Direct Sequencing of tRNA by 2D-HELS-AA MS Seq Reveals Its Different Isoforms and Dynamic Base Modifications

Abstract

Post-transcriptional modifications are intrinsic to RNA structure and function. However, methods to sequence RNA typically require a cDNA intermediate and are either not able to sequence these modifications or are tailored to sequence one specific nucleotide modification only. Interestingly, some of these modifications occur with <100% frequency at their particular sites, and site-specific quantification of their stoichiometries is another challenge. Here, we report a direct method for sequencing tRNA^Phe without cDNA by integrating a two-dimensional hydrophobic RNA end-labeling strategy with an anchor-based algorithm in mass spectrometry-based sequencing (2D-HELS-AA MS Seq). The entire tRNA^Phe was sequenced and the identity, location, and stoichiometry of all eleven different RNA modifications was determined, five of which were not 100% modified, including a 2′-O-methylated G (Gm) in the wobble anticodon position as well as an N², N²-dimethylguanosine (m²₂G), a 7-methylguanosine (m⁷G), a 1-methyladenosine (m¹A), and a wybutosine (Y), suggesting numerous post-transcriptional regulations in tRNA. Two truncated isoforms at the 3′-CCA tail of the tRNA^Phe (75 nt with a 3′-CC tail (80% abundance) and 74 nt with a 3′-C tail (3% abundance)) were identified in addition to the full-length 3′-CCA-tailed tRNA^Phe (76 nt, 17% abundance). We discovered a new isoform with A─G transitions/editing at the 44 and 45 positions in the tRNA^Phe variable loop, and discuss possible mechanisms related to the emergence and functions of the isoforms with these base transitions or editing. Our method revealed new isoforms, base modifications, and RNA editing as well as their stoichiometries in the tRNA that cannot be determined by current cDNA-based methods, opening new opportunities in the field of epitranscriptomics.

Graphical Abstract

As an essential component of protein synthesis machinery, tRNAs (tRNAs) are present in all living cells. More than 600 different tRNA sequences and a large breadth of different post-transcriptional base modifications have been reported.^1,2 Despite their significance, structural and functional studies to understand the underlying biochemistry of tRNAs themselves have been hindered due to the lack of efficient tRNA sequencing methods. tRNAs are the only class of small cellular RNAs that cannot be efficiently sequenced with current sequencing techniques,³ even though the first tRNA was sequenced early in 1965.⁴ Typically, current methods used to sequence tRNAs are indirect and require complementary DNA (cDNA) intermediates. However, cDNA synthesis results in the loss of endogenous base modification information originally carried by RNAs. Thus, cDNA-based methods are not able to accurately sequence a complete set of the rich and dynamic base modifications in tRNAs which are an intrinsic part of their structure and function. Other methods that do not involve cDNA can detect base modifications, but these techniques usually require harsh sample treatments such as intensive enzymatic or chemical hydrolysis, resulting in loss of sequence ladders and spatial modification information.^5-7

As opposed to indirect sequencing methods, direct sequencing of RNAs without cDNA synthesis would theoretically allow identification of both canonical and modified nucleotides in any RNA sample. Chemical and enzymatic degradation methods were previously developed to generate ladders for direct sequencing of RNA.^8,9 However, these methods rely on different reactions specific to each of the canonical bases (A, C, G, and U) to generate sequence ladders as well as electrophoretic separation of the resultant ladders in the same gel for sequence determination, limiting their application primarily to sequencing canonical bases, and having very limited applications in identifying and locating various base modifications, e.g., those carried by tRNA.

One potential direct RNA sequencing method that can be generally applied to sequence both canonical and modified bases is mass spectrometry (MS)-based sequencing, which directly correlates the mass of each base, rather than proxies such as fluorescence¹⁰ or fluctuations in electric current,¹¹ to sequence outputs. The four canonical ribonucleotides and most modified bases either have inherently unique masses or can be easily converted into different signature masses, allowing them to potentially be used for de novo MS-based direct RNA sequencing. Previously, top-down MS and tandem MS^12-14 have been used in attempts to directly sequence tRNAs. However, these traditional MS methods have significant methodological inadequacies in the preparation of high-quality sequence ladders and subsequently in the reduction of complicated data, resulting in inaccurate base-calling.^15,16 Therefore, a two-dimensional (2D) liquid chromatography (LC)-MS-based RNA sequencing method was established to produce easily identifiable mass-retention time (t_R) ladders,¹⁶ allowing de novo sequencing of short single-stranded RNAs. We recently developed a hydrophobic end-labeling strategy (HELS) for 2D LC-MS-based RNA sequencing by introducing 2D mass-t_R shifts for ladder identification of mixed RNAs,¹⁷ allowing for both the complete reading of a sequence from a single ladder and the sequencing of mixed RNA samples.¹⁶ Despite its success in model studies on short synthetic RNAs (~20 nt),¹⁷ it is still challenging to perform de novo MS sequencing of biological RNA such as tRNAs due to the data complexity caused by the increased length (typically ~76–90 nt) and abundance of base modifications. It is critical for the method to first demonstrate its feasibility to sequence real biological samples beyond previously assayed synthetic RNAs, as this will be a significant step forward to expand its applications in sequencing broader biological RNAs in the future. This helps to address the urgent demand in the field of epitranscriptomics for new technological development that can sequence all kinds of RNA modifications,^18,19 not just for a specific modification or a few modifications.²⁰

To process complex MS data derived from biological samples and to enable the method to sequence tRNAs, we developed a computational anchor-based algorithm which innovatively accomplishes automated MS sequencing of RNAs. The signature t_R-mass value of the hydrophobic tag specifies the exact starting data point, the anchor, for the algorithm to accurately determine data points corresponding to the desired ladder fragments needed for sequencing, significantly simplifying data reduction and enhancing the accuracy of sequence generation. The idea of using an anchor to identify sequence ladder start-points can be generalized and extended to any known chemical moiety beyond hydrophobic tags, e.g., PO₄⁻, at the beginning of the tRNA or any nucleotide with a known mass. We can program its mass as a tag mass and use our anchor algorithm for sequencing, addressing the issue of MS data complexity of biological RNA samples and making 2D-HELS MS Seq more robust and accurate. Here, we integrated 2D-HELS with the anchor-based algorithm into mass spectrometry-based sequencing (2D-HELS-AA MS Seq), and successfully applied the method to sequence a complete yeast tRNA^phe, including all eleven different RNA modifications. Interestingly, some of these modifications occur with <100% frequency at their particular sites, and site-specific quantification of their stoichiometries is another challenge.¹⁹ In addition to sequencing, we also quantified the base modification stoichiometries of all eleven modifications together in a single study for the first time, and showed that five modifications are not 100% modified at their particular sites of the tRNA. While directly sequencing the tRNA and its modifications, the new method also revealed two 3′-tail-truncated isoforms besides the full-length 3′-CCA-tailed tRNA^Phe and helped to discover a new base transition isoform with A─G transitions/editing at the 44 and 45 positions in the variable loop, demonstrating another aspect of the method’s efficacy in detection, sequencing, and quantification of RNA isoforms. Our technological advance promotes better understanding of functions of post-transcriptional modifications and isoforms including their correlations to human diseases.

RESULTS AND DISCUSSION

Development of an Anchor-Based Algorithm for 2D-HELS-AA MS Seq.

To extend the application of the 2D-HELS MS Seq approach from short synthetic RNAs¹⁷ to allow sequencing of tRNA, we developed a computational anchor-based algorithm to automate MS sequencing of RNAs. Due to the complexity of MS data derived from the tRNA, it is very challenging to process all data in a single LC-MS run simultaneously. Instead, data preprocessing was used to select a particular subset of the input data set for the algorithm to focus on initially. This is feasible because a hydrophobic tag was added to the terminus of each RNA to be sequenced, where it remained even after acid degradation. Additionally, the trends of t_R and mass of the tag-containing ladder fragments are known from our previous studies.^16,17 In the 2D mass-t_R plot of output LC-MS data sets, data points corresponding to tag-labeled RNA fragments are shifted spatially to a zone with larger t_Rs than those of their unlabeled counterparts, due to the tag’s hydrophobicity. Therefore, the algorithm can “zoom in” on one group, either labeled or unlabeled, in its specific zone of the 2D-plot, to read out the sequence of the selected group first. As such, we call the algorithm “anchor-based”, since it specifies the starting data point corresponding to the terminal tag, which latches down the data points corresponding to the specific ladder fragments that we aim to read out from the whole data set. The anchor-based algorithm significantly simplified the complicated MS data from the tRNA sample because it only read out the sequence for ladder fragments that had a hydrophobic tag or a specified tag with a known mass, and selectively filtered all non-tag/anchor-related data points out of the complicated MS data derived from the tRNA sample. More details related to the anchor algorithm are described in Supporting Information.

2D-HELS-AA MS Seq of Yeast tRNA.

As it was only possible to read segments of up to 35 nt long with a 40K mass resolution LC-MS,¹⁷ we incorporated a partial RNase T1 digestion step to sequence a tRNA that was commercially available, resulting in a reduction of the 76 nt tRNA to segments of sequenceable sizes for 2D-HELS-AA MS Seq. Subsequently, we directly sequenced the entire tRNA with single-base resolution in one single LC-MS run (Figure 1). To further verify the complete tRNA sequence obtained from the single run above, we labeled the three segments partially digested from the tRNA by RNase T1 and separated them one by one for 2D-HELS-AA MS Seq in three separate LC-MS runs (Figure S1). To obtain overlapping segment sequences for assembling the complete tRNA sequence, we included MS data of the tRNA generated without RNase T1 digestion, i.e., 31 nt of the tRNA read from the 5′-end using a phosphate (PO₄⁻) as the 5′-anchor, and 32 nt of the tRNA read from its 3′-end using a CCA tag as the 3′-anchor, respectively (Figure 1C). Taking all draft reads output by the anchor-based algorithm together (Table S1-S11), we assembled a full-length tRNA sequence which was a 100% match to the tRNA^Phe reference sequence with more than 2× coverage (Figure 1C).

Figure 1.

2D-HELS-AA MS Seq of Yeast tRNA^Phe. (A) (1–6): Sequencing workflow. More detailed information can be found in Materials and Methods in Supporting Information. (B) 2D plot of the entire tRNA sequenced from a single LC-MS run, showing the identity and location of all modifications. (C) Assembly of the full-length tRNA^Phe sequence based on overlapping sequence reads from different LC-MS runs, showing 100% coverage and accuracy as compared to the reported tRNA^Phe reference sequence. All output sequence reads are converted to FASTA format in the 5′- to 3′-order (44 and 45 AG conversion output reads not included). *Ts: the Supporting Information Table S where the sequencing data of that particular strand can be found.

Sequencing of All Eleven RNA Modifications.

During sequencing of the tRNA, we also successfully identified and located all eleven RNA modifications within the tRNA (Figure 2). Four of these modifications could be directly read out by their unique masses: dihydrouridine (D) at positions 16 and 17, N²N²-dimethylguanosine (m²₂G) at position 26, 5-methylcytidine (m⁵C) at position 40, and 5-methyluridine (T) at position 54. Methylation on the 2′-OH of C (Cm) at position 32 and G (Gm) at position 34 renders the adjacent 3′5′ phosphodiester linkage nonhydrolyzable, creating a mass gap in both the 5′- and the 3′-mass ladder families larger than 1 nt¹⁶ (Figure 1B). This gap can be filled in by collision induced dissociation (CID) MS, which determines which of the two unhydrolyzable nucleotides is methylated¹⁶ (Figure 2C and Figure S2). However, other RNA modifications such as pseudouridine (ψ) and U, N²-methylguanosine (m²G) and 7-methylguanosine (m⁷G), and 1-methyladenosine (m¹A) and N⁶-methyladenosine (m⁶A) share identical masses, and LC-MS alone cannot distinguish them. Additional enzymatic/chemical reactions were required to identify each base at their particular sites and differentiate them from their corresponding isomers with an identical mass, as shown in the Figure 2C. To differentiate m¹A at position 58 from its m⁶A isomer,²¹ we designed a reverse transcription/single base extension experiment (rtSBE), which indicates that m⁶A, but not m¹A, is able to form base-pairing interactions, thus causing a pause during reverse transcription at any m¹A.²² The rtSBE results proved that the nucleotide at position 58 is m¹A and not m⁶A (Figure S3). The demethylation experiment which employed ALKBH3, a m¹A and m³C demethylase of tRNA,²¹ to convert m¹A to A in tRNA^Phe followed by incorporation of ddT based on a positive MALDI result further confirmed that the nucleotide at position 58 is m¹A. In the absence of ALKBH3, we did not observe the ddT incorporation. To differentiate ψ from U, we treated the RNA with N-cyclohexyl-N′-(2-morpholinoethyl)-carbodiimide metho-p-toluenesulfonate (CMC) to convert ψ to its CMC adduct,²³ which has a different mass than U/ψ. The CMC-converted ψ (depicted as ψ*) results in a shift in both t_R and mass, allowing facile identification and location of ψ at positions 39 and 55 due to a single drastic shift each in the mass-t_R ladder at these sites¹⁷ (Figure S4 and Tables S12-S17). To differentiate m⁷G at position 46 from its isomeric m²G at position 10, we treated the tRNA with borohydride (NaBH₄) and aniline sequentially to generate a site-specific cleavage right after m⁷G.^24,25 The observed three major mass fragments after the cleavage measured by LC-MS were all a result of cleavage at m⁷G, but in three isoforms with 3′-tails of C, CC, or CCA, respectively (Figure S5), indicating that there is only one m⁷G in tRNA. We did not observe a mass fragment induced by a cleavage at m²G at position 10 from either the 5′-end or the 3′-end. However, the mass fragment from the 5′-end to m⁷G at position 46 after the cleavage was not observed (45 nt long), probably due to mass resolution limitation.¹⁷ Otherwise, no other mass fragments were observed. The unique masses of the cleaved 5′-segments were used to differentiate m⁷G at position 46 from m²G at position 10, which cannot be cleaved at the same reaction conditions.

Figure 2.

Sequencing of all eleven RNA modifications. (A) Proposed mechanism for the conversion of wybutosine (Y) to its depurinated form (Y′) in acidic conditions. (B) The mass of Y was found in the crude products after acid degradation. The relative percentages of Y and Y′ were quantified and can be found in Table S18. (C) Summary of all eleven RNA modifications sequenced by 2D-HELS-AA MS Seq. The relative percentages of modifications at each position were quantified by integrating the EIC peaks of their corresponding ladder fragments (Table S19). The percentages of partially modified nucleotides are highlighted in pink.

The primary task for sequencing is to determine the precise order of the four nucleotides. Our method thus extends this capacity to include nucleotide modifications beyond the four canonical nucleotides, based on the unique mass of each RNA modification, and we used this approach to expand the application of the technique beyond the synthetic RNA samples we examined previously, to directly sequence biological samples for the first time. Only in the case where modifications have isomers with identical masses but different chemical structures would one require a further RNA modification characterization method to differentiate these isomers following our 2D-HELS-AA MS Seq approach. However, the advantage of our method is that we already know the mass of the particular nucleotide modification and its location/order without any prior sequence knowledge. This is very different than other RNA characterizing methods that can identify RNA modifications, but must still rely on additional established sequencing methods for sequence/location information.^18,26-28

Stoichiometric Quantification of All Eleven RNA Modifications.

Relative stoichiometries/percentages of modified RNA vs nonmodified counterpart RNA can be quantified in partially modified synthetic RNA samples by our technique,¹⁷ and thus stoichiometries/relative percentages of all eleven RNA modifications were quantified at each position of the tRNA (Table S19), five of which were not 100% modified (Figure 2C). The data suggest that there is an abundance of post-transcriptional regulation that can occur in the tRNA at these different positions. For example, the wobble Gm at position 34 was partially modified (60% Gm vs 40% G), which has important regulatory implications since a lack of Gm could affect binding or stalling in the ribosome.²⁹ 2′-O-Methylation is essential for accurate and efficient protein synthesis, and a decreased level of 2′-O-methylation could lead to an increase in translational infidelity.^30,31

Our method also revealed unexpected nucleotides in tRNA. Position 26 in tRNA^Phe is thought to be m²₂G;^32-34 however, we found clear evidence that G coexists at this position, but there is no evidence for any monomethyled G (mG) coexisting at this position. The stoichiometries were quantified by integrating extracted-ion current (EIC) peaks of their corresponding ladder fragments,^17,35 which revealed that m²₂G and G were present at 58% and 42%, respectively (Figure 2C). Also, both m⁷G at position 46 (46% m⁷G vs 54% G) in the variable loop and m¹A at position 58 (94% m¹A vs 6% A) in the TψC loop were partially modified (Figure 2C), suggesting that the methylation process is highly regulated.³⁶ To the best of our knowledge, this is the first time the stoichiometry, identity, and location of these different RNA modifications were all directly measured together in a single study, something no currently available sequencing technologies are capable of, thus providing unique insights that call for further functional studies of these dynamic RNA modifications.³⁷

Identification and Quantification of a Dynamic Change from Y to Its Depurinated Y′ Form.

Upon analysis of the sequencing results, we noticed that the wybutosine (Y) at position 37 was converted to its depurinated product Y′ (ribose form) under acidic degradation conditions (Figure 2).^38,39 Without acid degradation, only 10% of the tRNA contained the depurinated Y′ form at this position, while 90% contained the standard Y form of the base (Table S18). However, no Y form was observed in any ladder fragments containing this position after acid degradation, and all of the Y bases were converted to Y′ due to depurination in the acidic conditions (Figure 2A). As another piece of evidence of the depurination, a mass of 376.1178 Da, corresponding to a cleaved Y nucleobase, was found in the crude products after acid degradation and subsequent MS analysis (Figure 2B), suggesting that Y′ was originally carried by the tRNA. The fact that our method can identify the dynamic change of Y to Y′ and quantify the relative Y/Y′ ratio could be useful for potential diagnostic assays, as such changes in the Y′/Y ratio could be used as a potential biomarker, e.g., in certain nervous system diseases,⁴⁰ where the common characteristics are decreased pH at both the tissue and cellular levels. Based on the same principle, our method could potentially probe dynamic changes of other base modifications, acid-labile or not, and quantify variations in their ratios, in particular, cells or tissues subjected to different biological processes or disease conditions.

Identification and Quantification of Two Other Truncation Isoforms (74 nt and 75 nt) at the 3′-End.

Unlike its nominal identity according to the supplier, upon sequencing, the commercially prepared tRNA^Phe (phenylalanine specific from brewer’s yeast) sample was revealed to be heterogeneous. When analyzing a biotinylated 3′-segment of the tRNA (58m¹A-76A), we found that there is more than one ladder that has the biotin tag as shown in Figure 3A, indicating that this segment contains more than one sequence. Besides the 76 nt tRNA with a complete post-transcriptionally modified CCA tail, two other incomplete isoforms of the tRNA that are missing an A and a CA at the 3′-CCA tail, respectively, were further identified in a 3′-segment of the tRNA (58m¹A-76A) (Figure 3) using the anchor algorithm and a revised Smith-Waterman alignment similarity algorithm (details in Materials and Methods). Surprisingly, the most abundant component was not the nominal 76 nt tRNA^Phe, which comprised only 17% of the sample as calculated by integration of the corresponding EIC peak (Table S24). Rather, the 75 nt tRNA^Phe with a missing A at the 3′-end was the major component of the sample at 80%, while the 74 nt tRNA^Phe with a missing CA at the 3′-end was a minor component at 3%. The two tail-truncation isoforms cannot be degraded products of longer tRNAs like the 76 nt tRNA^Phe; otherwise, they would not contain the free 3′-OH required for the 2D HELS chemistry.¹⁷ The data suggest that 2D-HELS MS Seq is not only able to sequence modified RNA, but can also identify tail-truncation isoforms that were primarily only studied by polyacrylamide gel electrophoresis methods previously.⁴¹ As stress-induced tRNA fragmentation has been implicated in cancers and other diseases,⁴² further studies into the relationship between the relative abundances of tRNA tail-truncation isoforms and various diseases will assist in understanding the potential role of such isoforms in disease-related biological processes and subsequent treatments.⁴³

Figure 3.

Identification of 3′-truncation isoforms. (A) 2D-HELS-AA MS sequencing of segment III, showing two other truncated isoforms of tRNA^Phe at the 3′-end (74 nt and 75 nt). t_R was normalized for ease of visualization of the 74 nt and 75 nt isoforms. (B) Terminal base of 76 nt tRNA^Phe and its two tail-truncated isoforms; all three isoforms contain a free OH at the 3′-end, which is required for introducing the biotin tag, suggesting that the isoforms were not generated during acid degradation but came together with the full-length 76 nt tRNA originally.

Discovering a New 44g45a Isoform at the tRNA’s Variable Loop.

We also observed a new isoform with an A to G transition at position 44 and a G to A transition at position 45, i.e., a 44A45G (wild-type, reported previously)⁴⁴ to 44g45a transition. Please note that the lower-case letters “g” and “a” in the isoform “44g45a” are used to represent the isomeric nucleotide that shares an identical mass with the canonical nucleotides G and A, respectively, but their exact structures remain to be confirmed. These two reads were revealed first by our anchor-based algorithm, and further verified manually in the original MFE files (Figure 4, Tables S4-S5, S8-S9, and S19-S22). We identified two distinct mass ladder fragments at position 44 when reading from the 5′-direction, apparently corresponding to sequences containing both 44A and 44g being simultaneously present. However, these two mass ladder fragments merged into one mass ladder fragment at position 45. Such an effect could only occur if two coexisting sequences contained a 45G or a 45a, respectively, thus confirming the coexistence of two coexisting isoforms (Figure 4A). This is consistent with the sequencing results when reading from the opposite direction when we performed bidirectional sequencing¹⁶ (Figure 4C). We observed these two isoforms in all reads which covered positions 44 and 45, and their relative percentages were consistent (~50% for wild-type, quantified by EIC) (Table S25). To further verify the coexistence of the two mass fragments, we employed full-spectral analysis provided by the commercial MassWorks software (Cerno Bioscience, Las Vegas, USA) to examine the corresponding ions of these two fragments simultaneously in one spectrum. When reading from the 5′-direction, two ions (m/z 778.1051 and 779.7068, both with 10 charge states) were found, corresponding to 44A and 44g. Full-spectral analysis also confirmed that 45G and 45a coexist when reading from the 3′-direction (Figure 4D). Furthermore, the ratios of 44A/44g as compared to 45G/45a quantified by the full spectral analysis⁴⁵ are consistent (Figure 4), indicating that the sequenced 44g and the 45a are indeed from the same RNA strand, while the 44A and 45G are also both from the same RNA strand. All these MS results support the existence of a new isoform, with the sequence 44g45a, coexisting with the wild-type RNA that contains the 44A45G sequence, and that these two isoforms occur at similar levels. To further confirm the coexistence of these two isoforms, we performed an rtSBE on the tRNA^Phe sample. For example, if tRNA^Phe has an A/g single-nucleotide polymorphism (SNP) at position 44, then the rtSBE assay would be able to incorporate both ddT and ddC, since the two isoforms exist at similar levels. However, the results showed that only ddT could be incorporated at position 44 (Figure S6A) and only ddC could be incorporated at position 45 (Figure S6B), indicating that the wild-type 44A45G was the only isoform present. The rtSBE results suggested that RNA reverse transcriptase could not recognize these edited bases well. It is also possible that the mass differences observed in the above A─G transitions at positions 44 and 45 may be caused by oxidation and reduction, e.g., oxidation of A to 2-oxoadenine (isoG) and/or 8-oxoadenine (8-oxo-A) at position 44 (Figure S7A), which both have a mass identical to G and would still allow canonical T incorporation. Complete acid digestion of the tRNA into single nucleotides followed by LC-MS analysis supports this possibility, as we found two different t_Rs in the EIC profile of the G monophosphate (Figure S7B), suggesting a coexisting nucleotide of the same mass as G, but a different structure. A similar mechanism could explain the putative G to A transition/editing at position 45. These findings raise interesting questions about how these newly identified isoforms emerge and affect the function of the tRNA.

Figure 4.

Discovering a new 44g45a isoform in the tRNA variable loop. (A) Schematic of sequence ladder fragments shows a transition/editing g (sharing an identical mass as G) coexists with A at position 44 when reading from the 5′-direction (Tables S4-S5 and S8-S9). (B) Least squares fitted mass spectrum to the calibrated mass spectrum t_R = 31.9–32.9 min) when reading from the 5′-direction. The full-spectral analysis confirms that the ions of the 44g and 44A fragments (with 10 charges) coexist and their relative abundances are 57% and 43%, respectively. The theoretical trace (black) of the two combined ion profiles fits well with the calibrated mass spectrum as observed (red), resulting in a good spectral accuracy of 87%. (C) Single transition/edited a (one oxygen less than G) coexists with G at position 45 when reading from the 3′-direction (Tables S19-S22). (D) Similar to B, full-spectral analysis confirms that the ions of the 45a and 45G fragments (both with four charges; spectral accuracy: 71%) also coexist and their relative abundances are 47% and 53%, respectively, when reading from the 3′-direction (t_R = 16.5–18.6 min).

CONCLUSION

The 2D-HELS-AA MS Seq technique expands RNA sequencing capacity beyond the four canonical ribonucleotides, and is able to determine the precise order of both canonical and nucleotide modifications potentially including any modification that an LC-MS instrument can detect. Unlike other successful sequencing technologies, we rely on mass differences of two adjacent ladder fragments to report identities of both canonical nucleotides and chemical modifications. Mass is an intrinsic nucleotide property that can be used to identity both known and unknown RNA modifications. This is very different than the use of proxies such as fluorescence or electronic signatures to report the identity of the four canonical nucleotides, which has limited capacity in discovering new and unknown base modifications. It is worth emphasizing that our method is a sequencing method, which includes both identification and location information on each nucleotide, canonical or not. This is very different than other RNA identification/characterization methods, which can only indicate the identity of RNA modifications but must rely on complementary established sequencing methods for sequence/location information. The primary purpose of the current work is to expand the sequencing capacity of this approach beyond the synthetic RNAs we reported on previously¹⁷ to achieve direct and de novo sequencing of biological RNA molecules like tRNA^Phe. Further characterization of RNA modifications was only needed when there were isomeric modifications that could not be differentiated by mass alone. We do not claim that our method can replace standard structural verification methods such as NMR, X-ray crystallography, and other chemical and enzymatic approaches that are specific to individual nucleotide modifications, which are designed to assess the chemical structure of such base modifications. Rather, these reliable methods are essential to further confirm the exact chemical structures of nucleotide modifications that we have revealed initially by their unique masses, such as isomeric base modifications.

Chemically, all RNAs consist of phosphodiester bonds that can be cleaved to generate mass ladders for our 2D-HELS-AA MS Seq. In this seminal study, the focus was to demonstrate that the approach is not limited to short synthetic RNAs (<35 nt) as described previously;¹⁷ we can indeed use it to sequence real biological samples such as tRNAs. However, in practice, the types of RNA that can be sequenced by this method are not only determined by our acid degradation chemistry for mass ladder generation, but by the capacity of LC-MS instrument to detect these mass ladders as well. The upper limit of RNA size that will give adequate resolution is LC-MS instrument-dependent, and the lower limit of the RNA sample loading amount is also instrument-sensitive. Both limits remain to be determined and will affect the utility of the approach. However, we are aiming to develop a general method that every user can tailor to their own instruments. Clearly, higher end LC-MS instruments provide higher mass resolutions (likely leading to higher read length) and/or higher sensitivity (likely leading to lower sample requirement). Once the method is fully developed, it will not be necessary for every end user to have a top-of-the-line instrument, since almost certainly companies offering the service will emerge, similar to many current vendors that provide NGS services. Nonetheless, the results of the 2D-HELS-AA MS Seq study revealed new isoforms, RNA base modifications, and editing, as well as their stoichiometries, in the tRNA that cannot be determined by cDNA-based methods (Figure 5), opening new opportunities in the field of epitranscriptomics.

Figure 5.

Summary of different RNA isoforms, base modifications, and base editing, as well as their stoichiometries, in the tRNA^phe.