RNA 5-methylcytosine marks mitochondrial double-stranded RNAs for degradation and cytosolic release

SUMMARY

Mitochondria are essential regulators of innate immunity. They generate long double-stranded RNAs (mt-dsRNAs) and release them to the cytosol to trigger immune response under pathological stress conditions. Yet, the regulation of these self-immunogenic RNAs remains largely unknown. Here, we employ CRISPR screening on mitochondrial RNA-binding proteins and identify NOP2/Sun RNA methyltransferase 4 (NSUN4) as a key regulator of mt-dsRNA expression in human cells. We find that NSUN4 induces 5-methylcytosine (m⁵C) modification on mitochondrial RNAs (mtRNAs), especially on the termini of light-strand long noncoding RNAs. These m⁵C-modified RNAs are recognized by complement C1q binding protein (C1QBP), which recruits polyribonucleotide nucleotidyltransferase to facilitate RNA turnover. Suppression of NSUN4 or C1QBP results in increased mt-dsRNA expression while C1QBP deficiency also leads to increased cytosolic mt-dsRNAs and subsequent immune activation. Collectively, our study unveils the mechanism underlying the selective degradation of light-strand mtRNAs and establishes a molecular mark for mtRNA decay and cytosolic release.

Graphical Abstract

eTOC Blurb

Through CRISPR screening of 89 mitochondrial RBPs, Kim et al. find that NSUN4 downregulates mitochondrial dsRNA expression by installing 5-methylcytosine modification on light-strand RNAs. The modified RNAs are recognized by C1QBP, which recruits PNPT1 to facilitate RNA turnover, and are released to the cytosol, where they activate the immune response.

INTRODUCTION

Mitochondria have a unique circular genome containing 22 tRNAs, two rRNAs, and 13 mRNAs that encode components for the oxidative phosphorylation (OXPHOS) system.¹ Although all OXPHOS genes except for NADH-ubiquinone oxidoreductase chain 6 (ND6) are encoded in the heavy (H)-strand, the mitochondrial genome is bidirectionally transcribed from canonical promoters located in H- and light (L)-strand (HSP and LSP, respectively).^2,3 Mitochondrial RNAs (mtRNAs) are transcribed as long polycistronic forms and undergo RNA processing by mitochondrial ribonuclease P (MRPP) family and elaC ribonuclease Z 2 (ELAC2) that excise tRNAs to release mRNAs and rRNAs.^1,4–6 Released RNAs undergo further maturation processes, including polyadenylation and RNA modifications.^7,8

The bidirectional transcription of the mitochondrial genome also generates long noncoding RNAs (lncRNAs), mostly from the L-strand, that are complementary to mt-mRNAs. These L-strand RNAs have been considered to act as primers for mtDNA replication⁹ and were subject to rapid degradation by mitochondrial degradosome composed of the RNA helicase suppressor of Var1,3-like 1 (hSUV3) and polyribonucleotide nucleotidyltransferase 1 (PNPT1).¹⁰ In this context, hSUV3 unwinds the mtRNA, which allows PNPT1 to degrade the RNA in a 3′-to-5′ direction.^10,11 Defects in hSUV3 or PNPT1 genes result in the accumulation of mitochondrial double-stranded RNAs (mt-dsRNAs).¹² Moreover, PNPT1-deficiency is also embryonic lethal and is associated with hyperactivation of the immune response system due to cytosolic efflux of mt-dsRNAs.^13,14 However, the underlying mechanism behind the selective degradation of L-strand RNAs by PNPT1 and the cytosolic release of mt-dsRNAs in PNPT1-deficient cells is not yet understood.

Recently, mt-dsRNAs have received increasing attention as an essential endogenous source of innate immune stimuli. In addition to PNPT1-deficiency, mt-dsRNAs can be released to the cytosol during stress and activate cytosolic dsRNA sensors such as protein kinase R (PKR), melanoma differentiation-associated protein 5 (MDA5), and retinoic acid-inducible gene I (RIG-I).^14,15 For example, disruption of the mitochondrial membrane potential due to mitochondrial dysfunction results in the cytosolic release of mt-dsRNAs and subsequent apoptosis via PKR activation.¹⁵ More importantly, cytosolic and even extracellular mt-dsRNAs were reported to play pivotal roles during the development of inflammatory and degenerative diseases.^16–18 To list a few, autoimmune Sjögren’s syndrome, osteoarthritis, and Huntington’s disease patients show elevated expression of mt-dsRNAs in their bodily fluids.^16,19,20 The emerging significance of mt-dsRNAs in human pathophysiology calls for a revisit to the regulatory mechanism of mtRNAs.

In this study, we performed CRISPR-Cas9 screening on RNA-binding proteins residing in mitochondria (mt-RBPs) to identify post-transcriptional regulators of mt-dsRNAs. We found that RNA modification factors, in particular 5-methylcytosine (m⁵C) RNA methyltransferase NOP2/Sun RNA methyltransferase 4 (NSUN4), are key determinants of mt-dsRNA levels. We further delineated the m⁵C modification on mtRNAs by NSUN4, identified the reader protein, and unveiled the molecular function of m⁵C modification in the selective degradation and cytosolic release of mtRNAs, ultimately regulating the activation status of innate immune response systems. Based on these results, we propose a revised model for the post-transcriptional regulation of mt-dsRNAs in humans.

RESULT

NSUN4 is a key regulator of mt-dsRNA expression

To identify modulators of mt-dsRNA expression, we first established a list of mt-RBPs for CRISPR screening by combining two published databases. One database contained a list of proteins in the mitochondrial matrix identified by ascorbate peroxidase (APEX) proximity labeling.²¹ The other database provided a list of RBPs captured by oligo-dT pull-down after UV or formaldehyde crosslinking.²² By intersecting these databases, we generated a list of 89 putative mt-RBPs located in the mitochondrial matrix. We then knocked out the expression of these mt-RBPs individually using the CRISPR-Cas9 system and analyzed the downstream effect on mt-dsRNA expression (Figure 1A). Of note, our list of mt-RBPs contains previously established regulators of mtRNA transcription and processing, such as mitochondrial RNA polymerase (POLRMT), mtRNA processing factors (MRPP1 and ELAC2), and mitochondrial degradosome components (hSUV3 and PNPT1), confirming the validity of our list.² For each target mt-RBP, we transduced a mixture of two sgRNAs to HEK-293T cells stably expressing Cas9 and analyzed the mtRNA levels derived from each strand of the mitochondrial genome using strand-specific RT-qPCR (Figure 1A).²³ We first validated our approach using MRPP1 and POLRMT as positive controls. We found that the expression of both proteins was remarkably downregulated by individual and a mixture of sgRNAs (Figure S1A). Of note, we could still detect weak expression of the target protein because we used the bulk knockout approach without selecting single-cell clones. For POLRMT-depleted cells, we extracted RNAs and analyzed mtRNA expression using strand-specific RT-qPCR. Consistent with our expectation, the levels of H- and L-strand mtRNAs, as well as the total mtRNAs, decreased significantly (Figure S1B).

Figure 1. mt-RBP CRISPR screening reveals NSUN4 as a pivotal regulator of mt-dsRNAs.

(A) Experimental scheme of the CRISPR screening. Eighty-nine mt-RBPs were individually downregulated using CRISPR, and extracted RNAs were analyzed by strand-specific RT-qPCR to assess the expression of both mt-mRNAs and their complementary lncRNAs. (B) Strand-specific and total mtRNA expression when the indicated mt-RBP was suppressed (n = 3). Each sample was normalized to cells transduced with sgNC. Due to decreased cell viability, siRNAs were used for GADD45GIP1, TACO1, TFAM, TRAP1, DDX20, and C1QBP (marked in blue). Group 1, increased both strand RNAs; Group 2, decreased both strand RNAs; Group 3, increased H-strand RNAs and decreased L-strand RNAs; Group 4, increased L-strand RNAs; Group 5, others. (C) Validation of the screening results using immunocytochemistry with anti-dsRNA (J2) antibody. The target genes were marked in green in (B). Scale bar indicates 50 μm. Quantification is shown on the right (n = 4). (D and E) mtRNA expression in NSUN4-deficient cells analyzed by strand-specific RT-qPCR (D, n = 3) and RNA-FISH (E). Scale bar, 50 μm. All error bars denote s.e.m. Statistical significance was calculated using one-tailed Student’s t-tests, n.s. not significant, *p ≤ 0.05, **p ≤ 0.01, and ***p ≤ 0.001. See also Figure S1.

Through this approach, we analyzed the effect of knocking out the 89 mt-RBPs on mt-dsRNA expression (Figure 1B). For GADD45GIP1, TACO1, TFAM, TRAP1, DDX20, and C1QBP, the gene knockout was lethal, and we used siRNAs instead to knock down their expression (Figure 1B, marked in blue). Overall, we found that the expression level of total mtRNAs reflected that of H-strand RNAs, consistent with the fact that most of the mtRNAs were derived from the H-strand.²⁴ We then classified mt-RBPs into five groups based on their effects on H- and L-strand RNAs (Group 1, increased both strand RNAs; Group 2, decreased both strand RNAs; Group 3, increased H-strand RNAs and decreased L-strand RNAs; Group 4, increased L-strand RNAs; Group 5, others). We focused our subsequent analysis on Group 1 mt-RBPs as they yielded the largest degree of effects. First, we verified the increase in mt-dsRNA expression using J2 antibody.^14,25–27 Consistent with the screening result (Figure 1B, marked in green), knockout of NSUN4 and TruB pseudouridine synthase family member 2 (TRUB2) both resulted in increased J2 signals (Figure 1C). As TRUB2 may mediate its effect through mt-tRNA modification,²⁸ we chose NSUN4 for further analysis.

NSUN4 is known to induce 5-methylcytosine at the C911 region of 12S mt-rRNA (encoded by the RNR1 gene) to assist mitochondrial ribosome assembly.²⁹ Yet, there is no report on the potential methylation of mt-mRNAs or their complementary lncRNAs. Considering that another NSUN family protein, NSUN2, can modify mRNAs and lncRNAs from the nuclear genome to regulate their nuclear export,^30–33 we investigated whether NSUN4 could epigenetically modify mtRNAs. We first confirmed that our sgRNAs for NSUN4 successfully downregulated the target protein expression (Figure S1C). We then performed strand-specific RT-qPCR and RNA in situ hybridization (RNA-FISH) to validate the screening result that NSUN4-deficient cells resulted in increased expression of mt-dsRNAs (Figures 1D and 1E). Given the conventional role of NSUN4 in inducing mt-rRNA methylation,²⁹ we further ruled out the potential effect of translation on observed mt-dsRNA expression. We specifically blocked mitochondrial translation using chloramphenicol (CAP)³⁴ and confirmed decreased expression of an OXPHOS subunit encoded by the mitochondrial genome (Figure S1D). Under this condition, we found that mtRNA expression levels showed no significant difference between control and CAP-treated NSUN4-deficient cells (Figure S1E), suggesting that increased mtRNA levels upon NSUN4 depletion were not due to disruption of mitochondrial translation.

Next, we investigated the mechanism of increased mtRNAs by NSUN4 downregulation. To test the potential effect on RNA stability, we inhibited mitochondrial transcription using IMT1, a noncompetitive inhibitor of POLRMT,³⁵ and monitored mtRNA expression patterns over time (Figure 2A). We found that the degradation of both strands of selected mtRNAs was delayed in NSUN4-deficient cells (Figures 2B and S2A). To rule out the potential indirect effect mediated by disruption of mitochondrial degradosome, we examined the expression of PNPT1 in response to the depletion of NSUN4 or IMT1 treatment. We found no difference in PNPT1 protein expression between sgNC and sgNSUN4-transduced cells as well as by IMT1 treatment (Figure S2B). Lastly, to show that the regulation of mtRNAs by NSUN4 was not limited to HEK-293T cells, we examined the effect of NSUN4 depletion in other cell lines. Similar to the results in HEK-293T cells, the knockdown of NSUN4 using shRNA in MCF7 breast cancer cells and HCT116 colorectal cancer cells resulted in increased stability of ND5 and ND6 mtRNAs from both strands (Figures S2C and S2D).

Figure 2. Regulation of mtRNA stability by NSUN4.

(A) Expression levels of mtRNAs upon IMT1 treatment over time (n = 3). (B) Stability of H- and L-strands of ND5 and ND6 in cells transduced with sgNC and sgNSUN4 (n = 3). (C) NSUN4 protein expression following transfection of plasmids with WT or catalytic mutant NSUN4 tagged with EGFP into NSUN4-deficient cells. (D) Comparison of dsRNA expression after transfection of WT or catalytic mutant NSUN4 into NSUN4-deficient cells using immunocytochemistry with J2 antibody. Quantification is shown on the bottom (n = 3). Scale bar indicates 20 μm. (E) Strand-specific RT-qPCR analysis of H- and L-strands of mtRNAs after transfection of WT or catalytic mutant NSUN4 into NSUN4-deficient cells (n = 3). All error bars denote s.e.m. Statistical significance was calculated using one-tailed Student’s t-tests, n.s. not significant, *p ≤ 0.05, **p ≤ 0.01, and ***p ≤ 0.001. See also Figure S2.

We further performed the rescue experiment and asked whether the catalytic activity of NSUN4 is required to regulate mt-dsRNA expression. We transfected NSUN4-deficient HEK-293T cells with plasmid encoding EGFP-tagged wild-type (WT) NSUN4 or catalytic mutant NSUN4 and then analyzed dsRNA expression using J2 antibody and strand-specific RT-qPCR. Of note, a catalytic mutant NSUN4 was generated by substituting two conserved cysteine residues to tryptophan (C258W/C310W).³⁶ We first confirmed that the level of overexpressed NSUN4 (both WT and catalytic mutant) was similar to that of the control cells (Figure 2C). Notably, the overexpression of WT NSUN4 significantly attenuated the dsRNA signals captured by J2 antibody, rescuing the upregulation of dsRNAs due to NSUN4 depletion (Figure 2D). However, this rescue effect was marginal in catalytic mutant-expressing cells (Figure 2D). Of note, by utilizing EGFP-tagged overexpressed NSUN4 proteins, we could distinguish NSUN4-expressing cells from non-expressing cells in the vicinity. Thus, we quantified J2 signals from EGFP-positive cells and compared them to those of the neighboring EGFP-negative cells to quantitatively analyze the rescue effect (Figure 2D, bottom panel). Similarly, strand-specific RT-qPCR analysis showed a significant rescue effect in attenuating the expression of mtRNAs from both strands when WT NSUN4 was expressed, while the degree of the effect was much smaller when mutant NSUN4 was expressed (Figure 2E). Thus, our data show that the catalytic activity of NSUN4 is needed for the protein to downregulate mtRNA expression.

NSUN4 selectively induces m⁵C modification on L-strand RNAs

To investigate the NSUN4-mediated methylation on mtRNAs at a single-nucleotide resolution, we employed bisulfite sequencing (Bis-seq). Bisulfite induces the deamination reaction to convert non-methylated cytosine to uracil while methylated cytosine remains intact (Figure 3A).³⁷ First, using the well-established RNA methylation sites on the 28S rRNA as a control, we verified the successful conversion of the non-methylated cytosine with high efficiency (Figure 3B). When we analyzed RNAs mapped to the mitochondrial genome, we detected methylation in mtRNAs from both strands (Figure 3C). The methylation level was slightly higher than that of mRNAs encoded by the nuclear genome, with median levels of 6.45% and 4.81%, respectively (Figures 3C and 3D). Using the Bis-seq data, we investigated a potential local structural preference for m⁵C modification on mtRNAs by comparing minimum free energy (MFE) between sequences containing m⁵C and scrambled sequences. The MFE of a given RNA sequence was calculated using RNAfold.³⁸ Of note, for scrambled sequences, we randomly permuted a given sequence 100 times, calculated MFE for each permuted sequence, and used the average value. The MFE difference between m⁵C and scrambled sequences was modest, although the MFE of sequences with m⁵C was slightly lower when a flanking length of 20 nt was used (Figure S3A).

Figure 3. m⁵C modification of mtRNAs by NSUN4.

(A) Scheme of bisulfite conversion. (B) Validation of bisulfite conversion using 28S rRNA through Sanger sequencing. (C and D) The overall methylation level of H- and L-strands of mtRNAs (C; Median, 6.45%) and of mRNAs encoded by the nuclear genome (D; Median, 4.81%). (E) The m⁵C enrichment of mtRNAs in sgNC or sgNSUN4-transduced cells analyzed by immunoprecipitation using anti-m⁵C antibody (n = 3). (F) The relative methylation level of mtRNAs in sgNC and sgNSUN4-transduced cells from Bis-seq (n = 2). The statistical significance was calculated using one-tailed paired Student’s t-tests. (G) A gene structure of ND5 L-strand and accumulation of Bis-seq reads at indicated methylation sites. (H) The m⁵C distribution along the RNA analyzed based on NSUN4-dependency. (I) Comparison of MFEs between NSUN4-dependent or independent m⁵C sites on mtRNAs with two different flanking lengths. The modified cytosine is positioned at the center of the analyzed sequences. All error bars denote s.e.m. Box plot features: center line, median; box limits, upper and lower quartiles; whiskers, 1.5x interquartile range; points, individual data. All statistical significances were calculated using one-tailed Student’s t-tests, *p ≤ 0.05, **p ≤ 0.01, and ***p ≤ 0.001. See also Figure S3.

We then examined NSUN4-dependency on the observed methylation on mtRNAs. We performed immunoprecipitation with an m⁵C antibody and analyzed the pulled-down mtRNAs with RT-qPCR. We found that the mtRNAs captured by the m⁵C antibody were reduced in NSUN4-deficient cells, although the effect was marginal for some mtRNAs (Figure 3E). Considering that the m⁵C antibody was originally designed for targeting m⁵C-modified DNA and can also recognize unmodified cytosine, we carried out Bis-seq analysis in control and NSUN4-deficient cells. We found that the methylation levels on mt-mRNAs and their complementary lncRNAs were remarkably reduced upon the depletion of NSUN4 (Figure 3F; Table S8). Of note, the m⁵C-modified sites were concentrated in L-strand RNAs. As an example, sequencing reads mapped to ND5 L-strand RNA is shown in Figure 3G. We found a total of four methylated sites, two of which showed a dramatic decrease in methylation levels in NSUN4-deficient cells in both biological replicates (Figure 3G, marked in orange). Notably, the RNA levels measured by Bis-seq reads mapped to the ND5 L-strand locus showed a significant increase, suggesting that the m⁵C modification may facilitate the decay of ND5 L-strand mtRNAs (Figure 3G).

To investigate the selectivity of NSUN4 on mtRNAs, we investigated the methylation level of mtRNAs following the downregulation of another NSUN family protein, NSUN2, by reanalyzing publicly available Bis-seq data.³⁰ A recent study showed that, in addition to mRNAs from the nuclear genome, NSUN2 can localize to the mitochondrial matrix and modify mt-tRNAs.³⁹ However, we found no difference in the methylation level of mt-mRNAs and their complementary mt-lncRNAs upon NSUN2 depletion (Figure S3B). In addition, we analyzed NSUN4-dependent methylation on mt-tRNAs. We found two NSUN4-dependent methylation sites on glycine mt-tRNA (TRNG) and one site on serine mt-tRNA (TRNS2), but there was no significant global downregulation of mt-tRNA methylation in NSUN4-deficient cells (Figure S3C). Considering that Bis-seq might not have efficiently captured methylation of mt-tRNAs due to their rigid structure and that m⁵C modification may affect the processing of premature polycistronic mtRNAs, we investigated the levels of preprocessed mtRNAs with intact flanking mt-tRNAs by RT-qPCR. Additionally, we examined the expression of premature mtRNAs linked to mt-rRNA because our Bis-seq data were produced from rRNA-depleted RNAs and one well-established target of NSUN4 is mt-RNR1. Our analysis showed that premature mtRNA levels were decreased for Phe-RNR1 while CO2-Lys, Gly-ND3, Glu-ND6, Leu-ND5, and ND5-ND6 levels were unaffected by NSUN4 depletion (Figure S3D). These findings suggest that NSUN4 may facilitate the processing of mt-rRNA, but it does not affect the maturation of mt-mRNAs.

Next, we performed meta-gene analysis on Bis-seq results to characterize the location of m⁵C modification along the RNA. Considering that mt-mRNAs have either no or very short UTRs,⁴⁰ we only used the coding region to determine the relative position of the modification site. The density plot reveals that the NSUN4-dependent sites with 10% or more methylation levels were found near the termini of mtRNAs (Figure 3H). The enrichment of the modification site near the 3′ ends is consistent with the 3′-to-5′ degradation of the L-strand RNAs by the mtRNA degradosome.¹² Lastly, we examined whether RNA secondary structure plays a role in determining m⁵C installation by NSUN4. In contrast to MFE analysis for all m⁵C-modified mtRNAs, the MFE of NSUN4-dependent m⁵C sites was slightly increased compared to that of NSUN4-independent sites (Figure 3I). To investigate the potential for base pairing near m⁵C-modified sites, we analyzed Watson-Crick (WC) pair co-occurrence frequencies of two positions flanking from NSUN4-dependent m⁵C sites.⁴¹ As shown in the tested model, the formation of hairpin structure exhibited a high WC-pair co-occurrence frequency (Figure S3E). We found that NSUN4-dependent m⁵C sites did not show increased WC-pair co-occurrence frequency compared to scrambled sequences (Figure S3E), suggesting that the secondary structure is not a determinant factor for the methylation by NSUN4.

Collectively, our findings implicate that NSUN4 acts as an m⁵C writer for mt-mRNAs and mt-lncRNAs and preferentially methylates the termini of L-strand mtRNAs, which may facilitate RNA degradation to suppress the formation of mt-dsRNAs.

C1QBP recognizes m⁵C-modified mtRNAs and recruits PNPT1 for degradation

Methylated RNAs are recognized by m⁵C-binding proteins, called readers, that determine the fate of the modified RNAs. To determine m⁵C readers in the mitochondria, we reanalyzed published data performed by mass spectrometry after RNA affinity chromatography to capture m⁵C-binding proteins (Figure 4A).³⁰ By intersecting the m⁵C RNA interactome and our mt-RBP list, complement C1q binding protein (C1QBP) emerged as a potential mitochondrial m⁵C reader (Figure 4A).

Figure 4. Degradation of m⁵C-modified mtRNAs by C1QBP and PNPT1.

(A) Replot of an m⁵C interactome from a published paper.³⁰ (B) Analysis of the interaction of C1QBP with synthetic mtRNAs containing m⁵C modification using biotin-streptavidin immunoprecipitation. (C) Analysis of mtRNAs pulled down with C1QBP in sgNSUN4-transduced cells normalized to that in sgNC-transduced cells (n = 3). (D) Bis-seq analysis on NSUN4-dependent methylation sites in siLuc or siC1QBP transfected cells (n = 2). (E) The strand-specific mtRNA expression in C1QBP downregulated cells (n = 3). (F) Stability analysis of selected mtRNAs in C1QBP- or PNPT1-deficient cells (n = 3). (G and H) Co-immunoprecipitation analysis for C1QBP-PNPT1 interaction in the presence (G) or absence (H) of nucleic acids. MNase was treated to digest nucleic acids. (I) Bis-seq analysis on NSUN4-dependent methylation sites in siLuc or siPNPT1 transfected cells (n = 3). All error bars denote s.e.m. All statistical significances were calculated using one-tailed Student’s t-tests, *p ≤ 0.05, **p ≤ 0.01, and ***p ≤ 0.001. See also Figures S4 and S5.

To test whether C1QBP can bind to m⁵C-modified RNAs, we performed an IP experiment using biotin-tagged synthetic RNA oligos. We found that C1QBP bound stronger to the m⁵C-modified RNAs compared to the unmethylated control oligos (Figure S4A). We also synthesized biotin-tagged m⁵C-modified RNAs for two selected mtRNAs with very different secondary structures and analyzed their interaction with C1QBP. We found that the binding affinity of C1QBP was similar among three synthetic RNAs with m⁵C, irrespective of their RNA structures (Figures 4B and S4B). We then examined whether C1QBP could interact with the m⁵C-modified mtRNAs in cells by using formaldehyde to crosslink the C1QBP-mtRNA complex and analyzing mtRNAs co-immunoprecipitated with C1QBP. We found that C1QBP could interact with mtRNAs in an NSUN4-dependent manner as NSUN4 knockdown cells exhibited significantly decreased C1QBP-mtRNA interaction (Figure 4C). We further performed Bis-seq analysis on RNAs extracted from C1QBP knockdown cells. Consistent with our RNA immunoprecipitation results, our Bis-seq results clearly showed an accumulation of methylation levels at NSUN4-dependent m⁵C sites in C1QBP-deficient cells (Figure 4D), while the effect of C1QBP knockdown was modest at NSUN4-independent m⁵C sites (Figure S4C).

Next, we asked whether C1QBP could promote the degradation of m⁵C-modified RNAs. Consistent with our earlier data that NSUN4 depletion increased mtRNA levels, the knockdown of C1QBP also resulted in a significant increase in mtRNAs (Figures 4E and S4D). Moreover, the degree of the effect was greater for L-strand mtRNAs, which is consistent with most m⁵C-modified sites concentrated on these RNAs. We further analyzed whether this increase was due to enhanced RNA stability. We inhibited mitochondrial transcription using IMT1 and examined the kinetics of mtRNA expression. We found that the knockdown of C1QBP resulted in reduced degradation rates of all four mtRNAs examined (Figure 4F). Of note, PNPT1 knockdown was used as a positive control. Our data implicate that NSUN4-mediated m⁵C installation on mtRNAs enhances C1QBP-mtRNA interaction that results in the destabilization of mtRNAs.

C1QBP belongs to the MAM33 family and is involved in immune response and energy homeostasis.^42–44 Yet, it does not contain any nuclease domain that may mediate RNA degradation or affect RNA stability. For this, we hypothesized that C1QBP may interact and recruit nucleases to degrade m⁵C-modified RNAs. To identify C1QBP-interacting nucleases, we utilized two web servers (PrePPI and PSOPIA) that predicted protein-protein interaction based on structure and homology, respectively.^45,46 Interestingly, we found that PNPT1 was one of the potential C1QBP interactors. A recent proximity-dependent biotinylation analysis also supported potential C1QBP-PNPT1 interaction.⁴⁷ Given that C1QBP has been reported as a highly promiscuous protein that interacts with 838 proteins based on the biological general repository for interaction datasets,⁴⁸ we experimentally examined potential C1QBP-PNPT1 interaction through immunoprecipitation. When we pulled down C1QBP, PNPT1 was co-immunoprecipitated and vice versa (Figure 4G). Of note, apoptosis-inducing factor (AIF) was used as a negative control. Moreover, the interaction between C1QBP and PNPT1 was still observed when we digested nucleic acids using micrococcal nuclease (MNase), indicating that C1QBP and PNPT1 interaction is RNA-independent (Figures 4H and S4E).

If C1QBP recruits PNPT1 to degrade mtRNAs, then PNPT1 should exhibit a preference toward m⁵C-modified RNAs. When we examined the effect of PNPT1 depletion in more detail, we found that L-strand mtRNAs were more sensitive (Figures S4F and S4G). This result is in line with previous studies that PNPT1 preferentially degrades L-strand mtRNAs.^12,14 To analyze whether the methylated RNAs were preferentially degraded by PNPT1, we carried out Bis-seq with RNAs extracted from PNPT1-deficient cells. We found that most NSUN4-dependent methylation sites showed upregulated methylation levels when PNPT1 was knocked down (Figures 4I and S4H). Considering that decreased methylation sites in NSUN4-deficient cells were mainly located in L-strand RNAs, our result implicates that m⁵C-modified L-strand RNAs undergo PNPT1-mediated degradation. Interestingly, our Bis-seq analysis showed the accumulation of only NSUN4-dependent methylation sites in C1QBP or PNPT1-deficient cells (Figures 4D, S4C, 4I, and S4H). One possibility of recognition of NSUN4-dependent methylation sites by C1QBP is that NSUN4, C1QBP, and PNPT1 work as a trimeric complex. However, NSUN4 did not show interaction with either C1QBP or PNPT1 (Figure S4I), suggesting that the recruitment of PNPT1 by C1QBP followed by the recognition of m⁵C-modified mtRNAs occurs independently from NSUN4.

Lastly, we comprehensively compared the effects on mtRNA expression upon NSUN4, C1QBP, and PNPT1 depletion using immunocytochemistry with J2 antibody, strand-specific RT-qPCR, and Sanger sequencing. Downregulation of these three proteins all resulted in an increase in dsRNA signals captured by the J2 antibody (Figures S5A). Notably, RT-qPCR analysis revealed that PNPT1 showed higher selectivity toward L-strand RNAs compared to NSUN4 or C1QBP (Figure S5B). To rule out the possibility that the increased mtRNA levels reflected changes in the expression of the normalization control (GAPDH), we confirmed the mtRNA expression upon NSUN4, C1QBP, or PNPT1 depletion using a different normalization factor, β-actin (ACTB) (Figure S5C). The Sanger sequencing result of NSUN4-dependent m⁵C site on ND5 L-strand mtRNA upon bisulfite conversion further demonstrated the role of NSUN4, C1QBP, and PNPT1 on regulating m⁵C methylation levels. In control cells, Sanger sequencing raw data showed mixed signals for C and T, indicating m⁵C installation (Figure S5D). In NSUN4-deficient cells, the partial C signal is almost completely abolished while the level of C signal was enhanced in C1QBP or PNPT1 knockdown cells (Figure S5D). Combined, these data support our model where the installation of m⁵C by NSUN4 is recognized by the C1QBP-PNPT1 complex that promotes the degradation of the modified RNAs.

m⁵C modification facilitates the cytosolic efflux and immune activation by mtRNAs

Depletion of PNPT1 results in the cytosolic accumulation of mt-dsRNAs and MDA5/RIG-I-dependent induction of interferon-stimulated genes (ISGs).¹⁴ Indeed, patients harboring hypomorphic mutations in the PNPT1 showed increased interferon and neopterin in cerebrospinal fluid, which is in line with hyperactivated immune response.¹⁴ Yet, the potential role of m⁵C modification in this process remains unknown. Considering that the knockdown of C1QBP or PNPT1 increased m⁵C-modified mtRNAs (Figures 4D and 4I) and that m⁵C modification facilitates the cytosolic export of nuclear mRNAs,³⁰ we asked whether m⁵C modification is also involved in the cytosolic release of mtRNAs.

We conducted subcellular fractionation and analyzed mtRNA levels from the free cytosolic compartment. We first confirmed successful subcellular fractionation using western blotting of selected marker proteins (Figure S6A). When C1QBP was knocked down, we found that the level of cytosolic mtRNAs was significantly increased (Figure 5A). The accumulated mtRNAs in the cytosol upon depletion of C1QBP or PNPT1 led to increased phosphorylation of PKR and interferon regulatory transcription factor 3 (IRF3), a key downstream factor of MDA5/RIG-I (Figure 5B). As the downstream response of PKR and IRF3 activation, phosphorylation of PKR substrate eIF2α was increased and the expression of several ISGs was induced (Figures 5B and 5C). More importantly, when we depleted mtRNAs using POLRMT inhibitor 2′-C-methyladenosine (2-CM), knockdown of C1QBP or PNPT1 did not induce PKR and IRF3 phosphorylation (Figures S6B and 5D). Consequently, eIF2α was no longer phosphorylated and ISG expression was dramatically decreased (Figures 5D and 5E). These results indicate that the accumulation of m⁵C-modified mtRNAs due to the C1QBP or PNPT1 depletion led to increased cytosolic mtRNAs and subsequent immune activation.

Figure 5. Cytosolic release and immune activation by m⁵C-modified mtRNAs.

(A) The expression of cytosolic mtRNAs in C1QBP-deficient cells (n = 3). (B and C) Analysis of innate immune response pathways through western blotting of key dsRNA sensors (B) and RT-qPCR of selected ISGs (C, n = 3) in PNPT1, C1QBP, or NSUN4-depleted cells. (D and E) Immune response pathways through western blotting of key dsRNA sensors (D) and RT-qPCR of selected ISGs (E, n = 3) in PNPT1, C1QBP, or NSUN4-depleted cells after depleting mtRNAs with 2-CM treatment. (F) The expression of cytosolic mtRNAs in NSUN4-deficient cells (n = 3). (G) Schematics of the m⁵C-modified RNA degradation and cytosolic release model via NSUN4-C1QBP-PNPT1 axis. All error bars denote s.e.m. Statistical significances were calculated using one-tailed Student’s t-tests, n.s. not significant, *p ≤ 0.05, **p ≤ 0.01, and ***p ≤ 0.001. See also Figure S6.

We then sought to determine the importance of m⁵C modification in the cytosolic release of mtRNAs. To do so, we analyzed the effect of the downregulation of the m⁵C writer NSUN4. Interestingly, in NSUN4-depleted cells, we did not detect a significant increase in mtRNAs in the free cytosolic fraction (Figure 5F) despite the overall increase in mtRNA levels, indicating that mtRNAs are not released to the cytosol without NSUN4-dependent m⁵C modification. Consistent with this, we found that, in NSUN4-depleted cells, the effect of increased mtRNA levels on eIF2α phosphorylation, IRF3 phosphorylation, and ISG induction were marginal compared to those observed in C1QBP or PNPT1 knockdown cells (Figures 5B–5E and S6C). Interestingly, we observed mtRNA-dependent increased phosphorylation of PKR in NSUN4-depleted cells (Figures 5B and 5D), which is consistent with our previous finding that mtRNAs can activate PKR inside mitochondria.¹⁵

DISCUSSION

Our findings highlight an important post-transcriptional regulation of mtRNAs via RNA methylation mediated by the NSUN4-C1QBP-PNPT1 axis in preventing the accumulation of dsRNAs that can trigger an aberrant immune response. According to our model, NSUN4 induces m⁵C modification on mtRNAs which is recognized by C1QBP that recruits PNPT1 to facilitate RNA turnover (Figure 5G). Considering that NSUN4 preferentially targets the termini of L-strand mtRNAs, m⁵C-mediated mtRNA turnover may account for the selective degradation of L-strand RNAs by PNPT1. The m⁵C modification also plays a role in the cytosolic release of mtRNAs. Depletion of C1QBP results in the cytosolic efflux of the RNAs and subsequent innate immune activation while, without NSUN4-dependent m⁵C modification, mtRNAs are retained inside mitochondria (Figure 5G).

Prior to the discovery of mt-dsRNAs as self-immunogenic molecules, a study investigated the regulation of the two strands of mtRNAs by mt-RBPs using a variant of nCounter Analysis System known as MitoString.⁴⁹ Our approach differs from the previous one in two aspects. First, we utilized two advanced techniques in identifying mitochondrial proteins via proximity labeling and UV/formaldehyde crosslinking to obtain an accurate list of putative mt-RBPs.^21,22 As a result, our study contains 55 mt-RBPs that were neglected in the previous study. In addition, the previous study only examined two probes that target the L-strand mtRNAs while we extensively examined L-strand mtRNAs in the region that are complementary to their H-strand counterparts in order to study how mt-RBPs may affect the expression of mt-dsRNAs. Through such approach, we identified NSUN4 as a pivotal regulator of mt-dsRNA expression and the downstream immune activation by installing m⁵C modification. Therefore, our mt-RBP screen provides valuable resources for understanding the post-transcriptional regulation of mt-dsRNAs.

Recent studies have shown a pivotal role of C1QBP in cells with a high demand for ATP. In the C9orf72 mutation model of amyotrophic lateral sclerosis, the depletion of C1QBP in human microglia cells induced the activation of the NLRP3 inflammasome, resulting in motor neuron apoptosis.⁵⁰ Moreover, the downregulation of C1QBP in mice hearts stimulated the expression of integrated stress response genes, such as Atf4, Sestrin2, and Chop, through the activation of mitochondrial unfolded protein response.⁵¹ In light of our findings, C1QBP deficiency may induce inflammation in these conditions via upregulation of m⁵C-modified mt-dsRNAs. Indeed, mt-dsRNAs are closely associated with neurodegenerative diseases and mitochondrial unfolded protein response.^20,52 Together, these studies highlight the importance of mtRNA regulation in pathophysiological conditions.

Our model raises important perspectives on mitochondrial epitranscriptome. Previous studies have focused on various types of RNA modification, including pseudouridine, N⁵-isopentenyladenosine (m¹A), N6-adenosine methylation (m⁶A), and m⁵C in rRNAs and tRNAs, to regulate the efficiency of mitochondrial translation.^53,54 Pseudouridylation or m¹A modification in mRNAs has been reported, although the biological function of these mtRNA modifications still remains unknown.^55,56 Consistent with these studies, our CRISPR screening for mt-RBPs yielded additional RNA-modifying enzymes, including pseudouridylase RPUSD4 and TRUB2. In particular, knockout of TRUB2 significantly elevated the level of mt-dsRNAs. Yet, it remains to be investigated whether mt-mRNAs and their complementary lncRNAs are directly modified by these enzymes as well as the potential downstream effects of these RNA modifications on mt-dsRNA formation.

Our study also provides valuable resources in expanding our knowledge of RNA regulatory systems in mitochondria. According to our screening data, depletion of mitochondrial aminoacyl tRNA synthetases (aaRSs), such as HARS, AARS, KARS, and LARS2, resulted in increased expressions of both H- and L-strand RNAs. A previous study showed that aaRSs could bind to mt-mRNAs or lncRNAs at a stem-loop structure with similar sequences to their anticodons.⁵⁷ Thus, it would be intriguing to study whether aaRSs could protect mtRNAs from mitochondrial degradosome by providing steric hindrance. Overall, our results demonstrate the fundamental role of RNA modification in facilitating RNA decay and transport and provide important resources for understanding the epigenetic regulation of mt-dsRNAs.

Limitations of the study

Although our study clearly shows NSUN4-dependent m⁵C modification on mtRNAs, with a preference toward L-strand RNAs, downregulating NSUN4 or C1QBP also has significant effects on H-strand RNA levels. This may indicate that the modification occurs on mt-dsRNAs and that the recognition of the L-strand RNAs may degrade H-strand RNA in complex with the L-strand RNA. However, it remains unclear whether NSUN4 recognizes ssRNAs or certain structured RNAs for m⁵C installation and whether there is any preference for C1QBP in recognizing m⁵C-modified ssRNAs or dsRNAs. In addition, our Bis-seq analysis revealed the existence of NSUN4-independent m⁵C sites. We reanalyzed the Bis-seq data in NSUN2-deficient cells, but NSUN2 did not affect m⁵C modification on mtRNAs, indicating that there might be additional methyltransferases mediating m⁵C modification on these NSUN4-independent sites. Another limitation of our study is the selectivity for the recognition of NSUN4-dependent m⁵C-modified cytosines by C1QBP. We showed that the RNA secondary structure is not a major determinant for the recognition by C1QBP, and that NSUN4 does not recruit C1QBP to the modified sites via protein-protein interaction. Yet, C1QBP clearly prefers NSUN4-dependent m⁵C RNAs as the depletion of C1QBP results in increased methylation of NSUN4-dependent sites while having only marginal effects on NSUN4-independent sites. Further studies using high-throughput techniques such as massively parallel reporter assays are required to extensively analyze C1QBP-m⁵C RNA interactions to better understand the regulation of m⁵C-modified mt-mRNAs and mt-lncRNAs. In addition, we attempted to identify the sequence motif of NSUN4 methylated sites using MEME Suite.⁵⁸ Our analysis revealed guanosine enrichment on the fourth nucleotide from the methylated cytosine (CNNNG), which is similar to the CNGGG motif of NSUN2.^30,59 However, we stress that the limited number of NSUN4 target sites makes this analysis unreliable to report in this study. In vitro methylation assay using recombinant NSUN4 on a large number of randomized oligonucleotides will provide a clue in elucidating the exact sequence preferences of NSUN4. Lastly, the potential recognition of m⁵C-modified RNAs by mitochondrial pore complexes needs to be analyzed, which would enhance our understanding of the cytosolic efflux of m⁵C-modified mtRNAs and mt-dsRNAs.

STAR★METHODS

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Yoosik Kim (ysyoosik@kaist.ac.kr).

Materials availability

All reagents and resources used for this study are available upon request to the lead contact.

Data and code availability

• Bis-seq data have been deposited at GEO and are publicly available as of the date of publication. Accession numbers are listed in the key resources table. Original western blot images and microscopy data have been reported at Mendeley and are publicly available as of the date of publication. The DOI is listed in the key resources table.

• This paper does not report the original code.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Cell lines and maintenance

HEK-293T cells were grown in Dulbecco’s modified eagle medium (DMEM) supplemented with 9.1% (v/v) fetal bovine serum (FBS). For HCT116 and MCF7, roswell park memorial institute (RPMI) 1640 medium supplemented with 9.1% (v/v) FBS was used. All cells were grown in an incubator maintained at 37°C with 5% CO₂, passaging at a 1:5 dilution every 2–3 days.

METHOD DETAILS

sgRNA plasmid production

For the CRISPR screening, two sgRNAs with the highest scores were selected from the Broad sgRNA designer (CRISPick; Table S1) for every gene of interest. The sgRNA guide vector (lentiGuide-Puro) was digested using BsmBI overnight. 100 μM forward and reverse oligos were phosphorylated and annealed using T4 polynucleotide kinase. The cleaved guide vector was incubated with 1:200 diluted annealed oligos and T4 ligase for 2 h at room temperature (RT). 5 μL of the mixture was added to Stbl3 chemically competent E. coli, and then heat-shock transformed. After 1 h incubation, the bacteria were spread in an ampicillin (Amp)-containing Luria broth (LB) plate with 1.5% agar. The plate was incubated at 37°C for 18~24 h.

Lentivirus production

Lentivirus production was performed following a published paper⁶⁰ with some modifications. HEK-293T cells were prepared with 90% confluency in a 100-mm dish with 5 mL of culture medium. The prepared transfection mixture (1 mL opti-MEM, 15 μL Lipofectamine 3000, 34.02 μL p3000, 5.67 μg psPAX2, 3.78 μg pMD2.G, and 7.56 μg transfer plasmid) was incubated for 15 min at RT and added to cells dropwise. Cells were then incubated for 6 h at 37°C with 5% CO₂. The medium was replaced with fresh medium with 1% (w/v) bovine serum albumin (BSA) to increase the transfection efficiency. Cells were incubated for an additional 48 h in an incubator maintained at 37°C with 5% CO₂. The medium was transferred to a 15 mL conical tube, centrifuged at 3,000 rpm for 10 min at 4°C, and filtered with a 0.45 μm filter. The cleaned medium containing the virus was aliquoted in 1.7 mL microtubes and stored at −80°C.

Lentivirus transduction

Cells were seeded in a 24-well plate and incubated overnight in an incubator at 37°C with 5% CO₂. The cell medium was replaced with virus medium containing 10 μg/mL of polybrene, and cells were incubated for 48 h. 10 μg/mL of blasticidin or puromycin was treated for 4 days, replacing the cell medium every 2 days. Cells were then kept for an additional 2 days in a fresh medium without antibiotics before harvesting them for further analysis.

Chemical treatment

To block mitochondrial transcription, 30 μM of Inositol 4-methyltransferase (IMT1) was treated for 1, 3, and 6 h. For mitochondrial translation inhibition, chloramphenicol (CAP) was treated at 20 μg/mL for 24 h. To degrade nucleic acids in cell lysates, micrococcal nuclease (MNase) was treated to cell lysates and incubated at 37°C for 5 min. To knock down C1QBP or PNPT1, siRNA was transfected twice in total, once every two days, using Lipofectamine 3000 following the manufacturer’s instructions. For GADD34GIP1, TACO1, TFAM, TRAP1, and DDX20, siRNAs were transfected with Lipofectamine 3000 and incubated for three days. Sequences of the siRNAs used in this paper are provided in Table S2.

RNA extraction and RT-qPCR

Total RNA was extracted using TRIzol following the manufacturer’s instructions. After chloroform extraction and precipitation, RNA was purified using DNase I followed by phase separation with Acid-phenol:Chloroform, pH 4.5. cDNA was prepared using RevertAid reverse transcriptase. For strand-specific RT-qPCR, cDNA was synthesized with SuperScript IV reverse transcriptase and gene-specific reverse transcription primers containing CMV or GST promoter sequences (Table S3). qPCR was performed using QuantStudio 1 Real-time PCR system. Primers used in the study are provided in Tables S4 and S5.

NSUN4 plasmid production and site-directed mutagenesis

To generate the wild-type (WT) NSUN4 expression plasmid, the pEGFP-C1 plasmid was used as a backbone. The backbone plasmid was cleaved with HindIII and EcoRI restriction enzymes by incubating at 37°C for 1 h, followed by concentrating the cleaved plasmid using an oligo clean and concentrator kit. The NSUN4 oligos were prepared by PCR on cDNAs extracted from HEK-293T cells. The digested plasmid and NSUN4 oligos were ligated for 1 h at RT with T4 DNA ligase. The ligation mix was transformed into DH5-α chemically competent E. coli and spread in an Amp-containing LB plate with 1.5% agar. The inserted sequence was confirmed by Sanger sequencing.

To generate a catalytic mutant NSUN4, primers were designed around 15 nucleotides centered around the target site. The NSUN4 WT plasmid was amplified with the primers with a point mutation using a high-fidelity polymerase (nPfu-Forte) following the manufacturer’s protocol. To remove the unmodified plasmid, Dpn I was used at 37°C for 15 min. After the enzyme was inactivated at 80°C for 20 min, the mixture was transformed into DH5-α competent E. coli. The point mutations were confirmed using Sanger sequencing. Primer sequences used to generate WT and mutant NSUN4 plasmids and Sanger sequencing are provided in Table S6. To overexpress WT or mutant NSUN4, 1 μg/mL of plasmids was transfected into cells with FuGENE HD Transfection Reagent following the manufacturer’s protocol.

shRNA plasmid production and lentivirus transduction

The pLKO.1 plasmid was cleaved using AgeI and EcoRI restriction enzymes following the manufacturer’s instructions. The cleaved DNA was purified via gel extraction. To design shRNA sequences, a shRNA pre-design tool provided on the Sigma website was used. Two or three target sequences were tested, of which the sequence showing the highest efficiency was used for further experiments. Annealed DNA oligos in a DNA annealing buffer (10 mM Tris pH 8.0 and 80 mM NaCl) were ligated with the cleaved plasmid for 1 h at RT using T4 DNA ligase. The plasmid was transformed into a Stbl3 chemically competent E. coli. As a negative control, pLKO.1 scrambled shRNA was used. Sequences of the shRNA used in this study are provided in Table S7.

Immunocytochemistry

HEK-293T cells were grown in a pre-coated confocal dish with 1% (w/v) gelatin for overnight. Cells were rinsed once with cold PBS and fixed in 4% (v/v) paraformaldehyde in PBS for 10 min at RT. The fixed cells were rinsed twice with PBS and washed twice in PBS for 5 min each. Cells were permeabilized in 0.1% (v/v) Triton X-100 in PBS for 10 min and rinsed twice with 0.1% (v/v) Tween-20 in PBS (PBST). After blocking in 1% (w/v) BSA dissolved in PBST, cells were incubated with primary antibody diluted in 1% BSA for 2 h at RT. Cells were rinsed twice and washed 3 times with PBST for 10 min each. The mixture containing 1:1,000 diluted Alexa fluor fluorophore-labeled secondary antibodies in 1% BSA, 500 nM of MitoGreen, and 450 nM of 4′, 6-Diamidino-2-Phenylindole, Dihydrochloride (DAPI) was added to cells and incubated for 45 min at RT while protected from light. Zeiss LSM 880 confocal microscope using a 63x objective (NA = 1.40) was used for imaging.

RNA fluorescent in situ hybridization (FISH)

RNA-FISH was performed according to the RNAscope Multiplex Fluorescent Assay following the manufacturer’s instructions. Cells were washed with PBS once and were fixed in 4% paraformaldehyde for 15 min at RT. The fixed cells underwent dehydration and rehydration and were hybridized with mtRNA probes (Hs-MT-ND5 and Hs-MT-ND5-sense). Amplification steps were performed following the manufacturer’s protocol. Zeiss LSM 880 confocal microscope using a 63x objective (NA = 1.40) was used for imaging.

Western blotting

To prepare total cell lysates, harvested cells were incubated in the lysis buffer (50 mM Tris-HCl, pH 8.0, 100 mM KCl, 0.5% NP-40, 10% Glycerol, and 1 mM DTT) for 10 min on ice and then sonicated for complete lysis. Protein lysates were quantified using a BCA protein assay kit. 30 μg of protein sample was separated on a 10% SDS-PAGE gel and transferred to a PVDF membrane using the Amersham semidry transfer system.

Bisulfite conversion and sequencing

rRNAs were depleted from 1 μg of total RNAs using Ribo-Zero Plus rRNA depletion kit following the manufacturer’s protocol. Cytosines were converted to uracil using an EpiNext RNA bisulfite conversion kit. The conversion was performed following the manufacturer’s instructions with some modifications. In the desulphonation step, the samples were incubated for 1 h at RT. To create cDNA libraries from converted RNAs, TruSeq Stranded Total RNA kit was used. Sequencing was performed on an Illumina NovaSeq 6000 platform.

RNA Bis-seq bioinformatics analysis

Analysis of the Bis-seq data was performed following previous papers.^30,61 Adaptor sequences and low-quality bases were removed using Trimmomatic.⁶² The processed reads with lengths greater than 35 nt were used for further analysis. The reads were aligned against the hg38 genome using Bismark (version 0.22.3)⁶³ with stringent parameters: –N 0 –X 500. To filter reads with incomplete conversion, rigorous criterion (a read with four methylated cytosines in a non-CG context) was employed. After the reads were confirmed using an integrative genomics viewer (IGV), the m⁵C site locations were extracted from the filtered bam file using Bismark. Sites showing at least 20 total cytosine reads and 5 methylated cytosine reads were considered modified. For more accurate methylated sites, only those detected in all samples were used for the study. The degree of methylation was calculated by taking the average of the two biological replicates. Information on m⁵C-modified sites with NSUN4 dependency is provided in Table S8.

RNA secondary structure analysis

The MFE analysis was performed using RNAfold in the Vienna package (version 2.6.4).³⁸ To assess the local structure surrounding the modified site, MFE values were calculated for sequences with flanking nucleotides of length 10 or 20, centered on the modified site. As a negative control, scrambled sequences were generated using the Python Random module, based on each paired sequence with m⁵C. To determine the MFE value of the scrambled sequence, a given sequence was permuted 100 times and the average MFE was used.

To determine the unbiased likelihood of base pairing near NSUN4-dependent m⁵C sites, Watson-Crick (WC) pair co-occurrence frequencies of two positions around the modified sites were compared to those of scrambled sequences.⁴¹ For sequences with 10 nt flanking length and the test model, the scramble sequence was permuted 1,000 times, and the average value was used. For sequences with 20 nt flanking length, the scramble was permuted 100 times and the average value was reported. The enrichment level was calculated as a log₂ ratio between the observed WC-pair co-occurrence frequency near m⁵C-modified sequences and the background scrambled sequences.

5-methylcytosine immunoprecipitation

25 μL of Pierce protein A plus agarose bead was washed 3 times with the lysis buffer (50 mM Tris-HCl, pH 8.0, 100 mM KCl, 0.5% NP-40, 10% Glycerol, and freshly added 1 mM DTT) at RT. 5 μg of m⁵C antibody in 300 μL of fresh lysis buffer was used to prepare antibody-conjugated bead. The bead-antibody mixture was gently rotated for 3 h at 4°C. To block non-specific binding, 250 μg of cytidine was added. The bead was washed 3 times with lysis buffer. Purified RNA was prepared using RNA fragmentation reagents following the manufacturer’s protocol to expose the modified region. 20 μg of fragmented RNA was added to the antibody-conjugated bead with 300 μL of lysis buffer containing 240 U of recombinant RNase inhibitor. The sample was gently rotated for 3 h at 4°C. The sample was washed 3 times with wash buffer (50 mM Tris-HCl, pH 8.0, 100 mM KCl, 0.1% NP-40, and 10% Glycerol). TRIzol was used to extract the RNAs for further analysis.

Immunoprecipitation

20 μL of Pierce protein A plus agarose bead was washed 3 times with the lysis buffer at RT. 5 μL of antibody in 300 μL of fresh lysis buffer was used to prepare the antibody-conjugated bead. The bead-antibody mixture was gently rotated for 3 h at 4°C. The antibody-conjugated bead was washed 3 times with lysis buffer. 300 μL of cell lysate in the lysis buffer was added to the antibody-conjugated bead with a protease inhibitor cocktail. The sample was gently rotated for 3 h at 4°C. The sample was washed 4 times with the wash buffer. 30 μL of SDS buffer was added to the washed bead and prepared western blotting.

C1QBP formaldehyde crosslinking and immunoprecipitation (fCLIP)

20 μL of Pierce protein A plus agarose beads were washed 3 times with the fCLIP lysis buffer (20 mM Tris-HCl, pH 7.5, 15 mM NaCl, 10 mM EDTA, 0.5% NP-40, 0.1% Triton X-100, 0.1% sodium dodecyl-sulfate (SDS), and 0.1% sodium deoxycholate). The beads were incubated with C1QBP antibody in the fCLIP lysis buffer for 3 h at 4°C after adjusting NaCl concentration to 150 mM. In the meantime, cells were harvested and fixed with 0.1% (v/v) paraformaldehyde for 10 min at RT and immediately quenched by adjusting glycine concentration to 250 mM. The crosslinked cells were lysed using 400 μL of fCLIP lysis buffer for 10 min on ice and then sonicated for complete lysis. The NaCl concentration of the lysate was adjusted to 150 mM, and cell debris was separated by centrifugation for 15 min at max speed. The lysate was added to the C1QBP-conjugated beads and incubated for 3 h at 4°C. The beads were washed 4 times with the fCLIP wash buffer (20 mM Tris-HCl, pH 7.5, 150 mM NaCl, 10 mM EDTA, 0.1% NP-40, 0.1% SDS, 0.1% Triton X-100, and 0.1% sodium deoxycholate), and C1QBP-RNA complex was eluted from the beads by incubating in the elution buffer (200 mM Tris-HCl pH 7.4, 100 mM NaCl, 20 mM EDTA, 2% SDS, and 7 M Urea) for 3 h at 25°C. The eluate was treated with 2 mg/mL proteinase K overnight at 65°C in a thermomixer. RNA was purified using acid-phenol:Chloroform, pH 4.5.

Streptavidin immunoprecipitation

Total cell lysate was prepared using 400 μL of cell lysis buffer (10 mM NaCl, 2 mM EDTA, 0.5% Triton X-100, 0.5 mM DTT, and 10 mM Tris-HCl, pH 7.5) containing protease inhibitor cocktail. 30 μL of streptavidin-conjugated magnetic nanobead was washed 3 times with the lysis buffer. To remove endogenous biotins, the lysate was incubated with washed streptavidin-conjugated magnetic nanobead for 1 h at 4°C. The biotin-labeled RNA oligonucleotides with (Oligo-m⁵C) or without m⁵C (Oligo-C): 5′-biotin-GAGGUAUGAAXUGUAAGTT-3′ (X = C or m⁵C) were prepared. The CYTB-m⁵C and ND6-m⁵C sequences were 5′-biotin-AAGGAGUGAGm⁵CCGAAGUUUCA-3′ and 5′-biotin-GCAUGGGGGUm⁵CAGGGGUUGAG-3′, respectively. 2 μg of biotin-labeled RNAs was gently rotated with 300 μL pre-cleared cell lysate supplemented with 0.4 U/μL of recombinant RNase inhibitor at 4°C. The mixture of biotin-labeled RNAs and pre-cleared cell lysate was added into 30 μL of fresh streptavidin-conjugated magnetic nanobead and then gently rotated for 2 h at 4°C. The bead was washed 4 times with cell wash buffer (50 mM Tris-HCl, pH 7.5, 250 mM NaCl, 0.4 mM EDTA, 0.1% NP-40, and freshly added 0.4 U/μL recombinant RNase inhibitor). 30 μL of SDS loading buffer was added to the washed bead and prepared western blotting.

Quantification and statistical analysis

RT-qPCR data, cell viability assay, and image quantification results were analyzed using the one-tailed Student’s t-test. Data were biologically replicated at least 3 times, and replicate numbers for specific experiments were indicated in the figure legend. The error bars indicate the standard error of the mean. P-values ≤ 0.05 were regarded as statistically significant. * denotes p-values ≤ 0.05, ** indicates p-values ≤ 0.01, and *** indicates p-values ≤ 0.001.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Rabbit polyclonal NSUN4	Abcam	Cat# ab101625; RRID: AB_10711878
Rabbit polyclonal MRPP1	Invitrogen	Cat# PA5-41873; RRID: AB_2607115
Rabbit polyclonal C1QBP	Invitrogen	Cat# PA5-80387; RRID: AB_2787709
Mouse monoclonal GAPDH	Santa Cruz Biotechnology	Cat# sc-32233; RRID: AB_627679
Mouse monoclonal MTCO1	Abcam	Cat# ab14705; RRID: AB_2084810
Mouse monoclonal J2	English and Scientific Consulting Kft	Cat# 10010500; RRID: AB_2651015
Mouse monoclonal 5-methylcytosine	Abcam	Cat# ab10805; RRID: AB_442823
Mouse monoclonal Beta Tubulin (D3U1W)	Cell Signaling Technology	Cat# 86298; RRID: AB_2715541
Rabbit polyclonal AIF	Cell Signaling Technology	Cat# 4642; RRID: AB_2224542
Rabbit polyclonal POLRMT	Abcam	Cat# ab32988; RRID: AB_873619
Rabbit polyclonal PNPT1	Abcam	Cat# ab157109; RRID: AB_2910229
Rabbit monoclonal p-PKR	Abcam	Cat# ab81303; RRID: AB_1640780
Mouse monoclonal PKR	Santa Cruz Biotechnology	Cat# sc-6282; RRID: AB_628150
Rabbit monoclonal p-IRF3	Cell Signaling Technology	Cat# 4947; RRID: AB_823547
Rabbit monoclonal IRF3	Cell Signaling Technology	Cat# 11904, RRID: AB_2722521
Rabbit monoclonal p-eIF2α	Cell Signaling Technology	Cat# 3398, RRID: AB_2096481
Rabbit monoclonal eIF2α	Cell Signaling Technology	Cat# 5324, RRID: AB_10692650
Mouse monoclonal Lamin A/C	Santa Cruz Biotechnology	Cat# sc-376248, RRID: AB_10991536
Rabbit polyclonal JNK	Cell Signaling Technology	Cat# 9258; RRID: AB_10694056
Horse Anti-Mouse/Rabbit IgG Antibody (H+L) (Universal), Biotinylated	Vector Labs	Cat# BP-1400-50; RRID: AB_2910231
Donkey mouse IgG (H+L) secondary antibody, Alex Fluor 555	Invitrogen	Cat# A-31570; RRID: AB_2536180
Bacterial and virus strains
DH5-α competent Escherichia coli (E. coli)	Enzynomics	Cat# CP010
Stbl3 competent Escherichia coli (E. coli)	Thermo Scientific	Cat# C737303
Chemicals, peptides, and recombinant proteins
Blasticidin S HCl	Gibco	Cat# R21001; CAS 3513-03-9
Ampicillin	Sigma	Cat# A5354; CAS 69-53-4
Dimethyl sulfoxide (DMSO)	Santa Cruz Biotechnolgy	Cat# sc-358801; CAS 67-68-5
IMT1; LDC195943	This paper	35
2′-C-Methyladenosine (2-CM)	Santa Cruz Biotechnology	Cat# sc-283467; CAS: 15397-12-3
Lipofectamine 3000 Transfection Reagent	Invitrogen	Cat# L3000075
Chloramphenicol	Sigma	Cat# C0378; CAS 56-75-7
Polybrene	Sigma	Cat# TR-1003-G; CAS 28728-55-4
2M KCl	Invitrogen	Cat# AM9640G; CAS 7447-40-7
5M NaCl	Invitrogen	Cat# AM9760G; CAS 7647-14-5
Urea	Sigma	Cat# U5378-500G; CAS 57-13-6
0.5M EDTA	Invitrogen	Cat# AM9260G; CAS 60-00-4
Sodium deoxycholate	Sigma	Cat# D6750-25G; CAS 302-95-4
Glycerol	Invitrogen	Cat# 15514011; CAS
DTT	VWR Life Science	Cat# VWRV0281-25G; CAS 3483-12-3
TRIzol	Invitrogen	Cat# 15596026
TRIzol LS reagent	Invitrogen	Cat# 10296028
Recombinant RNase Inhibitor (40 U/μL)	Takara	Cat# 2313A
Recombinant DNase I (RNase-free)	Takara	Cat# 2270
GlycoBlue coprecipitant (15 mg/mL)	Invitrogen	Cat# AM9515
Random hexamer primer	Thermo Scientific	Cat# SO142
dNTP mixture (2.5 mM)	Takara	Cat# 4030
RevertAid Reverse Transcriptase (200 U/μL)	Thermo Scientific	Cat# EP0442
SuperScript IV Reverse Transcriptase	Invitrogen	Cat# 18090200
Paraformaldehyde	Sigma	Cat# P6148; CAS 30525-89-4
Triton X-100, Molecular Biology Grade	Promega	Cat# H5141; CAS 9036-19-5
Tween-20	Biosolution	Cat# BN015; CAS 9005-64-5
Bovine serum albumin (BSA)	RMBIO	Cat# BSA-BSH; CAS 9048-46-8
4′,6-Diamidino-2-Phenylindole, Dihydrochloride (DAPI)	Invitrogen	Cat# D1306; CAS 28718-90-3
MitoGreen	Promokine	Cat# PK-CA707-70054
Phosphate Buffered Saline (PBS), pH 7.4	Takara	Cat# T9181; CAS 7758-11-4
OmniPur Polyoxyethylene (20)	Millipore	Cat# 9480; CAS 9005-64-5
Proteinase K	Sigma	Cat# 3115879001; CAS 39450-01-6
Acid-Phenol:Chloroform, pH 4.5	Invitrogen	Cat# AM9722
nPFU-Forte	Enzynomics	Cat# P410
Dpn I	Enzynomics	Cat# R0545
BsmBI	NEB	Cat# R0580S
EcoRI	NEB	Cat# R0101S
HindIII	Thermo Scientific	Cat# ER0501
Agel	NEB	Cat# R3552S
T4 polynucleotide kinase	NEB	Cat# M0201S
T4 DNA ligase	NEB	Cat# M0202S
Nuclease micrococcal from Staphylococcus aureus	Sigma	Cat# N3755-50UN
Gelatin	MP Biomedicals	Cat# 901771; CAS 9000-70-8
Cytidine	Sigma	Cat# C122106-1G
FuGENE HD Transfection Reagent	Promega	Cat# E2311
Trichloroacetic acid (TCA)	Sigma	Cat# 91228; CAS 76-03-9
Sulforhodamine B sodium salt	Santa Cruz Biotechnology	Cat# sc-253615; CAS 3520-42-1
Higel-Agarose clear	E&S	Cat# HB0100500
Pierce protein A plus agarose	Thermo Scientific	Cat# 22810
AccuNanoBead™ Streptavidin Magnetic Nanobead	Bioneer	Cat# TA-1015-1
Protease inhibitor cocktail set	Sigma	Cat# 535140
1.5M Tris-HCl, pH 8.8	Biosolution	Cat# BT021-1; CAS 1185-53-1
0.5M Tris-HCl, pH 6.8	Biosolution	Cat# BT033; CAS 1185-53-1
1M Tris-HCl pH 7.0	Invitrogen	Cat# AM9850G; CAS 1185-53-1
1M Tris-HCl pH 8.0	Invitrogen	Cat# AM9855G; CAS 1185-53-1
Glycine	Bio-basic	Cat# GB0235; CAS 56-40-6
20% SDS	Biosolution	Cat# BS003; CAS 151-21-3
TEMED	Biorad	Cat# 1610801; CAS 110-18-9
Ammonium persulfate	Thermo Scientific	Cat# 17874; CAS 7727-54-0
Acrylamide-Bis solution (40%, 37.5:1)	Biosolution	Cat# BA005
10% NP-40	Biosolution	Cat# BN015; CAS 9016-45-9
Ethanol	Alfa-Aesar	Cat# A9951; CAS 64-17-5
Methanol	Merck	Cat# 106009; CAS 67-56-1
SensiFAST SYBR	Bioline	Cat# BIO-94005
2-propanol	Merck	Cat# 818766; CAS 67-63-0
Critical commercial assays
Subcellular Protein Fractionation Kit for Cultured Cells	Thermo Scientific	Cat# 78840
Hybrid-Q Plasmid Rapidprep	GeneAll	Cat# 100-150
NucleoBond Xtra Midi	Macherey-Nagel	Cat# 740410.50
Expin Gel SV	GeneAll	Cat# 102-102
Cell counting kit-8	Dojindo	Cat# CK04-13
EpiNext RNA Bisulfite-Seq Kit (Illumina)	Epigentek	Cat# P-9006-12
Ribo-Zero Plus rRNA depletion kit	Illumina	Cat# 20040526
TruSeq stranded total RNA library prep	Illumina	Cat# 20020596
Oligo Clean and Concentrator	Zymo Research	Cat# D4061
Clarity Western ECL Substrate	Biorad	Cat# BR170-5062
Pierce BCA protein assay kit	Thermo Scientific	Cat# 23227
RNAscope Multiplex Fluorescent assay and mtRNA probes (Hs-MT-ND5 and Hs-MT-ND5-sense)	ACD	Cat# 323100, Cat# 539451-C2, Cat# 539321
RNA fragmentation reagents	Invitrogen	Cat# AM8740
Deposited data
Unprocessed images	This paper	DOI: 10.17632/9z7spn5372.1
Bisulfite sequencing data	This paper	GSE246026
Experimental models: Cell lines
293T	ATCC	Cat# CRL-3216; RRID: CVCL_0063
MCF7	ATCC	Cat# HTB-22; RRID: CVCL_0031
HCT116	ATCC	Cat# CCL-247; RRID: CVCL_0291
Oligonucleotides
See Table S1 for the list of sgRNAs	This paper	N/A
See Table S2 for the list of siRNAs	Bioneer	N/A
See Table S3 for the list of qPCR primers	This paper	N/A
See Table S4 for the list of primers for strand-specific reverse transcription	This paper	N/A
See Table S5 for the list of primers for strand-specific qPCR	This paper	N/A
See Table S6 for the list of primers for mutagenesis	This paper	N/A
See Table S7 for the list of sequences for shRNAs	This paper	N/A
Recombinant DNA
lentiGuide-Puro	Addgene	Cat# 53963
lentiCas9-Blast	Addgene	Cat# 52962
psPAX2	Addgene	Cat# 12260
pMD2.G	Addgene	Cat# 12259
pEGFP-C1	NovoPro Bioscience	Cat# V012024
pLKO.1 Puro	Addgene	Cat# 8453
pLKO.1-Scrambled	Addgene	Cat# 136035
Software and algorithms
ImageJ	ImageJ	https://imagej.nih.gov/ij/
Image Lab	Bio-rad	SOFT-LIT-170-9690-ILSPC-V-6-1
Illustrator	Adobe	N/A
Other
DMEM	Welgene	Cat# LM001-05
RPMI 1640	Welgene	Cat # LM011-01
FBS	Gibco	Cat# 26140079
Trypsin-EDTA	Welgene	Cat# LS015-01
Opti-MEM, Reduced Serum Medium	Gibco	Cat# 31985070
0.22 μm Millex-GP filter unit	Millipore	Cat# SLGPM33RS
0.45 μm syringe filter	Corning	Cat# CLS431220
Immobilon-P PVDF membrane	Millipore	Cat# IPVH00010
Luria broth (LB) broth (Miller)	Millipore	Cat# 110285
T100 thermal cycler	Biorad	Cat# 1861096
SureCycler 8800	Agilent	N/A
Thermomixer	Eppendorf	Cat# 5382
AriaMx Real-time PCR system	Agilent	Cat# G8830A
QuantStudio 1 Real-time PCR system	Thermo Scientific	Cat# A40427

Highlights.

• CRISPR screening of mt-RBPs reveals key regulators of mt-dsRNA expression

• NSUN4 installs m⁵C RNA modification on mitochondrial mRNAs and lncRNAs

• C1QBP recognizes m⁵C-modified mitochondrial RNAs and recruits PNPT1

• m⁵C modification marks the mitochondrial RNAs for decay and cytosolic release