 30 psi with a cell disruptor (Constant Systems). HsRNMT-RAM was initially purified by Ni-NTA affinity chromatography (HisTrap FF 5 mL, Cytiva) in the same buffer and eluted with an imidazole gradient. Imidazole was then quickly removed from the fractions of interest using a HiPrep 26/10 desalting column in the same buffer. The desalted fractions were later incubated with the TEV protease (1:20) overnight at 4°C prior to a second purification by Ni affinity chromatography performed in the same conditions. The flowthrough, containing the target protein, was concentrated and purified further with gel filtration chromatography on a Superdex 75 26/60, pre-equilibrated with the same buffer. Monomeric fractions were pooled and concentrated to 15 mg/mL before snapshot freezing in liquid nitrogen. The purification of HsRNMT catalytic domain (165–476) Δ416–455 GLGC was purified in a similar way with the following buffer: 20 mM HEPES, 350 mM NaCl, 50 mM KCl, 10% glycerol, 0.5 mM TCEP, pH 7.5.

A selection of 48,806 diverse drug-like compounds from our internal compound library were tested for inhibition against HsRMNT-RAM at a single concentration, using MTase-Glo, a bioluminescent-based assay from Promega (Madison, WI, U.S.A.). Briefly, compounds were incubated with 5 nM HsRNMT-RAM, 2 µM SAM and 2 µM cap, using buffer conditions specified in the MTase-Glo assay kit, for 60 minutes. Following this incubation, the MTase-Glo Reagent and Detection solution (at ratios of 5:2 and 1:1 of the final assay volume, respectively) were added to each well and the signal allowed to develop for a further 30 minutes in the dark. The resulting luminescence was read on a PHERAstar FS (BMG). A hit was defined as percent effect was greater than mean + 3 standard deviations of the screen results (50.8%). Following this screen, compounds that had been identified as hits were confirmed by retesting at 10-point dose response in the RapidFire assay. Potency was calculated by fitting data to a four-parameter logistical fit (model 203) using Sigmaplot Version 14.0.3.192. Data are presented as the mean of three independent experiments with 95% confidence intervals (95% CI).

An endpoint 384-well plate assay for HsRMNT-RAM activity was developed using the Agilent RapidFire 365 high-throughput system with integrated SPE interfaced with the Agilent 6740 triple quadrupole mass spectrometer. The assay was performed by mixing 5 nM HsRNMT-RAM in buffer (20 mM Tris, pH 8.0, 50 mM NaCl) supplemented with 1 mM EDTA, 1 mM DTT, 0.1 mg/mL BSA, 0.005% Nonidet P40 (Roche) and 3 mM MgCl2 with 2 µM of substrates (G(5ʹ)ppp(5ʹ)G Sodium Salt (cap) [New England Biolabs, Ipswich, MA]) and S-(5ʹ-Adenosyl)-L-methionine chloride dihydrochloride (SAM [Cayman Chemical, Ann Arbor, MI]). The enzymatic reaction incubated for 60 minutes at room temperature followed by quenching with 80 μL 1% formic acid (VWR, Radnor, PA) containing 0.03 µg/µL S-adenosylhomocysteine-d4 (d4SAH [Cambridge Bioscience, Cambridge, U.K]). Plates were centrifuged at 500×g for 1 minute after every addition and sealed using the PlateLoc thermal microplate sealer before analysis on the RapidFire. Samples were loaded onto the RapidFire system, and analysis was performed as described for the identification of inhibitors of the SARS-CoV-2 guanine-N7-methyltransferase [32].

Experiments to determine the K_m^app values were carried out using a two-fold dilution of cap or SAM from 40 to 0.08 µM and a saturating concentration (40 µM) of SAM (K_m^app cap) or cap (K_m^app SAM). HsRNMT-RAM was used at 5 nM. Data were analysed to calculate the initial velocity (V0). K_m^app and Vmax values were determined by nonlinear regression analysis fitted to the Michaelis–Menten equation [46] using Sigmaplot Version 14.0.3.192. Data are presented as the mean of three independent experiments with 95% CI.

Mode of inhibition experiments were carried out by testing enzymatic activity at a range of compound and substrate concentrations. Compounds were tested at five different concentrations selected around the pIC₅₀ against five substrate concentrations (between 2.2 and 20 µM). The data were fitted globally to equations describing inhibition [47] to determine the respective kinetic constants. All analyses were performed in Sigmaplot Version 14.0.3.192.

Best fit was determined using Akaike information criterion corrected [48], as well as reference to the standard error of the fit [49]. Where relevant, data are presented as the mean of three independent experiments with 95% CI.

Studies of binding kinetics were performed on a Biacore™ 8K + instrument (GE Healthcare). Full-length HsRNMT complexed to HsRAM 1–90 (Avi-HsRNMT-RAM) was immobilised in HBS-P + buffer (10 mM HEPES, 150 mM NaCl, 0.5 mM TCEP, 0.05% BSA and 0.05% Tween 20) on a Series S high-affinity streptavidin (SA) sensor chip by C-terminal Avi-tagcapture, to an immobilisation level of 3300 RU. Compounds were tested in duplicate in a three-fold, eight-point serial dilution series in immobilisation buffer supplemented with 1% DMSO, with a flow rate of 30 μL min^-1, a contact time of 60 s and a dissociation time of 120 s at 25°C.

All monitored binding resonance signals were referenced with responses from unmodified reference surfaces, corrected for bulk shifts arising from differences in DMSO concentrations between samples and running buffer (solvent correction) and blank-referenced. Analysis was performed using the Biacore™ Insight Evaluation Software 2.0 with data fitted using the Langmuir 1:1 model.

The full-length HsRNMT complexed with HsRAM 1–90 (Avi-HsRNMT-RAM) was immobilised onto a Series S high-affinity SA sensor chip via C-terminal Avi-tag capture in HBS-P + buffer (10 mM HEPES, 150 mM NaCl, 0.5 mM TCEP) to an immobilisation level of 2200 RU, and the titration was performed at 15°C.

Full-length HsRNMT-RAM was dialysed overnight at 4°C against 25 mM HEPES pH 7.5, 150 mM NaCl, 1 mM TCEP supplemented with 10 μM Sinefungin or 10 μM SAH. Working stock for ligands and protein complex was prepared at the required concentration by diluting the stocks with the dialysis buffer. The DMSO concentrations in the cell and syringe solutions were matched to 2.5% for the compound (DDD1060606 or DDD1870799) titrations. Experiments were performed using 13 injections of 3 μL of ligand using Malvern PEAQ-ITC system at 25°C, at 750 rpm with reference power set to 7 μcal/s. The thermograms were analysed using MicroCal PEAQ-ITC Analysis Software, blanked to ligand control and fit to One Set of Sites Model.

HsRNMT (165-476) Δ416–455 GLGC (HsRNMT) was buffer exchanged in PIPES 20 mM, NaCl 200 mM, glycerol 10% v/v, TCEP 1 mM, pH 6.5 using Zeba 10 kDa desalting columns (Thermofisher) and concentrated to 17 mg/mL. Co-complexes structures were achieved by mixing SAH (1 mM) with the appropriate inhibitor (5 mM), whereas the crystallisation of cap was only possible by a combination of Sinefungin (1 mM) with GMP-PnP (10 mM) and 2 mM MgCl₂. HsRNMT was crystallised at 17°C using the sitting drop vapor diffusion method by mixing 200 nL of the protein solution mixed with 200 nL of a reservoir solution containing 0.1M MES pH 6.3–6.9, 50 mM Na₂SO₄, 18–25% PEG 6000. Crystals grew in 48 hours and were flash-frozen in liquid nitrogen in the reservoir solution supplemented with 28% glycerol. Data sets were collected at 100K at beamline synchrotron ESRF-ID23 or using a Rigaku M007HF copper‐anode generator fitted with Varimax Cu‐VHF optics and a Saturn 944HG + CCD detector. Data were processed using XDS [50] in P21 and scaled using Aimless [51] to a resolution between 2.0 and 2.5 Å. The following structure solution and refinement activities were performed. The initial phases were determined by molecular replacement using Phaser [52] from the CCP4 suite [53] using the human catalytic domain HsRNMT (pdb code 5E8J) with HsRAM removed and the segment 416–455 as the search model. The initial density maps were further improved by solvent flattening and histogram matching using RESOLVE [54] as implemented in the Phenix suite [55]. The model was refined through iterative cycles of model building and computational refinement with COOT [56] and Phenix Refine (Phenix suite). Data measurements and model refinement statistics are presented in Supplementary Table S1. Coordinate files and associated experimental data have been deposited in the Protein Data Bank with accession codes 8Q9W, 8Q69 and 8Q8G.

All crystallographic data are deposited in the Protein Data Bank. All other data are contained within the manuscript.

The Authors declare that there are no competing interests associated with this manuscript.

We gratefully acknowledge financial support from the MRC Confidence in Concept award [MC_PC_19034], UKRI Impact Accelerator Award [MR/X502832/1] and a Wellcome Trust Centre Award [203134/Z/16/Z]. The authors would like to thank the European Synchrotron Radiation Facility (ESRF) for provision of synchrotron radiation facilities under proposal number MX-1902, and the staff of beamline ID-23 for assistance. We would also like to acknowledge the X-ray Crystallography Facility at the University of Dundee, which was supported by The Wellcome Trust [award no. 094090].

Lesley-Anne Pearson: Conceptualisation, Validation, Methodology, Formal analysis, Investigation, Data Curation, Writing – Original Draft, Visualisation. Alain-Pierre Petit: Conceptualisation, Validation, Methodology, Formal analysis, Investigation, Data Curation, Writing – Original Draft, Visualisation. Cesar Mendoza Martinez: Investigation, Writing – Original Draft, Fiona Bellany; Resources, De Lin: Methodology, Investigation, Sarah Niven: Investigation, Rachel Swift: Investigation, Thomas Eadsforth: Resources, Paul Fyfe: Writing – Original Draft, Resources, Marilyn Paul; Investigation, Vincent Postis; Conceptualisation, Supervision, Xiao Hu: Formal Analysis, Victoria H. Cowling: Conceptualisation, Writing – Original Draft, Writing – Review and Editing, David W Gray: Conceptualisation, Validation, Data Curation, Writing – Original Draft, Writing – Review and Editing, Supervision, Funding acquisition .

The authors recognise the valuable input from our Compound Management (Alex Cookson, Steve Bell, Kirsty Cookson, Fraser Hughes) and Data Management (Steve Thompson, Edan Gardner, Aren Lai, Kashish Sharma) Teams.

GMP-PnP

guanosine 5ʹ-[β, γ-imido]triphosphate

ITC

isothermal titration calorimetry

MIF

molecular interaction field

RAM

RNMT-activating miniprotein

RNMT

RNA guanine-7 methyltransferase

SAH

S-adenosylhomocysteine

SAM

S-adenosylmethionine

SPR

surface plasmon resonance