A chemical method to sequence 5-formylcytosine on RNA

Abstract

Epitranscriptomic RNA modifications can regulate biological processes, but there remains a major gap in our ability to identify and measure individual modifications at nucleotide resolution. Here we present Mal-Seq, a chemical method to sequence 5-formylcytosine (f⁵C) modifications on RNA based upon selective and efficient malononitrile-mediated labeling of f⁵C residues to generate adducts that are read as C-to-T mutations upon reverse transcription and PCR amplification. We apply Mal-Seq to characterize the prevalence of f⁵C at the wobble position of mt-tRNA(Met) in different organisms and tissue types and find that high-level f⁵C modification is present in mammals but lacking in lower eukaryotes. Our work sheds light on mitochondrial tRNA modifications throughout eukaryotic evolution and provides a general platform for characterizing the f⁵C epitranscriptome.

Graphical Abstract

The function of cellular RNA can be modulated by chemical modifications installed post-transcriptionally. Known as the epitranscriptome, over 150 distinct modifications have been reported to exist on RNA^1–2. A number of well-studied modifications have important roles in RNA metabolism, protein translation, and RNA trafficking³, however, we lack information on the function and distribution of most modifications. Further, RNA modification levels and their associated writer, eraser, and reader proteins can be dysregulated in certain disease states⁴, underscoring the need for a comprehensive understanding of epitranscriptomic mechanisms in biological systems.

A major challenge in the study of RNA modifications is the ability to map modifications at single-nucleotide resolution and measure their stoichiometry⁵. While Next-Generation Sequencing (NGS) has revolutionized transcriptomic studies, many modifications are “silent” upon RNA-seq analysis since they are reverse transcribed like the parent unmodified base, necessitating the development of alternative approaches for modification-specific sequencing. Approaches for modification mapping compatible with Illumina sequencing⁶ (the most commonly utilized NGS platform) generally fall into two categories: 1) antibody enrichment of modified RNAs^7–8 or 2) chemical/enzymatic conversion of the modified base to an adduct that can be identified based upon a distinctive reverse transcription (RT) signature^9–11. The second approach, particularly when the signature is a sequence mutation as opposed to an RT stop, is often preferred since it can provide higher resolution, less sequence bias, and modification stoichiometry; however, current strategies for RNA modification mapping mediated by chemical or enzymatic conversion are only applicable to a small number of modified bases, and there is a great need for the development of new approaches to characterize the epitranscriptome.

Herein, we develop a chemical approach to sequence 5-formylcytosine (f⁵C) on RNA at single-nucleotide resolution. 5-formylcytosine has been found on isolated tRNA isoacceptors^12–14, but we lack robust approaches to quantitatively sequence this modification and characterize its distribution across the transcriptome. Our strategy, which we name Mal-Seq, is based upon selective malononitrile-mediated labeling (Fig. 1a) and C-to-T conversion upon reverse transcription, amplification and sequencing. Chemical labeling with malononitrile is mild, efficient, and quantitative, and we exploit these properties to measure the levels of f⁵C at C34 on mt-tRNA(Met) in diverse organisms and tissue types.

Figure 1.

Malononitrile labeling of f⁵C on RNA. (a) Structure of malononitrile-f⁵C adduct and effects on Watson-Crick base pairing. (b) LC-QQQ-MS analysis of f⁵C levels in RNA oligo1 after treatment with malononitrile. Data are mean ± s.d. (n = 2). (c) LC-QQQ-MS analysis of f⁵C and hm⁵C levels in total RNA before and after treatment with malononitrile or 1,3-indandione. Data are mean ± s.d. (n = 3).

In order to sequence RNA f⁵C, we surveyed the literature for chemical transformations that would be selective for the modified base and generate a mutational signature. Notably, Yi and co-workers previously demonstrated that malononitrile¹⁵ and 1,3-indandione¹⁶ react with 5-formylcytosine in DNA to form adducts that induce C-to-T mutations upon DNA polymerase read-through (Fig. 1a and Fig. S1a). Therefore, we investigated the suitability of these reactions for sequencing f⁵C on RNA using total RNA and an artificial f⁵C-containing RNA transcript generated by in vitro transcription as model substrates. We started by quantifying depletion of f⁵C in RNA upon treatment with malononitrile or 1,3-indandione using nucleoside LC-MS/MS (Table S1). Gratifyingly, we measured 96.9±0.004% reduction in f⁵C levels upon treatment of our model f⁵C RNA with malononitrile (Fig. 1b). In addition, treatment of total cellular RNA with malononitrile or 1,3-indandione resulted in 90–92% (malononitrile: 91.5 ± 0.002%, 1,3-indandione: 90.0 ± 0.004%) reduction of f⁵C levels (Fig. 1c). Importantly, levels of related modifications in total RNA such as 5-methylcytidine (m⁵C) or 5-hydroxymethylcytidine (hm⁵C) remained unchanged (Fig. 1c and Fig. S1c), and analysis of RNA integrity using gel electrophoresis or Bioanalyzer assay demonstrated minimal RNA degradation (Fig. S1d and S1e), indicating that these transformations are selective for f⁵C and sufficiently mild for RNA sequencing.

Next, we tested whether the generated f⁵C adducts would produce a sequence mutation upon reverse transcription PCR (RT-PCR). Given comparable reaction efficiency between f⁵C and malononitrile or 1,3-indandione, we chose to work with malononitrile due to its enhanced solubility. We treated an in vitro transcribed RNA containing a single f⁵C site at 100% stoichiometry with malononitrile and performed RT-PCR. The f⁵C site was positioned in a Taqα1 digestion site such that mutation of C to another base could be monitored by restriction enzyme digestion¹⁷ (Fig. 2a). As shown, the RT-PCR products generated from untreated f⁵C RNA or an unmodified RNA are quantitatively digested by Taqα1, while malononitrile treatment of the f⁵C RNA inhibited digestion by ~50–60% (Fig. 2b and Fig. S2). To characterize the nature of the sequence change, we performed Sanger sequencing, which indicated that 60% of the transcripts contained a C-to-T mutation, while the remaining 40% contained a C (Fig. 2c). No other mutations were detected at the f⁵C site or surrounding residues. To further confirm our result and generate a calibration curve relating f⁵C stoichiometry and C-to-T conversion, we used high-throughput sequencing. Our data show a linear correlation between f⁵C levels ranging from 5–100% stoichiometry and malononitrile-induced C-to-T mutation (Fig. 2d and Table S2) with a conversion factor near 0.5 (i.e. ~50% C-to-T conversion after malononitrile treatment corresponds to 100% f⁵C). Given the depth of coverage afforded by high-throughput sequencing analysis, we could also detect mutations to G/A or deletions at the f⁵C site after malononitrile treatment, but the frequency of such events was low (0.1%). In addition, we tested RNA transcripts containing two or three f⁵C modification sites within different sequence contexts and observed similar levels of C-to-T conversion at each site upon malononitrile treatment (Fig. 2e, Fig. S2b and Table S3). Taken together, our results demonstrate that malononitrile-induced C-to-T mutations can be used to quantitatively sequence f⁵C modifications on RNA at nucleotide resolution and within diverse sequence contexts. We named this approach Mal-Seq.

Figure 2.

Quantitative sequencing of RNA f⁵C by Mal-Seq. (a) Schematic of Mal-Seq workflow. (b) Taqα1 enzymatic digestion assay to detect base mutations mediated by malononitrile labeling. (c) Malononitrile-mediated C-to-T conversion measured by colony picking and Sanger sequencing. Left: Representative Sanger sequencing traces; Right: Quantification analysis (n = 10). (d) Calibration curve relating f⁵C levels in RNA oligo1 and malononitrile-mediated C-to-T conversion. C-to-T mutations were measured by Illumina sequencing. Data are mean ± s.d. (n = 2). (e) C-to-T conversion at two distinct f⁵C-modified sites in RNA oligo2. Mutations were measured by Illumina sequencing. Data are mean ± s.d. (n = 2).

In order to demonstrate the utility of Mal-Seq, we applied our approach to characterize endogenous f⁵C modification levels in the anticodon loop of mt-tRNA(Met). Studies have indicated the presence of f⁵C at the C34 “wobble base” of mt-tRNA(Met) in a number of organisms, where it is proposed to facilitate the decoding of unconventional AUA and AUU Met codons among mitochondrial genes¹⁸. However, the lack of a unified, quantitative approach to characterize f⁵C modification levels has led to disparate findings regarding the prevalence and stoichiometry of f⁵C levels in biological systems. We started by applying Mal-Seq to quantify f⁵C on the wobble base of mt-tRNA(Met) from cultured human cells, where this modification has been best studied. Multiple groups have shown that f⁵C biogenesis at this position requires the sequential action of m⁵C methyltransferase NSUN3 followed by Fe(II), α-KG-dependent dioxygenase ALKBH1^{12−13, 19–20}, however quantification of modification levels has varied. Suzuki and co-workers used LC/MS analysis to show that C34 is fully modified to f⁵C²⁰, while two independent reports relying upon primer extension and bisulfite-based sequencing methods found a mixture of f⁵C and m⁵C at this position^{13, 19}. Therefore, we extracted total RNA from WT HEK293T cells and performed Mal-Seq using targeted RT-PCR of the anticodon region of mt-tRNA(Met). Our analysis shows 57.98±0.16% malononitrile-induced C-to-T conversion, indicating that mt-tRNA(Met) is fully modified with f⁵C at the wobble base (Fig. 3b, 3c and Table S4), consistent with Suzuki’s findings. In addition, we performed parallel analyses on RNA extracted from ALKBH1 or NSUN3 KO cells generated by CRISPR/Cas9 technology and found <0.3% C-to-T mutation (Fig. 3c, Fig. S3 and Table S4), confirming that both enzymes are required for f⁵C installation on mt-tRNA(Met).

Figure 3.

Mal-Seq reveals f⁵C34 in mt-tRNA(Met) is fully modified in HEK293T cells. (a) Schematic showing the formation of f⁵C by NSUN3 and ALKBH1 on mt-tRNA(Met). (b) RT-PCR of mt-tRNA(Met) amplified from WT, ALKBH1 KO, and NSUN3 KO cell lines. (c) C-to-T mutation at C34 on mt-tRNA(Met) detected by Mal-Seq using RNA from WT, ALKBH1 KO, and NSUN3 KO cells. Data are mean ± s.d. (n = 2).

We next characterized the presence of f⁵C on the wobble position of mt-tRNA(Met) in other eukaryotes. This modification has been found in organisms including squid²¹, flies²², chicken²³, -and cow¹⁴, but the extent of f⁵C modification in these species is largely unknown. We obtained total RNA from budding yeast, flies, C. elegans and mouse and characterized f⁵C levels by Mal-Seq using species-specific primers for each mt-tRNA(Met) (Fig. S4). In yeast and flies, we found no evidence of f⁵C on mt-tRNA(Met), indicating that this modification is absent or below our limit of detection (Fig. 4a and Table S5). Budding yeast lack a clear ALKBH1 homolog, which is consistent with low f⁵C modification. While f⁵C on mt-tRNA(Met) has been reported to occur in flies, modification levels were partial and quantitation was never performed²². In contrast, C. elegans showed 26.65 ±1.65 % Mal-Seq C-to-T conversion corresponding to ~50% of the mt-tRNA(Met) modified with f⁵C at the wobble position. This is in line with the recent characterization of a mitochondrial ALKBH1 homolog in this organism²⁴. We also measured f⁵C levels in different mouse tissues including heart, brain, and liver. In these tissues, we observed 37.8–42.9% (liver:37.8 ± 1.7%, heart: 41.5 ± 3.2%, brain: 42.9 ± 1.3%) C-to-T conversion corresponding to ~70–80% f⁵C modification, with a slight decrease in the liver (Fig. 4b and Table S5).

Figure 4.

Detection of f⁵C34 on mt-tRNA(Met) among different organisms. (a) Mal-Seq analysis of C34 on mt-tRNA(Met) using RNA extracted from C.elegans, budding yeast, and D. melanogaster. (b) Mal-Seq analysis of C34 on mt-tRNA(Met) in different mouse tissues. (c) Schematic showing f⁵C levels on mt-tRNA(Met) in different organisms detected by Mal-Seq. Data are mean ± s.d. (n = 2 for yeast, C. elegans, fly S2 cells; n = 4 for mouse tissues; n = 2 technical replicates for fly embryos). P values were determined using a two-sided unpaired student’s t-test.

In this work we develop a chemical sequencing approach, Mal-Seq, for detecting and quantifying 5-formylcytosine on RNA based upon malononitrile-mediated C-to-T conversion during RT-PCR. Using model f⁵C containing RNAs, we show that Mal-Seq can detect f⁵C sites in diverse sequence contexts and that C-to-T mutation frequency correlates linearly with f⁵C levels between 5 to 100% stoichiometry. An important limitation of our approach is that malononitrile-mediated C-to-T conversion at f⁵C sites is partial (~50%); therefore, identification and quantification of low stoichiometry f⁵C modifications (<10%) may pose a challenge.

We applied Mal-Seq to analyze f⁵C modification at the wobble base of mt-tRNA(Met) in different organisms and different tissues types. Our results show that mt-tRNA(Met) is fully modified with f⁵C in human HEK293T cell, and demonstrate ~70–80% modification levels in the mouse tissues that we assayed, consistent with the important role of this modification for mitochondrial translation in mammals^{12–13, 19}. Interestingly, modification levels are largely invariant in the different murine tissues that we sampled. Oxidation of m⁵C to f⁵C on mt-tRNA(Met) requires ALKBH1, which uses O₂ and α-KG as cofactors. In principle, ALKBH1 activity (and as a consequence mitochondrial translation efficiency) could be responsive to fluctuations in levels of these central metabolites; while this hypothesis remains to be tested explicitly, our data suggests that installation of f⁵C on mt-tRNA(Met) is largely independent of the physiological fluctuations in metabolite levels across different tissues under the conditions tested. In addition, we find that high-level f⁵C modification (>50%) is characteristic of mammals and absent in lower eukaryotes (Fig. 4c). Since recognition of mitochondrial Met AUA/AUU codons is important in many eukaryotes, other mechanisms must exist to support this role in systems lacking f⁵C. Alternatively, lower level f⁵C modification may be sufficient to satisfy the requirements of mitochondrial translation in these organisms.

Finally, the development of nucleotide resolution sequencing strategies for detecting f⁵C modification opens opportunities for mapping this modified base transcriptome-wide in different organisms. While our work was in review, two complementary strategies for chemical sequencing of f⁵C were reported by Meier²⁵ and Zhou²⁶ relying upon hydride reduction of f⁵C to dihydrouridine derivatives. In mammals, f⁵C has been detected in total RNA by LC-MS analysis²⁷, but it is unknown whether this modification occurs outside of the anticodon of mt-tRNA (Met). Interestingly, ALKBH1 has been shown to reside outside of the mitochondria in the nucleus²⁸, raising the possibility of f⁵C sites on non-mitochondrial RNAs. Chemical sequencing strategies, together with identification of the relevant writer enzymes, should enable comprehensive investigation of the f⁵C epitranscriptome and shed light on its role in biology. Such studies are underway and will be reported in due course.

ABBREVIATIONS

RT-PCR

reverse transcription polymerase chain reaction

f⁵C

5-formylcytosine

m⁵C

5-methylcytidine

hm⁵C

5-hydroxymethylcytidine

α-kG

alpha ketoglutarate