Engineering Bisubstrates to Target m⁶Am RNA Methyltransferases: Synthesis and Computational Studies

ABSTRACT

Phosphorylated CTD‐interacting factor 1 (PCIF1) has been recently discovered to introduce a methyl group at the N6 position of the first transcribed Am‐adenosine in nascent capped mRNA (m⁶Am modification), modulating mRNA stability, transcription, and translation. In addition, the activity of PCIF1 is involved in various diseases, making PCIF1 a potential therapeutic target. In the absence of complete crystal structures of PCIF1 with an RNA substrate and/or cofactor and with the aim of accessing potent inhibitors, the bisubstrate strategy was employed to synthesize a series of bisubstrate analogs of PCIF1 that contain an analog of the cofactor S‐adenosyl‐L‐methionine (SAM) covalently linked to an Am‐modified RNA substrate. The versatility of the synthesis was exploited to increase structural diversity at the amino acid chain of SAM, and a cap analog was introduced by click chemistry. Also, AlphaFold 3 was used to generate a complete structure of PCIF1 with an RNA substrate. Docking studies carried out with the bisubstrates indicate that the SAM analog occupies the cofactor binding site while the Am‐RNA substrate interacts with the putative RNA pocket. We think that these results could be a valuable starting point for the design of potent inhibitors of PCIF1.

Use of the convertible nucleoside approach to synthesize m6Am RNA methyltransferase bisubstrates. This methodology allows for the introduction of modifications on the SAM analog moiety and the RNA substrate part, including the introduction of a cap analog by click chemistry. Docking studies performed on full‐length methyltransferase PCIF1 generated by Alphafold 3 reveal interactions between the bisubstrates and the two substrate pockets.2

1. Introduction

Methylation at the N6 position of 2’‐O‐methyladenosine (m⁶Am) is a prevalent modification of RNA mostly found in capped mRNA. Phosphorylated CTD‐interacting factor 1 (PCIF1), also known as CAPAM, initially discovered for its ability to bind the phosphorylated C‐terminal domain (CTD) of RNA polymerase II [1], has been recently identified to catalyze the N6 methylation of 2’‐O‐methyladenosine (Am) at the first transcribed position of capped mRNAs (Figure 1A) [2, 3, 4, 5]. Since studies have demonstrated the reversibility of the m⁶Am modification [6, 7, 8] and have explored its roles in capped mRNA processing and translation, with controversial results [9, 10]. In parallel, the importance of PCIF1 in human health and disease has become increasingly apparent with studies linking its activity to cancer progression [11, 12, 13, 14], viral infections [15, 16, 17], and obesity [18]. Particularly in colorectal cancer, PCIF1 has been shown to modulate tumor malignancy and response to immunotherapy [11, 12]. Also, PCIF1 promotes infection by sustaining expression of the coronavirus receptors ACE2 and TMPRSS2 via m⁶Am‐dependent mRNA stabilization [16]. Altogether, these examples highlight the potential of PCIF1 as a therapeutic target. Though specific inhibitors of PCIF1 have not yet been identified. Recent works have explored alternative approaches to target PCIF1, such as the use of RNA interference to silence PCIF1 [11] or CRISPR/Cas9‐mediated deletion [12].

FIGURE 1.

(A) Methyl transfer reaction catalyzed by PCIF1. (B) Previous work. (C) Target molecules.

To date, very few crystal structures of PCIF1 are available. To decipher the enzymatic mechanism of m⁶Am deposition, Akichika et al. have reported crystal structures of partially resolved human PCIF1 and zebrafish PCIF1, as well as in complex with SAH [2]. Importantly, there is no experimental structure for hPCIF1 bound to RNA substrates. In the same study, structures of the ternary complexes zCAPAM/m⁷GpppA/SAH and zCAPAM/m⁷GpppAmG/SAH were reported, in which the position of the 2’‐O‐methyladenosine was not defined. All these structures are similar, revealing a methyltransferase domain (MTD) with a canonical Rossmann fold found in m⁶A methyltransferases (m⁶A MTases) that contains the substrate binding loop (SBL: P612‐N638), the catalytic loop (CL: N553‐C557), and the active site loop (ASL: F479‐P494). Close to the MTD catalytic loop, the helical domain forms a positively charged groove that is unique to PCIF1 to bind the capped mRNA, conferring its specificity (Figure S1) [2, 19]. In addition, PCIF1 has higher binding affinity and activity toward Am‐methylated RNA compared with unmethylated RNA, further explaining its substrate preference [2].

We have already designed bisubstrate analogs for m⁶A and m¹A MTases that simultaneously target the substrate and the SAM cofactor binding sites. Using versatile chemical strategies, we could access a diversity of RNA‐SAM conjugates by modulating the RNA size and sequence and the nature of the linker (Figure 1B) [20, 21, 22, 23]. These compounds revealed important tools for structural studies of bacterial and human m⁶A RNA MTases to explore RNA binding [24, 25, 26] and as modulators of RNA methylation in oncogenic contexts [23]. Bisubstrates of MTases were also developed by other groups, enabling the inhibition of coronavirus RNA N7‐G MTases with submicromolar IC₅₀ values [27, 28, 29, 30, 31, 32, 33, 34].

In this study, we designed and synthesized a series of bisubstrate analogs as potential inhibitors of the m⁶Am RNA MTase PCIF1. Since PCIF1 uses the SAM cofactor as the methyl donor to methylate capped Am mRNA, the structure of the target molecules contains a SAM analog connected to an Am‐containing RNA substrate (Figure 1C). In our previous work, we explored different chemistries, including aromatic nucleophilic substitution reaction (SNAr) [20, 25], isoprenylation [20], and CuAAC [21, 22] to connect SAM analogs to RNA substrate analogs, leading to various linker types (alkyl, urea, and triazole, respectively) (Figure 1B). From these studies, alkyl linkers (C2 and C3) were identified as optimal, as they allowed proper positioning of the bisubstrate analogs within both the SAM and substrate binding pockets [25, 26]. For this reason, in the present work, we focus on SNAr chemistry allowing alkyl linkers, and demonstrate that this strategy is compatible with the construction of more complex bisubstrate analogs. This includes (i) modifications at the ribose moiety (2′ and 5′ positions), (ii) functionalization of the α‐amino acid part of the SAM analog (introduction of an amine or a squaramide motif known to be a bioisostere of the amino acid moiety [35, 36, 37, 38]), and (iii) access to a cap analog, which is synthetically challenging to obtain (Figure 1C). In addition, to overcome the lack of structural data, modeling studies were performed to generate a complete structure of PCIF1 bound to an oligonucleotide substrate. The docking of two compounds in this model confirmed their binding in the two substrate sites.

2. Results and Discussion

2.1. Synthesis

First, we envisioned the synthesis of Am‐SAM conjugates containing a 2'‐O‐methyladenosine connected at the N6 position of the adenosine to a SAM analog. Based on the pioneering work of Wan and coworkers [39], we used the convertible nucleoside strategy to access these molecules by performing a SNAr between an activated 2’‐O‐methyladenosine and a SAM analog that contains a primary amine function. Thus, taking advantage of previous work to access m⁶A bisubstrates [20, 23], we first prepared O‐benzotriazolyladenosine 4 and 8 following the chemical routes reported in Scheme 1. First, commercially available 2’‐O‐methyladenosine 1 was converted into 2’‐O‐methylinosine 2 by a reaction of deamination in the presence of sodium nitrite in acetic acid and solvolysis (Scheme 1) [40]. Positions 3’ and 5’ were then protected using TBDMSCl, leading to intermediate 3 in 73% yield. Inosine 3 was then activated by reacting with benzotriazol‐1‐yloxy)tris(dimethylamino)phosphonium hexafluorophosphate (BOP reagent), affording compound 4 in 84% yield. To access the activated azido derivative 8, the 5’ position of inosine 3 was selectively deprotected in the presence of aqueous TFA, leading to compound 5. Subsequent activation in the presence of mesyl chloride followed by nucleophilic substitution afforded azido‐inosine 7, which was reacted with BOP reagent to afford the expected compound 8 in 71% yield (Scheme 1).

SCHEME 1.

Synthesis of activated inosines 4 and 8.

SAM‐adenosine conjugates 10a, 10b, and 11 were obtained by nucleophilic aromatic substitution between activated inosines 4 and 8 with SAM analogs 9a and 9b [20] with yields ranging between 80% and 89% (Scheme 2). Removal of the TBS groups in the presence of cesium fluoride afforded intermediates 12–13. In the case of the longer alkyl chain (n = 2), compound 12b could not be recovered as a pure compound after silica gel chromatography and was engaged as such in the following step to afford bisubstrate analog 14b in 13% yield over two steps. The deprotection of the amine and carboxylic acid functions for 12a and 13 was conducted in the presence of zinc bromide to afford 2’‐O‐methyladenosine‐SAM conjugates 14a and 15 in 47% and 41%, respectively, after HPLC purification (Scheme 2).

SCHEME 2.

Synthesis of 2’‐O‐methyladenosine‐SAM conjugates 14a, 14b, and 15.

The synthesis of squaramate‐based SAM‐adenosine conjugates 26a and 26b was achieved following the procedure described in Scheme 3. Convertible nucleosides 4 or 8 were reacted with SAM analogs 16a and 16b (see Supporting Information) to afford compounds 17–18 with yields between 68 and 75%. Using cesium fluoride for TBS groups removal, partial deprotection of the exocyclic N6 amine of adenosine was observed, leading to mixtures of mono‐ and di‐Boc derivatives. Di‐Boc compounds 19a, 21b, and 22 were engaged in the following deprotection step, affording the amino derivatives 24–25. Finally, intermediates 24a and 24b were reacted with dimethyl squarate to access target 2’‐O‐methyl‐adenosine‐SAM conjugates 26a and 26b in 24% and 38% yield respectively after HPLC purification.

SCHEME 3.

Synthesis of 2’‐O‐methyl‐adenosine‐SAM conjugates 26 with a squaramate unit.

We were then interested in the preparation of RNA‐SAM conjugates. Since zfPCIF1 was co‐crystallized with oligo m⁷Gppp‐Am‐G and since RNA substrates that contain a guanosine as the second transcribed nucleotide were used for the identification of PCIF1 as the m⁶Am MTase [2], we started with the synthesis of the dinucleotide‐SAM conjugate Am(SAM)‐G. In this aim, the activated phosphoramidite inosine 27 (see Supporting Information) was reacted with guanosine 28 to access dinucleotide 29 in 83% yield (Scheme 4). Compound 29 was further engaged in a nucleophilic aromatic substitution with SAM analog 9a, affording the dinucleotide‐SAM conjugate, which was subsequently deprotected to afford the final conjugate 30 in 10% overall yield (over four steps), after HPLC purification with aqueous NH₄OAc/MeCN as eluent mixture.

SCHEME 4.

Synthesis of Am(SAM)‐G conjugate 30.

Finally, we chose to introduce an analog of the RNA cap using click chemistry to enhance structural similarity with the native substrate of PCIF1. In this aim, we synthesized activated 7‐methylguanosine alkyne 31 as described in the literature [41], which was engaged with azido derivative 15 in CuAAC in the presence of copper(II) sulfate pentahydrate, sodium ascorbate, and tris((1‐hydroxy‐propyl‐1H‐1,2,3‐triazol‐4‐yl)methyl)amine (THPTA) to afford triazole 32 in 25% yield (Scheme 5).

SCHEME 5.

Access to triazole‐based bisubstrate 32.

2.2. Computational Studies

Molecular modeling studies were performed to predict the binding mode of our bisubstrates within PCIF1. The crystallographic structure of PCIF1 in complex with SAH is incomplete, with the C‐terminal region (M1–E173), the N‐terminal region (Q869–T874), and the G480–L489 segment of the active‐site loop remaining unresolved. To overcome these limitations, we exploited the capabilities of AlphaFold 3 (AF3) to model protein–small molecule and protein–nucleic acid complexes [42], and generated a full‐length PCIF1 model bound to the CGAA oligonucleotide shown to be an hPCIF1 RNA substrate [43]. Our model shows high reliability, supported by AlphaFold confidence metrics (ipTM = 0.82, pTM = 0.72) and a low RMSD of 2.03 Å relative to the experimental apo structure (PDB: 6IRV) (Figures 2A and S2). In the predictive model, guanosine of the CGAA sequence binds the putative cap‐binding site, establishing key contacts with conserved residues (E558 and W588) as observed for m7G in zfPCIF1 (PDB: 6IS0) (Figure 2B). Contrary to the m7G in the zfPCF1, the modeled guanosine does not form any contact with R235, P591, and P592 (Figure 2B, left). Interestingly, the 3’ adenosine of the sequence CGAA interacts with the catalytic residues N553, P554, P555, and F556 of the highly conserved NPPF motif, which is present in all adenine N6‐DNA MTases (Figure 2B, right) [44]. Furthermore, we validated our model by predicting SAH binding within the cofactor binding site using molecular docking. In our model, SAH adopts a binding mode similar to that observed in the crystal structure of hPCIF1 (PDB: 6IRW), with an RMSD of 0.58 Å between the docked pose and the experimentally determined structure (PDB: 6IRW) (Figure S3).

FIGURE 2.

(A) Superposition of the AF3 prediction (blue) with the experimental apo structure of hPCIF1 (red) (PDB: 6IRV). The AF3‐modeled loop G480–L489 in the active site is highlighted in yellow. The CGAA RNA substrate is omitted for clarity. (B) Structure of the substrate binding site of PCIF1. (Left) Superimposition of guanosine (pink) in the AF3 hPCIF1 model (blue) and m7G cap (gray) in zfPCIF1 (PDB: 6IS0). (Right) 3′ Adenosine in the AF3 hPCIF1 model. Nonbonded interactions (dashed lines) are shown as follows: hydrogen bonds (green), electrostatic interactions (orange), π–alkyl interactions (pale pink), and π–π interactions (dark pink).

We then used this AF3 model system to predict the binding mode of bisubstrates 14a and 14b. Docking of 2’‐O‐methyladenosine–SAM conjugates 14a and 14b using CDOCKER implemented in Biovia Discovery Studio 2025 revealed similar overall orientations within the active site, but with distinct binding patterns (Figures 3 and S4). In both cases, the SAM analog moiety occupies the cofactor binding site, forming hydrogen bonds and engaging in π‐stacking and hydrophobic interactions, consistent with the binding mode of SAH in hPCIF1 (Figure 3, top right). The 2’‐O‐methyladenosine moiety of compounds 14a and 14b adopts a positioning consistent with that observed for the CGAA oligonucleotide, and is located within a hydrophobic pocket formed by residues N553, P554, P555, and F556 of the NPPF motif, together with Y618 and Y633–H637 of the substrate‐binding loop (SBL) and Y484–G488 of the active‐site loop (ASL). Notably, 14a (n = 1 linker) achieves superior anchoring through additional contacts of the adenine substrate with Y618 and H637 of the SBL (Figure 3, bottom right).

FIGURE 3.

AF3 model bound to bisubstrates 14a, 14b, and CGAA oligonucleotide. Left: hPCIF1 is shown as a transparent ribbon. Compound 14a (orange), 14b (green), SAH (yellow), CGAA oligonucleotide (pink), AF3 residues (blue), and interacting residues of hPCIF1 (PDB: 6IRW) are represented as sticks. Top right: Close‐up on SAM analog moieties of 14a, 14b, and superimposed SAH bound to hPCIF1. Bottom right: Close‐up on Am moieties of 14a, 14b, and 3′ adenosine of CGAA oligonucleotide. Nonbonded interactions are indicated as dashed lines: hydrogen bonds (green), π–alkyl interactions (pale pink), and π–π interactions (dark pink). The alkyl linker between the SAM analog and Am moieties is omitted for clarity.

To complement the docking studies and further assess the binding propensity of compounds 14a and 14b, binding free‐energy calculations were performed using two independent computational methods, GNINA [45] and PRODIGY tools [46]. ΔG value of −10.50 kcal mol^{− 1} was calculated for the two bisubstrates, suggesting strong binding affinity.

3. Conclusion

In summary, we have reported the synthesis of bisubstrate analogs for the methyltransferase PCIF1 that catalyze the installation of a methyl group on the first transcribed adenosine on mRNA. These bisubstrates contain a SAM analog covalently linked via alkyl linkers to Am‐containing RNA mimics, including an adenosine, a dinucleotide, and a cap analog. Furthermore, our synthetic strategy allows for the modification of the 5’ position of the SAM‐adenosine conjugates with the introduction of an azide group for further functionalization. Molecular modeling studies were carried out to study the binding mode of the bisubstrate molecules. Since complete crystal structures of complexes PCIF1/RNA/SAM or SAH are lacking, AlphaFold 3 was first used to generate a full‐length hPCIF1 model with a CGAA oligonucleotide shown to be a good substrate for PCIF1 [43]. Importantly, the reliability of the AF3 model was validated by docking SAH within the cofactor binding site, resulting in a similar pose to SAH in the crystal structure of hPCIF1. Docking studies of two Am‐SAM conjugates confirm that the SAM analog binds the cofactor pocket while the 2’‐O‐methyladenosine moiety is positioned within a pocket formed by residues of the substrate‐binding, active‐site, and catalytic loops. Finally, these compounds could serve as chemical tools to get complete crystal structures of PCIF1.

METTL4 is the other enzyme that has been recently identified and shown to catalyze the N6 methylation of Am at an intern site in U2 small nuclear RNA (snRNA), which regulates pre‐mRNA splicing [47, 48, 49]. This is a member of the MT‐A70 family of MTases, which has recently gained attention for its dual role in RNA modification and mitochondrial DNA regulation. Structural studies reveal that METTL4 achieves substrate specificity through a unique L‐shaped catalytic cavity and a conserved Gate Loop2 domain, enabling precise recognition of 2'‐O‐methyladenosine in single‐stranded RNA [50]. Thus, we think that our bisubstrate analogs appear as critical tools for a better understanding of substrate specificity achieved by PCIF1 and METTL4.

Conflicts of Interest

The authors declare no conflict of interest.

Supporting information

Supporting File 1: Experimental procedures and characterization data are available in the Supporting Information. The authors have cited additional references within the Supporting Information [2, 3].