Cryo-EM structures of human m6A writer complexes

N6-methyladenosine (m6A) is the most abundant ribonucleotide modification among eukaryotic messenger RNAs. The m6A “writer” consists of the catalytic subunit m6A-METTL complex (MAC) and the regulatory subunit m6A-METTL-associated complex (MACOM), the latter being essential for enzymatic activity. Here, we report the cryo-electron microscopy (cryo-EM) structures of MACOM at a 3.0-Å resolution, uncovering that WTAP and VIRMA form the core structure of MACOM and that ZC3H13 stretches the conformation by binding VIRMA. Furthermore, the 4.4-Å resolution cryo-EM map of the MACOM–MAC complex, combined with crosslinking mass spectrometry and GST pull-down analysis, elucidates a plausible model of the m6A writer complex, in which MACOM binds to MAC mainly through WTAP and METTL3 interactions. In combination with in vitro RNA substrate binding and m6A methyltransferase activity assays, our results illustrate the molecular basis of how MACOM assembles and interacts with MAC to form an active m6A writer complex.


INTRODUCTION
Over one hundred different RNA modifications have been identified in a variety of RNAs, including mRNA, rRNA, tRNA, snRNA, microRNA, and long noncoding RNA. 1 Among these modifications, N 6 -methyladenosine (m 6 A) is the most prevalent and conserved ribonucleotide modification. 2 m 6 A modification is enriched in coding sequences and 3′ untranslated regions (3′ UTRs) of eukaryotic mRNAs and plays crucial roles in numerous physiological and pathophysiological processes. 3,4 Deposition of m 6 A onto RNAs is achieved by the m 6 A methyltransferase complex (m 6 A writer), a large holocomplex of 1000 kDa in size that contains methyltransferase-like 3 and 14 (METTL3/14), Wilms' tumor 1-associated protein (WTAP), KIAA1429 (VIRMA), Zinc finger CCCH domain-containing protein 13 (ZC3H13), RNA binding motif protein 15/15 paralog (RBM15/ 15B), and E3 ubiquitin ligase CBLL1 (HAKAI). [5][6][7][8][9][10][11] The heterodimer of METTL3 and METTL14 forms the catalytic subunit of the m 6 A writer, 12 also known as the m 6 A-METTL complex (MAC). 7 Structural and biochemical studies have shown that METTL3 is the catalytic component that binds to the cofactor SAM, while METTL14 plays a key role in stabilizing the conformation of METTL3 for optimal substrate binding. [13][14][15][16] In addition, WTAP, VIRMA, ZC3H13, HAKAI, and RBM15/RBM15B have been identified in various species to play regulatory roles in m 6 A writer activity and region-selective m 6 A methylation of mRNAs. [5][6][7][8][9][10] These components form a regulatory subunit called the m 6 A-METTL-associated complex (MACOM) 7,17 that is conserved in most eukaryotic species, such as flies, mice, humans, and plants. Protein-protein interaction analysis 18 has revealed that the four components of MACOM, HAKAI, WTAP, VIRMA, and ZC3H13 can form a stable functional complex (hereinafter referred to as HWVZ) in human cells, which has been validated by co-fractionation experiments. 7,8,10 WTAP was first identified as a METTL3-interacting protein in plants 19 and was shown to be critical for the anchorage of the METTL3/14 heterodimer in nuclear speckles, 5 probably through the interactions with Leader Helix (LH) of METTL3. 20 VIRMA has been shown to be closely associated with WTAP and HAKAI in plants 21 and to mediate preferential m 6 A mRNA methylation in the 3′ UTR. 10 ZC3H13 is required for the nuclear localization of MACOM in mouse embryonic stem cells (mESCs) 8 and for bridging WTAP-RBM15 interactions in flies. 7 HAKAI interacts with other components in MACOM via its RING domain, 17,22 which is necessary for the stability of MACOM components 22 and m 6 A methylation in plants and flies. 21,23 However, due to the lack of structural information of MACOM, how the components of MACOM are assembled and interact with MAC is not well understood at the molecular level.
Here, we use cryo-electron microscopy (cryo-EM) to resolve the structures of the complexes formed by human MACOM core components (HAKAI, WTAP, and VIRMA with or without ZC3H13) at 3.0-Å resolution, and the structure of the MACOM complexed with MAC at an overall resolution of 4.4 Å (the resolution of MAC region is relatively lower due to its flexibility). These threedimensional (3D) structural information, in combination with crosslinking mass spectrometry and GST pull-down analysis, enable us to propose a plausible model for MACOM and MAC assembly to form an active m 6 A writer complex for RNA substrate binding and modification. This model facilitates understanding of the molecular function of MACOM and drug development for m 6 A-related diseases such as human cancers. 24

RESULTS
MACOM directly binds to RNA substrates and is essential for m 6 A writer activity Previous studies have shown that each component of MACOM is essential for m 6 A modification in various species in vivo, 19,[21][22][23] and the catalytic subunit MAC is much less active when MACOM components are deleted (Fig. 1a). However, little is known about the properties of these components in vitro. After extensive trials of expression and purification, we obtained four complexes containing different MACOM components: HWVZ (containing HAKAI, WTAP, VIRMA, and ZC3H13 (1106-1668)), HWV (containing HAKAI, WTAP, and VIRMA), WVZ (containing WTAP, VIRMA, and ZC3H13 (1106-1668)), and WV (containing WTAP and VIRMA) (Fig. 1b, c). These four complexes possess different RNA substrate binding abilities according to electrophoretic mobility shift assay (EMSA) (Fig. 1d). HWVZ has the highest binding affinity (K d = 214.0 ± 2.6 nM), and removing HAKAI weakly attenuates the binding affinity of WVZ (K d = 256.3 ± 27.9 nM). In contrast, removing ZC3H13 has a dramatic effect, which reduces the binding affinities of HWV and WV complexes to K d = 562.7 ± 76.6 nM and 826.3 ± 336.0 nM, respectively. Therefore, all components of MACOM contribute to RNA substrate binding, especially for the ZC3H13 subunit. Then, we reconstituted the m 6 A writer complex by assembling MACOM components with MAC. The m 6 A writer complex reconstituted by HWVZ and MAC exhibits much higher enzymatic activity than MAC alone ( Fig. 1e-g). Removing HAKAI from the m 6 A writer complex has little effect on the activity ("MAC + WVZ" in Fig. 1g). However, removing ZC3H13 from the m 6 A writer complex dramatically reduces the enzymatic activity ("MAC + HWV" and "MAC + WV" in Fig. 1g). These results demonstrate that MACOM is essential for m 6 A writer activity and that ZC3H13 is a key component in MACOM. Notably, the RNA-binding ability of MAC is significantly weaker than the MACOM complex (Fig. 1d), suggesting that the lower enzymatic activity of MAC alone may be due to less RNA binding or the lack of active configuration in the absence of the other components. 10 Fig. S3 and Table S1). To our surprise, in the cryo-EM structure of the HWVZ complex, only three components (WTAP, VIRMA, and ZC3H13) can be found, while HAKAI is missing, although it is present in the SDS-PAGE analysis (Fig. 1c), indicating the high flexibility of HAKAI. Human   WTAP (residues 64-247), VIRMA (residues 334-1585), and ZC3H13 (residues 1492-1643) form a compact structure shaped like a "warhorse" that represents the core of human MACOM (Fig. 2a-d).
Notably, the overall structure of the HWV complex lacking the ZC3H13 component ( Fig. 2e-g) is similar to that of the HWVZ complex based on superimposition aligned by WTAP (Fig. 2h), although the horse body is slightly stretched out upon ZC3H13 binding ( Fig. 2i; Supplementary information, Video S1). These results suggest that WTAP and VIRMA form the structural core of MACOM with only minor conformational changes upon ZC3H13 binding, and HAKAI is not part of the structural core of MACOM.
WTAP forms a saddle-shaped homodimer through coiled-coil interactions Our cryo-EM structure reveals that WTAP forms a homodimer, 22 with each monomer consisting of four tandem helices (H1-H4/ H1'-H4' with coiled-coil interactions) and three linkers (L1-L3/ L1'-L3') ( Fig. 3a). Dimer formation depends on strong monomer-monomer interactions, where the interface area spans~4 620 Å 2 and contains three salt bridges, eight hydrogen bonds, and extensive non-bonded contacts (Supplementary information, Table S2), as revealed by PDBsum structure bioinformatics analysis. 28 The dimerization interactions of WTAP are corroborated by XL-MS data (Supplementary information, Fig. S4a, b and Table S3). Interestingly, the two copies of WTAP have a low crosscorrelation coefficient (CC = 0.47) between the maps and a low RMSD (15.9 Å) across all 184 pairs of the WTAP dimer, suggesting their conformational heterogeneity. There is a positively charged patch around the H2-L2-H3/H2'-L2'-H3' turn, which is composed of residues with relatively higher conservation ( Fig. 3b-d). A sharp turn was found between H2 and H3/(H2' and H3') at a 53°angle (Fig. 3a). Two hydrophobic cores play key roles in the sharp turn formation (Fig. 3d): W145, F147 in L2 and L142 in H2 form a hydrophobic ring around W145 from H2' coordinating with K155 in H3; the benzene ring of F147 engages in π-π stacking with P149 in L2' while binding to the hydrophobic patch formed by L157 and M158 in H3' and L157 in H3. Moreover, several hydrogen bonds in this area also stabilize the sharp turn of H2-L2-H3/H2'-L2'-H3'. These residues are highly conserved in WTAP orthologs (Fig. 3e).
VIRMA adopts a horse-shaped conformation composed of ARM-like (ARML) modules VIRMA is the largest component in MACOM and adopts a horseshaped conformation with 20 ARML modules ( Fig. 4a-d). The Belly (ARML1-4) and Chest (ARML5-7) are connected with the same shared helix (Connect Helix-1) (Fig. 4b), and the Chest and Back (ARML8-10) are connected by a similar connection, Connect Helix-2 ( Fig. 4c). Inter-repeat packing orientations are changed via Connect Helices (Fig. 4b, c). The Neck (ARML11-15) and Head (ARML16-20) are connected following the canonical stacking architecture of the ARML module (Fig. 4d). Notably, Back and Neck adopt different stacking orientations, mainly due to the hydrophobic interaction between ARML10 and ARML11 (Fig. 4e). Interestingly, a long loop after ARML20 extends along the Head domain and interacts with L1/L1' of the WTAP homodimer; therefore, it is called the Halter domain (Fig. 4a, d). Although the overall surface charge of VIRMA is mostly negative, several small regions are positively charged (Fig. 4f). These regions are also relatively conserved (Fig. 4g), suggesting that they may be important for interactions with other components of MACOM, especially WTAP.
WTAP and VIRMA form the structural core of MACOM through extensive interactions We then analyzed the intermolecular interactions between VIRMA and WTAP dimer. Strong interactions were found: a 2778 Å 2 interface area with three salt bridges and 15 hydrogen bonds for WTAP-a, as well as a 3303 Å 2 interface area with three salt bridges and eight hydrogen bonds for WTAP-b (Supplementary information, Table S2). Two positively charged patches in the Back and Belly domains of VIRMA are composed of highly conserved residues and located in the interface with H1/H1' and H4/H4' of WTAP, respectively (Fig. 4f, (Fig. 4j). Moreover, the side chain of R109 from L1' and the indole ring of W987 from the Neck domain form cation-π interactions in addition to the hydrogen bond between the side chain of R109 and the main chain of W987 (Fig. 4j). These specific interactions allow WTAP and VIRMA to form the MACOM core, and most of the interactions remain in the structure of the HWV complex when ZC3H13 is removed (Fig. 2d, g).
ZC3H13 stretches the MACOM conformation by interacting with VIRMA Unlike WTAP and VIRMA that form saddle-and horse-like conformations, respectively, the C-terminal fragment of ZC3H13 is only one-tenth of its full length and contains eight α-helices, but it does not exhibit a rigid conformation. Instead, it attaches to the Belly, Chest, and Neck domains of VIRMA and functions like a barding to protect the horse body (Figs. 2a-c, 5a; Supplementary informaiton, Video S1). Therefore, this fragment of ZC3H13 is called the Barding domain. Of particular interest, it is the most conserved fragment in ZC3H13 homologs across species, including flies (Supplementary information, Fig. S6b). The Barding domain of ZC3H13 has a strong interaction with VIRMA, and the interface is~4435 Å 2 , involving seven salt bridges and 19 hydrogen bonds (Supplementary information, Table S2). In contrast, the interaction between the Barding domain and WTAP is weak, with an interface area of~256 Å 2 in which highly conserved residues R1512 and F1521 of the first α-helix of the Barding domain form salt bridges and hydrophobic interactions with D206 and F207 in H4 of WTAP-a, respectively (Fig. 5b). Noteworthily, α2-4 of the Barding domain stack on ARML4 of the Belly domain mainly through hydrophobic interactions in a manner similar to the ARML module (Fig. 5c, d); therefore, α2-4 of ZC3H13 are called ARML-Z1. α5-6 of the Barding domain stack on ARML-Z1 in an ARM-like manner; thus, it is called ARML-Z2 (Fig. 5c). α7 of ZC3H13 interacts with the Neck domain and promotes the formation of a short helix in the Halter domain (Fig. 5e, f). After removing ZC3H13, the cryo-EM density of the short helix of the Halter domain is lost (Fig. 5f). α8, located in the C-terminus of ZC3H13, stacks on the ARML5-7 of the Chest domain mostly through conserved hydrophobic interactions (Fig. 5g). Overall, α1-6 and α8 of ZC3H13 interact with the Belly and Chest domains of VIRMA, while α6 and α7 interact with the Neck domain of VIRMA, thereby pushing the Neck domain away from the Belly and Chest domains (Fig. 5h-k).

MACOM and MAC assemble into the m 6 A writer complex mainly through WTAP and METTL3 interactions
We assembled the four components of MACOM (HWVZ) with MAC (the METTL3 and METTL14 heterodimer) and obtained a cryo-EM map at 4.4-Å resolution. However, due to its flexibility, the resolution of the MAC region is relatively low (Supplementary information, Fig. S7). At a low threshold (level = 0.013), there are extra densities near L2 and H4 of WTAP in the MACOM-MAC complex compared with the MACOM complex. At a high threshold (level = 0.1), the density near H4 of WTAP becomes clear (Fig. 6a, b; Supplementary information, Fig. S7f), which indicates that H4 of WTAP might play a vital role in MACOM-MAC interactions. Consistently, the GST pull-down results of different WTAP constructs show that H3 and H4 of WTAP are essential for MAC binding (Fig. 6c-f). Furthermore, we analyzed the interactions between MACOM and MAC using XL-MS. Data from bis (sulfosuccinimidyl) suberate (BS3)-based XL-MS indicate that METTL3 might interact with WTAP through its N-terminal LH and Zinc Finger 1-2 (ZF1-2) domains, while METTL14 interacts with all four components mainly through its N-terminal region (Supplementary information, Fig. S4c, d, Table S3 and Video S2). These data are consistent with the results of GST pull-down analysis, demonstrating that the N-terminal LH regions of METTL3 (1-69) and METTL14 (1-115) are essential for their interactions with WTAP ( Fig. 6g-j). The XL-MS data provide distance restraints and relative positions of MAC and MACOM. The crosslinked sites of METTL3 on WTAP are located at H3 (K155 and K160) and H4 (K192 and K230), while the crosslinked sites of METTL14 on WTAP are only located at H4 (K192) (Fig. 6k; Supplementary information, Fig. S4c, d and Table S3). Notably, K27 from the N-terminal LH of METTL3 is crosslinked with K230 from H4 of WTAP, suggesting that the extra density near H4 of WTAP may be LH of METTL3 (Fig. 6b, k).  Table S3). Taken together, we proposed a functional model of the m 6 A writer complex, including MACOM and MAC subunits, in which METTL3 and METTL14 bind H3 and H4 of WTAP and the Back and Belly domains of VIRMA through the LH domain of METTL3 and the N-terminal fragment of METTL14 (Fig. 6l). In such a way, MACOM can enhance the m 6 A writer activity (Fig. 1g).
A plausible model of the m 6 A writer complex for RNA substrate binding To understand the binding mode of RNA substrate by MACOM/MAC complex, the ACTB-1 RNA oligo is labeled by 4-thiouridine (s4U) and crosslinked with MACOM/MAC complex for mass spectrometry. Four identified s4U-crosslinking sites are located in the ZF2 of METTL3 (F321) and N-and C-terminus of VIRMA (P140, E1597, and F1706) ( Fig. 7a; Supplementary information, Fig. S8). The F321 in METTL3 plays a key role in direct interaction with RNA substrate, consistent with previously reported results. 16 The s4U-crosslinking of VIRMA is further verified by EMSA and methylation assays ( Fig. 7a; Supplementary information, Fig. S9a, b). MACOM containing C-terminally truncated VIRMA (1-1586) shows a significant decrease in RNA binding affinity and an almost complete loss of methylation activity (Supplementary information, Fig. S9a, b). Based on the protein-RNA crosslinking data and biochemical results, we propose a possible  path of the bound RNA in the m 6 A writer complex (Fig. 7b): MACOM binds the RNA substrate via the N-and C-terminus of VIRMA, enhancing the MAC binding affinity to the RNA substrate (Fig. 1d, h).

DISCUSSION
MACOM is a conserved subunit of the m 6 A writer RNA m 6 A modification is highly conserved in most eukaryotes. METTL3 and METTL14, the two MT-A70 family proteins in the catalytic core of the m 6 A writer, are also highly conserved from yeasts to humans. In addition to the catalytic core components of MAC, more components of MACOM have been identified in higher animals and plants, suggesting that more regulation is required for m 6 A modification in complicated biological systems. 29 Although the components of MACOM are not widely present, WTAP co-evolves with the METTL3 and METTL14 complex in most eukaryotic species. 29 VIRMA, ZC3H13, and HAKAI co-exist with WTAP in metazoans. 18 Interestingly, only VIRMA and HAKAI homologs are co-purified with WTAP in multicellular plants, while ZC3H13 has not been identified in plants by biochemical purification 21 or protein sequence BLAST (data not shown). Our structural study shows that ZC3H13 uses its C-terminal Barding domain to interact with VIRMA and induce the conformational change of MACOM (Fig. 5). It is possible that another protein different from ZC3H13 plays a similar role in multicellular plants.

MACOM is required for efficient m 6 A modification
Recent study reported that WTAP homologue Fl(2)d is present as homodimer through the region (125-536). 22 The WTAP dimer region (65-245) in our structure is corresponding to the Fl(2)d region (127-308) (Supplementary information, Fig. S6a), which is fitted well with previous biochemical results. Based on our structural study, WTAP and VIRMA constitute the MACOM core and form a warhorse-like shape, and ZC3H13 binds to VIRMA through its C-terminal Barding domain. The catalytic core component of MAC, METTL3, interacts with MACOM mainly through WTAP, as demonstrated by the XL-MS results showing that more than half of the crosslinking sites of METTL3 are related to WTAP. In vitro RNA substrate binding and m 6 A methyltransferase activity assays demonstrate that MACOM (HWVZ and WVZ) can directly bind to RNA substrates and significantly increase the m 6 A writer activity (Fig. 1). Specifically, removing ZC3H13 dramatically reduces substrate binding and methyltransferase activity (Fig. 1d, g), suggesting that ZC3H13 is a key component of MACOM for substrate binding and m 6 A writer activity. Furthermore, based on the protein-RNA crosslinking data and biochemical results, we propose a plausible model of the m 6 A writer complex for RNA substrate binding in which MACOM and MAC bind different regions of the RNA substrate (Fig. 7). Previous study demonstrated that VIRMA in MACOM mediates preferential m 6 A mRNA methylation in 3′ UTR and near stop codon through association with polyadenylation cleavage factors CPSF5 and CPSF6 in an RNA-dependent manner. 10 In addition, RBM15, another key component of MACOM, was shown to play an important role in binding the specific site of RNA substrate and recruit m 6 A writer complex. 7,9,30 Further studies of MACOM-MAC complex containing RBM15 would help to gain further insights into the mechanism of substrate selection of m 6 A writer complex.
Although HAKAI is absent in our cryo-EM structures, HAKAI is co-purified with the other three components in gel filtration (Fig. 1b, c) and contacts them using its pTyr-B domain based on the BS3-XL-MS data (Supplementary information, Fig. S4b, d and Table S3). Notably, the structure of HWVZ complex is fitted well in negative staining EM map of WVZ complex (Supplementary  information, Fig. S10). These results indicate that HAKAI is not missing but flexible in the HWVZ complex, which is consistent with the observation that the N-terminal SUN domain of VIRMA, a recently reported HAKAI-interacting region, is also absent in our cryo-EM structures. 22 Our RNA binding and enzymatic activity assays indicate that HAKAI is not crucial for the substrate binding and m 6 A modification in vitro (Fig. 1d, g), which might be caused by the lack of potential phosphorylation and ubiquitylation that may be important for HAKAI function. According to recent studies, HAKAI is essential for MACOM stability and maintenance of m 6 A levels in vivo. 22,23 Therefore, further studies are required to assess the role and molecular mechanism of HAKAI in MACOM.

Structures of the MACOM core provide a basis for the regulation of m 6 A modification in human diseases
It has been proven that m 6 A modification is involved in various biological processes, and deregulation of m 6 A modification leads to a variety of human diseases. The most conserved core component of MACOM, WTAP, was discovered due to its relationship with Wilms' tumor 1 (WT1) and was named as Wilms' tumor 1-associated protein. 31 WTAP has recently been shown to be upregulated in many tumors, such as acute myeloid leukemia. 32,33 Our cryo-EM structures, especially the highresolution structure of the MACOM core, provide a basis for the development of inhibitors or drugs targeting m 6 A modification to achieve potential therapeutic efficacy for human diseases such as cancers. 24

MATERIALS AND METHODS Cell lines and culture
Escherichia coli (E. coli) BL21 (DE3) (Novagen) and DH10Bac (Thermo Fisher Scientific) were grown in LB. Two kinds of insect cell lines, Spodoptera frugiperda (Sf9, Expression systems) cells and Trichoplusia ni (High Five, Expression systems) cells, were used for virus preparation and recombinant protein expression, respectively. Sf9 and High Five cells were grown at 27°C and 130 rpm in SIM SF expression medium and SIM HF expression medium (Sino Biological), respectively.
Cells were resuspended in Buffer A (150 mM KCl, 20 mM Tris-HCl, pH 8.0, 10% glycerol, 25 mM imidazole) with 0.5 mM PMSF and protease inhibitors, lysed by adding 0.5% Triton X-100 and shaken gently for 20 min at 4°C. After centrifugation at 18,000 rpm for 60 min, the supernatant was collected for further purification. The METTL3/METTL14 complex was Fig. 6 Cryo-EM and crosslinking mass spectrometry analyses of m 6 A writer complex suggest that MACOM interacts with MAC mainly through WTAP and METTL3 interaction. a Local resolution map of the HWVZ + M3/14 complex at high and low threshold levels. b Structure of the HWVZ complex in the HWVZ + M3/14 cryo-EM map at a high threshold level. The extra density besides the HWVZ complex is marked with a black dotted circle. c Schematic view of different WTAP N-and C-terminal truncation constructs. The MAC-binding affinities of these truncations are summarized from d-f and labeled on the right. d-f The N-and C-terminal truncation constructs of WTAP were used in GST pull-down assays to assess the interactions with GST-METTL3/METTL14 or mock control GST. g Schematic view of different METTL3 N-terminal truncation constructs. The WTAP-binding affinities of these truncation constructs are summarized from h and labeled on the right. h The Nand C-terminal truncation constructs of GST-METTL3 and mock control GST were used in GST pull-down assays to assess the interactions with WTAP (1-249). i Schematic view of different METTL14 N-and C-terminal truncation constructs. The WTAP-binding affinities of MAC formed by these truncation constructs and METTL3 (70-580) are summarized from j and labeled on the right. j MACs formed by N-and C-terminal truncation constructs of METTL14 and GST-METTL3 (70-580) were used in GST pull-down assays to assess the interactions with WTAP (1-249). purified by tandem affinity chromatography of Ni-NTA and GST. After removal of His-and GST-tags by TEV protease digestion at 4°C overnight, the protein solution was further purified by a Heparin HP column. The concentrated solution was loaded onto a Superdex 200 Increase 10/300 GL column (GE Healthcare cytiva) equilibrated in SEC buffer (100 mM KCl, 20 mM HEPES, pH 7.5, 10% glycerol, 1 mM DTT). Fractions of the METTL3/ METTL14 complex were collected and concentrated to 20 mg/mL. The purification procedures of METTL3 and METTL14 truncations were the same as those of the METTL3/METTL14 complex. The HWVZ complex was purified by tandem affinity chromatography of Ni-NTA and STREP beads. The concentrated solution was loaded onto a Superose 6 Increase 10/300 GL column equilibrated in SEC buffer. Fractions corresponding to the HWVZ complex were collected and concentrated to~10 mg/mL. The purification procedures of HAKAI/WTAP/VIRMA (HWV), WTAP/VIRMA/ ZC3H13 (WVZ), and WTAP/VIRMA (WV) complexes were the same as those of the HWVZ complex.
The cDNAs of truncated WTAP constructs were cloned into pET28a (with a 6× His-SUMO tag) for protein expression in E. coli BL21 (DE3). When E. coli were cultured at 37°C to an OD600 of 0.6, isopropyl β-D-1thiogalactopyranoside (IPTG) was added to a final concentration of 0.2 mM. Then E. coli were grown for~16 h at 18°C. The cells were collected by centrifugation, resuspended in buffer containing 20 mM Tris-HCl, pH 8.0, 100 mM KCl, 25 mM imidazole, and lysed with high-pressure homogenization. The truncated WTAP constructs were first purified by Ni-NTA affinity chromatography. Then protease ULP1 was added to remove the 6× His-SUMO-tag, and dialysis was applied to remove imidazole. Then the sample was subjected to the second Ni-NTA chromatography, and the flow-through was collected. The concentrated solution was loaded onto a Superdex 75 HiLoad 16/60 column (GE Healthcare) equilibrated in SEC buffer. Fractions corresponding to truncated WTAP constructs were collected and concentrated to~20 mg/mL.

Sample preparation for cryo-EM
To prepare HWVZ or HWV complex samples for cryo-EM study, the HWVZ, HWV or HWVZ + M3/M14 complex was crosslinked by 0.5 mM bis(sulfosuccinimidyl)suberate (BS3, Thermo Fisher) for 2 h at 25°C and quenched with 20 mM glycine. The complex was further treated by density gradient centrifugation. The gradient was generated from a light solution containing 10% (v/v) glycerol, 20 mM HEPES, pH 7.5, 100 mM KCl, 1 mM DTT, and a heavy solution containing 30% (v/v) glycerol, 20 mM HEPES, pH 7.5, 100 mM KCl, 1 mM DTT. Centrifugation was performed at 38,000 rpm in a Beckman SW41Ti swinging bucket rotor for 16 h at 4°C. The peak fractions were collected and further dialyzed with buffer containing 100 mM KCl, 20 mM HEPES, pH 7.5, and 1 mM DTT at 4°C overnight. The crosslinked HWVZ or HWV complex was concentrated to 1.0 mg/mL, and then applied to cryo-EM grids.

Preparation of RNA substrate
The RNA substrate ACTB-1 from the stop codon region of human ACTB mRNA containing two m 6 A sites and one RBM15 binding site 9 (5′-GGCCCCUCCAUCGUCCACCGCAAAUGCUUCUAGGCGGACUAUGACUU AGUUGCGUUACACCCUUUCUUGACAAAACCUAACUUGCACAGAAAACA-3′) was obtained by in vitro transcription using T7 RNA polymerase. A linearized recombinant pUC19 plasmid containing the target sequence with 5′-hammerhead ribozyme sequence and free 3′-end digested by EcoRI was used as the DNA template for in vitro transcription of ACTB-1 RNA substrate. In vitro transcription was performed at 37°C for 8 h in the reaction buffer containing 100 mM HEPES, pH 7.9, 10 mM MgCl 2 , 10 mM DTT, 6 mM NTP each, 2 mM Spermidine, 200 μg/mL linearized plasmid, and 100 μg/mL T7 RNA polymerase. The transcripts were purified by 8% denaturing urea PAGE, eluted from gel slices and precipitated with isopropanol. After centrifugation, the RNA precipitant was collected, washed twice with 70% ethanol and air-dried, and the RNA was dissolved in ultrapure water. FAM-labeled ACTB-1 RNA was produced by Silencer ® siRNA Labeling kit-FAMTM according to the product instructions.

EMSA
An aliquot of 5 nM FAM-labeled ACTB-1 RNA was mixed with increasing concentrations of protein from 0.25 to 256 times of RNA concentration with a total of 11 gradients in 10 μL buffer containing 20 mM HEPES, pH 7.5, 100 mM KCl, 10% glycerol, 1 mM DTT and incubated at 25°C for 20 min. Electrophoresis was performed with 8% native PAGE gels in running buffer containing 1× Tris-borate-EDTA (TBE) buffer at 4°C for 2.5 h. Gels were scanned using a Typhoon FLA-9000 (GE Healthcare). As the complexes of RNA and MACOM cannot run into gels and were held as black deposits in the gel-loading wells due to their biochemical properties, we quantified the RNA-binding ratios by measuring the reduction of free RNA. Free RNA was quantified using ImageJ. Binding curves were fit individually using GraphPad Prism 9.0 software fitting with "One site-Specific binding with Hill slope" (GraphPad Software). Curves were normalized as the percentage of bound oligonucleotides and reported as the means ± SD of the interpolated K d from three independent experiments.

BS3/EDC-mediated crosslinking mass spectrometry
The purified complexes were incubated with 0.5 mM BS3 (Thermo Fisher Scientific, 21580) in reaction buffer containing 50 mM HEPES, pH 7.5, 100 mM KCl, and 5% glycerol at 25°C for 2 h or 5 mM EDC (Thermo Fisher Scientific, PG82073) in reaction buffer containing 50 mM HEPES, pH 7.2, 100 mM KCl, and 5% glycerol at 25°C for 3 h. Crosslinked complexes were further purified to remove oligomers and glycerol by size exclusion chromatography. The proteins (10 μg) were precipitated and digested for 16 h at 37°C by trypsin at an enzyme-to-substrate ratio of 1:50 (w/w). The trypsin-digested peptides were desalted and loaded on an in-house packed capillary reverse-phase C18 column (40-cm length, 100-µM ID × 360-µM OD, 1.9-µM particle size, 120-Å pore diameter) connected to an Easy LC 1200 system. The samples were analyzed with a 120-min HPLC gradient from 6% to 35% of buffer B (buffer A: 0.1% formic acid in water; buffer B: 0.1% formic acid in 80% acetonitrile) at 300 nL/min. The eluted peptides were ionized and directly introduced into a Q-Exactive mass spectrometer using a nano-spray source. Survey full-scan MS spectra (300-1800 m/z) were acquired in the Orbitrap analyzer with resolution r = 70,000 at 400 m/z. Crosslinked peptides were identified and evaluated using pLink2 software. 34 The acquired MS/MS data were analyzed against a homemade database (including all the target proteins downloaded from UniProt) using pLink2. Cysteine alkylation by iodoacetamide was specified as a fixed modification with a mass shift of 57.02146 and methionine oxidation as a variable modification. Precursor mass tolerance and fragment mass tolerance were 20 p.p.m. BS3 (crosslinking sites K and protein N-terminus, xlink mass shift 138.0680796, mono-link mass shift 156.0786442) was defined as a crosslinker. Peptide length was set between 6 amino acids and 60 amino acids. Maximum missed cleavage sites were three. False discovery rate was set as 5%. We screened crosslinking sites that had at least 3 spectra with Evalues ≤ 0.01 for structural analysis.
GST pull-down assay GST pull-down assay was performed to detect protein-protein interactions using GST-tagged or untagged proteins purified from bacterial or insect cells. First, GST-tagged protein and untagged protein were mixed and incubated on ice for 30 min. Then the protein mixture was incubated with 15 μL GST beads in a total volume of 500 μL in binding buffer (100 mM KCl, 20 mM HEPES, pH 7.5, 10% glycerol, 1 mM DTT) at 4°C for 2 h with gentle rotation. After centrifugation at 300× g for 3 min, the supernatant was removed, and the beads were washed three times using wash buffer (100 mM KCl, 20 mM HEPES, pH 7.5, 10% glycerol, 1 mM DTT, 0.1% NP-40). Another three washes were performed using binding buffer, followed by SDS-PAGE analysis.

In vitro m 6 A methyltransferase activity assay
In vitro methyltransferase activity assay was performed as previously published 12 with several modifications. Briefly, the 50 μL reaction mixture contained 1 μM RNA substrate, 10 nM MAC, 15 nM MACOM, 30 μM d3-SAM, 100 mM KCl, 50 μM ZnCl 2 , 0.01% Triton X-100, 20 mM Tris, pH 7.5, 20 μg/μL BSA, 5 mM DTT, and 0.2 U/μL RRI (Recombinant RNase Inhibitor, Takara). The reaction was carried out by incubating at 37°C for 1 h and quenched by inactivating the enzyme at 95°C for 5 min. RNA was digested at 42°C for 2 h with 20 mM CH 3 COONH 4 and 1 U nuclease P1 (Wako). Afterwards, 1 μL BAP (Bacterial Alkaline Phosphatase, TOYOBO) and 6.4 μL 10× BAP Buffer were added, and the reaction was incubated at 37°C for 2 h. The sample was diluted to 120 µL and then filtered through a 0.22-μm filter (Millipore). A 5-μL aliquot of the sample was injected into the LC/MS/ MS system. The nucleosides were separated by reverse phase ultraperformance liquid chromatography on a C18 column with online MS detection using an Agilent 6460 QQQ triple-quadrupole LC mass spectrometer in positive electrospray ionization mode. The nucleosides were quantified by using nucleoside-to-base ion mass transitions of 282 to 150 (d3-m 6 A) and 268 to 136 (A). Quantification was performed in comparison with the standard curve obtained from pure nucleotide standards running on the same batch of samples. The ratios of d3-m 6 A to A and d3-m 6 A to probe were calculated on the basis of the calibrated concentrations.
UV crosslinking mass spectrometry of s4U-ACTB-1 and m 6 A writer complex The purified m 6 A writer complex was incubated with s4U-ACTB-1 for 30 min on ice to form the stable complex. Then the sample was UVcrosslinked on ice with 3 J/cm 2 at 365 nm wavelength with a UVPcrosslinker (analytik jena). Subsequently, the sample was treated with RNase I/RNase A/RNase T1 mixture at 1 U/μL concentration for 1 h at 37°C. Proteins were precipitated with acetone. The protein pellet was dried by using a SpeedVac for 1−2 min. The pellet was subsequently dissolved in 8 M urea, 100 mM Tris-HCl, pH 8.5. TCEP (final concentration is 5 mM) (Thermo Scientific) and Iodoacetamide (final concentration is 10 mM) (Sigma) for reduction and alkylation were added to the solution and incubated at room temperature for 30 min, respectively. The protein mixture was diluted four times and digested overnight with trypsin at 1:50 (w/w) (Promega). The digested peptide solutions were desalted using a MonoSpinTM C18 column (GL Science, Tokyo, Japan) and dried with a SpeedVac. The peptide mixture was analyzed by a homemade 30 cm-long pulled-tip analytical column (75 μm ID packed with ReproSil-Pur C18-AQ 1.9 μm resin, Dr. Maisch GmbH, Germany), the column was then placed inline with an Easy-nLC 1200 nano HPLC (Thermo Fisher Scientific) for mass spectrometry analysis. The analytical column temperature was set at 55°C during the experiments. The mobile phase and elution gradient used for peptide separation were as follows: 0.1% formic acid in water as buffer A and 0.1% formic acid in 80% acetonitrile as buffer B, 0-1 min, 5%-8% B; 1-114 min, 8%-35% B; 114-115 min, 35%-50% B; 115-116 min, 50%-100% B; 116-120 min, 100% B. The flow rate was set as 300 nL/min. Data-dependent tandem mass spectrometry (MS/MS) analysis was performed with an Orbitrap Eclipse mass spectrometer (Thermo Fisher Scientific). Peptides eluted from the LC column were directly electrosprayed into the mass spectrometer with the application of a distal 2-kV spray voltage. The cycle time was set to 3 s. Full scan resolution was set to 120,000 and MS/MS scan resolution was set to 30,000 with isolation window of 1.6 m/z. The dynamic exclusion settings used were as follows: charge exclusion, 1, and > 7; exclude isotopes, on; and exclusion duration, 30 s. MS scan functions and LC solvent gradients were controlled by the Xcalibur data system (Thermo Fisher Scientific).
The acquired MS/MS data were analyzed by RNPxl 35 using 10 ppm for MS1. The s4U-crosslinking peptides (mass adduct is same as U-H 2 O). Raw data were searched against a FASTA database consisting of the protein sequences of the proteins in the complexes using RNPxl node combined in PD software with 0.01 FDR. The result is filtered by score (> 5) and the MS2 spectrum. MS2 spectra are annotated in TOPPView. 36

Cryo-EM data acquisition
The samples of HWVZ, HWV, and HWVZ + M3/14 complexes were diluted at a final concentration of~1 mg/mL. Three microliters of the samples were applied onto glow-discharged 200-mesh R2/1 Quantifoil copper grids. The grids were blotted for 4 s and rapidly cryocooled in liquid ethane using a Vitrobot Mark IV (Thermo Fisher Scientific) at 4°C and 100% humidity. The samples were imaged in a Titan Krios cryo-electron microscope (Thermo Fisher Scientific) operated at 300 kV with a GIF energy filter (Gatan) at a magnification of 105,000× (corresponding to a calibrated sampling of 0.82 Åper pixel) for three samples. Micrographs were recorded by EPU software (Thermo Fisher Scientific) with a Gatan K3 Summit direct electron detector, where each image was composed of 30 individual frames with an exposure time of 2.5 s and an exposure rate of 22.2 electrons/s/Å 2 . A total of 12,745 movie stacks for HWVZ, 7642 movie stacks for HWV, and 14,168 movie stacks for HWVZ + M3/14 were collected.
Single-particle image processing and 3D reconstruction All micrographs were first imported into Relion 37 for image processing. The motion-correction was performed using MotionCor2 38 and the contrast transfer function (CTF) was determined using CTFFIND4. 39 Then the micrographs with "rlnCtfMaxResolution < 5" were selected using the "subset selection" option in Relion. All particles were autopicked using the NeuralNet option in EMAN2. 40 Then, particle coordinates were imported to Relion, where the poor 2D class averages were removed by several rounds of 2D classification. The initial maps were built and classified using the ab initio 3D reconstruction option in cryoSPARC 41 without any symmetry applied. The 3D homogeneous refinements, local and global CTF refinements, and non-uniform refinements were performed using the selected 395,916 particles for the HWVZ complex, 282,821 particles for the HWV complex, and 199,741 particles for the HWVZ + M3/ 14 complex, resulting in~3.0-Å,~3.0-Å, and~4.4-Å resolution maps, respectively. The resolution for the final maps was estimated with the 0.143 criterion of the Fourier shell correlation curve. Resolution map was calculated in cryoSPARC using the "Local Resolution Estimation" option. The figures were prepared using UCSF Chimera 42 or UCSF ChimeraX 43 (see more information in Supplementary information, Figs. S1-S3, S7 and Table S1).

Model building
Model building was first performed based on the 3.0-Å resolution cryo-EM map of the HWVZ complex. As none of these molecules has structural information, de novo model building was conducted. Phenix.map_to_model 9 was first used to generate the initial model. Coot 44 was then used to confirm the amino acid sequence registration of the initial model and assign amino acids to the cryo-EM density regions that were not resolved by phenix.map_to_model. Notably, the assignment of amino acid sequence by Coot was based on bulky residues (Trp, Lys, Arg, Phe, and Tyr). An atomic model composed of WTAP (residues 64-247), VIRMA (residues 334-1585), and ZC3H13 (residues 1492-1643) was obtained. The resulting model was refined using phenix.real_space_refine. 45 To build the atomic model for the HWV complex, the final model of the HWVZ complex without the ZC3H13 subunit was fitted into the 3.0-Å resolution cryo-EM map of the HWV complex, followed by the optimization using phenix.real_space_refine and Coot. The model quality was evaluated by MolProbity 25 and Q-scores. 27 Statistics of the map reconstruction and model optimization are summarized in Supplementary information, Table S1. PDBsum structure bioinformatics 28 was used to identify the key residues involved in interactions between subunits in our structures. All figures were made using UCSF Chimera or UCSF Chimera X.

Quantification and statistical analysis
For the quantification of the binding affinities of the HWVZ, WVZ, HWV, and WV complexes with FAM-labeled ACTB-1 RNA (Fig. 1d), measurements were carried out using ImageJ. GraphPad Prism was used to perform the statistical analysis of measuring results. Each data point represents the average of three independent experiments. Error bars represent SD. For the quantification of ACTB-1 RNA N 6 -adenosine methylation activity (Fig. 1g), GraphPad Prism was used to perform the statistical analysis of measuring results. Each data point represents the average of two independent experiments. Error bars represent SD.

Negative staining analysis
The purified crosslinked WVZ complex was applied to Superose 6 Increase 10/300 GL column (GE Healthcare) equilibrated with the gel filtration buffer. The complex was diluted to 35 ng/μL with SEC buffer. Five microliters of sample were applied to the glow-discharged 200 mesh carbon-coated copper grids. The samples were stained using 0.75% uranyl formate and air-dried. Data were collected on a Talos L120C transmission electron microscope equipped with a 4K × 4K CETA CCD camera (Thermo Fisher Scientific). Images were recorded at a nominal magnification of 92,000×, corresponding to a pixel size of 1.55 Å. CTF parameters of each micrograph were determined using CTFFIND4. 39 Particles were picked and subjected to two rounds of 2D classification, followed by de novo 3D model generation, 3D classification, and auto-refinement in Relion. 37

DATA AVAILABILITY
Cryo-EM maps of the HWVZ, HWV, and HWVZ + M3/14 complexes in this study have been deposited in the wwPDB OneDep System under EMD accession codes EMD-31946, EMD-31947, and EMD-34169, respectively. The associated atomic models of HWVZ and HWV complexes are under PDB codes 7VF2 and 7VF5, respectively. The crosslinking mass spectrometry data have been deposited in ProteomeXchange with the primary accession code IPX0003503000.