key: cord-0897254-63tnjy0k
authors: Matsuda, Alex; Plewka, Jacek; Chykunova, Yuliya; Jones, Alisha N.; Pachota, Magdalena; Rawski, Michał; Mourão, André; Karim, Abdulkarim; Kresik, Leanid; Lis, Kinga; Minia, Igor; Hartman, Kinga; Sonani, Ravi; Schlauderer, Florian; Dubin, Grzegorz; Sattler, Michael; Suder, Piotr; Popowicz, Grzegorz; Pyrć, Krzysztof; Czarna, Anna
title: Despite the odds: formation of the SARS-CoV-2 methylation complex
date: 2022-01-26
journal: bioRxiv
DOI: 10.1101/2022.01.25.477673
sha: c57a332d879b7ff0d3e392f76e46d280f595d05d
doc_id: 897254
cord_uid: 63tnjy0k

Coronaviruses protect their single-stranded RNA genome with the methylated cap added during the replication. This capping process is carried out by several nonstructural proteins (nsp) encoded in the viral genome. The methylation itself is performed consecutively by two methyltransferases, nsp14 and nsp16, which interact with nsp10 protein acting as a co-factor. The nsp14 protein also carries the exonuclease domain, which also serves as a part of the proofreading system during the replication of the large RNA genome. The available crystal structures suggest that the concomitant interaction between these three proteins is impossible due to the structural clash, and it is generally accepted that the nsp16 and nsp14 bind with the nsp10 separately. Here, we show that nsp14, nsp10, and nsp16 form a methylation complex despite the odds. Due to spatial proximity, this interaction is beneficial for forming mature capped viral mRNA. Further, it modulates the exonuclease activity of nsp14, protecting the viral RNA at the replication site. Our findings show that nsp14 is more amenable to allosteric regulation and may serve as a molecular target for the therapy.

The coronaviral genome is of positive polarity, and it serves as a substrate for the translational machinery of the cell after its release to the cytoplasm. The first and only product of the genomic mRNA translation is the large and non-functional 1a/1ab polyprotein. It maturates by autoproteolytic processing carried out by two viral proteases -M pro and PL pro .

This leads to the generation of a set of nonstructural proteins (nsp ), which are numbered consecutively and responsible for the whole viral replication process and remodelling of the intracellular environment. Once nsp's reshape the cell to form a viral factory, the genomic RNA is being copied, and a set of subgenomic (sg) mRNAs is produced in a peculiar discontinuous transcription process. These sg mRNAs are monocistronic and serve as templates for the production of structural and accessory proteins required for the formation, assembly and release of progeny viruses 1 . The activity of particular nsp's has been previously described, showing the complex network of interactions and presence of multi-functional nsp's. However, their coordinated action has not been fully understood.

Works by Gao et al. 2 , Yan et al. 3 , Wang et al. 4 and Kabinger et al. 5 shed light on the scaffold of the replicatory complex formed by nsp12 (polymerase) and two co-factors nsp7 and nsp8, that create functional machinery able to replicate the viral RNA. Next, an extended elongation complex was described, where the nsp12/7/8 is accompanied by nsp13 helicase. This complex is suggested to serve as the basic replicatory module 6 .

Coronaviruses are known for their large genomes requiring high replication fidelity to maintain their integrity. While the SARS-CoV-2 nsp12 polymerase is highly processive, it is error-prone and does not provide sufficient fidelity. Some time ago, it was shown that a proofreading system is encoded in coronaviral genomes. The nsp14 protein carries an Nterminal exonuclease (ExoN) domain that has been attributed to serve as a part of this system. The ExoN is a member of the DEDDh exonuclease superfamily, which possesses a 3'-5' exonuclease activity and is responsible for removing incorrectly incorporated nucleotides from the 3' terminus of the newly formed RNA. It has also been proposed to play a role during the discontinuous replication of coronaviruses. The nsp14 has b een shown to associate with the replicatory complex, with the nsp10 as a co-factor that modulates and enhances the nsp14 exoribonuclease activity [7] [8] [9] .

The nsp14, however, has yet another function. It takes part in the cap formation after the genome copying is finalized 7 . The capping of the viral mRNAs is essential for their function and integrity and enables translation initiation. Additionally, it protects the mRNA from recognition as foreign by cellular sensors and consequently induction of innate immune responses 10 . Cap formation is a tightly regulated process consisting of four independent enzymatic reactions. First, the nsp13 triphosphatase removes the γ-phosphate of the 5′triphosphate end (pppA) 11, 12 ; next, nsp12 guanylyltransferase (the nidovirus RdRp-associated nucleotidyltransferase (NiRAN) domain) 3 transfers the GMP to form the core structure of the cap (GpppA); the GpppA is methylated at the N7 position by the nsp14 N7-methyltransferase domain ( 7Me GpppA); subsequently, the ribose in the first ribonucleotide is methylated at the 2'-O-position by the nsp16 2'-O-methyltransferase [13] [14] [15] . This leads to the formation of a functional cap, and the genome replication is concluded ( 7Me GpppA 2'OMe ). Interestingly, this process is regulated by the nsp9 protein, which binds the nsp12 near the NiRAN active site 16 . While there is a good basis for understanding the part of the capping process performed by interacting nsp12, nsp13, and nsp14 proteins, the nsp16 remains an orphan.

Here, we studied the interaction between the two methyltransferases -nsp16 and nsp14.

While previous in silico results suggested that the interaction of nsp14 and nsp16 is impossible due to the steric hindrance on the nsp10 surface, the experimental data proves otherwise. Our data indicate the formation of a functional heterotriplex by nsp10, nsp16 and nsp14. The interaction occurs via the nsp10 protein, where the binding interfaces overlap to some extent.

This should render the triplex formation impossible unless structural rearrangement of the nsp14 N-terminal region occurs. Interestingly, in contrast to the remaining part of nsp14, this region is characterized by the absence of secondary structure. Additionally, the nsp10/14(Nterminal) interactions are devoid of "anchors" typical for protein -protein interactions (PPI).

There are no deeply buried lipophilic residues, solvent-protected hydrogen bonds and precise shape complementarity present. These anchors appear only in the other part of nsp14. There, the N-terminal α1′ helix of nsp10 forms an interaction with the deep, lipophilic cleft of nsp14 a very typical PPI interface (Figure 1) . Therefore, we hypothesized that the triplex formation is possible from the structural point of view, following a rearrangement of the first 50 Nterminal residues of nsp14. and nsp14 are mainly located within the exonuclease domain, where the α1 helix of nsp10 provides several deeply buried lipophilic and π-stacking interactions. This region is characterized by well-developed shape complementarity. The "lid" region of nsp14, composed of residues 1-50, covers the nsp16 binding site. This region is poorly structured and has poor shape complementarity with the nsp10. Therefore, it is freezing in nsp10-bound form carries a high entropy cost. Only several partially solvent-exposed hydrogen bonds can be identified there. (B) In contrast, nsp16 interacts in a similar region as the "lid" of nsp14 but with a well-developed interface involving deep lipophilic pockets and hydrogen bonds shielded from the solvent. The α1 helix of nsp10 is not visible in crystal structures of the nsp10/16 complexes (6W4H, 6YZ1), indicating its flexibility. (C) The formation of the ternary nsp10/14/16 complex must be accompanied by a "lid" opening. This can be plausible due to poor interface features and a lack of secondary structure that provides flexibility.

While the need for complex formation is understandable due to the complementary function of the two methyltransferases, we also observed that the nsp16 interaction with the nsp10/14 modulates the exoribonuclease activity. Consequently, we propose that the formation of the complex also regulates the nsp14 activity and limits the non-specific viral RNA degradation observed for the nsp10/14 complex. It is also to be elucidated to what extent such activity may tweak the degradation of cellular mRNA.

High-resolution crystal structures are available for the nsp14 and nsp16 with their common co-factor nsp10. We investigated the data to determine which residues of nsp10 contribute to the binding to each partner. From all the deposited structures of the nsp10/16 complex, we chose one with an almost intact N-terminal α1′ helix of nsp10 (PDB ID: 6WVN) and the nsp10/14 exoribonuclease domain complex (PDB ID: 7DIY). These structures were aligned in PyMol. The nsp10 structure in both complexes is almost identical with RMSD of 0.5, with the α1′ helix flipping as indicated by the arrow in Figure 2A .

The overlay of both complexes depicts that nsp16 (cyan) and the first 50 amino acids of nsp14 (green) occupy the same positions. Moreover, both proteins exhibit a high degree of overlap on the nsp10 interfaces, with nsp14 having almost twice as large interface surface area.

The nsp14 interface with nsp10 also involves the N-terminal α1′ helix of nsp10 structurally positioning it at ~130° angle compared to nsp16, which does not interact with it. However, the N-terminal region of nsp14 is flexible and poorly structured.

Interacting residues on nsp14 are mainly present in unstructured loop regions such as the Nterminal coil-coiled region that interacts with nsp10 α1' helix, β-turn 1, and loops between βsheet 2/3 and β-sheet 7/8. Many of those interactions are present within the region that would overlap with the nsp16 if those complexes are aligned (red region in Figure 2A and 2C) . The nsp10/16 interface, on the other hand, is based on a stronger hydrophobic interaction network between, among all, much more structurally rigid central antiparallel β1-sheet of nsp10 as well as helices α2, α3, α4 and a coiled-coil region connecting helix α1 and the sheet β1 and structured α helices 3,4 and 10 as well as β-sheet 4 ( Figure 2B ). Especially, Val42 and Leu45 of nsp10 seem to be essential for the interactions as they are immersed into hydrophobic pockets formed by helices α3, α4, and α10 of nsp16. We, therefore, hypothesize that within the nsp10/nsp14/nsp16 triplex, nsp16 displaces the N-terminal "lid" of nsp14 on the interface of nsp10 as depicted in Figure 1 . nsp14 Exon domain is shown in green and the associated nsp10 in purple. The structural clash between the nsp14 ExoN and nsp10/16 surfaces is shown in red. (B) The interface between nsp10 and nsp16. The strong hydrophobic interaction between the nsp10 and nsp16 is shown in blue. The nsp10 residues V42 and L45 are shown as sticks. (C) The interface between nsp10 and nsp14 ExoN. The hydrogen bonds between nsp10 and nsp14 are shown in green. The difference in orientation (~130°) of both nsp10 α1'-helices is evident when interacting with nsp16 or with nsp14. Important structural features involved in the interface are pointed with arrows.

Nsp10, nsp14, and nsp16 proteins were co-expressed in E. coli and purified. Size Exclusion Chromatography (SEC) was carried out, and the interaction of proteins was evaluated with no prior crosslinking or denaturation. As clearly visible in Figure 3A , all three proteins co-migrate, suggesting strong interaction and heterotriplex formation. To further ensure the content of the complex, it has been analyzed on SDS-PAGE gel, and protein identity was verified using mass spectrometry ( Figure 3B) . Obtained results suggested that the in silicopredicted complex is being formed.

The nsp10/14/16 complex formation was also evaluated by Native PAGE in mild conditions at 4°C, to support complex integrity (Supplementary Figure S1A) . The bands were cut out of the gel, denatured and analyzed by SDS-PAGE (Supplementary Figure S1B) . Additionally, mass spectrometry analysis of the bands excised from the native gel was carried out to verify the identity of proteins (Supplementary Table S1 ; highlighted in green). Obtained results confirmed the co-migration of all three proteins. Mass spectrometry was also employed to assess the stoichiometry of the complex. By analyzing the signals at 254 and 280 nm, it was established to be 1.2:1:1 (nsp10 to nsp14 to nsp16) (Supplementary Figure S2 and Supplementary Table S2 ). One may speculate that the used methodology may lead to biased results, as it is impossible to differentiate between the nsp14 and nsp16+nsp10 on a native gel.

To verify this hypothesis, nsp14 ExoN domain was co-expressed with nsp110/16. Obtained results show that the co-migration of all three proteins is maintained. The samples were treated with a nuclease to further confirm that the interaction between the proteins is not bridged by co-purified RNA. Again, the co-migration of proteins was not affected (data not shown).

To assess the binding between the proteins in the heterotriplex, MicroScale Thermophoresis (MST) was used. This technique relies on detecting the temperature-related intensity change (TRIC) of the fluorescently labelled target as a function of the concentration of a nonfluorescent ligand. Nsp14 or nsp16 were expressed with a His-tag and labelled with a fluorophore dye specific to the His-tag. As presented in Figure 3C , there is an increase in the fraction bound signal upon increasing the concentration of nsp10/16 complex on the nsp14 template that follows Kd fit, what represents the formation of the nsp10/14/16 complex. The binding affinity between nsp14 and nsp10/16 is similar to the one observed between nsp10 and nsp16 (0.28 ± 0.01 and 0.24 ± 0.00 µM, respectively). The binding between nsp10 and nsp14 is approximately an order of magnitude weaker (2.4 ± 0.2 µM). Nsp14 with 50 first N-terminal amino acids ("lid") removed has a similar kD towards nsp10 of 1.5 µM. This shows that "lid" region of nsp14 does not contribute to the binding as predicted from structural analysis. This further confirms the possibility of the triplex formation upon lid rearrangement. nsp16 did not interact with nsp14 alone at tested concentrations, excluding the possibility of nsp14/16 duplex formation.

To further corroborate the interaction between the three proteins, the thermal stability of the complex was investigated using nanoDSF. In this method, the intrinsic fluorescence of aromatic residues is monitored while increasing the temperature to determine the extent of secondary and tertiary structures preserved within the protein at a given temperature. The purity of all analyzed samples was verified with SDS-PAGE prior to the experiment. As depicted in Figure 3D , macromolecules are defined by a single peak indicating that the complex partners and domains are connected or have similar melting temperatures. From individually tested proteins, nsp16 is the least stable losing half of its secondary/tertiary structure at 45°C, nsp10 being most stable with a melting temperature of 50°C. Nsp10 and nsp14 form a stable complex with a single peak at 55°C, exhibiting a significant thermal stabilization over nsp14 alone, while nsp10/16 complex retained the melting temperature of 45.5°C. Upon the triplex formation, a single sharp peak at 51.7°C is formed, suggesting that all components of the triplex are structurally related to each other and confirming the nsp10/14/16 complex formation, as there are no other peaks present corresponding to the components of the triplex. The binding curve for nsp14 His-tag with nsp10/16 complex (blue), nsp14/His-tag with nsp10 (red), nsp16 His-tag with nsp10 (green), and nsp16 His-tag with nsp14 (gray). Experimental data presented in dots with error bars, Kd fits as solid lines. (D) The melting profiles for nsp10 (magenta), nsp14 (yellow), nsp16 (grey), nsp10/14 (red), nsp10/16 (blue), nsp10/14/16 (green) determined using nanoDSF.

Eukaryotic viruses modify their RNA to mimic the endogenous nucleic acids to avoid recognition by the innate immune systems. GpppA is first methylated at the N7 position by the nsp14 N7-methyltransferase domain yielding N7MeGpppA 17 .

In the assay, N-7 methyltransferase activity is assessed indirectly via quantifiable formation of the reaction product SAH using dedicated monoclonal antibodies ( Figure 4A ). The negative controls comprising no substrate (either SAM or RNA) or no protein all yielded maximal signal indicating no production of SAH from SAM. In the assay, nsp14 exhibits no preference for the nascent nucleotide methylating both GpppG and GpppA sub strates to the extent similar to nsp14/10 complex and triplex. Moreover, the activity is hindered upon the addition of panmethyltransferase inhibitor sinefungin, proving the activity of the enzyme ( Figure 4B ) 18 .

Nsp14 also harbours exoribonuclease activity, essential for proofreading during virus replication. However, in vitro, the protein acts as a highly processive ribonuclease, unspecifically degrading nucleic acids 19 . As the nuclease activity must be tightly controlled, we assessed whether the complex formation regulates this process. We evaluated the binding and nuclease activity of nsp10/14 compared to the nsp10/16 and the nsp10/14/16 complex.

Interestingly, we observed that the nuclease activity of nsp10/14 is reduced for the triplex, suggesting that the nsp14 activity is modulated nsp16, most likely by altering the RNA interaction with the nsp10 co-factor. Obtained data suggest that as the concentration of nsp16 becomes equimolar with the nsp10 and nsp14, the nuclease activity of nsp14 is abrogated ( Figure 4C ).

Both nsp14 and nsp16 rely on nsp10 as an activating co-factor, the degradation activity of nsp14 increasing 35-fold with the presence of nsp10 [19] [20] [21] . GpppG, GpppA and protein or their respective complexes along with negative controls (no SAM, no protein, no RNA) and positive control: sinefungin. The results were normalized using the SAH calibration curve and denoted µM of methylated product. All experiments were performed in triplicate, and average values with error bars (SD) are shown. (C, left) RNA is strongly degraded in vitro by nsp10/14 complex as seen in both native and denaturing RNA gels. This degradation is not seen for nsp16 as it is not a nuclease. Purified triplex is also not degrading the RNA. (right) Titration of nsp10/14 with increased amount of nsp16 causes reduction of exonuclease activity visible on a denaturing RNA gel. The concentration of nsp16 in each sample is indicated in blue. CoV-RNA1-A was used as a substrate.

Despite being a low-resolution technique, Small Angle X-ray Scattering (SAXS) is a valuable tool for investigating biomacromolecules under native conditions. Due to the nature of the measurements, SAXS averages over all conformational ensembles for proteins or protein complexes, accounting for the flexibility of the studied system. Moreover, it allows for a combined Size Exclusion Chromatography-SAXS (SEC-SAXS) measurements to increase understanding of the protein oligomerization or complex formation.

The measurements of nsp14, nsp10/14, nsp10/16 and nsp 10/14/16 complexes were performed with a BM29 beamline at the European Synchrotron Radiation Facility (ESRF) 14 in France, in an SEC-SAXS mode 15 . The retention times of tested biomolecules correlate with their molecular weights resulting from their chemical composition and assuming a 1:1(:1) molar ratio 16 The molecular envelopes that account for a low-resolution structure of tested systems in buffer were calculated using ten rounds of DAMMIF software 22 followed by models clustering and averaging and a final round of DAMMIN modelling. The resulting envelopes (Figure 5) present the experimental scattering profiles well. A high-resolution structure of nsp10/16, nsp14, nsp10/14, and nsp10/14/16 were fitted in generated envelopes using the SUPCOMB software. Despite an excellent fit to experimental data, the high-resolution models of nsp14 and nsp10/14 (Figure 5A and 5B , respectively) present a poor fit to molecular envelopes suggesting high flexibility of nsp14 alone or in complex with nsp10 in solution or multiple ensembles. The high-resolution structure of nsp10/16 fills the generated envelope tightly, similarly to the generated nsp10/14/16 model. The nsp10/14/16 model was generated by combining the high-resolution structure of nsp10 and nsp16 with nsp14 in PyMol, which was later subjected as a template for the normal mode analysis using the SREFLEX software 23 . Another classical method to investigate complex formation is negative-stained transmission electron microscopy (NS-TEM). We first monitored the full-length triplex. However, the generated particles yielded non-homogenous classes (Supplement Figure X) . Simultaneously, we tried visualizing the triplex with the truncated methyltransferase domain of nsp14. Here, particles were more homogenous. Obtained 2D classes from negative-stained transmission electron microscopy micrographs show the elongated shape particle of ~10 × 5 nm dimensions from three different viewing directions (Figure 6A) . The particle possesses a visible inter-band in its centre, indicating the presence of three proteins in the complex. We reconstituted the 3D structure of 20 Å using these class particles. The 3D reconstitution map shows an overall envelope of ~ 10 × 5 × 4.5 nm dimensions with a consensus elongated shape with 2D classes. 

It has been generally acknowledged that nsp16 and nsp14 bind to nsp10 separately. This was a logical and rational conclusion, based on the crystal structure data and the in silico modelling. In the available structure of nsp14/nsp10 and nsp16/nsp10 complexes, the overlap between nsp16 and nsp14 structures on the nsp10 interface was obvious. However, the biochemical and biophysical data we present here prove otherwise. It supports the hypothesis depicted in Figure 1 , presenting nsp10/14/16 ternary complex formation. We re-analyzed the existing data and re-formulated the model. The apparent structural clash between nsp14 and nsp16 on the nsp10 surface can be resolved by rearranging the flexible N-terminal "lid" of the nsp14. This fragment lacks a well-defined secondary structure and has only a few energetically favourable interactions with nsp10. Therefore, the rearrangement of the N-terminal end to prevent structural hindrances with nsp16 upon the triplex formation is plausible. The remaining parts of the nsp14 have a deep cleft accommodating an nsp10 helix with lipophilic anchoring residues typical for strong protein-protein interactions. Similarly, the nsp16 interface shows significant, well-defined binding elements interacting largely via hydrophobic bonds.

Interestingly, despite having a similar surface of the interfaces on nsp10, the affin ity of nsp14 to nsp10 was ten times lower than observed for nsp16 and nsp10, suggesting that some of the reported interactions in the crystal structure of nsp10/14 ExoN may have been artefacts resulting from the crystal formation. The expendability of the first 50 N-terminal amino acids of nsp14 for the interaction with the nsp10 was further confirmed using the deletion mutants.

We show that the affinity of truncated nsp14 without the first 50 N-terminal residues to nsp10 is similar to the full length nsp14/10 complex.

Our hypothesis was further validated using the SEC-SAXS experiment. We probed heterocomplexes of nsp10, nsp14 and nsp16 directly after eluting from the SEC column in native-like conditions. Not only do we see a correct trend in elution volumes from SEC data following the molecular weight of respective complexes, but we can also derive a similar trend from the SAXS scattering profiles calculated ab initio, without any model or assumption. These Figures 3 and 4) , We speculate that the spatial proximity of both methyltransferases involved in the viral RNA replication and the ExoN domain linked to the proofreading ability of the SARS-CoV-2 genome is well-justified, as the N-7 methylated mRNA product of nsp14 methyltransferase is a substrate for nsp16 O-2 methyltransferase, and the triplex formation does not abrogate the methyltransferase activity of the proteins. Further, the 3'➜5' exoribonuclease activity of nsp10/14 complex is evident and is diminished during the complex formation, most likely due to amended interaction between these two proteins. The modulation of the nsp14 nuclease activity by the nsp16 is beneficial to the virus, as it allows the nsp14 to switch from the proofreading mode to the methylation mode, preventing degradation of the viral nucle ic acids.

Constructs of nsp10 comprising amino acids 4254 -4392, nsp14 comprising amino acids 5926 -6452, and nsp16 comprising amino acids 6799 -7096 of SARS-CoV-2 polyprotein 1ab optimized for expression in E. coli were ordered from GeneArt and subcloned into expression vector pETDuet-1. For the nsp14 catalytic mutant, D90 and E92 were replaced with alanines 24 .

Plasmids were then co-transformed into E. coli strain Bl21. 

Samples were prepared, measured and analyzed as described in Pabis et al. 25 fragment mass tolerance: 20 mmu. Additionally, the SwissProt database restricted to E. coli taxonomy was searched to assess contamination with host proteins.

For protein quantitation, sample separation was carried out following a simple protocol using To confirm protein content under every chromatographic peak taken for protein quantitation, mass spectrometry-based identification was used. Our protocols for protein identification, applied with minor changes, are available elsewhere 26 

NanoDSF was performed in standard capillaries using Tycho equipment. Protein and their respective complexes were measured at 1 mg/mL using default ramp temperature. The resulting meting temperatures were reported as the first derivative of the fluorescence ratio.

The methyltransferase activity of wild type triplex complex nsp10/nsp14/nsp16 or triplex complex with mutated nsp14 protein (ExoN mutant and catalytic mutant -please add the info regarding the mutations) was measured using the EPIgeneous Methyltransferase kit from Cisbio as previously described 27 ratio of 665 to 620 nm wavelength was calculated. The data was background corrected on the averaged signal for the buffer control. Next, the data was normalized for each series individually on the wells not containing the enzyme.

The CoV-RNA1-A RNA was ordered from IDT as a PAGE-purified and desalted oligo. Protein complexes were incubated with 100 ng of RNA in a buffer containing 20 mM HEPES pH 7.5, 100 mM NaCl, 5% v/v glycerol, 10 mM MgCl 2 , and 5 mM β-mercaptoethanol, followed by analysis via a 20% urea-denaturing or a native 1% TBE agarose gel.

Samples were measured in SEC-SAXS mode at BM29/ESRF, Grenoble France on 12.12.2020.

(session ID MX2341). Samples (100 µL) were measured on Agilent AdvanceBio SEC 300 with 50mM Tris-HCl pH 8.5, 150mM NaCl, 5mM MgCl 2 , and 2 mM β-mercaptoethanol running phase at 0.16 mL/min flowrate. The measurements were performed at 0.99 Å wavelength. The sample to detector distance was set at 2.83 m with Pilatus2M detector for data acquisition.

Out of 100 randomly chosen micrographs, 1000 particles were manually picked without any structural knowledge about the complex to minimize the bias and assigned with 2D classes that were used in the ab initio model built. Out of generated 3D classes, one was manually picked and used for the training of TOPAZ neural networks, which in turned picked next interaction of particles that were used to retrain Topaz. With this approach approximately 0.5 M particles were selected to generate 50 2D classes, out of which 19 were manually picked.

Following SEC in HEPES buffer, the purified protein complex was crosslinked with bis(sulfosuccinimidyl)suberate (BS³, Thermo Scientific), an amine-to-amine crosslinker. The protein sample was incubated with 0.5 mM BS 3 (from 50 mM stock) for 30 min at RT. The crosslinking reaction was quenched with 1M Tris pH 7.5 to a final 50 mM Tris, and incubated 15 min at RT. The excess of BS 3 crosslinker was cleared by sequential dilution and concentration with Tris buffer. Concentrated samples were kept at 4°C or stored at -80°C for further measurements. Negative-stain transmission electron microscopy (NS-TEM) measurements was done in formvar/carbon supported 400 mesh copper grids, suspended in air with a negative lock tweezer. The purified protein-complex (0.03 mg/mL) was applied on glow-discharged Formvar/Carbon supported 400 mesh copper grids, and negatively stained with 1% neutralized uranyl-acetate. Grids were imaged using the JEOL JEM 2100HT electron microscope (Jeol Ltd, Tokyo, Japan) was used at accelerating voltage 200 kV. Images were taken by using 4kx4k camera (TVIPS) equipped with EMMENU software ver. 4.0.9.87.

Collected micrographs were processed using CryoSPARC 3.1.1. Initially, 9'350 particles were picked from micrographs using Blob Picker. Picked particles were subjected to a template-free 2D classification, from which 1'216 particles were selected and subjected for 3D reconstitution using Ab-initio reconstitution job. The nsp10/14/16 complex map derived from SAXS data was used for a rigid-body fit in such 3D-reconstituion map using Dock in map. Figure S1 . A: Native-PAGE of triplex; B: SDS-PAGE of the band excited from the native-PAGE (red box showing disintegration of the triplex into nsp14, nsp16, nsp10. Figure S2 . Mass spectroscopy of the triplex at 254 and 280 nm. 

To assess the structure of nsp10/14/16 triplex we performed the Transmission electron microscope (TEM) imaging. For this purpose, we crosslinked the triplex using amine reactive suberic acid bis sodium salt as previous attempts with the native complex yielded nonhomogenous samples. Crosslinking strengthened the conjugation so that it better withstood the deposition on copper grids. Samples were visualized using JEOL JEM 2100HT electron microscope at 80 kV accelerating voltage. Data was analyzed using cryoSparc software as described in the methodology section. Generated 2D classes clearly show lack of homogeneity of the sample on the grids. Figure S3 . 2D classes picked in cryoSparc for full length triplex.

We attempted to solve a high-resolution triplex structure using cryoEM technique. Highly purified triplex solution was vitrified on grids using Vitrobot under various conditions.

Resulting grids were measured using Titan Krios G3i at Solaris, Poland. Data analysis was performed using cryoSPARC software. We managed to train the TOPAZ neural networks, to pick approximately 0.5 M particles used to generate 50 2D classes, out of which 20 were manually picked ( Figure 6 ). Generated classes, though noisy, exhibit croissant-like shape with four distinct domains that correspond to nsp16, nsp10 and two domains of nsp14 (highlighted with arrows on the class that shows side-on view). Unfortunately, we were not able to reconstruct a high-resolution 3D structure from collected data. Our best attempt shown below in cyan has ca. 9 Å resolution. The fitted triplex hybrid model based on SAXS data (gold)

shows overall good fit with the extra volume that may arise from the flexible nature of the triplex. Figure S4 .: An overview of cryoEM data analysis in cryoSPARC. The overlay of generated 3D model from cryoEM (cyan) and SAXS model (gold).

Genome structure and transcriptional regulation of human coronavirus NL63

Structure of the RNA-dependent RNA polymerase from COVID-19 virus

Cryo-EM Structure of an Extended SARS-CoV-2 Replication and Transcription Complex Reveals an Intermediate State in Cap Synthesis

Structural Basis for RNA Replication by the SARS-CoV-2 Polymerase

Mechanism of molnupiravir-induced SARS-CoV-2 mutagenesis

Coupling of N7-methyltransferase and 3'-5' exoribonuclease with SARS-CoV-2 polymerase reveals mechanisms for capping and proofreading

Insights into RNA synthesis, capping, and proofreading mechanisms of SARS-coronavirus

RNA 3'-end mismatch excision by the severe acute respiratory syndrome coronavirus nonstructural protein nsp10/nsp14 exoribonuclease complex

High fidelity of murine hepatitis virus replication is decreased in nsp14 exoribonuclease mutants

2'-O methylation of the viral mRNA cap evades host restriction by IFIT family members

Multiple enzymatic activities associated with severe acute respiratory syndrome coronavirus helicase

Human coronavirus 229E nonstructural protein 13: characterization of duplex-unwinding, nucleoside triphosphatase, and RNA 5'-triphosphatase activities

Coronavirus nonstructural protein 16 is a cap-0 binding enzyme possessing (nucleoside-2'O)-methyltransferase activity

Functional screen reveals SARS coronavirus nonstructural protein nsp14 as a novel cap N7 methyltransferase

In vitro reconstitution of SARS-coronavirus mRNA cap methylation

Coronavirus replication-transcription complex: Vital and selective NMPylation of a conserved site in nsp9 by the NiRAN-RdRp subunit

Discriminating self from non-self in nucleic acid sensing

Synthesis of adenine dinucleosides SAM analogs as specific inhibitors of SARS-CoV nsp14 RNA cap guanine-N7-methyltransferase

Activation of the SARS-CoV-2 NSP14 3'-5' exoribonuclease by NSP10 and response to antiviral inhibitors

Structural analysis of the SARS-CoV-2 methyltransferase complex involved in RNA cap creation bound to sinefungin

Coronavirus Nsp10, a critical co-factor for activation of multiple replicative enzymes

DAMMIF, a program for rapid ab-initio shape determination in small-angle scattering

Deciphering conformational transitions of proteins by small angle X-ray scattering and normal mode analysis

Structural basis and functional analysis of the SARS coronavirus nsp14-nsp10 complex

Molecular basis for the bifunctional Uba4-Urm1 sulfur-relay system in tRNA thiolation and ubiquitin-like conjugation

Proteomic analysis of striatal neuronal cell cultures after morphine administration

Identifying SARS-CoV-2 antiviral compounds by screening for small molecule inhibitors of Nsp14 RNA cap methyltransferase

None.