Enhanced ac4C detection in RNA via chemical reduction and cDNA synthesis with modified dNTPs

Abstract

The functional analysis of epitranscriptomic modifications in RNA is constrained by a lack of methods that accurately capture their locations and levels. We previously demonstrated that the RNA modification N4-acetylcytidine (ac4C) can be mapped at base resolution through sodium borohydride reduction to tetrahydroacetylcytidine (tetrahydro-ac4C), followed by cDNA synthesis to misincorporate adenosine opposite reduced ac4C sites, culminating in C:T mismatches at acetylated cytidines (RedaC:T). However, this process is relatively inefficient, resulting in <20% C:T mismatches at a fully modified ac4C site in 18S rRNA. Considering that ac4C locations in other substrates including mRNA are unlikely to reach full penetrance, this method is not ideal for comprehensive mapping. Here, we introduce “RetraC:T” (reduction to tetrahydro-ac4C and reverse transcription with amino-dATP to induce C:T mismatches) as a method with enhanced ability to detect ac4C in cellular RNA. In brief, RNA is reduced through NaBH₄ or the closely related reagent sodium cyanoborohydride (NaCNBH₃) followed by cDNA synthesis in the presence of a modified DNA nucleotide, 2-amino-dATP, that preferentially binds to tetrahydro-ac4C. Incorporation of the modified dNTP substantially improved C:T mismatch rates, reaching stoichiometric detection of ac4C in 18S rRNA. Importantly, 2-amino-dATP did not result in truncated cDNA products nor increase mismatches at other locations. Thus, modified dNTPs are introduced as a new addition to the toolbox for detecting ac4C at base resolution.

INTRODUCTION

Chemical modification of RNA plays key roles in the regulation of posttranscriptional gene expression. Occurring in all four nucleobases and a variety of sequence contexts, more than 170 modifications have been detected in eukaryotic RNA (Boccaletto et al. 2022). However, efforts to understand their functions have been hampered by enzymatic redundancies and a general lack of methods to interrogate their localization in specific substrates at a global scale. Recent advances have overcome some of these obstacles by enabling high-resolution global mapping of specific modifications through antibody-based enrichment or chemical methods that alter the behavior of the modified RNA during sequencing (Motorin and Marchand 2021; Zhang et al. 2022b). Currently, these mapping approaches are restricted to a handful of modifications based on the availability of suitable antibodies and base chemical reactivity. Furthermore, the methodological limitations associated with the mapping approaches have introduced significant caveats: Antibodies may demonstrate cross-reactivity or preferential enrichment of specific sequence contexts, whereas chemical approaches are often inefficient and produce off-target effects that complicate computational pipelines (Dai et al. 2017; Zhu et al. 2017; Grozhik et al. 2019; Helm et al. 2019; Cui et al. 2021). Of particular concern to the RNA modification community, methodological differences have led to disparate conclusions regarding the distribution of modifications within substrates, particularly in relatively low abundance mRNAs (Schwartz et al. 2014; Linder et al. 2015; Li et al. 2017; Safra et al. 2017; Koh et al. 2019; Sendinc et al. 2019; Sas-Chen et al. 2020; Thalalla Gamage et al. 2021; Arango et al. 2022; Sturgill et al. 2022; Muthmann et al. 2023).

The RNA modification N4-acetylcytidine (ac4C) possesses features that have enabled its mapping through both antibody- and chemical-based approaches. ac4C represents the sole acetylation event in eukaryotic RNA and is uniquely installed by the NAT10 enzyme or its orthologs (Ikeuchi et al. 2008; Ito et al. 2014a, b; Sharma et al. 2015; Boccaletto et al. 2022). We were thus able to produce an antibody that displayed high specificity to ac4C for use in ac4C-RNA-immunoprecipitation and sequencing (acRIP-seq) comparing wild-type and NAT10 null cells (Sinclair et al. 2017; Arango et al. 2018). Through this approach, we effectively enriched the known acetylated regions within 18S rRNA and tRNA^Ser/Leu and uncovered an unexpected prevalence within mRNA (Arango et al. 2018). However, in the course of our studies, we determined that ac4C exerts location-specific impacts on mRNA translation, thus necessitating a higher-resolution mapping method (Arango et al. 2022). To that end, we developed RedaC:T-seq for base-resolution mapping of ac4C sites in transcriptome-wide results. In brief, sodium borohydride (NaBH₄) reduction of ac4C produces N4-acetyl-3,4,5,6-tetrahydrocytidine (tetrahydro-ac4C), which preferentially base-pairs with adenosine during cDNA synthesis, resulting in C:T mismatches in sequencing results (Thomas et al. 2018; Arango et al. 2022). Through RedaC:T-seq, we produced precision ac4C maps that solidified positional dependencies in mRNA translation. However, RedaC:T-seq is limited by its generally low efficiency, wherein a C:T mismatch rate of <20% was observed at a 100% modified ac4C site in 18S rRNA, thus potentially obscuring detection of ac4C in lower abundance and lower stoichiometry mRNA substrates (Arango et al. 2022).

Here, we introduce RetraC:T (reduction to tetrahydro-ac4C and reverse transcription with amino-dATP to induce C:T mismatches) as a chemical method for mapping ac4C at base resolution in RNA at improved efficiency. Based on the unique hydrogen bonding capabilities of ac4C after reduction to tetrahydro-ac4C, we reasoned that the inclusion of unnatural 2′–deoxy-NTPs with base-pairing features that complement tetrahydro–ac4C in reverse transcription (RT) reactions would lead to enhanced mismatch rates at ac4C sites in sequencing results. Based on structural properties and preliminary data, 2–amino-dATP emerged as our lead candidate. Through a detailed analysis encompassing distinct RNA treatments and RT conditions, 2–amino–dATP consistently improved C:T mismatch detection at two ac4C sites in 18S rRNA in Sanger, Illumina, and nanopore sequencing as compared to dNTPs alone. Importantly, we did not observe any off-target results in the form of truncation products or nonspecific mismatches associated with 2-amino-dATP inclusion in RT reactions. Our findings introduce modified dNTPs as novel tools for the detection of acetylated cytidine through complementary binding properties.

RESULTS

Inefficient detection of ac4C through current approaches

Detection of ac4C at base resolution in sequencing results is facilitated through its sensitivity to chemical reduction to generate tetrahydro-ac4C, which displays altered base-pairing properties (Miller and Cerutti 1967; Thomas et al. 2018). During cDNA synthesis, tetrahydro-ac4C preferentially base-pairs with adenosine, culminating in C:T mismatches during amplification and sequencing (Fig. 1A). Previous studies exploited this property to map ac4C within cellular RNA, with some distinctions: In RedaC:T-seq, ac4C is reduced through 1 h of exposure to NaBH₄ at 55°C in basic conditions, whereas ac4C-seq relies on 20 min of exposure to sodium cyanoborohydride (NaCNBH₃) at 20°C in acidic conditions (Sas-Chen et al. 2020; Arango et al. 2022). However, both approaches are subject to inherent inefficiencies, resulting in only partial C:T mismatch detection at acetylated 18S rRNA residues of known stoichiometry (Arango et al. 2022; Bortolin-Cavaillé et al. 2022). These findings underscore the need for a more robust and reproducible protocol for ac4C mapping.

FIGURE 1.

Acetylated cytidines are inefficiently captured through available methods. (A) Schematic depicting altered base-pairing following ac4C reduction to tetrahydro-ac4C (H4-ac4C). (B) Agarose gel analysis of total RNA treated with NaCNBH₃ for 20 min at 20°C or NaBH₄ for 1 h at 55°C. These conditions were used in the ac4C-seq and RedaC:T-seq studies, respectively. (C) Agarose gel analysis of total RNA integrity after treatment with NaCNBH₃ for the indicated times. (D) Sanger sequencing surrounding the ac4C site in 18S rRNA helix 34 with cDNA produced from the NaCNBH₃ time course shown in C. RT was performed with TGIRT. (E) Mass spectrometric analysis of ac4C and C levels in an RNA oligo containing a single ac4C after reduction with NaBH₄ or NaCNBH₃. An unmodified RNA oligo with a single C serves as control. Data represent the Mean ± SD as a percent of detected guanosine, n = 3. (F) NaBH₄ or NaCNBH₃ reduction of total RNA and cDNA synthesis with the NEBNext or TGIRT enzymes, followed by PCR amplification with primers flanking ∼90%–100% acetylated sites in 18S rRNA helices 34 and 45. Sanger sequencing results surrounding the known ac4C sites as indicated. (G) Reduction of an RNA oligo containing a single C or ac4C as depicted and cDNA synthesis with the NEBNext or TGIRT enzymes, followed by primer extension assay with a radiolabeled primer.

A central challenge in the detection of ac4C via reduction is balancing its natural lability against the harsh impact of reducing agents on RNA integrity. In comparing the conditions used in the RedaC:T-seq and ac4C-seq studies, we find that the two methods yield distinctive influences on these parameters. Whereas the alkaline conditions used in NaBH₄ reduction naturally fragment RNA to a size that is directly amenable to Illumina sequencing with no perceptible loss in total RNA abundance, NaCNBH₃ reduction results in widespread degradation and generally decreased RNA levels (Fig. 1B). Although the molecular basis supporting RNA losses upon exposure to NaCNBH₃ is unclear, degradation is not observed in HCl-treated controls, pointing to a direct relationship to reducing agent activity (Supplemental Fig. S1A). RNA degradation occurs within minutes of exposure to NaCNBH₃ and is concomitant with the emergence of C:T mismatches in Sanger sequencing of RT-PCR amplicons surrounding a fully acetylated 18S rRNA residue, illustrating the interconnectedness of these parameters (Fig. 1C,D). Considering that NaCNBH₃-treated RNA must undergo additional fragmentation to achieve an appropriate size distribution before sequencing, these losses have the potential to introduce unintended representational biases in ac4C-seq pipelines. Notably, ac4C detection is effectively ablated in the mass spectrometric analysis of an RNA oligo containing a single ac4C and no additional cytidines after 5 min of exposure to NaCNBH₃, raising the possibility that the impact on RNA integrity in ac4C-seq pipelines can be minimized through decreasing the time of treatment (Fig. 1E; Supplemental Table S1). We have accordingly adjusted the time of NaCNBH₃-treatment to 5 min for the purposes of our study, unless specified otherwise.

NaBH₄ and NaCNBH₃ reduction are also distinguished in their capacities to reduce ac4C to tetrahydro-ac4C. Specifically, the basic reaction conditions associated with NaBH₄ reduction result in unintended passive deacetylation, whereas the acidic conditions used in NaCNBH₃ reduction buffer against this off-target activity (Miller and Cerutti 1967; Thomas et al. 2018; Sas-Chen et al. 2020). Hence, the potential for ac4C detection through C:T mismatches in RedaC:T-seq workflows is effectively blunted before cDNA synthesis. This is illustrated in mass spectrometric analysis of an RNA oligo containing a single ac4C and no additional cytidines: Exposure to NaBH₄ for 1 h at 55°C effectively ablated ac4C detection, but also increased cytidine levels to the above background (Fig. 1E). These findings are consistent with efficient on-target reduction of ac4C to tetrahydro-ac4C, and undesired off-target deacetylation. Accordingly, we routinely observe only partial C:T mismatches at ∼90%–100% acetylated 18S rRNA nucleotide (nt) 1842 within helix 45 and nt 1337 within helix 34 in Sanger sequencing of RT-PCR amplicons from NaBH₄-treated RNA, irrespective of reverse transcriptase (RTase) choice (Fig. 1F; Taoka et al. 2018). Although the reaction kinetics of ac4C reduction is relatively robust as compared to deacetylation (Miller and Cerutti 1967), we were unable to shift the balance in favor of tetrahydro-ac4C production through decreasing NaBH₄-reaction time or temperature (Supplemental Fig. S1B; Supplemental Table S1).

Although NaBH₄ and NaCNBH₃ offer distinct advantages in ac4C mapping by maintaining sequencing library integrity and minimizing passive deacetylation, respectively, both methods have inherent limitations that temper their utility in transcriptome-wide sequencing. Significantly, both approaches exert pleiotropic impacts on RT that directly relate to ac4C reduction: In addition to C:T mismatches, tetrahydro-ac4C interferes with RT, resulting in a prominent truncation product in primer extension analysis of an oligo with a single ac4C after exposure to NaBH₄ or NaCNBH₃ (Fig. 1G). Although mitigated, this “RT stop” was observed in cDNA produced with the highly processive thermostable group II intron RTase (TGIRT) used in ac4C-seq, pointing to the potential for a combination of outcomes in sequencing results (Fig. 1G; Thomas et al. 2019). This notion is supported in the general decrease in full-length cDNA in primer extension of reduced ac4C-containing RNA as compared to controls (Fig. 1G). The net impact of these pleiotropic effects would be a general underestimation of ac4C levels through computational pipelines that rely on C:T mismatches to gauge its prevalence. Potentially related, neither method achieves stoichiometric detection of ac4C at fully acetylated sites in 18S rRNA, underscoring significant obstacles in identifying moderately acetylated locations (Fig. 1F). Given these challenges, we aimed to develop a method that improves our ability to reproducibly detect ac4C at base resolution in RNA-seq results.

Noncanonical dNTPs improve C:T mismatches at reduced ac4C sites

To improve ac4C detection in base-resolution sequencing, we investigated whether the unique molecular properties of tetrahydro-ac4C could be exploited to increase mismatch prevalence. This premise is rooted in the disconnect between the completeness of ac4C reduction in mass spectrometric analysis versus incomplete C:T mismatch detection in Sanger sequencing results, highlighting a possible avenue for improvement at the level of cDNA synthesis (Fig. 1E,F). Specifically, we explored whether incorporating modified NTPs with binding properties that complement tetrahydro-ac4C into RT reactions would augment mismatch detection at ac4C sites. To this end, we identified three commercially available noncanonical NTPs with molecular characteristics that are favorable to robust base-pairing with tetrahydro-ac4C during RT, while having limited potential to pair with cytidine or unreduced ac4C, thus boosting the potential for altered base-pairing opposite ac4C sites (Fig. 2A). Once incorporated into cDNA, the selected NTPs are predicted to specifically manifest in C:T mismatches at ac4C sites given chemical features that favor binding to dTTP during second-strand synthesis (Supplemental Fig. S2A,B). Specifically, 2–amino–purine–TP is predicted to form two stabilizing hydrogen bonds with tetrahydro-ac4C through its free hydrogen donor and acceptor, whereas 2–amino–dATP and 2–hydroxy–dATP are predicted to form three hydrogen bonds through their two hydrogen donors and single acceptor (Fig. 2A). In all cases, incorporation opposite unreduced ac4C or cytidine is disfavored because of poor thermodynamics, electrical repulsion, and/or steric hindrance (Fig. 2A).

FIGURE 2.

Improved detection of ac4C in total RNA through noncanonical dNTPs. (A) Predicted base-pairing interactions involving the noncanonical NTPs 2-amino-dATP (2-NH₂), 2-hydroxy-dATP (2-OH), or 2-amino-purineTP (2-AP). (B) RT-PCR of HeLa cell total RNA with primers flanking a fully acetylated ac4C site in 18S rRNA helix 45. PCR was performed with the canonical dNTPs or with dGTP either partially (75% analog:25% dGTP) or fully substituted by the indicated analog. (C) Reduction of an RNA oligo containing a single C or ac4C and cDNA synthesis with HIV RTase and the canonical dNTPs, or with dGTP either partially (75%:25%) or fully substituted by the indicated analog. Primer extension assay with a radiolabeled primer and polyacrylamide gel electrophoresis. (D) Quantification of the primer extension results from NaCNBH₃-treated RNA in C comparing the signal intensity associated with unincorporated primer or full-length cDNA as a percent of total lane volume. Volumes were derived from identically sized boxes in ImageQuant. (E) NaBH₄ and NaCNBH₃ reduction of total RNA and cDNA synthesis with TGIRT, SSIV, or HIV RTase with the canonical dNTPs (100% dGTP) or with dGTP partially substituted by the indicated analog (75% analog:25% dGTP). PCR amplification with primers flanking a ∼100% acetylated site in 18S rRNA helix 45 and Sanger sequencing results surrounding the known ac4C sites. (F) Quantification of C:T mismatch levels through chromatogram peak heights at 18S rRNA nt 1842 from E.

Supporting the predicted specificity, the selected analogs failed to substitute for guanosine in reactions lacking dGTP in gel-based visualization of RT-PCR amplicons encompassing the ac4C site in 18S rRNA helix 34 (Fig. 2B). In contrast, the omission of dGTP did not impede RT reactions involving tetrahydro-ac4C and the analogs. In brief, an RNA oligo containing a single ac4C was reduced with NaCNBH₃ for 5 min or NaBH₄ for 1 h and reverse transcribed in the presence of the canonical dNTPs or in the presence of dNTPs in which the dGTP component has been completely replaced with the queried analog or a 25%:75% mixture of dGTP and analog, respectively (Fig. 2C). RT was accomplished with HIV RTase given its enhanced processivity in the presence of modified NTPs (Werner et al. 2020). Primer extension analysis of the resulting RT products revealed several notable features. Significantly, full-length cDNA represents the predominant species in all cases and a prominent RT stop is not observed in any condition, including in the absence of dGTP (Fig. 2C). In addition to demonstrating the enhanced processivity of HIV RTase as compared to NEBNext and TGIRT enzymes upon encountering tetrahydro-ac4C (Figs. 1G and 2C, comparing dGTP lanes), the lack of a truncated RT product corresponding to the ac4C site in reactions omitting dGTP establishes that all three analogs were readily accepted by the RTase and incorporated into the templated cDNA (Fig. 2C). Markedly, the level of unincorporated primer was visibly decreased for lanes corresponding to NaCNBH₃-treated RNA that had been reverse transcribed in the presence of the analogs but lacking dGTP as compared to dGTP alone or a mixture of dGTP and analog (Fig. 2C). Densitometric analysis confirmed this observation and further established that analog alone produced more full-length cDNA relative to reactions including dGTP (Fig. 2D). The same trend was observed for NaBH₄-treated RNA (Supplemental Fig. S2C). These findings indicate that the analogs increase RT efficiency through tetrahydro-ac4C sites (Fig. 2D).

Armed with preliminary evidence bolstering the predicted binding between the unnatural NTPs and reduced ac4C, we next sought to determine the impact on C:T mismatch detection in sequencing results. To ensure broad applicability, total RNA was reduced with either NaBH₄ or NaCNBH₃, and RT was performed with a natural retroviral RTase (HIV), engineered retroviral RTase (Superscript IV), or group II intron RTase (TGIRT) in the presence of the canonical dNTPs, or dNTPs in which dGTP has been replaced with a 25%:75% mixture of dGTP and analog. To gauge the impact on a known physiological substrate, we focused on uniformly acetylated 18S rRNA nt 1842 within helix 45 and assessed C:T mismatch frequency through Sanger sequencing of RT-PCR amplicons from reduced RNA and mock-treated controls.

Supporting the predicted preferential association between the selected unnatural NTPs and tetrahydro-ac4C, C:T mismatches were largely increased in chemically reduced RNA reverse transcribed with the analogs as compared to dNTPs alone (Fig. 2E). This increase was particularly evident for NaBH₄-treated RNA, wherein mismatches at nt 1842 were modest and difficult to discern from background in chromatographic representations of the dNTP alone condition, but clearly visualized in the presence of the analogs (Fig. 2E). Considering that the RT reactions were derived from the same reduced RNA source, in addition to demonstrating the utility of modified NTPs for ac4C detection, the increase in mismatch rate relative to dNTPs alone bolsters the notion that preferential base-pairing between tetrahydro-ac4C and adenosine is incomplete and subject to potential improvement.

Quantification of the C:T mismatch rate at nt 1842 in the NaBH₄-treated samples revealed notable distinctions between the tested RTases and unnatural NTPs. Specifically, HIV RTase and 2-amino-dATP produced the largest gains in C:T mismatch rates, reaching ∼60% when used in combination (Fig. 2F). Given that passive deacetylation is a natural consequence of NaBH₄-treatment, affecting ∼20% of ac4C in our mass spectrometric analysis (Fig. 1E), this value reflects near complete ac4C detection through C:T mismatches. Supporting this premise, the combination of HIV RTase and 2-amino-dATP yielded ∼85% C:T conversion at nt 1842 in RNA reduced with the more effective reducing agent NaCNBH₃, which is less susceptible to passive deacetylation (Figs. 1E and 2E,F). These findings reveal 2-amino-dATP, particularly when used in conjunction with HIV RTase, as an important addition to ac4C detection at base resolution through firmly shifting tetrahydro-ac4C base-pairing toward C:T mismatch introduction in second-strand cDNA synthesis. We accordingly term our method RetraC:T to reflect this preference.

Linear detection of ac4C in Sanger sequencing through RetraC:T

Through examining C:T mismatch ratios in the context of uniformly acetylated nucleotides in 18S rRNA, we established RetraC:T as an effective method for pinpointing ac4C sites at base resolution. To further evaluate its capacity to quantify relative ac4C levels, we conducted RetraC:T followed by Sanger sequencing on RNA mixtures with known ac4C concentrations. Specifically, we combined total RNA from wild-type and NAT10^−/− HeLa cells in set ratios, followed by NaBH₄ reduction and cDNA synthesis with HIV RTase in the presence of 2-amino-dATP. We focused this quantitative analysis on nt 1337 to avoid potential RT artifacts associated with the proximity of nt 1842 to the 3′ end of 18S rRNA, which could result in inefficient random hexamer priming. Visual inspection of Sanger sequencing traces from RT-PCR amplicons encompassing nt 1337 revealed a discernible rise in C:T mismatch rates with increasing wild-type RNA content (Fig. 3A). To further quantify the relationship between ac4C abundance and mismatch rate, we extracted peak heights from the Sanger data and plotted percent C:T mismatches at nt 1337 versus percent wild-type RNA. In so doing, we noted a linear correlation between C:T mismatch level and proportion of wild-type RNA in the mix, with an R² value of 0.96 (Fig. 3B). From this, we can infer that reactions with the canonical dNTPs reach ∼55% of the C:T mismatch level of those including 2-amino-dATP, as assessed through comparison to cDNA produced from reduced wild-type RNA with dNTPs alone (Fig. 3B).

FIGURE 3.

Quantitative detection of ac4C with 2-amino-dATP in Sanger sequencing. (A) NaBH₄ reduction of total RNA from wild-type (WT) and NAT10^−/− HeLa cells mixed at the indicated ratios and cDNA synthesis with HIV RTase and the canonical dNTPs (100% dGTP) or with dGTP partially substituted with 2-amino-dATP (75% analog:25% dGTP). Sanger sequencing results of a PCR amplicon surrounding the ac4C site in 18S rRNA helix 34. (B) Quantification of C:T mismatch levels through Sanger chromatogram peak heights at 18S rRNA nt 1337 from A. The trendline depicts a linear increase in C:T mismatch detection with increasing percent WT RNA. The gray dotted arrow indicates the C:T mismatch level in parallel processed samples from 100% WT RNA reverse transcribed with dNTP alone. (C) Sanger sequencing traces of RT-PCR amplicons surrounding the helix 34 ac4C site from mixtures of untreated and NaCNBH₃-treated HeLa RNA, as indicated. RT was performed with HIV RTase and the canonical dNTPs (100% dGTP), or with dGTP partially substituted with 2-amino-dATP (75% analog:25% dGTP). (D) Quantification of C:T mismatch levels through Sanger chromatogram peak heights at 18S rRNA nt 1337 from C. The trendlines depict enhanced mismatch detection in reactions including 2-amino-dATP. (E) Quantification of C:T mismatch levels at 18S rRNA nt 1337 through Sanger chromatogram peak heights from NaBH₄ and NaCNBH₃-reduced WT HeLa RNA. cDNA synthesis was accomplished with HIV and SSIV RTases with decreasing concentrations of dGTP starting from the 500 µM used in canonical dNTP mixtures.

A challenge in precisely determining the linearity of 2-amino-dATP across mixtures including NAT10^−/− RNA arises from residual NAT10 activity, which results in the retention of ∼10% of ac4C and precludes a true null condition (Arango et al. 2018). At the opposite extreme, NaBH₄ reduction leads to significant passive deacetylation, rendering full conversion to tetrahydro-ac4C at uniformly acetylated sites unattainable (Fig. 1E). Thus, to accurately gauge the effectiveness of RetraC:T in capturing ac4C sites across the expected cellular stoichiometry range of >0%–100%, we reduced wild-type HeLa RNA with NaCNBH₃ for 20′ to achieve maximal conversion to tetrahydro-ac4C, and then mixed treated and untreated RNA at specific ratios. RT was accomplished with HIV RTase and the canonical dNTPs or a mixture including 2-amino-dATP. Inspection of Sanger sequencing traces of RT-PCR amplicons showed increasing C:T mismatches at nt 1337 as the proportion of treated RNA within the mixtures increases (Fig. 3C). Significantly, mismatches were evident with just a small percentage of treated RNA in the presence of 2-amino-dATP, whereas the same RNA pools did not show visible mismatches until reaching ∼25% treated RNA in parallel processed RT reactions generated with dNTPs alone (Fig. 3C). This distinction in relative efficiency is solidified through the extraction of peak heights from the Sanger data and plotting of percent C:T mismatches at nt 1337 versus percent treated RNA (Fig. 3D; Supplemental Table S2). Although the RetraC:T reactions achieved a robust linear correlation (R²= 0.99) that conformed to the expected range of 0% to ∼100%, reactions involving dNTPs alone displayed ∼50% efficiency (Fig. 3D). Notably, detection of ac4C through the canonical dNTPs was particularly compromised in the lower stoichiometry range: Mismatches were not detected until reaching 25% treated RNA where a 5.7% C:T rate was registered. This compares to 26.7% C:T mismatches for the same RNA sample in the presence of 2-amino-dATP (Fig. 3D; Supplemental Table S2). Importantly, increased detection of ac4C through RetraC:T is not attributable to the decreased concentration of dGTP in the dNTP mix. C:T mismatch rates in NaCNBH₃- or NaBH₄-reduced RNA reverse transcribed with HIV RTase or SSIV RTase were not altered through reducing dGTP levels from the 500 µM used in canonical dNTP mixes to the 125 µM used in RetraC:T (Fig. 3E; Supplemental Fig. S3A). These findings demonstrate both the linearity and robustness of RetraC:T in identifying ac4C sites at base resolution and have significant implications for the survey of ac4C in mRNA, where levels are expected to be lower than uniform rRNA substrates.

Detection of ac4C with high specificity in Illumina sequencing through RetraC:T

Although RetraC:T effectively enhances ac4C detection in reduced RNA, the introduction of a noncanonical NTP to cDNA reactions creates the potential for off-target sequencing variability. Indeed, just as 2-amino-dATP efficiently binds to dTTP during second-strand cDNA synthesis (Supplemental Fig. S2B), it can effectively bind to both tetrahydro-ac4C and uracil during first-strand production (Fig. 2A; Supplemental Fig. S4A). Although interactions with uracil would not manifest in mismatches during sequencing, they serve as an example of the expanded specificity of noncanonical NTPs in Watson–Crick base-pairing. To examine the influence of RetraC:T on overall sequencing integrity, we thus took an extended view through Illumina sequencing of total RNA. Although NaCNBH₃ was ∼2× more efficient than NaBH₄ at capturing ac4C sites through C:T mismatches in Sanger sequencing (Fig. 3B,D), the alkaline fragmentation associated with NaBH₄ reduction is advantageous from a sequencing perspective and was thus selected for this purpose (Fig. 1B,E). To explore both the specificity and linearity of RetraC:T in a variety of sequencing applications, reactions encompassed RT with HIV and SSIV RTases, and mixtures of wild-type and NAT10^−/− HeLa total RNA (Supplemental Fig. S4B,C). Libraries were constructed from double-stranded cDNA and sequenced on the NextSeq 2000 platform, resulting in a range of ∼4–5 million reads per sample (Supplemental Table S3; Supplemental Fig. S4C). Visualization of reads aligning to 18S rRNA comparing the impact of reduction and 2-amino-dATP on general sequencing quality demonstrated good 5′–3′ coverage with no apparent biases between conditions and major mismatches limited to several locations with known modified residues (Fig. 4A; Taoka et al. 2018).

FIGURE 4.

Linear and specific detection of ac4C through RetraC:T and Illumina sequencing. (A) Browser views detailing 18S rRNA read coverage in Illumina sequencing results from untreated and NaBH₄-treated RNA reverse transcribed with HIV and SSIV RTase with dNTPs alone or supplemented with 2-amino-dATP. Visible mismatches at a 0.15 allele threshold correspond to known modifications, as shown. (B) Plot showing percent C:T mismatches at 18S nt 1337 versus percent wild-type HeLa RNA in a mixture with NAT10^−/− RNA in RetraC:T-Illumina-seq with HIV and SSIV RTase. The gray dotted arrow indicates the C:T mismatch level in parallel processed samples from 100% WT RNA reverse transcribed with HIV or SSIV RTase and dNTP alone. (C) Browser views surrounding the ac4C site at 18S nt 1337 from Illumina sequencing of untreated or NaBH₄-treated RNA reverse transcribed in the presence of 2-amino-dATP or dNTPs alone, as shown. C:T mismatches are depicted in red. (D) Cumulative mismatch rates from sequencing of untreated or NaBH₄-treated RNA reverse transcribed in the presence of 2-amino-dATP or dNTPs alone, derived from each of the 1869 nt in 18S rRNA and segregated by modification status. (E) Browser views surrounding m1acp3Ψ at 18S nt 1248 from Illumina sequencing of untreated or NaBH₄-treated RNA reverse transcribed in the presence of 2-amino-dATP or dNTPs alone, as shown. T:C mismatches are depicted in blue.

Consistent with the Sanger sequencing results (Fig. 3B), RetraC:T displayed a robust linear association between ac4C abundance and C:T mismatch detection in Illumina sequencing (RetraC:T-Illumina-seq), as evidenced through comparing the percent C:T mismatches at nt 1337 to percent wild-type RNA in mixtures with NAT10^−/− RNA (Fig. 4B). Linearity was observed for both HIV and SSIV RTases with absolute mismatch levels elevated in the former (Fig. 4B). Notably, although marginally decreased as compared to Sanger sequencing, the maximal mismatch value achieved through RetraC:T-Illumina-seq substantially outperforms the ∼15% mismatch value registered at nt 1337 in previously performed RedaC:T-seq (Figs. 3B and 4B; Arango et al. 2018). In addition, control reactions generated with 100% wild-type HeLa RNA and dNTPs alone produced C:T mismatch values at nt 1337 that are comparable to the expected value from RetraC:T-Illumina-seq of a 59% wild type:41% NAT10^−/− mixture of HeLa RNA derived from the linear equation (Fig. 4B). Importantly, browser views surrounding 18S nt 1337 showed no signs of sequencing interruptions or RT truncations associated with the use of 2-amino-dATP (Fig. 4C). The same tendencies are seen at nt 1842, despite low sequencing coverage at the extreme end of the 18S molecule (Supplemental Fig. S4D; Supplemental Table S4). Critically, linear detection of ac4C is preserved in RNA mixtures with 25% wild-type RNA (Fig. 4B), emphasizing the utility of RetraC:T-Illumina-seq at lower stoichiometry sites in cellular RNA.

To further assess the specificity of 2-amino-dATP for reduced ac4C, we examined for mismatches at all other nucleotides in sequencing reads aligning to 18S and 28S rRNAs, where the locations and levels of modified residues are well defined (Supplemental Tables S4 and S5). In both molecules, mismatches were exceedingly low for the canonical nucleobases and were uninfluenced by the inclusion of 2-amino-dATP (Fig. 4D; Supplemental Fig. S4E). Likewise, modified nucleotides were generally unaffected as compared to untreated RNA with several notable exceptions, none of which showed increased mismatches in the presence of 2-amino-dATP. In brief, although the reduction of N7-methylguanosine (m7G) at 18S nt 1639 induces abundant mismatches in sequencing results, as previously established (Enroth et al. 2019), these were unaffected by the inclusion of 2-amino-dATP during RT (Fig. 4D; Supplemental Fig. S4D). Similarly, elevated basal mismatches associated with N1-methyladenosine (m1A) at 28S nt 1309 were unperturbed by NaBH₄ reduction or 2-amino-dATP inclusion (Supplemental Fig. S4E). In contrast, basal mismatches associated with hypermodified N1-methyl-3-(3-amino-3-carboxypropyl)pseudouridine (m1acp3Ψ) at 18S rRNA nt 1248 and N3-methyluridine (m3U) at 28S rRNA nt 4500 were specifically decreased in the presence of 2-amino-dATP (Fig. 4D,E; Supplemental Fig. S4E). Together, these observations from Illumina-sequencing results underscore both the linearity and specificity achieved through RetraC:T for mapping ac4C sites in cellular RNA without introducing significant artifacts.

RetraC:T displays utility in long-read nanopore sequencing

A primary limitation in identifying ac4C sites in cellular RNA using reduction chemistries stems from their nonspecific effects, which include unintended deacetylation, RNA degradation, and reactivity with other modified residues (Fig. 1B,E; Marchand et al. 2018; Lin et al. 2019; Draycott et al. 2022; Zhang et al. 2022a; Ammann et al. 2023). In evaluating the effectiveness of RetraC:T in Illumina sequencing, we deliberately chose NaBH₄ reduction for its nonspecific alkaline hydrolysis, which eliminates the need for additional fragmentation before library construction (Fig. 1B). However, deacetylation associated with the high pH achieved during NaBH₄ reduction significantly blunts ac4C detection, resulting in approximately half the mismatch level observed with NaCNBH₃ reduction (Fig. 3D,E; Miller and Cerutti 1967; Thomas et al. 2018). In contrast, the utility of NaCNBH₃ in ac4C-sequencing applications is constrained through its general influence on RNA degradation, which may leave treated RNA unsuitable for additional fragmentation (Fig. 1B,C). Nevertheless, given that we were able to achieve stoichiometric ac4C detection in Sanger sequencing after 20 min of exposure to NaCNBH₃, we investigated whether shortening the treatment time to 5 min would be sufficient to map ac4C in long-read sequencing, which does not require additional fragmentation (Fig. 3D). Supporting this premise, although agarose gel electrophoresis showed significant RNA degradation after exposure of wild-type and NAT10^−/− total RNA to NaCNBH₃ for 5 min as compared to HCl-treated controls, 18S transcripts remained detectable (Fig. 5A). Sanger sequencing of RT-PCR amplicons further confirmed significant C:T mismatches at 18S rRNA nt 1337 in the NaCNBH₃-treated samples, but not in NAT10^−/− and untreated controls (Supplemental Fig. S5A).

FIGURE 5.

Detection of ac4C in 18S rRNA through RetraC:T and nanopore sequencing. (A) Agarose gel analysis of total RNA integrity after treatment with NaCNBH₃ for 5 min. (B) Schematic of the nanopore-PCR sequencing strategy to evaluate the impact of 2-amino-dATP on 18S rRNA sequencing. (C) Browser views detailing 18S rRNA read coverage in sequencing results from untreated (HCl) and NaCNBH₃-treated RNA from wild-type and NAT10^−/− HeLa cells reverse transcribed with dNTPs alone or supplemented with 2-amino-dATP. At an allele threshold of 0.15, the ac4C site at nt 1842 is the only readily visible mismatch due to sequencing coverage drop-off. (D) Browser views surrounding the ac4C sites at 18S nt 1337 and 1842 from nanopore sequencing of untreated or NaCNBH₃-treated wild-type and NAT10^−/− HeLa RNA reverse transcribed in the presence of 2-amino-dATP or dNTPs alone, as shown. C:T mismatches are depicted in red. (E) Cumulative mismatch rates from nanopore sequencing of untreated or NaCNBH₃-treated wild-type and NAT10^−/− HeLa RNA reverse transcribed in the presence of 2-amino-dATP or dNTPs alone, derived from each of the 1869 nt in 18S rRNA and segregated by modification status. (F) Browser views surrounding m1acp3Ψ at 18S nt 1248 from nanopore sequencing of untreated or NaCNBH₃-treated RNA reverse transcribed in the presence of 2-amino-dATP or dNTPs alone, as shown. T:C mismatches are depicted in blue.

Based on preliminary evidence confirming the appropriateness of the time of treatment, we generated libraries for directed long-read sequencing of 18S transcripts through nanopore technology (Fig. 5B). Bioanalyzer profiles revealed a consistent size distribution across all libraries, with inserts ranging from truncated to full-length 18S (Supplemental Fig. S5B). Subsequent sequencing with a MinION mk1C nanopore yielded ∼100,000–120,000 reads per sample, with 100% of reads aligning to 18S rRNA (Supplemental Table S6). Browser view inspection of the mapped reads showed a consistent pattern across all conditions, marked by a pronounced 3′-bias, as is typical in nanopore sequencing (Fig. 5C; Soneson et al. 2019; Amarasinghe et al. 2020). Notably, overall read lengths were marginally increased in the untreated condition, as would be expected given the degradation associated with NaCNBH₃ treatment (Supplemental Fig. S5C).

Specific examination of the 18S ac4C sites confirmed the determined associations from Sanger and Illumina sequencing, wherein inclusion of 2-amino-dATP in RT reactions increased ac4C detection through C:T mismatches by ∼20% and ∼15% at nt 1337 and 1842, respectively, as compared to dNTPs alone (Fig. 5D). Consistent with the general sequencing drop-off observed in the browser views, coverage at nt 1337 was significantly attenuated as compared to nt 1842 in all conditions. Thus, although RetraC:T coupled to nanopore effectively captured 18S ac4C sites at base resolution, it may be unsuitable for long-read sequencing in lower abundance RNAs with 5′-localized ac4C, such as mRNAs.

Considering that nanopore sequencing is relatively error-prone as compared to Illumina, we further evaluated the impact of 2-amino-dATP on general mismatch rates in the nanopore results (Amarasinghe et al. 2020; Sahlin and Medvedev 2021; Burdick et al. 2023). As with the Illumina results, the impact of RetraC:T on general mismatch rates was low and similar to results produced with dNTPs alone (Fig. 5E). In further agreement with the Illumina findings, RetraC:T effectively reduced T:C mismatch detection at m1acp3Ψ at nt 1248 (Figs. 4E and 5E,F). The high basal T:C mismatch rate is consistent with previous findings that m1acp3Ψ induces T:C mismatches in RT-PCR and may reflect a response to the reduced dGTP used in RetraC:T reactions (Babaian et al. 2020). However, m7G represents a notable distinction between the sequencing approaches: Although prominent mismatches were associated with NaBH₄ reduction in the Illumina results, mismatches associated with NaCNBH₃ reduction in nanopore sequencing were similar to background, pointing to a potential difference in m7G sensitivity to the distinctive chemistries (Figs. 4D and 5E). All considered, RetraC:T through NaCNBH₃ reduction coupled to nanopore is an effective approach for mapping ac4C sites in cellular RNA at high sensitivity and high specificity through long-read sequencing, particularly for abundant RNAs. Overall, RetraC:T emerges as a valuable addition to the toolbox of chemical methods for mapping RNA modifications at base resolution in sequencing results.

DISCUSSION

A central challenge in the epitranscriptome field derives from the insufficient availability of high-quality methods to distinguish modified RNA residues from closely related bases. This problem is especially pronounced for mapping methods, wherein disparate results from antibody-based and chemical methods have fueled controversies that occlude the true distribution of modified residues within cellular RNA (Zhang et al. 2022b; Kong et al. 2023). ac4C represents one such modification, and studies exploring its occurrence in mRNA have produced conflicting conclusions (Arango et al. 2018, 2022; Sas-Chen et al. 2020). Complicating this narrative, maintaining ac4C levels in extracted RNA is particularly problematic given its natural tendency to deacetylation in conditions of elevated pH and heat (Miller and Cerutti 1967). Accordingly, methods that attempt to detect ac4C in RNA must be balanced against its propensity to deacetylate in the precise conditions that are used to improve nucleotide access through resolving inhibitory structures.

Recently, two methods were described for mapping ac4C at base resolution: ac4C-seq and RedaC:T-seq. These methods are distinguished through the chemistries by which they achieve ac4C reduction, NaCNBH₃ for ac4C-seq and NaBH₄ for RedaC:T-seq, but both approaches rely on the same principle. In brief, ac4C is reduced to tetrahydro-ac4C, which preferentially base-pairs with adenosine in cDNA synthesis, culminating in C:T mismatches at ac4C sites (Sas-Chen et al. 2020; Arango et al. 2022). However, both methods present limitations that temper ac4C detection in cellular RNA. For example, substantial RNA degradation is observed within minutes of exposure to NaCNBH₃ (Fig. 1C), whereas the alkaline conditions associated with NaBH₄ reduction result in passive RNA deacetylation that is unrelated to NaBH₄ (Miller and Cerutti 1967; Arango et al. 2018). Although the kinetics of ac4C reduction to tetrahydro-ac4C through NaBH₄ is rapider than deacetylation (Miller and Cerutti 1967), the unavoidable losses dampen the capacity to detect ac4C in rare or incompletely modified substrates, presumably including mRNAs.

We noticed a disconnect between expected tetrahydro-ac4C levels and C:T mismatches in sequencing results. In brief, complete loss of ac4C in mass spectrometry after reduction, even in the absence of evidence for deacetylation, did not produce equivalent gains in C:T mismatch levels in sequencing results (Fig. 1E,F). This suggested that tetrahydro-ac4C is capable of binding to dGTP during cDNA synthesis, raising the prospect for potential improvement. With this in mind, we questioned whether we could present tetrahydro-ac4C with a preferred substrate during cDNA synthesis that would enforce mismatches. To that end, we identified unnatural NTPs with molecular characteristics that complement tetrahydro-ac4C, and also complement dTTP, to compel mismatches during RT and PCR amplification. Acknowledging that not all enzymes would readily accept the modified NTPs, we further tested several distinct RTases. In so doing, we found that the combination of HIV RTase and 2-amino-dATP produced a robust and reproducible increase in C:T mismatch detection at known ac4C sites (Fig. 2).

Evaluation of the effectiveness of RetraC:T in detecting ac4C sites in sequencing results further established its specificity and sensitivity. Our analysis focused on 18S and 28S rRNAs given their relative abundances and known locations of modified nucleotides (Taoka et al. 2018). The resounding consensus across Sanger, Illumina, and nanopore sequencing results showed that the inclusion of 2-amino-dATP in RT reactions from reduced RNA improves ac4C detection through increased C:T mismatches as compared to dNTPs alone, without enhancing mismatches elsewhere, including at other modified nucleotides, thus validating the specificity of the method (Figs. 3–5). Comparison of C:T mismatch rates in Sanger and Illumina sequencing results with varying ac4C levels—achieved through mixing treated wild-type HeLa RNA with either treated NAT10^−/− or untreated wild-type RNA—further confirmed its sensitivity and linearity in quantifying ac4C (Fig. 3). Importantly, RetraC:T outperformed dNTPs alone in detecting ac4C at stoichiometries of <25%, highlighting its potential for pinpointing ac4C discovery in substrates with lower modification frequencies (Figs. 3 and 4). However, it is notable that mismatch rates at acetylated 18S rRNA sites were lower in Illumina and nanopore as compared to Sanger results (Figs. 3–5). Although the determination of mismatch rates through Sanger sequencing is imprecise, among the tested approaches Sanger uniquely uses forward and reverse gene-specific primers, hinting at possible avenues for additional optimization during library construction and sequencing. For example, whether mitigated ac4C detection in RetraC:T deep-sequencing derives from RT stops associated with tetrahydro-ac4C, and how to minimize such truncations, requires evaluation. Overall, our findings endorse RetraC:T as an effective strategy for enhancing ac4C identification in cellular RNA.

Although our method specifically focuses on ac4C, we envision the use of modified dNTPs as a generalized strategy to increase the stringency of modified RNA detection in approaches that involve cDNA synthesis. Indeed, the principles we apply herein can be adapted to other modified RNA nucleotides that alter Watson–Crick base-pairing and can be readily used in Illumina sequencing platforms. To this point, unnatural thymidine derivatives that differ from the canonical dTTP by replacement of selenium or sulfur in position 4 have been shown to induce RT stops in m6A sequencing (Hong et al. 2018). Unnatural dNTPs may further prove useful as an additional layer of validation in nanopore sequencing, wherein unique signatures in direct RNA-seq can be compared to correspondingly altered signatures in direct cDNA-sequencing involving modification-tailored dNTPs. In sum, through harnessing fundamental concepts in nucleotide chemistry, RetraC:T offers a prelude into the enhanced detection of RNA modifications and their functional implications through exploiting the expanded specificity conferred through base-pairing interactions involving modified dNTPs.

MATERIALS AND METHODS

Cell culture

Wild-type and NAT10^−/− HeLa cells from Arango et al. (2018) were maintained in a DMEM medium supplemented with bovine calf serum (10%) and glutamine (50 µM).

RNA extraction

Cell pellets were resuspended in TRIzol. Chloroform (1:5) was added and samples were incubated for 5 min at room temperature. The upper phase (RNA) was further precipitated with 100% ethanol, sodium acetate, and linear acrylamide for 2–16 h at −20°C. RNA was pelleted at 13,200 rpm for 15 min at 4°C, washed twice with 75% EtOH, and dissolved in water. RNA quantity and quality were assessed with a Nanodrop spectrophotometer (ND-1000) and/or Agilent bioanalyzer (2100).

Chemical reduction of ac4C

Reduction with NaBH₄ (100 mM) was performed in a total volume of 40 µL for 1 h at 55°C, unless specified otherwise. Reactions were quenched with 4 µL HCl (1 N). Reduction with NaCNBH₃ (100 mM in HCl) was performed in a total volume of 40 µL for 5 min at 20°C, unless specified otherwise. Reactions were placed on ice and quenched with 30 µL TRIS (1 M, pH 8.0). For both conditions, the RNA was further precipitated with 100% ethanol, sodium acetate, and linear acrylamide for 2–16 h at −20°C. RNA was pelleted at full speed for 15 min at 4°C, washed twice with 75% EtOH, and dissolved in water.

Reverse transcription

In total, 1–10 µg of RNA was denatured for 5 min at 65°C in the presence of random hexamers and 0.5 mM of dNTPs. When analogs were used (2-aminopurine riboside-5′-triphosphate, N-1067-1, Trilink; 2-hydroxy-dATP, NU-1209S, Jena Bioscience; 2-amino-dATP, N2003-1, Trilink), dNTPs were supplemented with 375 µM of analogs while dGTP was reduced to 125 µM. RTs were performed with 1 U of RTase in 1× first-strand buffer (Thermo Scientific) and 10 mM DTT. Superscript IV RT was performed as follows: 10 min at 25°C, 10 min at 50°C, and 5 min at 85°C. When using TGIRT, samples were incubated for 10 min at 25°C and 60 min at 57°C. RT with HIV RTase was performed by incubating samples for 10 min at 25°C, 50 min at 42°C, and 5 min at 75°C. The conditions for NEBNext RT were as follows: RT reactions were initiated by adding one volume of a mixture containing 0.5 mM dNTPs, 1 µL NEBNext First-Strand Synthesis Enzyme Mix (NEB E7761AVIAL), 40 U Murine RNase Inhibitor (NEB), 1× First-Strand synthesis buffer (Thermo Fisher Scientific), and 10 mM DTT in a 20 μL reaction and incubated for 10 min at 25°C, 15 min at 48°C, and 15 min at 70°C.

Polymerase chain reaction and Sanger sequencing

PCR was carried out using 1× Q5 Hot MasterMix (NEB), 0.2 µM of forward and reverse primer pairs (Supplemental Table S7). PCR cycles were as follows: 30 sec at 98°C, 35 cycles with 15 sec at 98°C, 30 sec at 60°C, 30 sec at 72°C, final extension for 5 min at 72°C, and hold at 4°C. PCR products were resolved on 1% agarose gel for 50 min at 100 V. PCR bands were extracted using Zymo Research gel extraction kit and Sanger sequencing (Psomagen) was performed with forward primers (Supplemental Table S7).

Mismatch quantification from Sanger sequencing

Data from the ab1 file were extracted using a custom script written with Biopython. Briefly, ab1 data were parsed using SeqI0 to extract the x-coordinate of each peak corresponding to G, T, A, or C. The fluorescence values were extracted by calling the data from the four different channels that correspond to G, T, A, or C. At every peak location (x-coordinate) the fluorescence values for the four bases were extracted to determine the peak height of the cognate base and the level of the noncognate base. The following formula was used to determine the percentage of a designated mismatch: noncognate base/sum of all bases × 100 or specifically T/(C + T) × 100.

RNA digestion and mass spectrometric analysis

In vitro RNA samples (∼600 fmol) were diluted to a total volume of 20 µL with nuclease-free water. A master mix of snake venom phosphodiesterase (0.2 U), calf intestinal phosphatase (2 U), and benzonase (2 U) was prepared with 1 mM MgCl₂, 5 mM TRIS (pH 8), 10 nmol butylated hydroxytoluene, and 5 µg tetrahydrouridine (modified from Heiss et al. 2021) containing isotope analogs of cytidine, N6-methyladenosine, 5–methylcytidine, and N4–acetylcytidine as internal standards. After the addition of 15 µL of the master mix, samples were incubated at 37°C for 2 h. The samples were diluted with 15 µL LC-MS buffer A (0.0075% formic acid in ultrapure water) and filtered for 35 min at 3273g and 4°C through 0.2 µm Supor AcroPrep Advance 96-well plates (Pall corporation). Of the filtrate, 39 µL was subjected to mass spectrometric analysis on an Agilent triple quad 6495C mass spectrometer. For HPLC, buffer A (described above) and buffer B (0.0075% formic acid in acetonitrile) were used in combination with a C–8 reversed-phase column (Agilent Poroshell 120 SB-C8, 2.1 × 150 mm, 2.7 µm LC). All data were analyzed according to the published protocol using the isotope dilution technique (Traube et al. 2019).

Primer extension analysis

To evaluate whether ac4C affects RT, first-strand cDNA synthesis was performed on the NaBH₄-treated C- or ac4C-RNA probes (Supplemental Table S7). For this purpose, a primer complementary to the probe sequence (10 pmol; Supplemental Table S7) was radiolabeled with 20 U T4 polynucleotide kinase (NEB) and 10 μCi ³²P-γATP (3000 Ci/mmol, PerkinElmer) in a 5 μL reaction for 10 min at 37°C. The radiolabeled primer was diluted with 45 µL nuclease-free water and the reaction heated for 5 min to 95°C. To remove excess of nonincorporated ³²P-γATP, reactions were filtered using Illustra MicroSpin G-50 Columns (GE Healthcare). RT reactions including 2 μl of reduced RNA and 2 μl radiolabeled primers were initiated by adding one volume of a mixture containing 0.5 mM dNTPs, 200 U TGIRT (In-Gex), 6 U HIV RT (EMD Millipore) or 1 µL NEBNext First-Strand Synthesis Enzyme Mix (NEB E7761AVIAL), 40 U Murine RNase Inhibitor (NEB), 1× First-Strand synthesis buffer (Thermo Fisher Scientific), and 10 mM DTT in a 20 μL reaction for 10 min at 25°C and 1 h at 57°C (TGIRT), 10 min at 25°C, 15 min at 48°C, and 15 min at 70°C (NEBNext enzyme) or 10 min at 25°C, 50 min at 42°C and 5 min at 80°C (HIV RTase). Template RNA was digested with 2 U RNase H (Thermo Fisher Scientific) for 30 min at 37°C and reactions were stopped by adding one volume of 2× loading dye (95% formamide, 0.025% bromophenol blue, 0.025% xylene cyanol, 0.5 mM EDTA). RT products were resolved in 7 M Urea/10% PAGE gels and examined through phosphorimager analysis (ImageQuant TL 10.2, Cytiva Life Sciences).

Illumina library preparation

Total RNA was reduced using NaBH₄ as described previously. Untreated samples were fragmented using NEB fragmentation buffer (E6150S) for 3 min at 94°C. Fragmentation was stopped by adding NEB fragmentation stop solution and RNA was ethanol precipitated. One hundred nanograms of RNA was denatured at 65°C for 5 min in the presence of random primers (NEBNext Ultra II Directional RNA Library kit, NEBE7760) and 0.5 mM of dNTPs. In reactions with 2-amino-dATP (Trilink N2003-1), dGTP was reduced to 125 µM dNTPs, and 375 µM of analog was added. RTs were performed with 1 U (SSIV) or 2.84 U (HIV) of RTase and 40 U Murine RNase Inhibitor (NEB) in 1× first-strand buffer (Thermo Scientific) and 5–10 mM DTT (SSIV and HIV RTase, respectively) in a 20 μL reaction. Superscript IV RT was performed as follows: 10 min at 23°C, 10 min at 50°C, and 5 min at 80°C. RT with HIV RTase was conducted by incubating samples for 10 min at 25°C, 50 min at 42°C, and 5 min at 80°C.

Eighteen microliters of the first-strand synthesis product was subjected to second-strand synthesis by the addition of 8 µL NEBNext second-strand synthesis reaction buffer with dUTP (NEBNext Ultra II Directional RNA Library kit, NEB E7760), 4 µL NEBNext second-strand synthesis enzyme mix (NEBNext Ultra II Directional RNA Library kit, NEB E7760), and 50 µL of nuclease-free water on ice and thorough mixing. The mixture was incubated in a thermal cycler for 1 h at 16°C with the heated lid set at 40°C.

Double-stranded cDNA was purified by the addition of 144 µL (1.8×) AMPure XP beads (Beckman Coulter A63880) and thorough mixing. Samples were incubated at room temperature for 5 min before magnetic separation. The beads were subsequently washed twice with freshly prepared 80% EtOH and air-dried for 5 min at room temperature. The DNA was finally eluted in 53 µL 0.1× TE buffer (NEBNext Ultra II Directional RNA Library kit, NEB E7760) through thorough mixing and incubation for 2 min at room temperature. Of the cDNA, 50 µL were subjected to the adapter ligation reaction (NEBNext Ultra II Directional RNA Library kit, NEB E7760). All subsequent steps were performed exactly according to the manufacturer's instructions using 25-fold dilutions of the adapters (NEBNext Multiplex Oligos for Illumina, NEB E6440) and eight cycles for final PCR amplification. The libraries were subjected to 60 bp paired-end NextSeq 2000 sequencing.

Illumina data analysis

Adapters and low-quality bases were removed using Cutadapt (version 4.4) (Martin 2011) based on the following option settings: ‐‐minimum-length 30 -a AGATCGGAAGAGCACACGTCTGAACTCCAGTCA -A AGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT. Trimmed reads were mapped to the 18S rRNA sequence, obtained from the UCSC Table Browser (Karolchik et al. 2004), using Bowtie2 (version 2-2.5.1) (Langmead and Salzberg 2012) with the default settings. Unstranded mismatch calling was performed based on concordantly mapped fragments with a minimum mapping quality of 10, using a custom Python script. In case of overlapped fragment mates, mismatch-supported fragments were counted once. 18S rRNAs-related mismatches with a minimum Phred quality score of 20 (equivalent to 99% base calling accuracy), were used in downstream analysis. Browser views were visualized using integrative genome viewer (IGV) with an allele threshold of 0.15.

Nanopore sequencing

Three micrograms of total RNA were treated with 100 mM NaCNBH₃ in 100 mM HCl for 5 min at 20°C in a thermocycler. Reactions were quenched with 30 µL of 1 M Tris-HCl pH 8.0 and RNA precipitated overnight at −20°C. Two hundred nanograms of treated and untreated RNA were used to generate the Nanopore-PCR library (SQK-PCS111) according to the manufacturer's instructions with a few adjustments. RNA was mixed with 0.2 µM of specific primer toward 18S, dNTPs ±375 µM of 2-amino-dATP (Trilink Biotechnologies). The RNA was denatured for 5 min at 65°C. The strand-switching reaction was initiated by adding 1 µL of strand-switching primer (SQK-PCS111), 5× FS buffer, and RNase inhibitor, and samples were incubated for 2 min at 42°C in a thermocycler. Then, 13 units of HIV reverse transcriptase (Millipore) were added. The samples were incubated for 90 min at 42°C. The HIV enzyme was heat-inactivated for 5 min at 85°C before proceeding to PCR amplification. cDNA products were amplified using cPRM primer (SQK-PCS111 kit) and 1× Long-Amp Taq polymerase (NEB) under the following PCR cycles: (95°C ×1, [95°C: 15 sec/62°C: 15 sec/65°C:90 sec] × 14, 65°C 6 min, 4°C hold). PCR products were digested by 1 U of Exonuclease I (NEB) for 15 min at 37°C, ExoI was further denatured for 15 min at 80°C, and 25 fmol of purified PCR products were ligated with 1 µL of Rapid Sequencing Adapter (SQK-PCS111 kit) for 5 min at room temperature. The sequencing was performed using a MinION mk1C until the acquisition of 150,000 reads.

Oxford Nanopore data analysis

Adapter trimming and base calling were performed using Guppy (version 6.5.7) (Wick et al. 2019) with the default settings. The resulting FASTQ files were mapped to 18S rRNA using minimap (version 2.24) (Li 2018). Stranded mismatch calling was performed based on mapped nanopore reads with minimum mapping quality, and minimum Phred quality score of zero. Browser views were visualized using IGV with an allele threshold of 0.15.

DATA DEPOSITION

Sequencing data have been deposited at SRA and are publicly available as of the date of publication under accession number PRJNA1071147. All other data and code will be shared by the lead contact upon request. There are no restrictions on any data or materials presented in this paper.

SUPPLEMENTAL MATERIAL

Supplemental material is available for this article.

MEET THE FIRST AUTHOR

Sarah Schiffers.

Meet the First Author(s) is an editorial feature within RNA, in which the first author(s) of research-based papers in each issue have the opportunity to introduce themselves and their work to readers of RNA and the RNA research community. Sarah Schiffers is the co-first author of this paper, “Enhanced ac4C detection in RNA via chemical reduction and cDNA synthesis with modified dNTPs.” Sarah received her PhD in organic chemistry from the LMU in Munich, Germany under the supervision of Professor Dr. Thomas Carell. Afterward, she joined Dr. Shalini Oberdoerffer's laboratory at the NCI as a postdoctoral fellow. She studies the RNA modification N4-acetylcytidine (ac4C) and how it impacts protein translation. This manuscript describes a new method for its study.

What are the major results described in your paper and how do they impact this branch of the field?

Our study describes an enhanced method for the detection of N4-acetylcytidine (ac4C) in RNA. Previous methods rely only on the reduction of the modification to tetrahydro-ac4C (H4-ac4C) by two different chemical reagents and subsequent misincorporation of adenosine during reverse transcription. In this improved method, we show that either reagent, in combination with the addition of a H4-ac4C-specific modified adenosine triphosphate in the reverse transcription reaction, increases C:T mismatches with multiple different sequencing methods.

What led you to study RNA or this aspect of RNA science?

I first started my career studying the role of epigenetic modifications in the DNA of mouse embryonic stem cells. Although we still do not know everything about regulating gene expression on this level, I got more and more intrigued by the variety of modifications in RNA and how modifications in mRNA impact protein synthesis directly.

What are some of the landmark moments that provoked your interest in science or your development as a scientist?

Science has been on my mind ever since I did an internship in ninth grade. At the time, I was convinced I wanted to work as a technician in the pharma industry. In high school and college, I studied both organic chemistry and biochemistry and could not decide between the two. On paper, I loved organic synthesis, but on the bench, I was more successful with biological methodologies. During my PhD studies, I used mass spectrometry to study epigenetic modifications and realized that I could take the best of both worlds because they complement each other.

Are there specific individuals or groups who have influenced your philosophy or approach to science?

Both my PhD supervisor Dr. Thomas Carell, as well as my postdoc supervisor Dr. Shalini Oberdoerffer, had a great impact on my growth as a scientist. The former showed me how to apply chemistry to biology, and the latter is teaching me how to translate my results in biology toward the development of RNA therapeutics. I was always encouraged to look beyond the obvious, and in my opinion, a multidisciplinary view is necessary to make the transition from “bench to bedside.”

What are your subsequent near- or long-term career plans?

In the long term, I plan to apply what I learned from basic research to develop RNA-based technologies for challenges in the clinic. I like to be actively involved in the research and want to continue working on the bench.

How did you decide to work together as co-first authors?

The original idea to use an adenosine analog complementary to H4-ac4C was mine. Our entire laboratory is working on RNA modifications and is interested in sequencing technologies to develop our projects. With my background in chemistry, I opened a new “toolbox” and suggested multiple strategies for different problems. While Dr. Relier started working on evaluating the strategy of using modified dNTPs with Sanger sequencing and achieved promising results, I applied my expertise in mass spectrometry and primer extension studies to validate the choice of reagent and enzyme.