L1 retrotransposons exploit RNA m6A modification as an evolutionary driving force

L1 retrotransposons can pose a threat to genome integrity. The host has evolved to restrict L1 replication. However, mechanisms underlying L1 propagation out of the host surveillance remains unclear. Here, we propose an evolutionary survival strategy of L1, which exploits RNA m6A modification. We discover that m6A ‘writer’ METTL3 facilitates L1 retrotransposition, whereas m6A ‘eraser’ ALKBH5 suppresses it. The essential m6A cluster that is located on L1 5′ UTR serves as a docking site for eukaryotic initiation factor 3 (eIF3), enhances translational efficiency and promotes the formation of L1 ribonucleoprotein. Furthermore, through the comparative analysis of human- and primate-specific L1 lineages, we find that the most functional m6A motif-containing L1s have been positively selected and became a distinctive feature of evolutionarily young L1s. Thus, our findings demonstrate that L1 retrotransposons hijack the RNA m6A modification system for their successful replication.

L ong interspersed element-1 (L1) is currently an active autonomous retrotransposon, and constitutes~17% of the human genome 1 . The average human genome contains 80-100 copies of retrotransposition-competent L1s 2,3 . A retrotransposition-competent L1 is 6 kb in length and consists of a 5′ untranslated region (UTR) containing an internal promoter 4 , two open reading frames (ORF1 and ORF2), and a short 3′ UTR. ORF1 encodes a nucleic acid chaperon protein (ORF1p) 5 , while ORF2 encodes a protein with endonuclease and reverse transcriptase activity (ORF2p) 6,7 . ORF1p and ORF2p associate preferentially with their parental mRNA to form an L1 ribonucleoprotein (RNP) particle 8 . The L1 RNP enters the nucleus and then generate the progeny through de novo insertion of its cDNA 9,10 . The mobility of L1s contributed to a source of genetic variation, but also pose a threat to genome integrity [11][12][13] . Although several host factors have evolved to suppress L1 retrotransposition, the youngest L1 subfamilies are still active and replicated continuously [14][15][16] . To date, however, the mechanism of how L1s have propagated under host surveillance remains unknown.
N6-methyladenosine (m 6 A) is the most prevalent internal modification in eukaryotic mRNAs, which determines RNA function and fate 17 . Several enzymes dynamically process the m 6 A modification of mRNA. The methyltransferase-like enzyme METTL3, which is the catalytic subunit of the RNA methyltransferase complex, adds m 6 A at the consensus motif DRA m CH (where D = G/A/U, R = G/A, and H = U/C/A) 18,19 . Conversely, m 6 A is removed by the demethylases α-ketoglutarate-dependent dioxygenase AlkB homolog 5 (ALKBH5) or fat mass and obesityassociated protein (FTO) 20,21 . Emerging studies have revealed that m 6 A modifications in viral transcripts affect the gene expression and replication of viruses such as HIV-1 22 . Despite the critical role of m 6 A in pathogenic viral transcripts, it remains unclear whether m 6 A participates in the regulation of the endogenous parasites, L1 retrotransposons.
Here, we show that L1 retrotransposon exploits m 6 A modification to facilitate its mobility. We figured out that m 6 A machinery plays a role in L1 regulation and identified the functional m 6 A cluster located on 5′ UTR of retrotranspositioncompetent full-length L1. Our results show that L1 5′ UTR m 6 A cluster recruits eukaryotic initiation factor 3 (eIF3) for efficient translation and promotes the formation of L1 RNP, which are essential for L1 mobility. Lastly, we traced a recent episode of human-and primate-specific L1 evolution and revealed that the most functional m 6 A site (A332 residue in L1 5′ UTR) first appeared~12 million years ago. During the primate evolution, A332 m 6 A-positive L1s have been selected and became a distinctive feature of evolutionarily young L1s, which suggests that the acquisition of m 6 A motif has acted as an evolutionary driving force for L1 retrotransposons.

Results
METTL3 and ALKBH5 regulate L1 retrotransposition. To determine whether RNA m 6 A modification affects L1 retrotransposition, we evaluated the effects of the RNA m 6 A machinery on L1 retrotransposition using a cell-based engineered L1-reporter assay 23 . For the assay, we used the pJJ101-L1-dn6 2.2 construct (hereafter referred to as pL1Hs) that contains a blasticidin S deaminase gene (mblastI) within the 3′ UTR antisense to the SV40 promoter 24,25 (Fig. 1a). When L1 is successfully integrated into the host chromosome, the cells acquire resistance to blasticidin (Fig. 1a).
We depleted the m 6 A methyltransferase METTL3, RNA demethylase ALKBH5, and FTO using small-interfering RNA (siRNAs) in HeLa cells and transfected pL1Hs vector. In METTL3-depleted cells, the number of blasticidin S-resistant colonies, which represent successful L1 retrotransposition, was reduced by >2-fold compared to that of control siRNA ( Fig. 1b  and Fig. S1a). Conversely, the silencing of ALKBH5 increased L1 mobility, while the silencing of FTO did not affect L1 retrotransposition (Fig. 1b). The depletion of the m 6 A machinery did not vitiate cell viability (Fig. S1b). In a reciprocal experiment, we performed an L1 retrotransposition assay with the ectopic expression of RNA m 6 A demethylase ALKBH5 or FTO. Notably, the overexpression of ALKBH5 inhibited L1 mobility by~4-fold, whereas FTO overexpression did not affect L1 mobility compared to that in AcGFP-expressing negative control cells ( Fig. 1c and Fig. S1c). We hypothesized that ALKBH5 may function as an L1 restriction factor by removing essential m 6 A for L1 mobility. To examine whether the enzymatic function of ALKBH5 is critical for L1 mobility suppression, we performed L1 assays using the plasmid-encoding catalytically inactive mutant of ALKBH5 (ALKBH5 H204A ). As anticipated, ALKBH5 could successfully restrained L1 mobility to levels that were comparable to that suppressed by a reverse transcription inhibitor (stavudine; d4T), whereas ALKBH5 H204A overexpression did not result in the restriction of L1 mobility ( Fig. 1d and Fig. S1d). The viability of transfected cells remained unaffected (Fig. S1e).
The pL1Hs plasmid encodes reporter L1 downstream of the CMV promoter and L1 5′ UTR promoter. Since the presence of the CMV promoter might affect L1-associated m 6 A modification, we used a pYX014 L1-luciferase vector driven only by the L1 5′ UTR promoter. Using pYX014, the firefly luciferase reporter within the 3′ UTR allowed us to assess L1 mobility by measuring luminescence as previously reported 26 (Fig. S1f). Overexpression of ALKBH5 impaired L1 retrotransposition, regardless of the presence of the CMV promoter (Fig. S1g). In-line with this result, depletion of METTL3 or ALKBH5 regulates L1 mobility, whereas FTO knockdown did not affect (Fig. S1h). These results indicate that ALKBH5-specific m 6 A substrates are necessary for L1 expansion. To summarize, our data support the functional role of the m 6 A machinery in regulating L1 retrotransposition.
RNA m 6 A metabolism regulates gene expression at posttranscriptional levels. Therefore, we speculated that the m 6 A machinery would influence the protein expression of L1. Immunoblot analysis of HeLa cells devoid of m 6 A enzymes revealed that m 6 A enzymes regulate the expression of ORF1p (Fig. 1e). Overexpression of ALKBH5 inhibited ORF1p expression, while the ectopic overexpression of FTO and ALKBH5 H204A did not affect the ORF1p expression ( Fig. 1f, g). In each condition, the transfection efficiency of pL1Hs was not affected by siRNA or plasmids transfection (Fig. S2a, b). Furthermore, neither the depletion of RNA m 6 A machinery nor the overexpression of ALKBH5 altered the levels of expression of the control EGFP (Fig. S2c, d), which indicates that m 6 A enzymes do not affect transfection efficiency. These results suggest that m 6 Amediated L1 regulation affects both retrotransposition and L1 protein expression. L1 RNA is modified by m 6 A. Although the possibility of L1 m 6 A modification was demonstrated in recent studies 27,28 , it remains unclear whether m 6 A modification occurs in retrotransposition-competent full-length L1, and if so, which region of the L1 transcript is modified by m 6 A. To validate whether m 6 A modifies L1 RNA, we performed methyl-RNA immunoprecipitation (MeRIP) using human embryonic stem cells (H9 hESCs) that express endogenous L1 at sufficient levels 29,30 . Through qRT-PCR analysis of the MeRIP eluates, we detected the enrichment of L1 RNA at a level comparable to that for known m 6 A-modified SON and CREBBP mRNA, but much more than negative control HPRT1 mRNA (Fig. 2a). To minimize bias resulting from primers in L1 RNA detection, we used three different primer sets that targeted the 5′ UTR, ORF1, and ORF2 regions and did not observe significant differences in the results obtained for these primers (Fig. 2a). Similar to the endogenously expressed L1 RNA in hESCs, MeRIP-qPCR analysis clearly demonstrated that the L1 RNA exogenously expressed in HeLa cells undergoes m 6 A modification (Fig. S3a). We then evaluated if the silencing of METTL3 or ALKBH5 would alter the extent of m 6 A modification of the L1 RNA. Indeed, MeRIP-qPCR with METTL3-depleted cells revealed lower enrichment of the m 6 A-modified L1 than of siCtrltreated cells, whereas ALKBH5 knockdown augmented the levels of m 6 A-positive L1 (Fig. 2b). These results indicate that METTL3 can install m 6 A modification in L1 transcripts, while ALKBH5 plays a role in removing the modification.
To examine the m 6 A-modified regions in the L1 transcripts, we analyzed the m 6 A transcriptome of hESCs reported previously 31 and mapped reads to the consensus sequence of L1Hs, the youngest L1 32 . We identified 18 peaks across the L1Hs sequence using two biological replicates (Fig. 2c). Given that the reads from L1s may yield false-positive results, we narrowed down and selected the peaks that are likely to contain m 6 A motifs from 18 peaks through the m 6 A prediction score algorithm (SRAMP) 33 . SRAMP analysis revealed that the 9 peaks found in the ORF1, ORF2, and 3′ UTR regions do not contain m 6 A motifs, and that only the 6 peaks located at 5′ UTR have potential m 6 A motifs (Table S1a).
Next, we mapped the sites of m 6 A modifications in reporter L1-transfected HeLa cells using MeRIP-seq. Consistent with findings from previous studies 18 found in 3′ UTR and CDS, but not in the 5′ UTR (Fig. S3b). By mapping the reads on reporter L1, we obtained five candidate peaks (Fig. S3c), and further sorted according to the approach based on the m 6 A prediction as described above. All five peaks were classified as m 6 A-putative regions with high scores ( Table S1b). Two of the featured peaks were located in the 5′ UTR, the other two were located in the ORF1, and another was in ORF2. The m 6 A modification sites commonly detected in endogenous and exogenous L1 RNA are A332 and A839, both located in the 5′ UTR (Table S1). This is a notable phenomenon since m 6 A modification typically occurs near the stop codon and at the 3′ UTR, and this gives rise to the possibility that the L1 5′ UTR acts as the regulatory hub for L1 mobility via m 6 A modification.
5′ UTR m 6 A cluster is critical for L1 activity. Given that the L1 5′ UTR has a potential m 6 A cluster, we next examined whether the L1 5′ UTR is necessary for m 6 A-dependent L1 regulation. We transfected the 5′ UTR-deleted pL1Hs (pL1Hs Δ5′ UTR) into HeLa cells treated with m 6 A machinery-targeting siRNA and monitored ORF1p expression. Intriguingly, the knockdown of m 6 A enzymes did not affect ORF1p expression in the absence of 5′ UTR (Fig. S4a). Furthermore, using the codonoptimized synthetic L1 construct that encoded ORFs with the same amino acids yet different nucleotide sequences, we examined whether alterations in ORF1 and ORF2 nucleotide sequences could affect m 6 A machinery-mediated L1 regulation.
Remarkably, silencing of METTL3 or ALKBH5 regulates L1 ORF1p expression only when 5′ UTR is contained in synthetic L1, which indicates that m 6 A machinery regulates L1 expression in a 5′ UTR-dependent manner (Fig. S4b). These results suggest that the L1 5′ UTR contains functional m 6 A motifs for successful ORF1p expression.
To identify the site of functional m 6 A in L1 5′ UTR, we selected six adenosine candidates of m 6 A modification (332, 495, 569, 600, 679, and 839, numbering based on L1PA1 consensus sequence 32 ) through MeRIP-seq analysis in either hESCs or L1-reporter-expressing HeLa cells (Table S1). We generated a set of firefly luciferase reporter plasmids encoding L1 5′ UTR or its m 6 A-silencing A to T mutants (Fig. 3a). To quantify the effect of L1 5′ UTR m 6 A mutation without the bias from transfection efficiency, we normalized the firefly luciferase activity to that of Renilla luciferase. The dualluciferase reporter assay revealed that a single A332T, A495T, or A600T mutation reduced the expression of firefly luciferase, compared to that of native 5′ UTR (Fig. 3b). However, the weak effect of these single mutants led us to hypothesize that multiple m 6 A modifications may function synergistically. Indeed, the double mutation of A332/600T and the triple mutation of A332/495/600T exerted significantly more synergistic and potent effects (Fig. 3c).
We next performed the L1 retrotransposition assay using the 5′ UTR m 6 A mutants of the pL1Hs construct. Mutations at each m 6 A motif of A332, A495, and A600 showed a marginal effect on L1 retrotransposition, whereas A332/A495/A600 triple mutation (hereinafter referred to as pL1 m 6 A mut) markedly inhibited L1 mobility ( Fig. 3d and Fig. S5a-c). We validated the effect of the m 6 A cluster using the L1-luciferase reporter construct pYX014. Indeed, the triple m 6 A mutant of the L1-luciferase construct (pYX014 L1 m 6 A mut) induced approximately 50% decline in L1 mobility compared to that induced by the wild-type L1 (Fig. S5d).
To assess the effect of the triple mutation in the m 6 A modification level of L1, we performed MeRIP-qPCR for comparing m 6 A enrichments between cells that expressed pL1Hs and pL1 m 6 A mut. Surprisingly, the triple mutation reduced the enrichment of m 6 A-modified L1 by~50%, while it did not affect the m 6 A levels of the endogenous controls SON and CREBBP ( Fig. 3e and Fig. S5e). These results indicate that A332, A495, and A600 are the essential adenosines for L1 mobility and serve as m 6 A modification sites.
Based on our finding that ALKBH5 inhibits L1 mobility, we attempted to determine whether ALKBH5 could restrict the mobility of the L1 m 6 A mutant. L1 assays with co-transfection of pL1 vectors and FH-ALKBH5 revealed that the ectopic expression of ALKBH5 impaired the retrotransposition of pL1Hs (Fig. 3f, and Figs. S1e, 5f). However, ALKBH5 overexpression caused only marginal effects in pL1 m 6 A mut-expressing cells (Fig. 3f). Moreover, silencing the triple m 6 A modification led to the suppression of L1 mobility in AcGFP-expressing cells, but not in ALKBH5-expressing cells (Fig. 3f). In a reciprocal experiment, we measured the L1 retrotransposition frequency of pL1Hs and pL1 m 6 A mut in ALKBH5-lacking cells. Notably, ALKBH5 knockdown led to the enhancement of L1 mobility in pL1Hs-expressing cells, whereas no measurable changes were observed in pL1 m 6 A mut-expressing cells (Fig. 3g, and Fig. S5g, h). Consistent with this result, ALKBH5 was not able to suppress L1 ORF1p expression in the absence of the m 6 A cluster (Fig. S5i, j). In summary, we demonstrated that ALKBH5 suppresses L1 expression in the 5′ UTR m 6 A cluster-dependent manner, which suggests that the L1 5′ UTR m 6 As serve as the substrates for ALKBH5 demethylation. (n = 3 independent samples, mean ± s.e.m., one-way ANOVA and Dunnett's multiple comparisons test). b MeRIP-qPCR analysis of pL1Hstransfected HeLa cells with m 6 A machinery knockdown. Eluted RNAs were quantified using primers specific for reporter L1. The enrichment of RNA was normalized to that of the control. (n = 5 independent samples, mean ± s.e.m., unpaired two-tailed t-test, *p < 0.05). c Map of m 6 A modification sites in full-length L1Hs from previously reported MeRIP-seq data for H1 hESCs (GSE52600). Read coverage was normalized to the total number of reads mapped to the L1Hs consensus sequence. The plot presents data from MeRIP-seq in red and input RNA-seq in blue. Bars (in red or black) indicate the m 6 A peaks identified by manual inspection in two replicates. m 6 A peaks in red correspond to peaks containing high score m 6 Aprediction sites. Source data are provided as a Source data file. m 6 A modification promotes the translational efficiency of L1 RNA. Given that m 6 A regulates L1 ORF1p expression, we investigated the stages in the L1 replication cycle that are regulated by m 6 A modification. First, we quantified L1 RNA expression in the presence or absence of the 5′ UTR m 6 A cluster using two different plasmids, pL1Hs and pYX014. Irrespective of the vectors used, L1 m 6 A mutation did not influence the levels of L1 RNA expression through northern blot and qRT-PCR ( Fig. S6a-c). We next assessed the stability of reporter L1 mRNAs with or without the 5′ UTR m 6 A mutation using the transcription inhibitor, actinomycin D. L1 RNA was more stable in both pL1Hs-and pL1 m 6 A mut-expressing HeLa cells when compared to positive control, cMYC mRNA (Fig. S6d). We did not observe any significant difference in L1 RNA stability by m 6 A mutation (Fig. S6d). We next examined the distribution of reporter L1 mRNAs in the nuclear and cytoplasmic fractions. In comparison to that of GAPDH (abundant in the cytoplasm) and MALAT1 (abundant in the nucleus), over 80% of the L1 mRNA was present in the cytoplasmic fraction and the m 6 A-deficient mutation did not affect the cellular localization of L1 RNA (Fig. S6e). Several recent studies have linked 5′ UTR m 6 A modification to translational efficiency in the context of cellular stress [34][35][36] . Besides, a previous study raised the possibility that the presence of the L1 5′ UTR determines the quality of L1 RNA 37 . Therefore, we reasoned that the L1 5′ UTR m 6 A cluster could modulate the translation of L1 RNA. To test this hypothesis, we performed an immunoblot assay in HeLa cells that expressed a single to triple m 6 A mutant of the pL1 construct. The expression levels of ORF1p gradually decreased as the number of mutations increased ( Fig. 4a and Fig. S7a). In addition, through polysome profiling, we captured polysome-bound RNA to assess the translational efficiency of L1 RNA. The deletion of the m 6 A cluster significantly reduced the enrichment of polysome-bound L1 RNA compared to that of pL1Hs ( Fig. 4b and Fig. S7b). To validate these results, we investigated whether m 6 A regulates the translational efficiency of endogenous L1 mRNAs in PA-1 human embryonic carcinoma cells. Consistent with the effects of m 6 A machinery depletion in pL1Hs-expressing HeLa cells (Fig. 1e), ALKBH5 knockdown augmented the production of endogenous ORF1p while METTL3 knockdown reduced ORF1p synthesis ( Fig. 4c and Fig. S7c). The comparable levels of L1 mRNA in PA-1 cells with or without ALKBH5 depletion suggests that the enhanced production of ORF1p is a consequence of translational upregulation (Fig. S7d). Consistent with this result, the levels of polysome-associated L1 RNA substantially increased in ALKBH5-depleted PA-1 cells in comparison to the control cells ( Fig. 4d and Fig. S7e), which indicates that ALKBH5 regulates L1 retrotransposition by suppressing the efficiency of L1 RNA translation. . Firefly luciferase luminescence reflected the effect of 5′ UTR and of its mutations. b, c Dual-luciferase assay using HeLa cells transfected with pFR-L1Hs 5′ UTR or its A to T m 6 A-abrogating mutant. The ratio of the luminescence of firefly and Renilla luciferase (Fluc/Rluc) was normalized to pFR-L1Hs 5′ UTR-expressing cells. (mean ± s.d., four (b) or five (c) independent samples). d L1 assays using the triple m 6 A mutated L1 construct (pL1 m 6 A mut) in HeLa cells (n = 3 independent samples, mean ± s.d.). e MeRIP-qPCR analysis for evaluating the effect of the triple m 6 A mutation construct (pL1 m 6 A mut). m 6 A antibody-bound L1 RNA was normalized to that of pL1Hs-transfected cells. (n = 3 independent samples, mean ± s.e.m.) f, g Retrotransposition assay using pL1Hs-or pL1 m 6 A mut-expressing HeLa cells ALKBH5 overexpression (f) or silencing (g) (n = 3 independent samples, mean ± s.d.). In b-g, Statistical significance was calculated by one-way ANOVA with Dunnett's (b), Tukey's multiple comparisons test (c, f, g), and unpaired two-tailed t-test (d, e) (****p < 0.0001, ***p < 0.001, and **p < 0.01, *p < 0.05, ns not significant). Source data are provided as a Source data file. eIF3 is an m 6 A-binding protein and promotes the selective translation of mRNAs that bear m 6 A in 5′ UTR 34 . These characteristics of eIF3 lead us to hypothesize that the L1 5′ UTR m 6 A cluster serves as a docking site for eIF3 to promote translation. To define the functional relationship between eIF3 and the L1 m 6 A cluster, we analyzed previously reported data from photoactivatable ribonucleoside-enhanced crosslinking and immunoprecipitation sequencing (PAR-CLIP seq) of eIF3 subunits a, b, d, and g 38 . By mapping the reads from PAR-CLIP of the eIF3 subunits along the endogenous L1Hs, we revealed that eIF3 exhibits preferential binding to the L1 5′ UTR (Fig. S8a). Furthermore, the PAR-CLIP clusters were significantly enriched in the A332 m 6 A region in all four eIF3 subunits, while the A495 m 6 A region contained PAR-CLIP clusters of three eIF3 subunits: eIF3a, d, and g (Fig. 4e). We were unable to detect the comparable eIF3-binding sites in the A600 m 6 A region (Fig. 4e). To verify the interaction between eIF3 and the L1 m 6 A cluster, we transfected pL1Hs or pL1 m 6 A mut into HeLa cells and performed UV crosslinking immunoprecipitation using eIF3b antibody (Fig. S8b). Through RT-qPCR analysis of the . e Identification of eIF3-binding sites in L1Hs 5′ UTR using the previously reported eIF3 PAR-CLIP data set (GSE65004) 38 . The red boxes indicate the m 6 A sites-containing region. f eIF3 UV-CLIP-qPCR using pL1Hs-or pL1 m 6 A mut-expressing HeLa cells. IgG-IP and PSMB6 served as negative controls (n = 4 independent samples, mean ± s.e.m., unpaired two-tailed t-test; *p < 0.05). Source data are provided as a Source data file.
immunoprecipitated eluates, we observed the enrichment of L1 RNA comparable to c-JUN, a known eIF3-bound mRNA, in pL1Hs-expressing cells (Fig. 4f). PSMB6 and eluates from IgG immunoprecipitation served as negative controls. Remarkably, the silencing of the m 6 A cluster reduced the quantity of eIF3bound L1 RNA by~70%, which indicates that the L1 5′ UTR m 6 A cluster bears the eIF3 docking site (Fig. 4f). Indeed, eIF3 knockdown suppressed endogenous L1 ORF1p expression in PA-1 cells and L1 retrotransposition in HeLa cells (Fig. S8c, d).
Since another m 6 A binding protein, YTHDF1, regulates translation efficiency of m 6 A-modified RNA and interacts with eIF3 39 , we tested whether YTHDF1 also binds to L1 5′ UTR m 6 A cluster. Through RNA immunoprecipitation and qPCR analysis, we confirmed that L1 RNA associates with YTHDF1 and another YTH protein, YTHDF2, (Fig. S8e, f). However, 5′ UTR m 6 A cluster mutation did not impair interaction between YTHDFs and L1 RNA (Fig. S8f). These data suggest that YTHDFs bind to L1 RNA via m 6 A in region other than 5′ UTR. Collectively, the 5′ UTR m 6 A cluster specifically recruits eIF3 for the efficient translation of L1 RNA.
5′ UTR m 6 A cluster is necessary to produce a functional unit for L1 retrotransposition. For successful L1 retrotransposition, both ORF1p and ORF2p are required to generate the L1 RNP with the encoding L1 RNA 40 . Though we observed m 6 Amediated regulation in ORF1p synthesis, it is necessary to determine whether m 6 A modification at the 5′ UTR influences L1 RNP formation. To investigate L1 RNP regulation by m 6 A, we obtained the cellular RNP fraction as previously reported 41 . Briefly, we prepared lysates from pL1-transfected cells and purified L1 RNPs using sucrose cushion ultracentrifugation (Fig. 5a). We detected comparable levels of L1 RNA in the RNP fractions from pL1Hs-and pL1 m 6 A mut-expressing cells ( Fig. 5b and  Fig. S9a). cDNA synthesis reaction in absence of reverse transcriptase revealed that neither genomic DNA nor plasmid contamination was present in the RNP fraction (Fig. 5b). Immunoblotting of the RNP fraction showed that the levels of RNP-associated ORF1p were diminished by L1 5′ UTR m 6 A mutation (Fig. 5c). This indicates that the m 6 A cluster mutation abolished the sufficient production of ORF1p for L1 RNP formation.
Since ORF2p expression level is too low to observe changes in the m 6 A mutant 42,43 , we introduced the L1 element amplification protocol (LEAP) to gauge the reverse transcriptase activity of ORF2p 41 (Fig. 5a). Incubation of RNPs with LEAP primer facilitates ORF2p-mediated L1 cDNA synthesis. We amplified LEAP products using PCR with specific primers for reporter L1 and RACE adapter, which yielded products of 300-400 base pairs (bp) (Fig. 5d). However, m 6 A-abrogated L1 RNP produced cDNA at significantly lower levels than the wild-type L1 RNP did (Fig. 5d). These results reveal that the m 6 A cluster is necessary for L1 cDNA production, which suggests that the m 6 A cluster regulates ORF2p expression or its activity.
ORF1p oligomerization is critical for successful L1 retrotransposition 44 . We examined whether inefficient ORF1p synthesis results in a failure of L1 RNP formation. For a quantitative assessment of individual L1 RNP formation, we introduced the pAD3TE1 construct carrying T7-tagged ORF1p and MS2-stemloop structures in the L1 3′ UTR (Fig. 5e). We performed RNA fluorescence in situ hybridization (FISH) with fluorescent Q670labeled probes complementary to the linker regions between the MS2 loops and immunofluorescence experiments with anti-T7 antibody ( Fig. 5f and Fig. S9b). Through z-stack analysis, we obtained the coordinates for the fluorescent signals of L1 RNA and ORF1p and identified the L1 RNPs by sorting out colocalizing particles within an intermolecular distance of 330 nm between L1 RNA and ORF1p. Consistent with the previous study 45 , we observed colocalizing signals of L1 RNP as cytoplasmic aggregates (Fig. 5f). However, L1 m 6 A mutexpressing cells showed a significant reduction in both the number of L1 RNP foci and the signal intensity of colocalizing ORF1p (Fig. 5f-h). These data indicate that the abrogation of the m 6 A cluster reduces the levels of ORF1p in L1 RNP and causes a concomitant decrease in the number of L1 RNP particles. m 6 A is a driving force for L1 evolution. Over the last 40 million years of human evolution, L1 subfamilies have frequently acquired novel 5′ UTRs 32 . Since a new L1 lineage will emerge only through its successful replication, the genetic novelty that promotes L1 mobility must remain preserved in the genomic fossils of L1s 46 . Considering that RNA methyltransferase installs m 6 A in a sequence-specific manner, we speculated that nucleotide mutations might lead to the acquisition or loss of the m 6 A consensus motif during L1 evolution. To unravel the evolutionary history of L1 5′ UTR m 6 A cluster regions, we analyzed 443 human-specific full-length L1s 47 and compared the three m 6 A motif sites, A332, A495, and A600. Given that adenosine residue should be followed by cytosine residue to form the m 6 A consensus motif DRA m CH, A332 m 6 A-positive L1s constitute a considerably small part in the L1PA3 lineage (12.4%). In L1PA2 and younger lineages, the number of A332 m 6 A positive L1s increased drastically (92.9%) (Fig. 6a), and the same was observed in the youngest L1Hs (Fig. S10a). On the contrary, A495 and A600 are tightly conserved in all human-specific L1 subfamilies (Fig. S10b). We investigated this tendency of the A332 m 6 A motif in L1s of chimpanzee and gorilla, which share L1PA2 and L1PA3 lineages with humans. Comparative analysis of chimpanzee-or gorilla-specific full-length L1s revealed the seismic shift toward the population of A332 m 6 A positive L1s, while the chimpanzeeor gorilla-specific L1s continue to harbor the m 6 A motifs of A495 and A600 (Fig. 6a, and Fig. S10c-f). As in the L1Hs subfamily, the majority of the youngest chimpanzee-specific L1 subfamily (L1Pt) harbor the A332 m 6 A motif (Fig. S10C). In summary, we found that A332 m 6 A motif acquisition by single nucleotide substitution (T333C) first appeared in L1PA3 or older lineages, which indicates that the productive potential of m 6 A has allowed positive selection of A332 m 6 A-positive L1s during the evolution from the common ancestor.
To evaluate the consequence of A332 m 6 A acquisition in ancestral L1 5′ UTR, we generated a chimeric pL1 construct that contained L1PA2 5′ UTR and L1Hs ORF1/2 with the mblastI reporter (Fig. 6b). Based on the m 6 A consensus motif DRA m CH, T333 of pL1PA2 5′UTR does not allow m 6 A modification at A332, whereas T333C point mutation enables A332 m 6 A modification (Fig. 6b). The retrotransposition assay revealed that T333C mutation enhanced the mobility of pL1PA2 5′ UTR , while the mutagenesis control (T333G) did not exert the same effect ( Fig. 6c and Fig. S11a, b). As expected, the T333C m 6 A-gain mutation enhanced ORF1p synthesis of pL1PA2 5′ UTR (Fig. 6d). Although the acquisition of A332 m 6 A motif only led to a 1.4-fold increase in the cultured cell-based L1 retrotransposition assays (Fig. 6c), the 12 million years of L1 evolution would have been sufficient to amplify the profound effect of m 6 A. These results suggest that m 6 A modification in the L1 5' UTR region may have played a crucial role in the L1 evolution of primates.

Discussion
The role of m 6 A modification in pathogenic viral transcripts has been reported in the past decade 48  study, we demonstrated that the proper formation of the m 6 A cluster in 5′ UTR of L1 RNA is essential for L1 retrotransposition. The evolutionary history of the m 6 A cluster in primate-specific L1s revealed the most influential m 6 A region (A332) that was obtained in the past 12 million years. This suggests the potential role of m 6 A as a driving force in L1 evolution (Fig. 7).
Two recent studies have revealed that the m 6 A modification decreases the stability of L1 RNA with respect to R-loop or chromatin regulation 27,28 . However, our results revealed that the L1 5′ UTR m 6 A cluster did not affect RNA stability but promoted translation. Abakir et al., and Liu et al. observed the role of m 6 A in genome-wide L1 repetitive elements, which are mostly inactive by 5′ truncations or inversions 3,49 . Considering that our study focused on the functions of m 6 A in the replication cycle of retrotransposition-competent L1s, which have intact 5′ UTR, this difference in the scope of L1 RNA types may contribute to the discrepancy.
Indeed, m 6 A enzymes regulate L1 expression only when L1 contains its 5′ UTR. The presence of 5′ UTR in L1 transcripts affects retrotransposition efficiency 37    characteristics of L1 5′ UTR that is lengthy, GC rich, and exhibits promoter activity, its regulatory function at the posttranscriptional level has posed a long-standing question. Our findings demonstrated that L1 5′ UTR m 6 A modification is essential for L1 translation, L1 RNP formation, and thus retrotransposition. Therefore, we provide a new perspective on the regulatory function of L1 5′ UTR as a hub for RNA modification. We demonstrated that m 6 A promotes not only ORF1p production via enhancing the translational efficiency, but also L1 cDNA synthesis. Since ORF2p can proceed reverse transcription regardless of association with ORF1p 41,45 , it remains to clarify whether m 6 A modification upregulates ORF2p translation or m 6 A-modified L1 RNA indirectly influences reverse transcriptase activity of ORF2p. The unconventional translational mechanism of ORF2p, which relies on the translation of the upstream ORF 50 , suggests that enhanced ORF1p translation rates by m 6 A cluster successively stimulate ORF2p synthesis. In addition, m 6 A modifications could alter RNA-protein interactome 51,52 or RNA secondary structure 53 , which might affect L1 ORF2p enzymatic activity. Therefore, future studies could reveal the role of m 6 A in ORF2p regulation. By adopting a microscopic approach, we confirmed that m 6 A is critical for the formation of L1 RNP aggregates. The rate of ORF1p oligomerization is the limiting factor in the production of successful L1 RNPs 44 . Therefore, we speculated that 5′ UTR m 6 As enable L1 RNA to produce sufficient ORF1p, which further accelerates the oligomerization of ORF1p. Since the process of L1 RNP formation is more complicated than the biochemical interaction between L1 RNA and its protein, the process by which m 6 A orchestrates the assembly of retrotransposition-competent L1 RNP remains to be understood. eIF3 recognizes an m 6 A residue in the 5′ UTR and promotes the translation of mRNAs 34 . We assumed that the L1 5′ UTR recruits eIF3 to the m 6 A cluster for efficiently translating the L1 mRNA. Indeed, eIF3 PAR-CLIP-seq data reveal the interaction between eIF3 and L1 5′ UTR m 6 A residue. We also demonstrated that the eIF3-bound portion of L1 decreases in the absence of the 5′ UTR m 6 A. A single m 6 A residue is sufficient to induce eIF3mediated translation 34 . This could explain the synergetic effects of triple m 6 A residues in L1 5′ UTR, which suggests that each m 6 A residue can serve as a docking site for eIF3. Moreover, under cellular stress, the 5′ UTR m 6 A facilitates the cap-independent translation of mRNA [34][35][36] . These studies raise the possibility that m 6   Phylogenetic tree of gorilla, chimpanzee, and human L1s with predicted age and the corresponding L1 subfamily lineages (left). Changes in the A332-m 6 A motif region from L1PA3 or older L1s to L1PA2 and a younger L1 (right). The substitution site wherein the residue converts from T to C (333) is highlighted in yellow. The percentage indicates the proportion of m 6 A motif-positive L1s with nucleotide C to total L1s. b A schematic of retrotransposition assay using pL1PA2 5′ UTR construct that is generated by substituting 5′ UTR of pL1Hs with A332 m 6 A negative 5′ UTR of L1PA2. A schematic of T333C m 6 A acquisition mutagenesis in L1 5′ UTR 328-336 region was indicated in red. c Retrotransposition assays for assessing the effect of A332 m 6 A acquisition in pL1PA2 5′ UTR with T333C mutation. T333G mutation served as negative control (n = 3 independent samples, mean ± s.d., one-way ANOVA and Tukey's multiple comparison test; **p < 0.01). d Immunoblot assay showing L1 ORF1p expression in the indicated pL1-transfected HeLa cells. HSP70 served as a loading control. The immunoblot images are representative of three independent experiments. Source data are provided as a Source data file.
is important to determine whether m 6 A modification enables the cap-independent translation of L1 and whether m 6 A acts as a molecular switch for L1 expression under cellular stress. L1s have been continuously active since the origin of mammals 55 . One of the previous studies on L1 evolution revealed that several distinct L1 lineages coexisted and were in a simultaneously activated state in the ancestral primate genome. However, since the emergence of the L1PA lineage, the L1 subfamily has evolved and maintained itself as a single lineage in the last 25 million years of the evolution of human and its close relatives 32 . The study proposed that the competition between or coexistence of L1 lineages is determined by the status of the 5′ UTR of L1s and acquisition of novel 5′ UTR is a fundamental feature in mammalian L1 evolution 32 . Given that m 6 A methyltransferase marks m 6 A in a sequence-specific manner, the accumulation of mutations in L1 might cause the loss or acquisition of putative m 6 A motifs. To further elucidate the history of m 6 A in L1 evolution, we analyzed species-specific full-length L1s from the human, chimpanzee, and gorilla genome. Notably, the A332 m 6 A motif first appeared in L1PA3 or older L1 lineages more than 12 million years ago. During the evolution of the three different primates, humans, chimpanzees, and gorillas, the A332 m 6 A-positive L1s have propagated their progenies and have become the dominant L1 subfamilies. As m 6 A modification promotes L1 mobility, the acquisition of m 6 A would have resulted in the positive selection of A332 m 6 A-containing L1s. Over the extended periods of L1 evolution, L1s have competed for survival against host restriction 15,16,32 . KRAB-zinc finger proteins (ZFP), which have evolved with L1s, suppress the old L1 transcription in a sequencespecific manner 15,16 . However, L1Hs, which is the youngest L1 lineage in the human genome, escapes KRAB-ZFP restriction and is not recognized by any KRAB-ZFPs 16 . Instead, the host defense utilizes post-transcriptional suppression mechanisms, such as small RNA interference (e.g., piRNA) or APOBECs, to restrict the replication of L1s 56 ; however, the youngest L1s are still active. Our findings provide clues on how the youngest L1s continuously replicate under host surveillance. The emergence and the propagation of the A332 m 6 A-positive L1s suggest that 5′ UTR m 6 A modification was a countermeasure against the host posttranscriptional restriction.
Plasmids. The FLAG-HA-pcDNA3.1-derived plasmids used in this study were named using FH as a prefix with the respective protein names specified. AcGFP, ALKBH5, and FTO cDNA were cloned into FLAG-HA-pcDNA3.1 (Addgene, 52535) for overexpressing N-terminally FLAG-HA-tagged protein. FH-based plasmids were generated by restriction enzyme cloning using XbaI and PmeI (NEB). The site-directed mutagenesis of FH-ALKBH5 to catalytically inactive mutant (H204A) construct was performed using the Phusion High-Fidelity polymerase kit (Thermo Fisher Scientific).
pAD3TE1 is an L1.3 plasmid containing the T7 gene 10 epitope tag on the carboxyl-terminus of ORF1p, TAP tag on the carboxyl-terminus of ORF2p, and 24 copies of the MS2 loop repeat in the 3′ UTR 45 . pAD3TE1 was gift from Aurélien J. Doucet. The generation of L1 5′ UTR m 6 A mutant constructs of pAD3TE1 was performed according to the method for pL1Hs mutant construct generation.
L1-firefly luciferase-tagged plasmids pYX014 and pYX015 were gifts from W. An 26 . pYX014 encodes L1 constructs under the L1 native 5′ UTR promoter. pYX015 carries a retrotransposition-defective mutation in L1 ORF1. pYX014 and pYX015 plasmids contain a Renilla luciferase cassette to normalize transfection efficiency levels. To generate pYX014 L1Hs and m 6 A mut constructs, NotI-PmlI fragments that were 2166 bp in length, including those spanning from the L1 5′ UTR to the forepart of ORF2 in pL1Hs and pL1 m 6 A mut, were subcloned into pYX014 via restriction enzyme cloning.
pDEST HA-derived plasmids were named using HA as a prefix with the respective proteins, YTHDF1 and YTHDF2. YTHDF1 and YTHDF2 cDNA were cloned into pDEST HA vector using pENTR/D-TOPO vector (Invitrogen) and Gateway® LR Clonase® II Enzyme mix (Invitrogen). HA tag sequence is located in 5′ end of insert for overexpressing N-terminally HA-tagged protein. Full-length young L1s RNA have m 6 A in 5′ UTR A332 residue (lower part of scheme, m 6 A in red circle and m 6 A-gain mutation at T333C in red). 5′ UTR m 6 A cluster recruits eIF3 complex (green) and promotes translation of L1 ORF1p (yellow). Increased L1 ORF1p synthesis leads efficient production of L1 ribonucleotide particle (RNP) formation with its parental mRNA (line with poly A) and the reverse transcriptase, L1 ORF2p (magenta). The L1 RNP enters the nucleus and then generate the progeny through insertion of its cDNA. Old L1s with no A332 m 6 A motif have lower efficiency of translation and replication than those of A332 m 6 A-positive L1s (upper part of scheme). Since the A332 m 6 A motif first appeared~12 million years ago, m 6 A-stimulated L1 replication has allowed m 6 A-positive L1s to survive during evolution, but made old L1s out of competition.
RNA extraction and RT-qPCR. Total RNA was extracted using TRIzol reagent (Invitrogen) according to the manufacturer's instructions. For the removal of genomic or plasmid DNA, total RNA was treated with recombinant DNase I (Takara) for 1 h at 37°C, followed by purification using the NucleoSpin RNA Clean-up kit (Macherey-Nagel). cDNA synthesis was performed using the ReverTra Ace qPCR RT Kit (Toyobo) according to the recommended protocol. TOPreal TM qPCR 2X PreMIX (Enzynomics) was used for subsequent qPCR reactions. The qPCR primers used in this study are listed in Supplementary  Table 2.
Polysome fractionation. Ten milliliter of 10-50% linear sucrose gradients in base solution (100 mM NaCl, 20 mM Tris pH 7.5, 10 mM MgCl 2 , 100 μg/ml of cycloheximide) were prepared a day before polysome fractionation. For polysome fixation, 7-10 × 10 6 cells were incubated for 10 min in a media containing 100 μg/ ml of cycloheximide at 37°C and were collected by scrapping with PBS containing 100 μg/ml of cycloheximide. After centrifugation at 1200 × g, 4°C for 5 min, the cell pellets were lysed in 100 μl of polysome extraction buffer (20 mM Tris pH 7.5, 100 mM KCl, 5 mM MgCl 2 , 0.5% NP-40) supplemented with RNase inhibitor (Enzynomics), protease/phosphatase inhibitor cocktail (Cell signaling), and 1 mM DTT. The cells were incubated in the buffer for 10 min at 4°C and centrifuged at 12,000 × g, 4°C for 10 min to remove debris and nuclei. The protein concentration was measured using the Pierce BCA Protein Assay kit (Thermo Fisher Scientific). Five hundred to 600 μg of the lysate was introduced at the top of the linear sucrose gradient and centrifuged at 222,000 × g, 4°C for 2 h using the SW41Ti rotor of the Beckman ultracentrifuge. Fifty microgram of the lysate was saved as input RNA. After centrifugation, 1 ml fractions were collected from the top to the bottom of the gradient using the BioLogic LP system and fraction collector (BioRad) with UV absorbance at 260 nm. Next, 250 μl of each fraction was mixed with 750 μl of TRIzol LS reagent (Invitrogen) and 20 ng of spike-in RNA (synthesized firefly luciferase mRNA). RNA extraction and qPCR were performed as described above. The levels of RNA in each fraction were normalized to those of spike-in RNA and input RNA.
L1 mblastI retrotransposition assay. HeLa cells were plated at 8 × 10 4 cells in 12well plates. After 18 h, the cells were transfected with L1 plasmid (pJJ101-L1-dn6 2.2; pL1Hs) at 800 ng per well using Lipofectamine 3000 (Invitrogen) following manufacturer's instructions. Two days later, 200 μg/ml hygromycin B (Invitrogen) was added to the media to select the transfected cells. Cell selection continued for 4 days, and the hygromycin B-resistant cells were reseeded at 2.5 × 10 4 per well in a 6-well plate. The next day, blasticidin S (Invitrogen) was added to a final concentration of 8 μg/ml and the cells were cultured for 7-9 days in its presence. The colonies were stained with crystal violet and counted using Colony, version 1.1 (Fujifilm). Retrotransposition assays were performed using RNA interference targeted toward METTL3, ALKBH5, and FTO with slight modifications in the process described above. For this, 6 × 10 4 HeLa cells were seeded into 12-well plates with 40 nM siRNA-Dharmafect1 (Dharmacon) mixture. After 24 h, the cultures were divided equally and plated into two wells in 12-well plates. The next day, the cells were transfected with pL1Hs at 500 ng per well. Four days after transfection, the cells were plated at 6 × 10 4 cells per well in 6-well plates and selected using 8 μg/ml of blasticidin S. Retrotransposition assays with the overexpression of AcGFP, ALKBH5, and FTO were performed as described above. Briefly, HeLa cells were transfected 500 ng FH-plasmid and 700 ng of L1 plasmid, and 4 days after transfection, 6 × 10 4 cells were reseeded into a well in a 6-well plate and selected after treatment with blasticidin S for 7-9 days.
Luciferase assay. HeLa cells were plated at 8 × 10 4 cells per well in 12-well plates. The next day, the cells were transfected with 800 ng per well of the pFR vector (pCMV-L1 5′ UTR-firefly luciferase) using Lipofectamine 3000 (Invitrogen). Two days later, the transfected cells were harvested and luminescence was measured using the Dual-Luciferase Reporter Assay System (Promega) according to manufacturer's instruction. Briefly, 250 μl of passive lysis buffer was used to lyse cells in each well in 12-well plates. Next, 20 μl of the lysate was mixed with 100 μl of the Luciferase Assay Reagent II, and the luminescence of firefly luciferase was measured using a microplate luminometer (BERTHOLD). Renilla luciferase activity was subsequently measured after administering 100 μl Stop & Glo Reagent.
Crosslinking immunoprecipitation and qPCR (CLIP-qPCR). eIF3-RNA CLIP-qPCR was performed as described previously 34 with some modifications. For each experiment, 1.2 × 10 6 HeLa cells were plated on two 100 mm dishes each. The next day, the cells were transfected with 6 μg of L1 plasmid per dish using Lipofectamine 3000 (Invitrogen). Two days later, the cells were washed twice with cold PBS, and allowed to form UV crosslinks on ice under 150 kJ/cm 2 of UV 254 nm light (XL-1500, Spectrolinker). The cells were scraped and transferred to PBS and pelleted by centrifugation at 1000 × g, 4°C for 3 min. The pellets were resuspended in 1 ml of lysis buffer (50 mM Tris pH 7.5, 100 mM NaCl, 1% NP-40, 0.1% SDS, 0.5% sodium deoxycholate, 1× cOmplete protease inhibitor cocktail, 1 mM DTT, 80 unit/ml RNase inhibitor). The lysate was passed through a 21 G needle ten times and shock-frozen using liquid nitrogen. The lysate was thawed on ice and centrifuged at 15,000 × g for 15 min.  Table 3. For MeRIP-qPCR analyses, we followed this procedure, except the poly (A) + RNA fragmentation step. Eluted RNA was purified using the NucleoSpin RNA Clean-up kit (Macherey-Nagel), and subjected to cDNA synthesis. For MeRIP-seq analysis, the adapters were trimmed using Cutadapt 60 (cutadapt -g TACAGTCCGACGATC -A TGGAATTCTCGGGTGCCAAGG). The 3′ and 5′ adapter sequences in the first and second read in a pair (owing to the short insert size) were further trimmed and the read pairs with either reads < 18 bp were discarded. The remaining reads were then aligned to the combined human genome (hg19), and reporter L1 (pL1Hs) sequence using Spliced Transcripts Alignment to a Reference (STAR) 61 and peak calling was performed using MACS2 62 . For analyzing the m 6 A modifications in endogenous L1, the sequence reads from human embryonic stem cells were retrieved 31 (accession code: GSE52600) and NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-21197-1 ARTICLE were aligned against L1Hs consensus sequence using STAR. The codes are available from https://github.com/hastj7373/merip-seq.
RNA FISH and immunofluorescence. The L1-MS2-stem-loop constructs pAD3TE1 L1Hs and pAD3TE1 L1 m 6 A mut were transfected into HeLa cells. The following day, the cells were reseeded on sterile coverslips where 200 μg/ml hygromycin B was added for selection. After 3 days, the cells were fixed with 3.7% formaldehyde in PBS for 10 min, and the fixation was quenched by adding 0.1 M glycine in PBS for 10 min. The fixed cells were permeabilized in 70% ethanol for at least 3 h to 1 week at 4°C. The cells were then rehydrated with PBS for 30 min and incubated in a prehybridization solution (10% formamide, 2× SSC solution) for 30 min at 37°C. Hybridization was performed overnight at 37°C in 50 μL of hybridization solution containing 10% formamide, 2× SSC, 10% dextran sulfate, 50 μg yeast tRNA, 0.2% BSA, 0.1 M DTT, 50 units RNase inhibitor (Enzynomics), and 10 ng of MS2-Q670 probe (generously provided by Hye Yoon Park 63 ; listed below) at 37°C. Next, the cells were washed twice with a prehybridization solution for 30 min. For the immunofluorescence experiment, the hybridized cells were incubated in the blocking solution (10% formamide, 2× SSC, 0.2% BSA) for 1 h, followed by incubation with anti-T7 primary antibody (ab9138, Abcam) diluted in the blocking solution (1:200) for 2 h. The cells were washed twice with the prehybridization solution for 15 min and incubated with FITC-conjugated anti-goat secondary antibody (305-095-047, Jackson immunoresearch) diluted in blocking solution (1:200) for 1 h. The cells were washed twice as described above, and the coverslips were mounted on slide glasses using the Vectashield antifade medium with DAPI (Vector Laboratories). The samples were imaged using an inverted microscope Nikon Eclipse Ti2 equipped with a 1.45 numerical aperture Plan apo λ ×100 oil objective and a sCMOS camera (Photometrics prime 95B 25 mm). For each field of view, stacks of images of 6 μm were captured at every 0.3 μm in the DAPI395, GFP488, and Alexa647 channels using the NIS-Elements software.
Co-localization analysis of RNA FISH and IFA microscope image. Binary masks of cells were generated using the ROI manager in ImageJ 57 . Protein and mRNA particles from z-stack images were detected using the TrackNTrace software 64 . After the detection of particles, the protein-mRNA pairs with an intermolecular distance of 330 nm (three pixels) were considered as colocalizing pairs. The intensities of proteins colocalizing with mRNA were determined based on the amplitude of the fitted 2D Gaussian function from the TrackNTrace software.
LEAP assay. The LEAP assay was performed as described previously 41 . Briefly, HeLa cells were plated (4 × 10 6 cells in 60 mm dishes); the following day, the cells were transfected with 3 μg of L1 plasmid (pL1Hs) using Lipofectamine 3000 (Invitrogen). After 48 h, 200 μg/ml hygromycin B was added to the media to select the cells carrying the L1 plasmid. After 2 days of selection, the cells were lysed with CHAPS lysis buffer (10 mM Tris pH 7.5, 0.5% CHAPS (w/v), 1 mM MgCl 2 , 1 mM EGTA, and 10% glycerol) supplemented with 1 mM DTT and the cOmplete protease inhibitor cocktail (Roche) and cleared by centrifugation (4°C, 20,000 × g for 15 min). The cleared lysates were loaded on a sucrose cushion (20 mM Tris pH 7.5, 80 mM NaCl, 8 mM MgCl 2 , 1 mM DTT, 1× protease inhibitor, 4 ml of 8.5% sucrose (from the top) and 6 ml of 17% sucrose (from the bottom) solutions) in 13.2 ml Ultra-Clear tubes (Beckman Coulter) and centrifuged at 178,000 × g, 4°C for 2 h in a SW41Ti rotor of Beckman ultracentrifuge. The colorless pellets were suspended by pipetting in 100 μL of RNase-free water supplemented with 1× protease inhibitors. Pierce BCA Protein Assay (Thermo fisher Scientific) was conducted to determine the protein concentration. Three microgram of the RNP samples were retained and used later in RNA isolation and immunoblotting experiments. Seven hundred and fifty nanogram of each RNP sample was mixed with the LEAP assay reaction buffer (50 mM Tris pH 7.5, 50 mM KCl, 5 mM MgCl 2 , 0.05% Tween-20, 0.2 mM dNTPs, 1 mM DTT, 0.4 μM 3′ RACE adapter (5′ -GCG AGC ACA GAA TTA ATA CGA CTC ACT ATA GGT TTT TTT TTT TTV N -3′), 40 units of RNase inhibitor (Enzynomics), total reaction volume: 50 μl) and incubated at 37°C for 1 h. One microliter of LEAP cDNA products were subsequently amplified using 0.4 μM of L1 LEAP primer with the Phusion High-Fidelity polymerase kit (Thermo Fisher Scientific). PCR amplicons were separated and visualized in EtBr-stained 2% agarose gel.
eIF3 PAR-CLIP analysis. We utilized previously published PAR-CLIP data for eIF3a, b, d, and g 38 . Briefly, the authors immunoprecipitated eIF3b from 4-thiouridine-and-UV-treated 293 T cells to capture the eIF3-RNA complex. After high-salt washing and RNase digestion, they separated individual eIF3-RNA complexes through denaturing gel electrophoresis. eIF3a, b, d, and g were identified from four separate bands using mass-spectrometry and the interacting RNAs were purified and sequenced. Although three replicates were generated for each protein, only the first replicate was used for each. After retrieving the raw sequence files from NCBI (accession code: GSE65004), reads with low basecall qualities were excluded using the fastq_quality_filter from FASTX Toolkit (-q 25 -p 80; version 0.0.13.2; http:// hannonlab.cshl.edu/fastx_toolkit/). PCR duplicates were also excluded using fas-tx_collapser. Moreover, we excluded the reads that were shorter than 10-nt after trimming primer IDs and 3′ adapters from further analysis (cutadapt -a TGGAATTCTCGGGTGCCAAGG -u 12 -m 10; version 2.3 60 ;). The remaining reads were mapped to the L1Hs consensus sequence, wherein upto three mismatches were allowed using bowtie2 65 (--local -norc -score-min L,-18,2; version 2.2.4). For mean coverage analysis of 5′ UTR, ORF1, ORF2, and 3′ UTR, the number of reads that begin and end within each region were counted and the number was divided by the length of the corresponding region. The codes used for analyzing PAR-CLIP and mapping data are available from https://github.com/schanbaek/eif3_par-clip.
Comparison analysis of species-specific m 6 A site. To identify the species-specific full-length L1s in human, chimpanzee, and gorilla genome, we used BLAT-based and liftOver-based methods 47 with a computational approach. Only L1s of which insertion sites and two flanking regions are supported to be unique to the human, chimpanzee or gorilla genome by BLAT and liftOver were included in the further analyses. Then, we eliminated certain ambiguous elements containing gap sequence in the reference genome data and those that were less than 5.5 kb. The flanking sequences (2 kb, both upstream and downstream) of each species-specific L1 candidate were manually compared to the orthologous loci in human (GRCh37/hg19; Feb. 2009), chimpanzee (CSAC Pan_troglodytes-3.0/panTro5; May. 2016), gorilla (GSMRT3/gorGor5; May. 2016), and orangutan (Susie_PABv2/ponAbe3; Jan. 2018) genomes. The flanking sequences were used to identify the orthologous positions in the other genomes using BLAST-Like Alignment Tool (BLAT). We collected and retrieved the species-specific full-length L1s. We then classified the L1 subfamilies (L1Hs, L1PA2~L1PA5) using RepeatMasker utility 66 . Multiple sequence alignment of species-specific full-length L1s in each genome was performed using MUSCLE (MUltiple Sequence Comparison by Log-Expectation) under the default option 67 . The conserved sequence motifs at the three sites (A332, A495, and A600) were visualized using the program Weblogo 68 . Species-specific L1 loci are listed in Supplementary data 1-3.
Statistical analysis. GraphPad Prism 7.00 was used for statistical analysis. Two-sided student's t-test was used for unpaired data. Two-sided Kolmogorov-Smirnov test was used to assess the quantification of number and intensity of colocalizing puncta in Fig. 5g, h. For multiple comparisions, one-way ANOVA with Tukey's or Dunnett's multiple comparison test were used. P-values <0.05 were considered significant.