The RNA methyltransferase METTL8 installs m3C32 in mitochondrial tRNAsThr/Ser(UCN) to optimise tRNA structure and mitochondrial translation

Modified nucleotides in tRNAs are important determinants of folding, structure and function. Here we identify METTL8 as a mitochondrial matrix protein and active RNA methyltransferase responsible for installing m3C32 in the human mitochondrial (mt-)tRNAThr and mt-tRNASer(UCN). METTL8 crosslinks to the anticodon stem loop (ASL) of many mt-tRNAs in cells, raising the question of how methylation target specificity is achieved. Dissection of mt-tRNA recognition elements revealed U34G35 and t6A37/(ms2)i6A37, present concomitantly only in the ASLs of the two substrate mt-tRNAs, as key determinants for METTL8-mediated methylation of C32. Several lines of evidence demonstrate the influence of U34, G35, and the m3C32 and t6A37/(ms2)i6A37 modifications in mt-tRNAThr/Ser(UCN) on the structure of these mt-tRNAs. Although mt-tRNAThr/Ser(UCN) lacking METTL8-mediated m3C32 are efficiently aminoacylated and associate with mitochondrial ribosomes, mitochondrial translation is mildly impaired by lack of METTL8. Together these results define the cellular targets of METTL8 and shed new light on the role of m3C32 within mt-tRNAs.

N aturally occurring modifications in RNA represent an important and dynamic layer of gene expression regulation [1][2][3] . Recent advances in RNA modification detection techniques have uncovered in excess of 150 differently modified ribonucleosides in cellular RNAs, and some of the enzymes responsible for installing these modifications have been identified and characterised [4][5][6][7] . Modified ribonucleotides are present throughout the transcriptome and are broadly implicated in modulating RNA structure, stability and function either directly or by influencing RNA-protein interactions 3,[8][9][10][11] . However, an accurate and comprehensive overview of modified sites in the transcriptome remains elusive and the precise functions of many individual modification marks in RNAs are still unclear.
Transfer (t)RNAs are the most extensively and diversely modified class of RNA, with more than 80 different modification types present across the domains of life 7,[12][13][14] . Introduction of specific base and ribose modifications into eukaryotic tRNAs is an important step during tRNA maturation, which can not only regulate their roles as amino acid adaptors during translation but also influence other extra-translational functions, such as the production of tRNA-derived small RNAs [15][16][17] . Modifications present in the T-and D loops of tRNA cores generally act cooperatively to regulate tRNA folding and mediate stabilisation of tRNA tertiary structure [18][19][20][21] . In contrast, modifications within the anticodon loop typically influence tRNA function by expanding codon recognition, improving decoding efficiency and accuracy during translation or preventing frameshifting events [22][23][24][25] . Consistent with this, positions 34 (wobble position) and 37 are "hotspots" of diverse and complex (hyper-)modifications 26 . The importance of tRNA modifications within the cellular context is highlighted by the numerous human diseases arising from defects in diverse tRNA modifying enzymes as well as the pathogenic effects of mutations in tRNA sequences that impair modification installation 13,27,28 .
Methylation of N3 of cytidine (3-methylcytidine; m 3 C) is present at position 32 of eukaryotic, cytoplasmic tRNAs Thr and tRNAs Ser in species ranging from yeast to humans and has also been identified in mitochondrial (mt-)tRNA Thr/Ser(UCN) in mammals [29][30][31][32][33][34][35][36] . In Saccharomyces cerevisiae (S. cerevisiae), the m 3 C methyltransferase Trm140 mediates modification of both cytoplasmic tRNA Thr/Ser whereas in Schizosaccharomyces pombe (S. pombe) two alternative enzymes, Trm140 and Trm141, install m 3 C 32 modifications in cytoplasmic tRNAs Thr and tRNAs Ser respectively 30,37 . The mammalian homologues of the S. pombe proteins, METTL2 (with paralogues A and B in humans) and METTL6 are likewise responsible for installing these modifications into the human tRNAs Thr and tRNA  . The m 3 C 32 modification is also present in mammalian tRNA Arg(CCU) and tRNA Arg(UCU) , and METTL2A/B, together with the cofactor protein DALRD3, has recently been identified as the methyltransferase responsible for introducing these modifications 40 . The precise functions of the m 3 C 32 modifications in tRNAs Thr/Ser remain unknown but potential roles in fine-tuning translation have been discussed 37,42 . Deletion of trm140 in yeast cells lacking another tRNA methyltransferase, Trm1, are sensitive to the translation inhibitor cycloheximide, as are Trypanosoma brucei (T. brucei) lacking TbTrm140 37,43 , and treatment with alkylating agents has been shown to increase m 3 C levels in yeast tRNA Thr leading to enhanced translation of threonine-rich membrane proteins encoded by degenerate threonine codons 42 . Furthermore, the importance of METTL2A/B-and METTL6-mediated m 3 C 32 modification of tRNAs Thr/Ser is emphasised by the findings that lack of METTL6 retards tumour cell growth and impairs pluripotency 39 , and that a pathogenic mutation in DALRD3, which impairs m 3 C 32 installation in tRNA Arg(CCU)/(UCU) , is found in patients with developmental delay and early-onset epileptic encephalopathy 40 . Interestingly, in mammals, a third putative m 3 C methyltransferase METTL8, also homologous to yeast Trm140, has been identified 30,38 . METTL8 remains poorly characterised; although it has been linked to potential m 3 C modifications in cytoplasmic mRNAs, no specific target sites have thus far been identified 38 .
Here we reveal that METTL8 is an active m 3 C RNA methyltransferase that localises to the mitochondrial matrix and is responsible for installing m 3 C 32 modifications in both mt-tRNA Ser(UCN) and mt-tRNA Thr . Using in vivo UV crosslinking and analysis of cDNA (CRAC) and in vitro assays, we discover that METTL8 interacts directly with many mt-tRNAs but displays robust methylation activity only on mt-tRNA Thr and mt-tRNA Ser (UCN) . Our in vitro and in vivo data identify prior installation of other modifications at position 37 and specific nucleotides in the anticodon loops of these tRNAs as important specificity determinants for METTL8-mediated methylation. We provide evidence that the presence of m 3 C 32 influences the structure of mt-tRNA Thr/Ser (UCN) and propose that the m 3 C 32 modifications installed in mt-tRNA Thr/Ser(UCN) by METTL8 may optimise tRNA structure to fine-tune mitochondrial translation.

Results
METTL8 localises to the mitochondrial matrix. While METTL2A/B and METTL6 have recently been identified as the methyltransferases responsible for introducing m 3 C 32 modifications into the human, cytoplasmic tRNAs Ser/Thr/Arg(CCU/ UCU) [38][39][40][41] , the third human m 3 C methyltransferase METTL8 (Fig. 1a) has remained poorly characterised. METTL8 is extensively alternatively spliced ( Supplementary Fig. 1a), however, the two most common transcripts expressed in HEK293 cells only differ in the sequences encoding the protein C-terminus (Supplementary Fig. 1a, b; Supplementary Table 1) and the mRNA encoding the full-length (407 amino acid) protein is the topranked isoform in most human tissues 44 . To explore the subcellular localisation of the longest METTL8 isoform, a stably transfected HEK293 cell line for the tetracycline-inducible expression of C-terminally GFP tagged METTL8 (METTL8-GFP) was generated. Confocal fluorescence microscopy revealed a punctate distribution of METTL8-GFP in the cytoplasm colocalising with MitoTracker, indicating that METTL8 localises to mitochondria (Fig. 1b, upper panel), which is consistent with its detection in high-throughput mitochondrial proteome analysis 45 . Prediction of potential mitochondrial targeting sequences (MTSs) in METTL8 and the other known human m 3 C methyltransferases using the MitoFates algorithm 46 revealed a high probability, 20 amino acid MTS at the N-terminus of METTL8 that is lacking in METTL2A/B and METTL6 (Fig. 1a, c). The functionality of this MTS was demonstrated by analysing the localisation of a C-terminally GFP-tagged version of METTL8 lacking the predicted MTS (METTL8 21-407 -GFP). In contrast to the full-length protein, METTL8 21-407 -GFP was present throughout the cytoplasm and nucleus, and accumulated in nuclear foci likely corresponding to nucleoli (Fig. 1b, lower panel).
A protease protection assay was then performed to verify mitochondrial import of METTL8 and determine its submitochondrial localisation. Mitochondria and mitoplasts isolated from cells expressing METTL8-His 6 -2xFLAG were left intact or lysed by sonication and then treated with different concentrations of Proteinase K. Western blot analyses demonstrated that all examined proteins were susceptible to protease digestion after sonication (Fig. 1d). As expected, the outer mitochondrial membrane protein MFN2 47 was degraded in protease-treated mitochondria and mitoplasts, while TIM23, which resides in the inner mitochondrial membrane (IMM) 48  The predicted mitochondrial targeting sequence (MTS) is depicted in blue and its amino acid sequence is given with the predicted mitochondrial processing peptidase cleavage site highlighted in red and residues with probability of forming an amphipathic alpha-helix in blue. The predicted methyltransferase domain (MTase domain; pfam 13649) is indicated in purple. Amino acid substitutions affecting catalytic activity (D230A and D309A) are indicated with green lines. b Fluorescence microscopy was performed on HEK293 cells expressing METTL8-GFP or GFP-tagged METTL8 lacking the predicted MTS (METTL8 21-407 -GFP). Mitochondria were visualised with MitoTracker (red) and nuclear material stained with DAPI (blue). An overlay image (Merge) is shown and the scale bar represents 10 µm. Schematic representations of the expressed proteins are shown with colours as in (a) and the GFP tag in green. Representative images from three independent experiments. c Probabilities of the indicated human methyltransferase-like (METTL) proteins containing a mitochondrial targeting signal were calculated. Range 0-1. d, e Mitochondria isolated from HEK293 cells expressing METTL8-His 6 -2xFLAG (d) or METTL8 21-407 -His 6 -2xFLAG (e) were converted to mitoplasts or sonicated (Sonic.), then treated with different concentrations of Proteinase K (PK). Proteins were analysed by western blotting using the indicated antibodies. MFN2, TIM23 and uS14m are proteins of the outer mitochondrial membrane, inner mitochondrial membrane and mitochondrial matrix respectively. Schematic representations of the tagged proteins are given below using colours as in (a) and the His 6 -2xFLAG tag in red. f Mitochondria isolated from HEK293 cells expressing METTL8-His 6 -2xFLAG were resuspended in homogenisation buffer supplemented with Triton-X-100 or buffer containing 0.1 M sodium carbonate at the indicated pHs and then centrifuged. Samples of the total extract prior to centrifugation (T), the pellet (P) and supernatant (S) were analysed by western blotting with the indicated antibodies. TIM23 is an integral membrane protein of the inner mitochondrial membrane, TIM44 and uL3m are peripherally associated with the inner mitochondrial membrane. A schematic representation of the tagged protein shown as in (d, e). For d, three independent experiments were performed and representative data are shown, and for e and f, a single experiment was performed. Source data are provided as a Source Data file.
towards the intermembrane space, was protected in mitochondria but not mitoplasts. Similar to the mitochondrial ribosomal protein uS14m present in the mitochondrial matrix 49 , METTL8-His 6 -2xFLAG, but not METTL8 21-407 -His 6 -2xFLAG, was resistant to digestion in both protease-treated mitochondria and mitoplasts, indicating that METTL8 is present within the mitochondrial matrix (Fig. 1d, e). To further refine the localisation of METTL8 within the mitochondrial matrix by determining if it is associated with the IMM, mitochondria isolated from METTL8-His 6 -2xFLAG cells were treated with sodium carbonate at different alkaline pH values before centrifugation (Fig. 1f). As the pH increases, only integral membrane proteins, such as TIM23, are retained and pelleted, whereas those more peripherally associated with the IMM or soluble in the matrix, such as TIM44 50 and uL3m 51 , are (partially) released into the supernatant. Analogous to TIM44, METTL8 was found predominantly in the pellet after sodium carbonate treatment at pH 10.5 and centrifugation, but was partially extracted into the soluble fraction at higher pH values. Together, these data demonstrate that METTL8 is a mitochondrial matrix protein peripherally associated with the IMM.

METTL8 interacts with mt-tRNA ASLs in vivo and in vitro.
Based on its close homology to METTL2A/B and METTL6, METTL8 is a predicted m 3 C RNA methyltransferase but no targets have yet been defined. Therefore, a UV cross-linking and analysis of cDNA (CRAC) experiment 52-54 was performed to identify cellular RNAs directly contacted by METTL8. Cells expressing METTL8-His 6 -2xFLAG, or just the His 6 -2xFLAG tag as a control, were UV crosslinked and protein-RNA complexes were purified. Bound RNAs were trimmed, radiolabelled with [ 32 P] and ligated to adaptors. Protein-RNA complexes were separated by denaturing polyacrylamide gel electrophoresis (PAGE), transferred to a membrane and radiolabelled RNAs were detected by autoradiography (Fig. 2a). No radioactive signal was detected in the sample derived from the cells expressing the His 6 -2xFLAG tag while the strong signal at and above the migration position of METTL8-His 6 -2xFLAG indicated association of METTL8 with cellular RNAs (Fig. 2a). The region of the membrane containing the radiolabelled RNAs, and a corresponding area of the membrane from the control lane, were excised and extracted RNAs were copied into cDNA libraries that were subjected to Illumina deep sequencing. Consistent with the protein localisation analyses (Fig. 1), mapping of the obtained sequencing reads on the human genome revealed a strong enrichment of mitochondrial-encoded RNA species (mitoRNA) in the METTL8-His 6 -2xFLAG sample compared to the control. In the METTL8-His 6 -2xFLAG sample, 71% of the total reads obtained were derived from mitoRNA, compared to only 6% for the FLAG control ( Fig. 2b; Supplementary Table 2). Among the mitoRNAderived reads, those arising from mt-tRNAs were the most abundant and were also most enriched in the METTL8-His 6 -2xFLAG sample compared to the control (Fig. 2b). After normalisation, the numbers of sequencing reads mapping to each of the 22 mt-tRNA genes in both the METTL8-His 6 -2xFLAG and control datasets were determined, and the fold difference between the numbers in the METTL8-His 6 -2xFLAG and control samples calculated ( Fig. 2c; Supplementary Table 3). Enrichment of reads derived from all mt-tRNAs in the METTL8-His 6 -2xFLAG sample compared to the control was observed but those most strongly enriched with METTL8-His 6 -2xFLAG were those derived from mt-tRNA Ser(UCN) , mt-tRNA Thr and mt-tRNA Ile (Fig. 2c). To confirm the interaction of METTL8 with mt-tRNAs, an anti-FLAG immunoprecipitation after UV crosslinking was performed using lysates from cells expressing METTL8-His 6 -2xFLAG or the His 6 -2xFLAG tag. Expression of METTL8-His 6 -2xFLAG and its specific enrichment were confirmed by western blotting (Fig. 2d, upper two panels). RNAs present in the lysates and co-precipitated RNAs were analysed by northern blotting using probes hybridising to a selection of mt-tRNAs and, as a control, the U6 small nuclear (sn)RNA. The U6 snRNA was recovered with neither the His 6 -2xFLAG tag nor METTL8-His 6 -2xFLAG, while mt-tRNA Ser(UCN) , mt-tRNA Thr and mt-tRNA Ile were retrieved with METTL8-His 6 -2xFLAG but not in the control sample (Fig. 2d, lower four panels). The numbers of sequencing reads and nucleotide substitutions mapping to each nucleotide of the two most enriched mt-tRNAs (mt-tRNA Thr and mt-tRNA Ser(UCN) ) were plotted to generate profiles of METTL8-His 6 -2xFLAG crosslinking ( Fig. 2e; Supplementary Tables 4 and  5). Nucleotide substitutions can arise due to reverse transcriptase errors when nucleotides crosslinked to amino acids are encountered and can therefore indicate sites of protein-RNA crosslinking. These profiles indicate that METTL8-His 6 -2xFLAG predominantly contacts the ASL region of mt-tRNA Thr/Ser(UCN) (Fig. 2e).
To further explore interactions between METTL8 and RNA, fluorescence anisotropy experiments were performed using fluorescein-labelled model substrates together with recombinantly expressed His 14 -MBP-METTL8 purified from E. coli. Consistent with the CRAC data, His 14 -MBP-METTL8 bound both the mt-tRNA Ser(UCN) and mt-tRNA Met ASLs with high affinity (Fig. 2f).
To assess the specificity of these interactions and determine whether METTL8 preferentially binds the ASL structure, further experiments were performed using two unstructured RNA substrates of comparable length with diverse nucleotide sequences and also a DNA oligonucleotide (Fig. 2g). His 14 -MBP-METTL8 did not bind the DNA oligonucleotide, demonstrating its specificity for RNA binding (Fig. 2g). However, both unstructured RNAs were bound with similar affinities to the mt-tRNA ASLs (Fig. 2f, g), implying that the robust RNA binding of METTL8 is not strongly influenced by sequence or structure.
METTL8 installs m 3 C 32 modifications on mt-tRNA Thr and mt-tRNA Ser(UCN) . METTL8 is a putative m 3 C methyltransferase 38 and among all mt-tRNAs, only mt-tRNA Thr and mt-tRNA Ser(UCN) are reported to contain m 3 C 32 33,36,55 . As these two mt-tRNAs were the most enriched RNAs with METTL8-His 6 -2xFLAG in our crosslinking experiments, we explored if they are substrates of METTL8 methylation activity. Using the CRISPR/Cas9 genome editing system, two HEK293 cell lines (KO1 and KO2) carrying 15 and 1 nucleotide (nt) deletions within exon 3 of METTL8 that do not express functional full-length METTL8 were generated (Fig. 3a, b). Small RNAs recovered from wild-type (WT) HEK293 cells and those lacking METTL8 (KO1 and KO2) were analysed by primer extension to detect m 3 C 32 in mt-tRNA Thr and mt-tRNA Ser(UCN) 38,40,43 . During primer extension, the presence of m 3 C impedes progression of the reverse transcriptase leading to stalled cDNA synthesis, whereas in the absence of m 3 C, the reverse transcriptase continues until another blocking RNA modification is encountered or the 5′ end of the RNA is reached ( Fig. 3c and Supplementary Fig. 2a). Consistent with the reported presence of m 3 C 32 in mt-tRNA Thr / Ser(UCN)33,36,55 , primer extension on small RNAs derived from WT cells showed strong reverse transcription stalling at position 32 and minimal read-through for both mt-tRNA Thr and mt-tRNA Ser(UCN) (Fig. 3d). In contrast, primer extension on small RNAs derived from the cells lacking METTL8 showed minimal reverse transcriptase stalling at C 32 and clear signals corresponding to the m 2 G 10 and/or m 1 A 9 modifications in mt-tRNA Thr or the 5′ end of the tRNA in the case of mt-tRNA Ser(UCN) , which does not contain any modifications 5′ of the m 3 C 32 site (Fig. 3d). In parallel, m 3 C 32 levels in mt-tRNA Thr/ Ser(UCN) in WT, KO1 and KO2 cells were monitored using deoxyribozymes that either preferentially cleave adjacent to m 3 C nucleotides compared to unmodified cytosines (AL112) or those whose cleavage activity is impaired by the presence of methylated cytosines (AK104) 56  To verify that METTL8 is directly responsible for methylation of C 32 in these tRNAs, small RNAs extracted from WT, KO1 and KO2 cells were subjected to in vitro methylation using S-adenosylmethionine (SAM) as a methyl group donor and recombinant His 14 -MBP-METTL8 or METTL8-derivatives carrying amino acid substitutions in the catalytic domain predicted to impair methylation activity (His 14 -MBP-METTL8 D230A and His 14 -MBP-METTL8 D309A ; Fig. 1a) 29 . Primer extension performed with primers specific for mt-tRNA Thr and mt-tRNA Ser(UCN) showed that treatment of RNAs from WT cells with recombinant His 14 -MBP-METTL8 lead to slightly decreased read-through signals, consistent with the methylation of endogenously unmethylated mt-tRNA Thr/Ser(UCN) (m 3 C modification level: mt-tRNA Thr 85 -94% 36,57 ) (Fig. 3e, f). In the presence of SAM, addition of His 14 -MBP-METTL8 to small RNAs derived from the METTL8 KO cell lines allowed methylation of C 32 of both mt-tRNA Thr and mt-tRNA Ser(UCN) to be fully restored (Fig. 3e, f). In contrast, primer extension of RNAs incubated with His 14 -MBP-METTL8 D230A showed C 32 methylation levels similar to untreated RNAs from METTL8 KO cells and only a low level of methylation of C 32 on RNAs from METTL8 KO cells treated with His 14 -MBP-METTL8 D309A (Fig. 3e, f). Together these results demonstrate that the methylation activity of METTL8 targets C 32 of mt-tRNA Thr/Ser(UCN) and that amino acid substitutions in the methyltransferase domain impair its catalytic activity.
As these methylation assays were performed on RNAs isolated from human cells, to investigate whether METTL8 was able to methylate nascent transcripts or whether other elements installed within the cellular context are required for C 32 methylation, the activity of His 14 -MBP-METTL8/METTL8 D230A /METTL8 D309A on in vitro transcribed mt-tRNA Thr and mt-tRNA Ser(UCN) was examined using [ 3 H]-SAM as the methyl group donor. His 14 -MBP-METTL8 was readily able to methylate mt-tRNA Thr , but only minimal methylation of mt-tRNA Ser(UCN) was observed ( Fig. 3g), despite similar levels of C 32 methylation of these two mt-tRNAs being observed in cellular RNAs (Fig. 3e, f). Specificity of the observed methylation activity of His 14 -MBP-METTL8 on mt-tRNA Thr was confirmed as neither His 14 -MBP-METTL8 D230A nor His 14 -MBP-METTL8 D309A was able to substantially methylate the in vitro transcript that, in contrast to the endogenous mt-tRNA Thr (Fig. 3e, f), was present at an equimolar level to the recombinant protein (Fig. 3h).
Modifications at position 37 promote installation of m 3 C 32 in mt-tRNAs Thr/Ser(UCN) by METTL8. The finding that METTL8 efficiently methylates cellular mt-tRNA Ser(UCN) but not in vitro transcribed mt-tRNA Ser(UCN) suggests the requirement of other RNA modifications for efficient METTL8-mediated methylation of C 32 . In yeast, interdependence of m 3 C 32 and modifications at position 37 has been observed 30,31,58 , raising the possibility that (ms 2 )i 6 A 37 in mt-tRNA Ser(UCN) and t 6 A 37 in mt-tRNA Thr may stimulate C 32 methylation by METTL8. Unmodified ASLs or those containing appropriate combinations of the modified nucleotides m 3 C, i 6 A, ms 2 i 6 A and t 6 A were prepared by solid phase synthesis using ribonucleoside phosphoramidite building blocks, which were chemically synthesised from canonical nucleosides in five or six linear steps ( Fig. 4a and Supplementary  Fig. 3). Methylation assays with recombinant His 14 -MBP-METTL8 and [ 3 H]-SAM showed that for mt-tRNA Thr , the unmodified ASL was sufficient for METTL8-mediated methylation and that the presence of m 3 C 32 prevented methylation, confirming this as the target nucleotide also in vitro (Fig. 4b).
Strikingly, the presence of the t 6 A 37 modification increased significantly the level of METTL8-mediated methylation of the mt-tRNA Thr ASL (Fig. 4b). For mt-tRNA Ser(UCN) , very little methylation of the unmodified ASL was observed, as previously seen for the unmodified in vitro transcript (Fig. 3g), but the presence of either i 6 A 37 or ms 2 i 6 A 37 facilitated methylation by METTL8 (Fig. 4c). Interestingly, the ms 2 i 6 A 37 modification, which is predominantly present in cellular mt-tRNA Ser(UCN)36 , stimulated methylation significantly more than i 6 A 37 , an intermediate formed during ms 2 i 6 A hypermodification. ASLs containing m 3 C 32 as well as i 6 A 37 or ms 2 i 6 A 37 were not methylated, consistent with C 32 of mt-tRNA Ser(UCN) being the METTL8 target nucleotide (Fig. 4c).
As prior modification of A 37 of mt-tRNA Thr/Ser(UCN) strongly enhances C 32 methylation in vitro, the effect of depleting enzymes responsible for modifying A 37 on the levels of m 3 C 32 in cellular mt-tRNA Thr/Ser(UCN) was examined. RNAimediated depletion of the N 6 -threonylcarbamoyltransferase to mt-tRNAs in cells. a HEK293 cells expressing METTL8-His 6 -2xFLAG or the His 6 -2xFLAG tag were UV crosslinked. Protein-RNA complexes were retrieved, and co-purified RNAs trimmed, radioactively labelled and ligated to adaptors. Protein-RNA complexes were separated by denaturing PAGE, transferred to a nitrocellulose membrane and detected by autoradiography. Areas of the membrane excised are indicated with red boxes. Data presented in a-c and e derive from a single experiment. b RNAs eluted from the membrane areas indicated in (a) were reverse transcribed and the cDNA library deep sequenced. Doughnut charts show the relative distribution of reads derived from different classes of RNA in the His 6 -2xFLAG and METTL8-His 6 -2xFLAG samples. Abbreviations: mRNA -messenger RNA, tRNA -transfer RNA, lncRNA -long non-coding RNA, mt-tRNA -mitochondrial tRNA, mt-rRNA -mitochondrial rRNA, mt-mRNA -mitochondrial mRNA. c The normalised numbers of sequencing reads mapping to each mt-tRNA gene in the METTL8-His 6 -2xFLAG and His 6 -2xFLAG datasets are depicted as a heatmap in logarithmic scale (left panel). The fold-enrichment of reads derived from mt-tRNAs in the METTL8-His 6 -2xFLAG compared to His 6 -2xFLAG control is depicted in the right panel. d Lysates from crosslinked cells expressing METTL8-His 6 -2xFLAG or the His 6 -2×FLAG tag were used for immunoprecipitation experiments. Proteins and RNAs in input and eluate samples were analysed by western and northern blotting respectively. Three biologically independent experiments were performed and representative blots are shown. e The numbers of sequencing reads and nucleotide substitutions (Sub.) mapping to each nucleotide of the mt-RNA Thr and mt-tRNA Ser(UCN) genes in the METTL8-His 6 -2xFLAG and control sample (His 6 -2xFLAG) are shown. The nucleotide sequence of each mt-tRNA is given with nucleotides of the anticodon stemloop (ASL) indicated in red. RPM -reads per million mapped reads. f, g Fluorescence anisotropy measurements were taken to determine the affinity of recombinant His 14 -MBP-METTL8 for different fluorescein-labelled oligonucleotides: ASLs of mt-tRNA Ser(UCN) and mt-tRNA Met (f), unstructured RNAs and DNA oligonucleotide (g). Data from three independent experiments are shown as mean ± standard deviation and dissociation constants (K d s) are given. Source data are provided as a Source Data file.
OSGEPL1 and the N 6 -isopentenyltransferase TRIT1, which target mt-tRNA Thr and mt-tRNA Ser(UCN) respectively, was established 57,59 . Treatment of cells with two independent siRNAs against OSGEPL1 and TRIT1 decreased protein levels to approximately 11% and 33-38% respectively, of those observed in cells treated with non-target siRNAs ( Fig. 4d and Supplementary Figs. 4a, b). Analysis of m 3 C 32 levels in mt-tRNA Thr/Ser(UCN) by primer extension showed that the  percentage of read-through of m 3 C 32 was mildly, but significantly, increased on RNAs obtained from cells treated with siRNAs against OSGEPL1 or TRIT1 compared to those treated with the non-target siRNA (Fig. 4e). Hypomodification of tRNAs has, in some cases, been shown to destabilise tRNAs thus increasing turnover 20,21 . To determine whether the mild effects of these knockdowns on m 3 C 32 levels in cellular mt-tRNA Thr/Ser(UCN) compared to the effects in vitro arise due to degradation of hypomodified mt-tRNAs in the knockdown cells, the levels of mt-tRNA Thr/Ser(UCN) in these samples were determined by northern blotting. No significant differences were observed ( Supplementary Fig. 4c, d), implying that the mild effects, especially of the TRIT1 depletion, may rather be caused by residual protein remaining in the siRNA-treated cells. Furthermore, although the presence of t 6 A 37 /(ms 2 )i 6 A 37 stimulates METTL8-mediated methylation in vitro, some methylation activity, especially on mt-tRNA Thr , is observed in the absence of these modifications. Thus it is also possible that some m 3 C 32 installation may take place even when OSGEPL1/ TRIT1 are lacking. Our earlier finding that METTL8 binds not only the ASLs of its modification substrate mt-tRNAs, but also other RNAs (Fig. 2c, d), together with the discovery that modification of A 37 in mt-tRNA Thr/Ser(UCN) stimulates C 32 methylation by METTL8, raised the possibility that the affinity of METTL8 for its methylation substrates is enhanced by the presence of the modified nucleotides at position 37. Fluorescence anisotropy experiments were therefore performed using recombinant His 14 -MBP-METTL8 and fluorescein-labelled mt-tRNA Ser(UCN) ASLs without modified nucleotides, or with the i 6 A 37 or i 6 A 37 and m 3 C 32 modifications present (Fig. 4f). The highest affinity, with a K d of 0.2 μM, was observed for His 14 -MBP-METTL8 interacting with the i 6 A 37 -containing ASL. The presence of both i 6 A 37 and m 3 C 32 lead to approximately 2.5-fold weaker binding of His 14 -MBP-METTL8 (Fig. 4f). Although the effects are small, these data may indicate that the presence of i 6 A 37 is more important for the methylation activity of METTL8 than for binding of the methyltransferase, but that upon installation of m 3 C 32 , METTL8 has lower affinity for its substrate, potentially promoting its dissociation.
Among the mt-tRNAs, eight have a cytidine in position 32 and a modified adenosine at position 37, yet only mt-tRNA Thr and mt-tRNA Ser(UCN) are reported to have m 3 C 32 ( Supplementary   Fig. 5). To verify that other mt-tRNAs containing the features thus far identified as important for C 32 methylation are not targets of METTL8, in vitro methylation assays were performed using [ 3 H]-SAM, His 14 -MBP-METTL8 and the ASLs of mt-tRNA Ser(AGY) containing t 6 A 37 and mt-tRNA Phe containing ms 2 i 6 A 37 . Neither of these ASLs were methylated by His 14 -MBP-METTL8 (Fig. 4g), suggesting that METTL8 also relies on other elements within mt-tRNA Thr and mt-tRNA Ser(UCN) for substrate recognition. U 34 and G 35 in the anticodon loop are recognition elements for METTL8. It has previously been shown that a G 35 U 36 t 6 A 37 motif in tRNA Thr is required for substrate recognition by the yeast m 3 C methyltransferase Trm140, whereas a distinctive variable loop and the seryl-tRNA synthetase are specificity determinants important for m 3 C 32 installation in tRNA Ser(CGA/UGA)31 . As the mitochondrial seryl-tRNA synthetase (SARS2) has been recovered with METTL8 60 , a potential role in methylation of C 32 of mt-tRNA Ser(UCN) by METTL8 was explored. However, RNAimediated depletion of SARS2 lead to only a very mild decrease in the amount of m 3 C 32 detected in mt-tRNA Ser(UCN) and the addition of His 10 -SARS2 to in vitro methylation assays did not significantly stimulate methylation by METTL8, suggesting that SARS2 is not strictly required for C 32 methylation of mt-tRNA Ser(UCN) (Supplementary Note 1 and Supplementary  Fig. 6b, d, e).
We therefore focused instead on analysing the influence of individual nucleotide substitutions within the anticodon loop on METTL8-mediated installation of m 3 C 32 (Fig. 5a, b). In vitro methylation assays performed using His 14 -MBP-METTL8 and [ 3 H]-SAM and various mt-tRNA Thr transcripts revealed that nucleotide substitutions at positions 32, 34, 35, 36 and 38 abolished methylation by METTL8 (Fig. 5a, right), indicating that the U 34 G 35 U 36 t 6 A 37 A 38 motif is important for substrate recognition. Only exchange of U 33 for A did not significantly affect methylation of C 32 by METTL8. For mt-tRNA Ser(UCN) , as the presence of (ms 2 )i 6 A 37 is an important pre-requisite for substantial METTL8-mediated methylation of C 32 , nucleotides 34, 35 and 38 were exchanged in the context of the ms 2 i 6 A 37containing mt-tRNA Ser(UCN) ASL (Fig. 5b, left). Although methylation by METTL8 was reduced when U 34 was changed to A, only substitution of G 35 for A significantly decreased mt-tRNA Ser(UCN) ASL methylation (Fig. 5b, right). Interestingly, Fig. 3 METTL8 methylates C 32 of mt-tRNA Thr and mt-tRNA Ser(UCN) . a Schematic representation of the human METTL8 gene and a magnified view of exon 3 showing the genome editing to generate the two METTL8 knockout cell lines (KO1 and KO2). Both cell lines are homozygous so only one DNA strand is represented. b Proteins extracted from wild-type (WT) HEK293 cells and two METTL8 knockout cell lines (KO1 and KO2) were analysed by western blotting. Representative data from three independent experiments is shown. c Schematic views of m 3 C 32 -and C 32 -containing mt-tRNA Thr are shown with the binding site of the primer used for primer extension analyses (solid line) and the expected primer extension products (dashed lines). d Small RNAs extracted from wild-type (WT) cells and those lacking METTL8 (KO1 and KO2) were used for primer extension with [ 32 P]-labelled primers specific for mt-tRNA Thr (left panel) or mt-tRNA Ser(UCN) (right panel). Resulting cDNA fragments were separated by denaturing PAGE alongside [ 32 P]labelled DNA oligonucleotides of the indicated lengths. Representative image of three independent experiments. e Small RNAs as in (d) incubated with purified His 14 -MBP-METTL8, His 14 -MBP-METTL8 D230A or His 14 -MBP-METTL8 D309A in presence of S-adenosylmethionine or left untreated were used in primer extension reactions as in (d). Representative image of three independent experiments. f Signal intensities of the extension products were quantified and the percentage of read-through (Rt) of m 3 C 32 are given as a percentage of the total extension signal. Data from n = 3 independent experiments are shown as mean ± standard deviation. Statistical analysis was performed using one-way ANOVA (F = 35.18, p < 0.0001 for mt-tRNA Thr ; F = 36.04, p < 0.0001 for mt-tRNA Ser(UCN) ) and significance calculated using Tukey's multiple comparisons test. g In vitro transcribed mt-tRNA Thr and mt-tRNA Ser(UCN) were incubated with purified His 14 -MBP-METTL8 and [ 3 H]-SAM. Tritium incorporated into the RNA was measured by scintillation counting. Bar plots show mean counts per minute (CPM) of n = 3 independent experiments ± standard deviation. Statistical analysis was performed using two-tailed unpaired Student's t test (p < 0.0001). Pelleted RNA was resuspended, separated by denaturing PAGE and stained with ethidium bromide (EtBr). Labelled RNAs ( 3 H-Me) were detected by autoradiography. h n = 3 independent in vitro methylation assays were performed as in (g) using purified His 14 -MBP-METTL8, His 14 -MBP-METTL8 D230A and His 14 -MBP-METTL8 D309A . Data were analysed using one-way ANOVA (F = 207.3; p < 0.0001) and significance calculated using Tukey's multiple comparisons test. p values for data in f, g and h are given in source data; *p < 0.05, ***p < 0.001, ****p < 0.0001 and ns not significant. Source data and p values are provided as a Source Data file.   Tritium incorporated into the RNA was measured by scintillation counting. Bar plots show mean counts per minute (CPM) of n = 3 independent experiments ± standard deviation. Statistical analysis was performed using one-way ANOVA (F = 190.5; p < 0.0001) and significance calculated using Tukey's multiple comparisons test (d) or using a two-tailed unpaired Student's t test (ns non-significant) (e). RNA was also separated by denaturing PAGE, stained with ethidium bromide (EtBr) and labelled RNAs ( 3 H-Me) were detected by autoradiography. Representative images from three independent experiments. f CD spectra of unmodified mt-tRNA Ser(UCN) ASL (black) (top), its ms 2 i 6 A 37 modified analogue (red) (bottom), and the analogous ASL containing A38C (green) (bottom) and G35A (blue) (top) substitutions were recorded at 5 µM in 10 mM Na-phosphate buffer pH 7.0, 1 mM MgCl 2 . CD spectra were recorded at least twice for each sample. p values for data in a, b, d, e are given in source data; ***p < 0.001, ****p < 0.0001, ns not significant. Source data and p values are provided as a Source Data file. the A38C substitution in mt-tRNA Ser(UCN) lead to increased METTL8-mediated methylation, similar to the recently observed higher methylation activity of METTL6 on cytoplasmic tRNA-Ser(GCU) carrying an A38C substitution 41 . These data suggest that METTL8 relies less on the sequence context for recognition of mt-tRNA Ser(UCN) than mt-tRNA Thr . However, in both cases G 35 and U 34 , which are notably absent from the other t 6 A 37 -and ms 2 i 6 A 37 -containing mt-tRNAs ( Supplementary Fig. 5a), play important roles in the recognition of both substrate mt-tRNAs.
To determine if the identified sequence motif is sufficient to induce methylation of non-substrate mt-tRNAs that have a C 32 residue, chimeric mt-tRNA Asn transcripts containing the anticodon loop or anticodon stem loop of mt-tRNA Thr were generated (Fig. 5c, top). In vitro methylation assays demonstrated that, in contrast to mt-tRNA Asn (Supplementary Fig. 6f), the chimeric transcripts were methylated by METTL8, but at a significantly lower level than METTL8-mediated methylation of mt-tRNA Thr (Fig. 5d). These data support that the U 34 G 35 U 36 A 37 A 38 motif is recognised by METTL8 and indicate that the anticodon stem does not contribute to substrate recognition, but that other elements of mt-tRNA Thr are recognised/bound by METTL8. To explore whether the same is true for mt-tRNA Ser(UCN) , a synthetic ASL composed of the anticodon loop of mt-tRNA Ser(UCN) (containing ms 2 i 6 A 37 ) and a mt-tRNA Phe stem was generated (Fig. 5c,  bottom). In this case, the chimeric ASL was methylated by METTL8 as efficiently as the mt-tRNA Ser(UCN) ASL (Fig. 5e).
Interestingly, a mutation identified in patients with mitochondrial disorders and myoclonic epilepsy with ragged red fibres (MERRF) syndrome (m.15923A>G) 57 leads to exchange of A 38 for G in mt-tRNA Thr . It has previously been shown that this nucleotide substitution impairs N 6 -threonylcarbamoylation of A 37 , implying that loss of t 6 A 37 in mt-tRNA Thr might have pathological consequences. As m 3 C 32 can be installed in mt-tRNA Thr in the absence of t 6 A 37 , we explored how this nucleotide substitution affects METTL8-mediated methylation of m 3 C 32 . Both in the context of an unmodified full-length mt-tRNA Thr transcript and a synthetic mt-tRNA Thr ASL containing t 6 A 37 , exchange of A 38 for G strongly reduced METTL8 methylation ( Fig. 5a and Supplementary Fig. 5b), implying that lack of m 3 C 32 in mt-tRNA Thr may also contribute to the disease phenotype observed in patients carrying this mutation.
The sensitivity of the anticodon loop to point mutations was also reflected in the circular dichroism (CD) spectra of mt-tRNA Ser(UCN) ASLs (Fig. 5f), thus supporting the functional findings by structural aspects. Strikingly, introducing the G35A point mutation that prevents METTL8-mediated methylation into the ms 2 i 6 A 37 -modified ASL resulted in a CD signature reminiscent of the unmodified ASL, which is also a poor substrate for METTL8 (Fig. 5f, upper panel). In contrast, the A38C substitution that does not impede m 3 C 32 installation, maintained the overall shape of the ms 2 i 6 A 37 -containing wildtype sequence, which is also efficiently methylated (Fig. 5f, lower  panel). Furthermore, separation of these ASLs by native gel electrophoresis revealed that the ms 2 i 6 A 37 -containing ASLs with either the wild-type sequence or the A38C substitution migrated similarly, whereas the ASL containing the G35A mutation and the unmodified ASL both migrated faster ( Supplementary  Fig. 5c). m 3 C 32 affects the structure of mt-tRNA Thr/Ser(UCN) . Characterisation of METTL8 as the m 3 C methyltransferase responsible for methylating C 32 of mt-tRNA Thr/Ser(UCN) enables the functional relevance of these RNA modifications to be explored. The presence of modified nucleotides in tRNAs can contribute to their stability and achieving correct folding. Therefore, the influence of m 3 C 32 on the stability of mt-tRNA Thr/Ser(UCN) was analysed by northern blotting.
RNAs derived from wild-type and METTL8 KO cells showed no significant differences in the levels of mt-tRNA Thr/Ser(UCN) (Fig. 6a), indicating that the hypomodified mt-tRNAs are not degraded. To explore the possibility that the m 3 C 32 modifications contribute to the folding of mt-tRNA Thr/Ser(UCN) , RNAs derived from the wild-type and METTL8 KO cell lines were separated by native gel electrophoresis. Strikingly, the migration behaviour of both mt-tRNA Thr and mt-tRNA Ser(UCN) from cells lacking METTL8 was altered compared to those from wild-type cells, especially for mt-tRNA Ser(UCN) , whose migration was notably retarded (Fig. 6b). This indicates conformational differences in these mt-tRNAs depending on the presence of m 3 C 32 . To examine in more detail whether structural differences arise within the ASLs and dissect the contributions of the modifications at both positions 32 and 37 make to mt-tRNA Thr/Ser(UCN) structure, the unmodified, m 3 C-, t 6 A/ ms 2 i 6 A-, and m 3 C and t 6 A/ms 2 i 6 A-containing ASLs of these mt-tRNAs were separated on native gels. Portions of both the mt-tRNA Thr/Ser(UCN) ASLs containing m 3 C 32 showed retardation indicated by the presence of slower migrating bands, but predominantly migrated similar to the unmodified ASLs. In contrast, the presence of t 6 A 37 or ms 2 i 6 A 37 alone lead to slower migration behaviour, which was notably reduced by the additional presence of m 3 C 32 (Fig. 6c). The faster migration of the ASLs containing both m 3 C and t 6 A/ ms 2 i 6 A compared to those containing only t 6 A/ms 2 i 6 A is analogous to the behaviour of mt-tRNA Thr/Ser(UCN) from wild-type and METTL8 KO cells, which have the equivalent modification status. Together, these data indicate m 3 C 32 -dependent structural differences within the ASLs of these mt-tRNA Thr/Ser(UCN) .
Earlier studies found that i 6 A and ms 2 i 6 A destabilise base pairing with uridine and, when incorporated into synthetic anticodon hairpins, resulted in lower thermodynamic stability compared to unmodified RNA 61 . Similarly, melting curves of the synthetic mt-tRNA Ser(UCN) ASLs containing (ms 2 )i 6 A 37 showed slightly reduced stability compared to the unmodified ASL, while the insertion of m 3 C 32 in addition to (ms 2 )i 6 A 37 partially rescued the mild effect ( Supplementary Fig. 7a, b). Conformational differences within these ASLs were further investigated by proton nuclear magnetic resonance ( 1 H-NMR) spectroscopy, which reveals hydrogen-bonded base pairs as a fingerprint of imino protons with a chemical shift between 10 and 14 ppm. The imino proton resonances of unmodified, single and double modified mt-tRNA Ser(UCN) ASLs showed distinct chemical shift changes upon introduction of m 3 C 32 (Fig. 6d) as well as variable solvent exchange rates, seen as different extents of line broadening at increasing temperatures ( Supplementary Fig. 7c-e). The central resonances in the fully modified ASL showed narrower line-width and higher intensity, reflecting the unique chemical environments generated by the presence of m 3 C and i 6 A. While i 6 A 37 affects π-π stacking and introduces a bulky hydrophobic substituent that restricts the position of the nucleobase by Van der Waals contacts, the m 3 C 32 modification introduces a positive charge to enhance electrostatic stabilisation, a role that is otherwise taken by metal ions such as Mg 2+ .
Further support of conformational differences in the mt-tRNA ASL as a result of the modification state was obtained from CD spectroscopy (Fig. 6e). The unmodified ASL showed a distinct maximum at 269 nm, which was slightly shifted to 271 nm and reduced to half the intensity due to broadening of the peak and the appearance of a shoulder at 250 nm in the ms 2 i 6 A 37 modified ASL. The double modified m 3 C 32 -and ms 2 i 6 A 37 -containing mt-tRNA Ser(UCN) ASL showed two distinct peaks at 250 and 270 nm. These differences in the ellipticity of CD spectra indicate changes in base stacking. Collectively, these data strongly suggest that m 3 C 32 , both independently and acting in concert with (ms 2 )i 6 A 37 and t 6 A 37 modifications, influences the structure of the anticodon stem loop of mt-tRNA Ser(UCN) and mt-tRNA Thr . mt-tRNA Thr/Ser(UCN) aminoacylation and mitoribosome association are not affected by lack of METTL8 but mitochondrial translation is impaired. The finding that mt-tRNA Thr/Ser(UCN) lacking m 3 C 32 are structurally impaired suggests that their functional capacity may be compromised. The importance of m 3 C 32 of mt-tRNA Thr/Ser(UCN) on the cellular level was first investigated by comparing the growth rate of WT cells and those lacking METTL8. Equal numbers of cells were seeded and counted every 24 h for 72 h during exponential growth. Compared to WT, both KO1 and KO2 cells grew slower (Fig. 7a), indicating a physiological relevance of METTL8-mediated m 3 C 32 of mt-tRNA Thr/Ser(UCN) . Functionally, tRNAs serve as adaptors between mRNA codons and amino acids, and therefore aminoacylation is a key step in achieving functional competence. As some tRNA modifications influence the association and/or activity of aminoacyl-tRNA synthetases 62,63 , the aminoacylation status of mt-tRNA Thr/Ser(UCN) in the presence or absence of METTL8 were determined. RNAs purified from WT and METTL8 KO cells under acidic conditions to retain covalently bound amino acids were either subjected to alkaline conditions to drive deacylation or left untreated. Separation by acidic PAGE in which deacylated tRNAs migrate faster than their aminoacylated counterparts revealed that mt-tRNA Ser(UCN)/Met are almost constitutively aminoacylated in WT cells and only a very minor fraction of mt-tRNA Thr is non-acylated (Fig. 7b). As expected, mt-tRNA Met from METTL8 KO cells migrated as the one obtained from WT cells, both with and without alkaline treatment, consistent with no effect on aminoacylation. In contrast, in the case of mt-tRNA Thr/Ser(UCN) , both the aminoacylated and deacylated mt-tRNAs from METTL8 KO cells migrated differently to those from WT cells (Fig. 7b), reminiscent of the altered migration patterns of these mt-tRNAs observed by native gel electrophoresis (Fig. 6b). However, under acidic conditions both mt-tRNA Thr/Ser(UCN) from METTL8 KO cells were (almost) fully aminoacylated demonstrating that lack of m 3 C 32 and the consequent structural alterations do not impede charging of these mt-tRNAs with the respective amino acids.
Aminoacylated (mt-)tRNAs are recruited to ribosomes during translation so we investigated the influence of lack of m 3 C 32 on the mitoribosome-association of mt-tRNA Thr/Ser(UCN) . Extracts from mitochondria isolated from WT or METTL8 KO cells were separated by sucrose density centrifugation and western blotting was used to identify fractions containing mitoribosomal small subunits (28S), large subunits (39S) and monosomes (55S) (Fig. 7c). Fractions containing either non-mitoribosomeassociated mt-tRNAs (1-2) or mitoribosome-associated  mt-tRNAs (11)(12) were pooled and analysed by northern blotting (Fig. 7d). Similar to the non-m 3 C 32 -containing mt-tRNA Met , mt-tRNA Thr/Ser(UCN) from both WT and METTL8 KO cells were readily detectable in fractions containing 55S monosomes, demonstrating that the lack of m 3 C 32 and the associated conformational changes do not prevent their recruitment to mitoribosomes. Interestingly, compared to mt-tRNA Met , a mild accumulation of mt-tRNA Thr/Ser(UCN) on ribosomes was observed in the METTL8 KO cells (Fig. 7e). The mitochondrial translation machinery is responsible for synthesising 13 mitochondrial-encoded proteins. Production of nascent mitochondrial proteins was therefore monitored in WT, KO1 and KO2 cells using [ 35 S]-methionine incorporation (Fig. 7f). Despite the normal aminoacylation and mitoribosome-association of mt-tRNA Thr/Ser(UCN) in cells lacking METTL8, compared to WT, METTL8 KO cells showed decreases in the levels of the some de novo synthesised mitochondrial proteins under these conditions (Fig. 7g). ND5, COX1, ND2, ND3 and ND4L/ATP8 were most affected by lack of METTL8, while production of CYTB, ND1, COX2/COX3, ATP6 and ND6 was less reduced (Fig. 7g). Overall, these data are in line with the model that m 3 C 32 fine-tunes mt-tRNA Thr/Ser(UCN) structure to support optimal mitochondrial translation.

Discussion
The extensive landscape of modified nucleotides within cellular RNAs expands the functionality of the four basic nucleosides and allows gene expression to be delicately fine-tuned at many levels for optimal accuracy and efficiency. The development of novel RNA modification detection methods has rapidly expanded our  64 . It is important to note that in the study by Zhang et al., METTL8 was expressed with an N-terminal tag, which would impair mitochondrial targeting and consistent with this, we observe that METTL8 lacking its MTS accumulates in nucleoli (Fig. 1b). METTL8 has further been proposed as a regulator of mouse embryonic stem cell differentiation, based on the interaction of an N-terminally tagged version of the protein with the MAPKBP1 mRNA, which encodes a protein component of the c-Jun N-terminal kinase signalling pathway 65 . A potential role for METTL8 in installing m 3 C modifications in cytoplasmic mRNAs is suggested by the lower level of m 3 C detected in mRNAs enriched from METTL8 KO mouse liver or human cell lines compared to control 38 . mRNAs encoding METTL8 isoforms containing the MTS are the most highly expressed in human cells, but expression of an mRNA isoform encoding a protein containing the METTL8 methyltransferase domain but lacking the MTS is also detectable. It is possible therefore that a nonmitochondrial targeted population of METTL8 indeed methylates cytoplasmic mRNAs. Recent transcriptome-wide mapping approaches for m 3 C have, however, failed to detect modification sites outside tRNAs, raising questions about the presence of m 3 C in mRNAs 33,55 .
Here, we characterise METTL8 as the long-sought mitochondrial m 3 C methyltransferase responsible for methylation of C 32 of mt-tRNA Thr and mt-tRNA Ser(UCN) . Evolutionary and functional analysis of yeast m 3 C methyltransferases 30 revealed a gene duplication event in fission yeast that separated the methylation activity of the S. cerevisiae Trm140 enzyme, which targets both tRNA Thr and tRNA Ser , into two distinct enzymes (S. pombe Trm140 and Trm141). In vertebrates, this distinction between the cytoplasmic tRNA Thr and tRNA Ser m 3 C methyltransferases (METTL2A/B and METTL6 respectively) is maintained. Intriguingly, human METTL8, which is most closely related to METTL2A/B, the homologues of S. cerevisiae/S. pombe Trm140 30 , has the ability to recognise and methylate both mt-tRNA Thr and mt-tRNA Ser(UCN) . Both S. cerevisiae Trm140 and METTL2A/B recognise specific nucleotides within the anticodon loop of tRNA Thr , whereas the presence of i 6 A 37 /t 6 A 37 , a distinct variable loop and the seryl-tRNA synthetase all influence m 3 C 32 installation in tRNA Ser by Trm140 or METTL6 30,31,41 . We discover that a U 34 G 35 U 36 t 6 A 37 A 38 motif in the anticodon loop of mt-tRNA Thr is recognised by METTL8 and that U 34 and G 35 within the anticodon loop of mt-tRNA Ser(UCN) are likewise necessary for METTL8 methylation of C 32 . Analogous to Trm140, the presence of (ms 2 )i 6 A 37 dramatically increases the extent of m 3 C 32 installation in the mt-tRNA Ser(UCN) ASL in vitro implying that modification of position 37 may be a pre-requisite for m 3 C 32 to be installed in this tRNA. In contrast, while the presence of t 6 A 37 in mt-tRNA Thr increases methylation of C 32 , METTL8 can bind and modify this mt-tRNA without prior installation of any other modifications. This is consistent with the previous finding that in cells lacking OSGEPL1 m 3 C 32 is reduced but not abolished 57 . These recognition elements within the METTL8 methylation targets are especially important as our unbiased search for RNAs crosslinked to METTL8 in cells and our in vitro analysis of RNA binding by METTL8 show that METTL8 robustly interacts with both substrate and non-substrate RNAs. An induced-fit binding model has been proposed for the Trypanosoma brucei Trm140 homologue (TbTrm140), in which the methyltransferase forms weak/transient interactions with tRNAs and only forms the high-affinity interaction necessary for methylation in the presence of a cofactor [66][67][68] . Our anisotropy data suggest that binding of METTL8 substrate RNAs is largely independent of sequence, secondary structure and the presence of i 6 A 37 , implying that for METTL8, these features are determinants of methylation rather than binding.
In both yeast and humans, the seryl-tRNA synthetase stimulates Trm140/METTL6-mediated methylation of tRNA Ser(UCN)38,41,60 . Furthermore, DALRD3, which acts as an adaptor directing MET-TL2A/B to tRNA Arg(CCU/UCU) , contains an anticodon binding domain present in the arginyl-tRNA synthetase 40 . Although an interaction between METTL8 and the mitochondrial seryl-tRNA synthetase SARS2 has been reported 60 , our data imply that SARS2 is dispensable for METTL8-mediated methylation of C 32 in mt-tRNA Ser(UCN) , highlighting a difference to yeast. This difference may arise due to the distinct ways in which the cytoplasmic and Fig. 7 Effect of m 3 C 32 on mt-tRNA Thr/Ser(UCN) aminoacylation and ribosome-association, and mitochondrial translation. a The numbers of cells in equally seeded cultures of wild-type HEK293 cells (WT) and those lacking METTL8 (KO1 and KO2) were determined over time. Points represent mean total number of cells of n = 3 independent experiments ± standard deviation. Statistical analysis was performed using one-way ANOVA (F = 0.3165, not significant for 24 h; F = 5.6, p < 0.05 for 48 h; F = 7.425, p < 0.05 for 72 h) and significance calculated using Dunnett's multiple comparisons test. b Aminoacylation levels of mt-tRNA Thr/Ser(UCN)/Met in WT and METTL8 KO cells were analysed using acid urea electrophoresis comparing deacylated (alk; alkaline treated) and acylated (aci; acidic conditions) mitochondrial RNA samples. Arrows indicate bands corresponding to the acylated and nonacylated mt-tRNAs for WT (left) and METTL8 KO (right) cells. Representative images of three independent experiments. c-e Mitochondrial extracts from WT and METTL8 KO cells (KO1 and KO2) were separated by sucrose density gradient centrifugation and fractions collected. Proteins in each fraction were analysed by western blotting (c). Fractions 1-2 (non-mitoribosome-associated mt-tRNAs) and fractions 11-12 (55 S monosomes) were pooled and RNAs extracted from these pools were analysed by northern blotting (d). Representative images of three independent experiments. Ratios of non-mitoribosomeassociated (free) and mitoribosome-associated (bound) mt-tRNAs were determined in n = 3 experiments and are shown as mean ± standard deviation (e). Statistical analysis was performed using one-way ANOVA (F = 35.22; p < 0.001 for mt-tRNA Thr and F = 13.54, p < 0.01 for mt-tRNA Ser(UCN) ) and significance calculated using Tukey's multiple comparisons test. f Nascent mitochondrial proteins in WT, KO1 and KO2 cells were labelled with [ 35 S]methionine, separated by denaturing PAGE, transferred to a membrane and detected using a phosphorimager. Tubulin, detected by western blotting, served as a loading control. Representative images of eight independent experiments. g Levels of de novo synthesised mitochondrial proteins were quantified in n = 8 experiments as in (f) and are shown as mean ± standard deviation. Statistical analysis was performed using one-way ANOVA (F = 18.91, p < 0.0001 for ND5; F = 16.59, p < 0.0001 for COX1; F = 7.345, p < 0.01 for CYTB; F = 16.08, p < 0.0001 for ND2; F = 9.867, p < 0.001 for ND1; F = 6.991, p < 0.01 for COX2/COX3; F = 10.6, p < 0.001 for ATP6; F = 4.442, p < 0.05 for ND6; F = 22.45, p < 0.0001 for ND3; F = 22.35, p < 0.0001 for ND4L/ ATP8) and significance calculated using Tukey's multiple comparisons test. p values for data in a, e and g are given in source data; *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001, ns not significant. Source data and p values are provided as a Source Data file. mitochondrial seryl-tRNA synthetases interact with their substrate tRNAs; the cytoplasmic seryl-tRNA synthetases interacts with tRNA Ser via the large variable loop, which is absent in the mitochondrial counterparts and SARS2 instead requires the TΨC and D loops to associate with mt-tRNA Ser(UCN) 69 . Importantly, we observe that aminoacylation of mt-tRNA Ser(UCN) is unaffected by lack of METTL8-mediated m 3 C 32 , implying that functional relevance of the association between METTL8 and SARS2 may be independent of SARS2 catalytic activity. It is possible that the recovery of SARS2 with METTL8, and the very mild decrease in m 3 C 32 observed in SARS2-depleted cells, reflect an indirect role for the synthetase in optimising mt-tRNA Ser(UCN) structure for METTL8-mediated methylation.
While m 3 C 32 modifications are conserved features of eukaryotic (mt-)tRNAs, the precise function of this modification has remained elusive. Strikingly, the migration behaviours of mt-tRNA Thr/Ser(UCN) ASLs on native polyacrylamide gels and thermal melting temperatures vary slightly depending on the modifications at positions A 37 and C 32 , while additional evidence from imino proton NMR and CD spectra further support m 3 C 32dependent alterations in ASL structure. The chemical shift changes of imino protons and ellipticity signatures demonstrated that the modified nucleotides act together to cause distinct alterations in base stacking and electrostatic interactions, which in turn modulate the accessibility of hydrogen bond donors and acceptors to the environment. The migration patterns of fulllength mt-tRNA Thr/Ser(UCN) lacking m 3 C 32 on native and acidic polyacrylamide gels also differ from those of the wild-type mt-tRNAs. As the extent of the migrational shifts observed is different for the two mt-tRNAs, this implies that regions outside the ASL are also structurally affected by the presence or absence of m 3 C 32 , and that the exact composition of nucleotides outside the anticodon loop may influence how these structural changes manifest. A notable difference between mt-tRNA Ser(UCN) and mt-tRNA Thr is the modifications they carry outside the anticodon loop. As modifications that influence RNA folding/stability often function as a co-operative network, it may therefore be that m 3 C 32 differentially affects the structures of these tRNAs due to their different modification content. It is also possible that m 3 C 32 is a pre-requisite for installation of another modification present in mt-tRNA Ser(UCN) but not mt-tRNA Thr , which would be similar to the previous observation that yeast lacking both Trm140 and the guanine dimethyltransferase Trm1 are sensitive to cycloheximide 37 .
On the cellular level, the functional relevance of m 3 C 32 modifications has remained challenging to address. Lack of MET-TL2A/B or METTL6 does not influence polysome formation 38 and while ribosome profiling revealed changes in ribosome occupancy in cells lacking METTL6, these alterations mostly correlated with changes in transcript level, suggesting they arise due to an altered metabolic state of METTL6 KO cells rather than a direct effect of m 3 C 32 on translation 39 . Likewise, Ribo-tRNAseq results indicate no significant differences in translation between wild-type yeast and a Δtrm140 strain 70 . We observe that neither mt-tRNA Thr/Ser(UCN) charging nor recruitment to the ribosome are impaired by lack of METTL8-mediated m 3 C 32 . However, the levels of most nascent mitochondrial-encoded proteins are mildly affected by loss of METTL8, implying a role for m 3 C 32 in mt-tRNA Thr/Ser(UCN) during mitochondrial translation. Consistent with this, and the finding that mt-tRNA Thr/ Ser(UCN) mildly accumulate on mitoribosomes in the absence of METTL8, a parallel study revealed the enrichment of codons recognised by mt-tRNA Ser(UCN) in ribosome profiling data from METTL8 KO cells 71 . Together, these data suggest that the presence of m 3 C 32 fine-tunes mt-tRNA Thr/Ser(UCN) structure to optimise its functionality.
As mitochondria are the hubs of many metabolic processes and responsible for most cellular energy production, even very mild effects on mt-tRNA structure and mitochondrial translation can have profound effects on the cellular level. Impaired t 6 A 37 and m 3 C 32 modification of mt-tRNA Thr in cells caused by exchange of A 38 of mt-tRNA Thr for G is associated with mitochondrial myopathy 57 (Supplementary Fig. 5b). Interestingly, METTL8 expression is upregulated in cancer 72 , when the demand for mitochondrial function is increased. Quantitative analysis of m 3 C 32 levels in mt-tRNA Thr/Ser(UCN) and our reconstitution of METTL8 methylation on cellular RNAs indicate that these modifications are not fully stoichiometric, raising the possibility that mitochondrial translation could be regulated by differentially modified mt-tRNA Thr/Ser(UCN) . m 3 C in RNAs can be actively demethylated by ALKBH3 or ALKBH1 73,74 . However, ALKBH3 is not present in mitochondria 45 and no changes in m 3 C 32 levels in mt-tRNA Thr/Ser(UCN) were observed in in vitro demethylation assays using purified ALKBH1 and m 3 C-containing synthetic anticodon stem loops (Supplementary Fig. 8). Therefore, if the levels of m 3 C 32 vary in different conditions to dynamically finetune mitochondrial translation, regulation likely occurs via differential expression and/or methylation by METTL8.
In summary, this work firmly establishes the role of the mitochondrial methyltransferase METTL8 in the biosynthesis of fully functional mt-RNA Thr/Ser(UCN) . Several independent lines of evidence reveal the molecular requirements for METTL8 targets and identify a role of m 3 C 32 in fine-tuning tRNA structure and function in mitochondrial translation.

Methods
Molecular cloning. The METTL8 coding sequence (CDS; NM_024770.5) and a truncated version lacking the 5′ 60 nucleotides were amplified using primers listed in Supplementary Table 6, and cloned into pcDNA5 vectors for inducible expression of proteins with a C-terminal GFP or His 6 -2xFLAG (Hexahistidine-PreScission protease cleavage site-2x FLAG) tag in HEK293 T-Rex cells (Supplementary Table 7). For CRISPR/Cas9 constructs, guide sequences (Supplementary Table 6) were cloned into the PX459 plasmid pSpCas9(BB)-2A-Puro (Addgene #62988) using the BbsI restriction sites as previously described 75 . For recombinant expression of His 14 -MBP-METTL8 in E. coli, the CDS was cloned into a pQE80derivative vector 76 and site-directed mutagenesis (Supplementary Table 6) was employed to introduce mutations to induce substitution of aspartates 230 and 309 with alanine. Plasmids containing tRNA sequences including a 3′ CCA tail were generated by cloning a recursive PCR product generated using four overlapping oligonucleotides (Supplementary Table 6) into a pQE-derivative vector 77 (Supplementary Table 7). All constructs were verified by Sanger sequencing (Eurofins Genomics).
Human cell culture. HEK293 Flp-In T-Rex cells (ThermoFisher Scientific) were cultured at 37°C in 5% CO 2 in DMEM (Gibco) supplemented with 10% foetal calf serum (Merck) and 1% penicillin/streptomycin (Gibco) according to standard protocols. Cell lines for expression of C-terminally His 6 -2xFLAG or GFP tagged METTL8 or METTL8 21 -407 were generated by transfection of the pcDNA5-base constructs (Supplementary Table 7) into HEK293 Flp-In T-Rex cells using X-tremeGENE 9 DNA Transfection Reagent (Roche) according to the manufacturer's instructions. Stably transfected cells were selected using hygromycin (100 µg/ml) and blasticidin (10 µg/ml) and expression of the transgene was induced by addition of 1 µg/ml tetracycline for 24 h before harvesting. Human METTL8 knock-out cell lines were generated by CRISPR/Cas9-mediated mutagenesis. Briefly, HEK293 Flp-In T-Rex cells were transfected with 1 µg of PX459-based plasmid (Supplementary Table 7) and transfectants were selected with puromycin (1 μg/ml). Genomic cleavage efficiency was assessed using the GeneArt ® Genomic Cleavage Detection Kit (ThermoFisher Scientific) according to the manufacturer's instructions. Following selection, cells were seeded at single cell density in 96-well plates and cultured to generate a clonal population. Genomic DNA was extracted using PureLink Genomic DNA kit (ThermoFisher Scientific) according to the manufacturer's instructions. The target region of the genome in each clone was PCR-amplified using primers listed in Supplementary Table 6 and genomic mutations were detected by sequencing. Lack of the METTL8 protein was analysed by western blotting using antibodies listed in Supplementary Table 8.
RNAi. Cells were seeded at a density of 3 × 10 6 per 10 cm dish and transfected on the following day with siRNAs (50 nM; Supplementary Table 9) using Lipofectamine RNAiMAX reagent (ThermoFisher Scientific) according to the manufacturer's instructions. Cells transfected with siRNAs against TRIT1 were harvested 72 h after transfection whereas those transfected with siRNAs against OSGEPL1 were serially transfected twice for 72 h. Proteins extracted were used for western blotting and extracted RNAs were analysed by primer extension.
Cell counting by analytical flow cytometry. Viable cells were counted by trypan blue staining in a hemocytometer chamber. Increasing cell concentrations were measured for 10 s at medium flow using a FACSCanto II to generate a standard growth curve. Cells were seeded at a density of 7.5 × 10 4 cells in a 12 well dish and, at 24 h intervals over a 72 h period, cells were harvested by trypsinisation, transferred to flow cytometry tubes, washed with PBS and counted. Data acquisition and gating were performed using FACS Diva software (version 6.1.1) and data were exported using FlowCore (version 2.6.0). Total numbers of cells per well were extrapolated from the average from three technical measurements of three biological replicates (see Supplementary methods for further details).
Fluorescence microscopy. HEK293 cells for expression of METTL8-GFP or METTL 21-407 -GFP were treated with 1 µg/ml of tetracycline for 24 h. Cells were treated with MitoTracker ® Deep Red FM. (ThermoFisher Scientific) in media without FCS for 30 min at 37°C, washed twice in PBS and fixed with 4% formaldehyde in PBS for 10 min at room temperature. Cells were mounted on coverslips using Vectashield ® (Vector labs) and fluorophores were visualised using a Nikon Ti-EEclipse inverted/UltraVIEW VoX spinning disc confocal microscope.
Isolation of mitochondria and protease protection assays. HEK293 cells for expression of METTL8-His 6 -2xFLAG and METTL 21-407 -His 6 -2xFLAG were treated for 24 h with 1 µg/ml of tetracycline to induce protein expression and mitochondria were isolated 51 . Cells were resuspended in homogenisation buffer (300 mM trehalose, 10 mM KCl, 10 mM HEPES-KOH pH 7.4, 1 mM PMSF and 0.2% BSA) and homogenised using a Homogenplus Homogeniser (Schuett-Biotec). Differential centrifugation was performed and mitochondria were pelleted by centrifugation at 11,000 × g for 10 min. After washing with homogenisation buffer, mitochondria were either resuspended in homogenisation buffer or in swelling buffer (10 mM HEPES-KOH pH 7.4, 1 mM EDTA) with a final protein concentration of 1 μg/μl. After 15 min incubation on ice, samples were treated with Proteinase K for 15 min. Protease digestion was arrested by addition of 2 mM PMSF and samples were analysed by western blotting using antibodies listed in Supplementary Table 8. To determine whether proteins are integrated into, or associated with, the IMM, mitochondria isolated from cells expressing METTL8-His 6 -2xFLAG were resuspended in 0.1 M Na 2 CO 3 at pH 10.5, 11.5 and 12.5, then incubated on ice for 20 min. Samples were centrifuged at 100,000 × g for 60 min at 4°C and supernatant and pellet fractions were analysed by western blotting.
Total RNA extraction and small RNA enrichment. Total RNA and total mitochondrial RNA were isolated from cells or isolated mitochondria using TRI Reagent® (Sigma-Aldrich) according to the manufacturer's instructions. Small RNAs (<200 nt) were enriched from 100 μg of total RNA using the mirVana™ miRNA isolation kit (ThermoFisher Scientific) following the manufacturer's instructions. RNA concentrations and purity were measured on a Nanodrop One c (ThermoFisher Scientific).
Denaturing polyacrylamide gel electrophoresis and northern blotting. RNA for northern blotting were separated on 10% denaturing (7 M urea) polyacrylamide gels and transferred to Hybond-N+ membranes (cytiva). RNAs were chemically crosslinked with N-(3-dimethylaminopropyl)-N′-ethylcarbodiimide hydrochloride and 1-methylimidazole (Sigma-Aldrich) for 2 h at 60°C 78 . Membranes were incubated with pre-hybridisation buffer (250 mM sodium phosphate pH 7.4, 7% SDS and 1 mM EDTA pH 8.0) for 30 min and hybridised with [ 32 P]-labelled DNA oligonucleotides antisense to target RNAs (Supplementary Table 6) overnight at 37°C in the same buffer. Membranes were then washed with 6× saline sodium citrate (SSC) buffer and 2× SSC buffer supplemented with 0.1% SDS each for 30 min at 37°C. Membranes were exposed to phosphorimager screens and signals detected using a Typhoon FLA 9500 (cytiva). The acquired images were analysed and quantified using the Image Studio 5.2.5 software (LI-COR).
Native polyacrylamide gel electrophoresis. Two hundred nanograms extracted small RNAs or 4 pmol synthetic ASLs containing different modified nucleotides were resuspend in 10 mM Tris pH 7.4 and 50% glycerol, supplemented with bromophenol blue, and separated on 20% non-denaturing polyacrylamide gels. RNAs were visualised using sybr gold staining or transferred to Hybond-N + membranes (cytiva) and further analysed by northern blotting.
Aminoacylation. For acid-urea electrophoresis, total mitochondrial RNAs were extracted under acidic conditions using 10 mM NaOAc (pH 5) at 4°C 79 . To distinguish uncharged from charged tRNAs, a sample of the extracted mitochondrial RNAs was deacylated by heating at 37°C for 60 min at pH 8.9. 2 μg of total mitochondrial RNAs deacylated or stored in acidic conditions (10 mM NaOAc at pH 5 and 1 mM EDTA at pH 8) were separated in an acidic (pH 5) 10% polyacrylamide 7 M urea gel at 4°C for 20 h at 170 V. Analysis of mt-tRNA aminoacylation status was done by northern blotting.
Protein-RNA crosslinking experiments. For UV crosslinking and analysis of cDNA (CRAC) 52,53,80 experiments, HEK293 cells for expression of His 6 -2xFLAGtagged METTL8 or the His 6 -2xFLAG tag were treated with 1 µg/ml of tetracycline for 24 h to induced protein expression then UV (254 nm) crosslinked 3× at 800 mJ/cm 2 using a Stratalinker (Stratagene). Cells were lysed in a buffer containing 50 mM Tris/ HCl pH 7.6, 150 mM NaCl, 0.1% NP-40, 5 mM β-mercaptoethanol and protease inhibitors (Roche). Protein-RNA complexes were affinity purified in non-denaturing conditions using anti-FLAG magnetic beads (Sigma-Aldrich) and eluted overnight with 3X FLAG Peptide (Sigma-Aldrich). RNAs were partially digested with 1 unit of RNAce-IT (Agilent) for 30 s at 37°C and complexes were immobilised on Ni-NTA (Qiagen) in denaturing conditions (6 M guanidium-HCL). Co-purified RNAs were 5′ labelled with [ 32 P] and ligated to 3′ and 5′ adaptors. The 5′ adaptor contained a unique molecular identified (UMI; NNNNNAGC) to allow consolidation of sequencing reads derived from the same RNA template during bioinformatic analyses. After elution, protein-RNA complexes were separated by NuPAGE (ThermoFisher Scientific), transferred to a nitrocellulose membrane and exposed to an X-ray film. Relevant areas of the membrane were excised and RNAs released by Proteinase K (Roche) treatment. Isolated RNAs were reverse transcribed with Superscript III (ThermoFisher Scientific), and cDNA libraries were amplified by PCR and subjected to Illumina sequencing.
Bioinformatic analyses were based on the pyCRAC approach (version 1.4.5; https://git.ecdf.ed.ac.uk/sgrannem/pycrac/; see Supplementary methods for details 81 ). In brief, the obtained sequencing reads were first subjected to barcode removal, then the reads were trimmed of the 3′ adaptor and quality filtered. Reads containing identical sequences and the same UMI were collapsed. Sequencing reads were aligned to the human genome ensemble (GRCh38.p13 Ensembl release 104) and for visualisation of aligned sequencing reads, bedgraph files were generated. Individual nucleotide substitutions within mapped reads were counted using a custom script (see Supplementary methods). Aligned reads were counted and count matrices generated. Read counts were normalised using the TMM method and fold changes in the numbers of reads mapping per transcript were calculated between the His 6 -2x-FLAG and METTL8-His 6 -2xFLAG libraries.
Alternatively, crosslinked protein-RNA complexes from cells as above were isolated by tandem affinity purifications on anti-FLAG magnetic beads and Ni-NTA under native and denaturing conditions respectively. Untrimmed, co-purified RNAs were eluted from the Ni-NTA by Proteinase K treatment, isolated and analysed by northern blotting.
Recombinant expression of proteins. For recombinant expression of proteins in Escherichia coli Rosetta 2 (DE3), cells containing necessary plasmids (Supplementary Table 7) were grown in LB media with appropriate antibiotic selection until OD 600 of 0.6 was reached. Protein expression was induced by addition of 1 mM IPTG for 16 h at 18°C. Cells were harvested by centrifugation at 6000 × g for 20 min at 4°C and washed with PBS. Cell pellets were resuspended in Lysis buffer (30 mM phosphate buffer pH 7.3, 300 mM KCl, 10% (v/v) glycerol, 5 mM imidazole, 0.1 mM DTT) and lysed using an Emulsiflex-C3 (Avestin). The lysate was cleared by centrifugation at 30 000 × g for 20 min twice and the clear lysate was incubated with cOmplete His-Tag Purification Resin (Roche In vitro transcription of RNA. For in vitro transcription, 50 μl reactions containing 1 μg PCR products (generated using plasmid templates (Supplementary Table 7) and primers listed in Supplementary Table 6), 2 mM NTPs (ATP, UTP, GTP, CTP; ThermoFisher Scientific), T7 RNA polymerase (ThermoFisher Scientific), 1× transcription buffer (ThermoFisher Scientific) and 1 U/μl Ribolock (ThermoFisher Scientific) were incubated for 1 h at 37°C. After transcription, samples were treated with Turbo DNase (ThermoFisher Scientific) and purified over Quick Spin RNA columns (Roche), according to the manufacturer's instructions. The RNA concentration was determined using a Nanodrop One c (ThermoFisher Scientific).
Chemical synthesis of RNAs containing modified nucleotides. RNA oligonucleotides were prepared by solid-phase synthesis using appropriately protected nucleoside phosphoramidites (structures shown in Supplementary Fig. 3). The 5′-O-DMT-2′-O-TBDMS i 6 A 3′-CEP and the 5′-O-DMT-2′-O-TOM m 3 C 3′-CEP building blocks were synthesised from 6-chloropurine riboside and uridine, respectively, as previously described 82 Figs. 9 and 10). Briefly, TBDMS-protected threonine p-nitrophenylethylester was coupled to the N 6 -amino group of persilylated adenosine via activation with 1-N-methyl-3-phenoxycarbonyl-imidazolium chloride, in analogy to previous reports 83,84 . The nucleobase modifications in ms 2 i 6 A were introduced by sequential substitutions on peracetylated 6-chloro-2aminopurine riboside. Diazotation of the amino group and displacement with methylsulfide 85 was followed by substitution of chloride by isopentenyl amine 83 . The final phosphoramidite building blocks were used as 100 mM solutions in dry acetonitrile together with 0.25 M ethylthiotetrazole as activator and a coupling time of 12 min. Detritylation, capping and oxidation were performed under standard conditions. Cleavage from solid support and alkaline deprotection was specific for each type of modified nucleotide: ASLs containing m 3 C 32 were treated with 25% NH 4 Table 10) was added. The reactions were incubated at 22°C for 2 h before addition of 1 μg/μl of Proteinase K (Roche) at 22°C for 30 min. RNAs were ethanol precipitated and resuspended in nuclease-free water (QIAGEN). Half the sample was separated by denaturing (7 M urea) polyacrylamide (10% for full-length tRNA transcripts, 15% for synthesised ASLs) gel electrophoresis. The gel was stained with ethidium bromide and RNAs visualised using UV light. After fixing, the gel was immersed in Amplify solution (cytiva) for 1 h and dried before exposure to an X-ray film for 24 h to 2 weeks at −80°C. The remainder of the reaction was analysed by scintillation counting by mixing with 4 ml of scintillation liquid (Carl Roth) and measuring in a Hidex 300SL (HIDEX).
For non-radioactive RNA methylation, 1 μM of recombinant His 14 -MBP-METTL8 was mixed with 1.6 mM SAM (NEB) and used to methylate 1 μg of small RNA extracted from human cells using the same protocol as above. After Proteinase K treatment, the RNA was recovered using RNA Clean and Concentrator columns (Zymo Research) and the methylation was analysed using primer extension.  Table 10) in the same buffer were incubated at room temperature for 5 min. Anisotropy measurements were performed at 22°C using a FluoroMax-4 spectrofluorometer (Horiba Scientific). The fluorophore was excited at 495 nm and emission was measured at 517 nm. Binding affinity (A T or total anisotropy) was calculated using Eq. (1) where A R is the anisotropy for unbound (free) RNA, A PR is the anisotropy of the complex of protein-RNA, and [protein] tot and [RNA] tot are the total concentrations of protein and RNA, respectively.
De novo synthesis of [ 35 S]-methionine labelled mitochondrial-encoded proteins. To analyse nascent protein synthesis, cells were depleted of methionine and cytoplasmic translation was inhibited in WT and METTL8 KO cells by treatment with 100 μg/ml emetine for 10 min. Cells were then incubated in media supplemented with 0.2 mCi/ml [ 35 S]-methionine (Hartmann Analytic) for 1 h. Proteins were extracted, separated by denaturing PAGE and [ 35 S]-labelled proteins were detected using a phosphorimager.
Imino proton NMR spectroscopy. Samples for NMR spectroscopy were prepared from 36 nmol of RNA oligonucleotides (concentrated and purified by precipitation from acetone/LiClO 4 , 2% w/v), in a final volume of 180 µL containing 10 mM sodium phosphate buffer pH 7.0, 100 mM NaCl in H 2 O/D 2 O 9/1. The solution was heated to 90°C for 2 min and annealed for 15 min on ice and transferred to a 3 mm NMR tube. As internal standard, 0.4 µL DSS (sodium trimethylsilylpropanesulfonate) was added. 1 H NMR spectra with water suppression by excitation sculpting were recorded on a Bruker Avance III HD 600 MHz spectrometer at 283, 290, 298, and 306 K.
Circular dichroism spectroscopy. CD spectra were recorded on a JASCO Spectropolarimeter J-715 equipped with a Xenon lamp and power supply PS-150J, in quartz cuvettes with 1 cm path lengths at ambient temperature, with a band width of 10 nm and a scanning speed of 500 nm/min. The ASL RNA samples (5 µM) were annealed in 10 mM sodium phosphate buffer (pH 7.0) and supplemented with 1 mM MgCl 2 . All data were baseline corrected using a control containing buffer only. Spectra are shown as the average of duplicate data collection.
Data normalisation and statistical analyses. All statistical analyses and plotted graphs were generated using the GraphPad Prism software version 9. Where applicable, error bars represent standard deviation and dots represent individual data points. At least three independent replicates were performed for each experiment where statistical analyses were conducted. Statistical analyses were performed using one-way ANOVA for groups of 3 or more and significance was calculated using Tukey's multiple comparisons test. To compare between two groups, statistical significance was calculated using a two-tailed unpaired Student's t test. For quantification of northern and western blot band signal intensities, data were normalised with respect to the signal intensity of the loading controls detected on the same blot. For quantification of scintillation counting, primer extension, northern blot signal intensities and [ 35 S]-labelling experiments the values from each set of experimental replicates were normalised by mean.
Synthetic methods. Details of the chemical synthesis are provided in Supplementary methods.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The data that support this study are available from the corresponding authors upon reasonable request. The CRAC datasets for METTL8-His 6 -2xFLAG and His 6 -2xFLAG generated in this study, and their analyses are deposited in Gene Expression Omnibus