A cryptic pocket in METTL3-METTL14 regulates m6A conversion and sensing

The nuclear METTL3-METTL14 enzyme complex transfers a methyl group from S-adenosyl-L-methionine (SAM) to the N6 amino group of an adenosine (A) base in RNA to convert it to m6A and in ssDNA to 6mA. m6A marks are prevalent in eukaryotic mRNAs and lncRNAs and modulate their stability and fate in a context-dependent manner. The cytoplasmic METTL3 can act as a m6A reader to regulate mRNA translation. However, the precise mechanism that actuates the switch from m6A writer to reader/sensor is unclear. Here, we present a ~2.5Å crystal structure of the methyltransferase core of human METTL3-METTL14 in complex with the reaction product, N6-methyladenosine monophosphate (m6A), representing a state post-catalysis but before the release of m6A. m6A occupies a novel evolutionarily conserved cryptic pocket in METTL3-METTL14 located ~16Å away from the SAM pocket that frequently mutates in cancer. We propose a two-step model of swiveling of target A upon conversion to m6A and sensing its methylation status by the cryptic pocket, enabling it to actuate enzymes’ switch from writer to an m6A-sensor. Cancer-associated mutations cannot distinguish methylated from unmethylated adenine and show impaired RNA binding, de-stacking, and defective m6A writing and sensing.

m 6 A marks are also present in genomes of RNA viruses such as hepatitis C, Zika, dengue, West Nile, yellow fever, and SARS-CoV-2, and modulate viral replication and host immune response 27 .
Thus, METTL3 has emerged as an attractive drug target for anti-cancer and anti-viral therapy development. Consistently, pharmacologic inhibition of METTL3 limits the growth of acute myeloid leukemia 26 and SARS-CoV-2 28,29 . The first METTL3 inhibitor STC-15 that targets its SAM pocket has entered the Phase I clinical trial (NCT05584111). Despite significant advancement in the m 6 A field and interest in targeting it for therapy, the structural details of RNA recognition and catalysis by METTL3-METTL14 are lacking. Here we present a ~2.5Å crystal structure of the methyltransferase core of METTL3-METTL14 bound to methyladenosine monophosphate (m 6 A), a product mimic of the methylation reaction (Fig. 1a).
We show that m 6 A occupies a novel cryptic pocket constituted to a large extent by residues from METTL3 and an interface arginine (R298) of METTL14. This pocket is evolutionarily conserved in mammals, plants, and yeast (Fig. 1b). Importantly, residues that partake in interaction with m 6 A are mutated in gynecologic, stomach, kidney, and bladder cancers 32 (Fig. 1b). When introduced into wild-type METTL3-METTL14, the mutant enzymes exhibit a significant loss in catalysis, perturbed RNA binding, and compromised ability of de-stacking of the target adenine for presentation to the active site. Our data suggest that the target base swivels ~180º after methylation for sensing by the cryptic pocket located ~16Å away from the point of methyl transfer. METTL3-METTL14 uses this unique mechanism to sense the methylation status before releasing the substrate RNA. This arrangement will require de-stacking of the target base during catalysis and sensing. We also show that the wild-type METTL3-METTL14, but not the mutant, binds more tightly to an m 6 A-modified RNA to distinguish it from the unmethylated RNA. Moreover, the enzyme harboring R298P mutation, the most frequent mutation in endometrial cancer 32 , exhibits sub-optimal RNA binding, catalysis, and base de-stacking ability. Our results uncover entirely unexpected operating principles underlying methyl transfer and m 6 A-sensing by METTL3-METTL14.
We succeeded in resolving its structure in the presence of N 6 -methyladenosine 5'-monophosphate (m 6 AMP, referred to as m 6 A), a product of methylation reaction, by soaking m 6 A into apo crystals ( Fig. 1 a, Extended data Fig. 1a-c). The difference omit map showed clear and unbiased electron density for m 6 A, which was refined well with no discrepancies for the ligand, surrounding regions, or the rest of the protein (Extended data Fig. 1d-g and table 1). METTL3-METTL14-m 6 A model was refined to ~2.5Å resolution, with excellent stereochemistry and Rfree and Rwork of ~26.2 and 22.9%, respectively (Extended data Table 1). The final model contains one molecule each of METTL3 (aa 369 -579), METTL14 (aa 116 -402), one m 6 A, 90 water and two ethylene glycol.
The overall fold of METTL3-METTL14 is similar to those reported previously [14][15][16] , except for notable changes in the region around the m 6 A binding pocket. MTases adopt a β-class of MTase fold with a central β-sheet of seven parallel and one antiparallel β-strands flanked by three helices on each side. Three major loops (gate loops 1 and 2, and an interface loop) emanating from the central β-sheet of METTL3 participate in SAM, RNA, and METTL14 binding. While the two gate loops exhibit high flexibility upon SAM or SAH binding and release, the interface loop remains rigid due to extensive protein-protein contacts from METTL14 MTase (Fig. 1a).
Evolutionarily conserved m 6 A pocket plays an essential role in m 6 A sensing Strikingly, m 6 A occupies a cryptic pocket ~16Å away from the methyl donor SAM pocket with its N 6 -methyl moiety in an energetically favored syn conformation, facing outward (Fig. 1a).
Previously, this region was postulated to bind RNA due to its positive charge and polar nature [14][15][16] . m 6 A is stabilized by a vast network of specific interactions, mostly from METTL3 and R298 of METTL14. The purine ring of m 6 A is sandwiched between the side chain of M402 and the backbone atoms of the interface loop residues, R471, T472, G473, and H474. The two arginine residues (R471 of METTL3 and R298 of METTL14) act like a clasp to hold the N 6 -methyl moiety in place through a direct h-bond between R298 and N 1 , van der Waals and hydrophobic interactions between N 6 -methyl and its aliphatic portion, and the amino group of the R471 side chain, respectively. The carbonyl oxygen of G473 appears to neutralize the positive charge of the R298 residue. The carbonyl moiety of R471 embraces the N 6 atom of m 6 A via a direct h-bond while the opposite side is stabilized by the side chain of H474 via a π-π interaction. Altogether, the arginine clasp, interface loop residues R471-H474 and M402, forms a partial closure around the methylated purine ring of m 6 A. The ribose in the C3'-endo conformation is stacked between the backbone atoms of G473 and H478 and the side chains of I400 and H478. The phosphate group of m 6 A is locked in place by multiple direct h-bonds with its phosphoryl oxygens and side chains of H478, E481, T433, and K459 (water-mediated) -all from METTL3, and another water-mediated interaction with E257 of METTL14. The side chain of H478 holds the m 6 A phosphate on one side and E257 of METTL14 on the other, thus acting as a hinge (Fig 1a, Extended data Fig. 1g).
Strict conservation of the extensive interaction network of m 6 A in human, animal, plant, and yeast suggests that m 6 A sensing by this cryptic pocket is an evolutionarily conserved mechanism (Fig.   1b). Several key residues that partake in m 6 A binding, such as R471 and R298 of the arginine clasp, E481, and H478 that stabilize the N 6 -methyl and phosphate groups are recurrently mutated in endometrioid and adenocarcinoma 32 (Fig. 1b). We introduced the R298P mutation in METTL14, a recurrent mutation event in endometrioid carcinoma 32,33 , and the R471H, E481A, T433A, K459A, and H478A mutations in METTL3. In addition, we generated two deletion mutants (∆472-473, ∆472-474) in which three residues of METTL3 (T472, G473, H474) that stack against the purine ring of m 6 A were deleted to shorten the interface loop.
We co-purified the full-length wild-type human METTL3-METTL14 and eight mutant enzymes from insect cells and probed their RNA methylation and binding activities. We used a 30-mer RNA oligo (NEAT2*) consisting of one canonical GGACU motif. Consistently, R298P and R471H mutants significantly reduced (up to 85%), whereas T433A resulted in ~20% loss in methyltransferase activity, agreeing with the reduced m 6 A levels observed in endometrial tumors harboring the R298P mutation 33 . The other five mutations in METTL3 (∆472-473, ∆472-474, K459A, E481A, and H478A) completely abolished the RNA methyltransferase activity of METTL3-METTL14 (Fig. 1c). Thus, the evolutionarily conserved m 6 A binding pocket is essential for efficient conversion of A to m 6 A.
Next, we quantitatively determined the binding affinities of wild-type (WT) and mutant enzymes to the substrate and a product RNA, wherein the target A base within GGACU is replaced by m 6 A.
We covalently attached a fluoresceine moiety to the 5'-end of both the substrate NEAT2* (A-RNA) and product (m 6 A-RNA) RNAs and performed fluorescence polarization-based assays. The WT enzyme binds the m 6 A-RNA with a 2-fold stronger affinity than A-RNA (Kd = 9 vs. 20 nM) ( Fig.   1 d,e), corroborating previous studies attributing the m 6 A-reader function to METTL3 in vivo 23,25,31,34 . In contrast, the mutants, although bound to both RNAs with weaker, yet nanonmolar affinity (Kd range = 18 -30 nM), failed to distinguish m 6 A from A. One exception was the R298P mutant, and to some extent R471H (both mutate in cancers and belong to the arginine clasp motif that stablizes the m 6 A), that not only showed a significant loss in binding to both RNAs, but also switched the binding preference to unmodified (A-RNA) compared to the WT enzyme ( Fig. 1e, Extended data Fig. 1h). Thus, its altered specificity (inability to sense and distinguish m 6 A) coupled with a significant loss in RNA methylation and binding affinity by the R298P mutation could promote tumorigenicity and growth of endometrial tumors as observed previously 33 . The nanomolar affinity of mutant enzymes suggests a significant contribution of flanking accessory motifs such as zinc fingers of METTL3 and RGG repeats of METTL14.

Base swiveling facilitates m 6 A sensing
The two loops in METTL3 (gate loops 1 and 2) surrounding the methyl donor SAM and acceptor base A pockets show varying degrees of flexibility upon SAM and SAH binding from their original positions in a ligand-free (apo) form [14][15][16] . Thus, we compared the m 6 A structure with three states (SAM, SAH, and apo). These loops also move in opposite directions upon m 6 A binding from their original positions in the SAM-bound METTL3 (Fig. 2a). Gate loop 1 (aa 398-409) moves ~ 5.7Å inward to the direction of m 6 A, whereas the gate loop 2 (aa 506-512) moves ~ 7.8Å outward, with several residues in this region, including H512, that flips ~180º. The invariant T433 and G434 from a small loop between β3 and α2 move ~ 2.1Å with the side chain of T433 rotating ~90º to stabilize the phosphate and ribose of m 6 A (Fig. 2a). This region remains unperturbed in SAHbound METTL3, suggesting the m 6 A binding to this pocket occurs after hydrolysis of SAM ( The side chain of M402 from gate loop 1 in the m 6 A structure stacks over the purine ring of m 6 A. In the SAM-bound form, this region is moved >5Å away, but in the SAH and apo forms, the M402 side chain will sterically clash with m 6 A ribose (distance between C of M402 and C4' of ribose ~1.2 Å). To avoid this clash, the side chain of M402 in m 6 A-METTL3 rotates > 45º, resulting in a ~3.8Å gain in the distance for the C atom compared to its position in the apo structure. As a result of this repositioning, the inter-gate area between interface loop (H474) and gate loop 1 (M402) becomes wider, from 6.8Å in apo to 8.0 Å in the m 6 A structure (Fig. 2d) Such a rotation may necessitate the de-stacking of the target base for its presentation to catalytic pocket and or base swiveling.
A water molecule at the putative site of the substrate A base is present in the m 6 A structure to compensate for the loss in binding energy in the emptied site by rotation of m 6 A from this site post-catalysis. This water coordinates with K459, and its mutation to alanine abolishes the methylation activity (Fig. 1c). SAM-dependent DNA methyltransferases, including the ancestral members of MTA-70 family MTases such as EcoP15I, efficiently flip the target adenine base out of the DNA helix into the catalytic pocket 36 . Although METTL3-METTL14 does not methylate dsDNA and dsRNA 11,12 , it can still de-stack the target base from a single-stranded RNA into the catalytic pocket, similar to the m 6 A/m 6 Am eraser enzyme, FTO 37,38 , and the m 6 A reader, YTHDC1 39 . To test this activity, we replaced the target A (6-aminopurine) in a GGACU in a 14mer ssRNA with 2-aminopurine (2Ap), a fluorescent nucleobase used as a conformational probe due to its high sensitivity to changes in the local environment induced by DNA 40

METTL3-METTL14 acts as an atypical m 6 A sensor
The m 6 A-METTL3-METTL14 structure allowed us to gain valuable insights into how m 6 A writer (METTL3-METTL14), eraser (FTO), and reader (YTHDC1) proteins accommodate m 6 A. We examined their m 6 A pocket in detail (Fig. 3a-c). Despite the lack of obvious resemblance at the protein sequence, domain, and structure levels, we observed high similarity in the interaction networks of m 6 A in METTL3-METTL14 to the binding mode of 6mA in FTO (PDB: 5ZMD) and m 6 A in YTHDC1 (PDB: 4R3I) (Fig. 3). Of note, the purine ring of 6mA in FTO stacks between a hydrophobic amino acid, L109 (the equivalent of M402 in m 6 A), from the top and the backbone atoms of V228, S229, and H231 (the equivalent of T472, G473, and H474 in m 6 A) from the bottom. Interestingly, the arginine clasp we found in m 6 A-METTL3-METTL14 is also present in 6mA-FTO. Notably, the side chain of R96 in FTO forms a direct h-bond to N 1 of 6mA, while the guanidino group of its R322 residue forms a van der Waals interaction with N 6 methyl group, akin to identical interactions by R298 and R471 to stabilize m 6 A in m 6 A-METTL3-METTL14. Stacking interactions that lock the sugar moieties in place also display similarities. For example, the sugar of 6mA in FTO stacks between I85 and H231, whereas the sugar of m 6 A stacks between I400 and H478 of METTL3 (Fig. 3a, b).
We found that m 6 A in METTL3-14 and YTHDC1 (PDB: 4R3I) had many similarities and striking differences, mainly in the orientation of the base (Fig. 3c). The N 1 of m 6 A forms an h-bond with N367, whereas an h-bond with carbonyl of S378 akin to carbonyl of R471 of METTL3 stabilizes the N 6 . Additional hydrophobic interactions from W377 and W428 also support the N 6 methyl group in YTHDC1. The nature of stacking interactions for the purine ring is also similar, i.e., hydrophobic residues M434, L380, and L439 on one side and backbone atoms of K361, S362, and N363 on the other. However, the orientation of the m 6 A base in YTHDC1 is reversed by 180º compared to 6mA in FTO and m 6 A in METTL3-14. As such, when the direction of sugars and phosphates of modified bases is aligned in three structures (facing downward in Fig. 3a, b, and upper panel of c), the hydrophobic residues (M434/L380/L439) in YTHDC1 stack from the bottom side and the backbone atoms of K361, S362, and N363 stack from the top side, in contrast to the base orientation in FTO and METTL3. A ~180º rotation of YTHDC1 will place the interacting residues in all three proteins in the same plane. However, the orientation of ribose and phosphate of m 6 A in YTHDC1 will also be reversed (facing upward, Fig. 3c lower panel). Thus, a m 6 A reader protein approaches the m 6 A entirely differently than a writer or eraser. This unique geometric 13 difference may allow the reader to avoid clashes with a writer or eraser enzyme acting simultaneously on same transcript.
We show that METTL3 possesses features that enable it to act as an atypical m 6 A sensor/readera function ideally suited for its emerging non-catalytic functions, including crosstalk with eIF3H to promote mRNA circularization, thereby enhancing RNA translation as observed in lung cancer 23,25 and bone marrow mesenchymal stem cells 21 . Consistently, METTL3 showed more robust binding to a methylated (m 6 A) RNA form, corroborating with previous results describing it as a 'm 6 A reader' for alternative mode of translation initiation during oncogenic translation 23,25 and cellular stress (e.g., heat shock) 34 .

Full-length METTL3-METTL14 and mutants
The full-length human METTL3 and METTL14 (wild-type and mutants) were expressed in insect cells (ExpiSF Expression System, Thermo Fisher) and purified using a protocol published earlier 12 .
In brief, the METTL3 and METTL14 plasmids were transformed individually into Max Efficiency All results are reported as the means from three independent experiments (n=3) for each group, with one standard deviation (s.d.).

Fluorescence polarization
The reactions were carried out in a buffer containing 10 mM HEPES pH 7.5 and 50 mM KCl. The

Data availability
The coding sequences of METTL3 (NCBI reference sequence GI: 33301371) and METTL14 (NCBI reference sequence GI: 172045930) used in this study are available at NCBI. We have provided source data as a separate Source Data file. Requests for additional material and information should be directly addressed to Y.K.G. (guptay@uthscsa.edu).   suggesting that reader proteins approach RNA from the opposite direction. The m 6 A pocket of METTL3-METTL14 harbors features that enable it to act as an atypical m 6 A sensor/reader during its switch from writer to reader. Dashed lines, h-bonds.

Movie 1 | Loop dynamics during conversion of A to m 6 A and sensing by METTL3-METTL14
An animation shows the motions in two gate loops and the interface loop of METTL3-METTL14.
The model for catalysis and m 6 A sensing was generated by ChimeraX (