Progress in Tandem Mass Spectrometry Data Analysis for Nucleic Acids

Abstract

Mass spectrometry (MS) has become a critical tool in the characterization of covalently modified nucleic acids. Well-developed bottom-up approaches, where nucleic acids are digested with an endonuclease and the resulting oligonucleotides are separated before MS and MS/MS analysis, provide substantial insight into modified nucleotides in biological and synthetic nucleic. Top-down MS presents an alternative approach where the entire nucleic acid molecule is introduced to the mass spectrometer intact and then fragmented by MS/MS. Current top-down MS workflows have incorporated automated, on-line HPLC workflows to enable rapid desalting of nucleic acid samples for facile mass analysis without complication from adduction. Furthermore, optimization of MS/MS parameters utilizing collision, electron, or photon-based activation methods have enabled effective bond cleavage throughout the phosphodiester backbone while limiting secondary fragmentation, allowing characterization of progressively larger (~100 nt) nucleic acids and localization of covalent modifications. Development of software applications to perform automated identification of fragment ions has accelerated the broader adoption of mass spectrometry for analysis of nucleic acids. This review focuses on progress in tandem mass spectrometry for characterization of nucleic acids with particular emphasis on the software tools that have proven critical for advancing the field.

1 |. Introduction

Mass spectrometry has proven extremely versatile for the characterization of many categories of biomolecules—including glycans, (Delafield and Li 2021) lipids, (Cajka and Fiehn 2014; Rustam and Reid 2018; Züllig, Trötzmüller, and Köfeler 2020) proteins, (Wysocki et al. 2005; Domon and Aebersold 2006; Han, Aslanian, and Yates 2008; Gillet, Leitner, and Aebersold 2016; Noor et al. 2021) and nucleic acids (Schürch 2016; Wetzel and Limbach 2016; Limbach and Paulines 2017; Li, Yuan, and Feng 2018; Santos and Brodbelt 2021; Herbert et al. 2024)—providing critical insight into their biological roles. Recent developments in the analysis of nucleic acids by mass spectrometry have focused on localization of both biological and synthetic covalent modifications (Schürch 2016; Wetzel and Limbach 2016; Limbach and Paulines 2017; Li, Yuan, and Feng 2018; Santos and Brodbelt 2021; Herbert et al. 2024; Chen, Yuan, and Feng 2019; Deng et al. 2024). These have been observed across many classes of biological nucleic acids, (Roundtree et al. 2017; Cappannini et al. 2024) including genomic DNA, (Klose and Bird 2006) tRNAs, (Hopper 2013; Jackman and Alfonzo 2013; Torres, Batlle, and Ribas de Pouplana 2014; Suzuki 2021) rRNAs, (Sergiev et al. 2011; Higa-Nakamine et al. 2012; Sergiev et al. 2018) miRNAs, (Newman, Mani, and Hammond 2011; Vitsios and Enright 2015; Creugny, Fender, and Pfeffer 2018; Gebert and MacRae 2019) and mRNA (Gilbert, Bell, and Schaening 2016; Zhao, Roundtree, and He 2017; Boo and Kim 2020; Anreiter et al. 2021; Flamand, Tegowski, and Meyer 2023; Warminski et al. 2023; Zhang, Dai, and He 2024). Covalent modification of these nucleic acids causes nuanced effects on the structure and function of the modified molecules, often dependent on the full complement of modifications occurring on each molecule (Roundtree et al. 2017; Gebert and MacRae 2019; Boo and Kim 2020; Lee et al. 2016; Li, Qu, and Yang 2023). Over 150 biological modifications have been identified to date, (Cappannini et al. 2024) each of which can influence the binding characteristics, degradation, and secondary structure of the modified nucleic acid. Additional noncanonical modifications can be readily incorporated into synthetic oligonucleotides, including a burgeoning class of therapeutic molecules, where changes to the pentose sugar, phosphodiester backbone, and nucleobases all impart differential functionality, including resistance to degradation by cellular processes and enhanced binding affinity to target molecules (Warminski et al. 2023; Prakash 2011; Eckstein 2014; Moumné, Marie, and Crouvezier 2022). A subset of potential nucleic acid modifications are shown in Figure 1. Development of robust data acquisition and processing workflows to analyze these modified nucleic acids is critical to assigning sequences and modifications and understanding their biological effects (Limbach and Paulines 2017; Herbert et al. 2024; Chen, Yuan, and Feng 2019).

FIGURE 1 |.

Chemical structures of selected biological and synthetic nucleic acid modifications and their locations. The noncanonical nucleobases dihydrouridine, inosine, and wybutosine are encoded by the symbols D, I, and yW, respectively.

Implementation of mass spectrometry analyses of nucleic acids is often orthogonal to traditional sequencing methods (Wetzel and Limbach 2016; Herbert et al. 2024; Chen, Yuan, and Feng 2019). Routinely used next-generation sequencing platforms provide rapid, accurate, and highly sensitive sequencing information owing to their use of polymerase chain reaction amplification to produce an ample amount of DNA complementary to the target oligonucleotide sequence (Deng et al. 2024; Behjati and Tarpey 2013; Muzzey, Evans, and Lieber 2015). Although this approach is exceptional for providing accurate sequence information, its dependence on complementary DNA eliminates the possibility of characterizing covalent modifications directly (Chen, Yuan, and Feng 2019; Ozsolak et al. 2009; Ozsolak and Milos 2011; Kellner, Burhenne, and Helm 2010; Motorin and Marchand 2021). Several approaches have been developed to capture these modifications and their locations within a nucleic acid sequence, including chemical derivatization, immunoprecipitation workflows, and direct sequencing methodologies (Herbert et al. 2024; Chen, Yuan, and Feng 2019). Chemical derivatization and immunoprecipitation both enable highly accurate measurement of a specific modification or subset of chemically similar modifications (Limbach and Paulines 2017; Herbert et al. 2024; Chen, Yuan, and Feng 2019; Kellner, Burhenne, and Helm 2010; Motorin and Marchand 2021). These rely on either specific chemical reactions or a specially designed antibody whose epitope corresponds to the modification of interest (Chen, Yuan, and Feng 2019; Kellner, Burhenne, and Helm 2010). Both methods subsequently sequence the modified or isolated nucleic acid utilizing next generation sequencing methods, generating information about the modification for which the assays are designed, but not for any other potential modifications present in the original target sequence (Kellner, Burhenne, and Helm 2010). Furthermore, direct sequencing methodologies, like nanopore sequencing, can capture modified nucleotides, however these workflows are not yet as robust as next-generation sequencing techniques (Herbert et al. 2024).

Mass spectrometry instead directly analyzes nucleic acid molecules by mass (Wetzel and Limbach 2016; Limbach and Paulines 2017; Li, Yuan, and Feng 2018; Herbert et al. 2024; Liu et al. 2024). Mass analysis here is critical, both in its sensitivity to any modifications that exhibit a difference in mass – the most notable exception to this being uridine vs pseudouridine, which is a nucleobase linkage isomer – and in its capability for direct analysis of the nucleic acid molecules (Wetzel and Limbach 2016; Limbach and Paulines 2017; Li, Yuan, and Feng 2018; Herbert et al. 2024). Analogous workflows are well-developed for analysis of posttranslational modifications in proteins and are poised for continued development in analysis of nucleic acid modifications (Chen et al. 2018; Fornelli et al. 2018, 2017). Compared to proteins, mass spectrometry approaches for the analysis of nucleic acids tend to be less advanced (Schürch 2016; Wetzel and Limbach 2016; Limbach and Paulines 2017; Herbert et al. 2024). In part, this is owed to the difficulty in MS analysis of nucleic acids relative to proteins; nucleic acids ionize less readily than peptides and proteins and often require using the negative polarity, which is more prone to corona discharge (Cole and Harrata 1993). In addition, the less heterogeneous monomer population of nucleic acids relative to proteins (5 canonical nucleobases vs. 20 canonical amino acids) leads to a higher likelihood of isomeric or isobaric fragment ions generated during tandem mass spectrometry (MS/MS) (McLuckey, Van Berkel, and Glish 1992; McLuckey and Habibi-Goudarzi 1993; Ammann et al. 2023). Both of these aspects present significant challenges to MS-based analysis of nucleic acids; however, new developments in the field alongside improvements in modern mass spectrometry platforms are enabling more routine analysis of progressively larger and more complex oligonucleotides (Herbert et al. 2024).

The two major regimes of mass spectrometric workflows for nucleic acid analysis are categorized based on the size of the analytes and associated sample preparation steps. In bottom-up workflows, nucleic acids are enzymatically digested to create oligonucleotides or individual nucleotides before separation with high-performance liquid chromatography and introduction to the mass spectrometer (Schürch 2016; Santos and Brodbelt 2021; Oberacher, Wellenzohn, and Huber 2002; Thüring et al. 2016; Wolf et al. 2022). This workflow enables rapid characterization of nucleic acids as large as many thousand nucleotides based on identification of specific digestion products (Wolf et al. 2022). Additionally, complete digestion of a nucleic acid population to single nucleotides can provide valuable information about the types of modified sugars and nucleobases present, and their relative abundances (Xie et al. 2023; Tost and Gut 2002). With these workflows, however, digestion removes each nucleoside or oligonucleotide from the overall context of the intact nucleic acid (Herbert et al. 2024). Given the important dynamics of combinatorial covalent modifications on the function of a nucleic acid molecule, direct and simultaneous interrogation of covalent modifications is critical. The alternative regime utilizes the top-down workflow for analysis of intact nucleic acids.

2 |. Top-Down Mass Spectrometry

To characterize combinatorial modifications, top-down methodologies aim to ionize and analyze intact nucleic acids (< 30 kDa, ~100 nt) without digestion (Wetzel and Limbach 2016; Herbert et al. 2024; Taucher and Breuker 2010, 2012; Hossain and Limbach 2007). This strategy presents unique benefits in observation of the various isoforms of each nucleic acid molecule, direct observation of modifications, and simultaneous characterization of modifications that exhibit a difference in mass (Schürch 2016; Wetzel and Limbach 2016; Limbach and Paulines 2017; Herbert et al. 2024; Taucher and Breuker 2010, 2012). Top-down workflows also have the potential of retaining the secondary and tertiary structures of nucleic acid molecules after their transition into the gas phase, providing an additional avenue for structural analysis (Rosu et al. 2010; Marchand et al. 2018; Gabelica 2021; Largy et al. 2022). Recent progress in top-down analysis of nucleic acids has focused on three specific areas: automated sample preparation, assessment and optimization of MS/MS characterization using different ion activation modes, and development of automated approaches for analysis of the wealth of data produced by mass spectrometry experiments (Limbach and Paulines 2017; Herbert et al. 2024).

2.1 |. Sample Introduction

A critical obstacle to analysis of nucleic acids by mass spectrometry is their affinity for binding small cations, such as ammonium, sodium, magnesium, and potassium, present in common electrospray ionization solutions (McLuckey, Van Berkel, and Glish 1992; Covey et al. 1988; Stults, Marsters, and Carr 1991; Potier et al. 1994; Little et al. Thannhauser, and McLafferty 1995, 1996; Birdsall et al. 2016). These small cations tend to interact with the acidic phosphodiester backbone characteristic of nucleic acids while in solution and can be retained during the electrospray ionization process (Stults, Marsters, and Carr 1991; Potier et al. 1994). This kind of adduction is problematic for mass spectrometry because stochastic adduction dilutes signal intensity across the m/z domain, prohibiting effective analysis, prohibiting interpretation of intact mass and isolation for MS/MS activation (Stults, Marsters, and Carr 1991; Potier et al. 1994). Some methods use matrix-assisted laser desorption ionization to ionize nucleic acids; however, these are not amenable to analysis or characterization of large nucleic acids and are outside the scope of this chapter (Gao et al. 2013). A recent review extensively summarized the challenges of analyzing nucleic acids using ESI and MALDI methods and described some of the key methods to mitigate the issues (Sharin, Floro, and Clark 2023). For introduction of nucleic acids by ESI, strategies such as adoption of nanospray and using metal chelators to minimize metal adduction or various solution additives to modulate charge states were discussed (Sharin, Floro, and Clark 2023). For analysis of nucleic acids by MALDI, the importance of selecting a matrix that minimized formation of clusters and metal adducts and increased pulse-to-pulse reproducibility were key practices (Sharin, Floro, and Clark 2023). MALDI in general has been less widely used for analysis of large nucleic acids, primarily owing to the generation of singly-charged ions that are less effective for characterization by MS/MS, in addition to the limited resolving power of TOF mass analyzers which have impeded confident assignment of large fragment ions.

Early MS analyses of nucleic acids often depended on substantial sample preparation to eliminate retention of adducts during ESI. One of the first techniques developed to mitigate adduction of nonvolatile cations was to treat oligonucleotides with high-concentration (10 M) ammonium acetate followed by treatment with cold ethanol to replace any adducts (often sodium) with more volatile ammonium ions (Stults, Marsters, and Carr 1991; Potier et al. 1994). These techniques yielded interpretable mass spectra of intact nucleic acids up to 72 nt in length, although incomplete desalting was observed for the largest species analyzed (Potier et al. 1994). Investigation into chemical additives for mitigation of adducts during ESI proved to be an effective strategy for generation of high-abundance nucleic acid anions (Greig and Griffey 1995).

Inclusion of organic acids and bases enabled generation of multiply charged anions of small, synthetic DNA molecules (Greig and Griffey 1995; Muddiman et al. 1996). At millimolar concentrations, piperidine and triethylamine both individually provided substantial suppression of sodium adduction, but exhibited decreased overall signal abundance of the precursor ions compared to imidazole (Greig and Griffey 1995). In combination, piperidine or triethylamine and imidazole were found to yield high-abundance precursors ions alongside a significant reduction of sodium adduction (Greig and Griffey 1995). Additional acetonitrile in the electrospray solution aided in minimizing potential sodium adduction during ESI (Muddiman et al. 1996). Furthermore, addition of organic acids such as formic and acetic acid effectively reduced the average charge of nucleic acid precursor anions observed after ESI, which has substantial implications on fragment ions generated by subsequent MS/MS (Taucher and Breuker 2012; Little, Thannhauser, and McLafferty 1995; Muddiman et al. 1996; Gawlig and Rühl 2023). Precursor ions in high charge states exhibit higher degrees of intramolecular Coulombic repulsion, which lowers the energy required for fragmentation and often leads to multiple secondary cleavage events (Wu and McLuckey 2004; Sun et al. 2023). This process tends to generate many uninformative internal fragments, containing neither the original 5′ or 3′ terminus, complicating the MS/MS spectrum and making assignment of terminal fragments more difficult (Lyon et al. 2018; Kenderdine et al. 2023).

To alleviate the burden of sample preparation, on-line desalting has risen in popularity as a sample preparation technique for isolated nucleic acids owing to its capability for high-throughput analysis and quality desalting. The earliest implementations of HPLC for desalting of nucleic acids employed similar mobile phase additives outlined above, (Greig and Griffey 1995) featuring triethylammonium acetate in an acetonitrile-rich solution to achieve near-complete desalting upon elution (Little, Thannhauser, and McLafferty 1995). HPLC techniques were further extended to employ ion pairing reagents to achieve desalting (Birdsall et al. 2016). While these techniques often employed an organic base, they often also included accompanying counterions to improve desalting, including solution-phase additives such as 1,1,1,3,3,3-hexafluoro-2-propanol (Birdsall et al. 2016; Macias et al. 2023). These techniques likewise provide near-complete desalting, including for nucleic acids containing 100 or more nucleotides.

A recent trend in sample desalting and introduction has incorporated the principles of on-line buffer exchange to preparation of nucleic acids (Figure 2), further reducing the manual sample preparation requirements for mass spectrometry analysis using ESI (VanAernum et al. 2020). These on-line techniques utilize the principles of size-exclusion chromatography, where the precursor molecules in solution do not interact with the stationary phase functionalization and are too large to enter the stationary phase pores (typically < 300 Å) (Macias et al. 2023; VanAernum et al. 2020; Crittenden, Lanzillotti, and Chen 2023; Lanzillotti and Brodbelt 2024a). In this way, nucleic acids can be desalted using basic (Crittenden, Lanzillotti, and Chen 2023; Lanzillotti and Brodbelt 2024a) or ion-pairing (Macias et al. 2023) mobile phases while in solution, then the large nucleic acid molecules elute with the stationary phase’s exclusion volume while small salts and other adducts are retained and can be diverted to waste (VanAernum et al. 2020). These techniques can effectively desalt > 100 nt nucleic acid molecules in a high-throughput fashion, enabling rapid acquisition of high-quality top-down mass spectrometry data with low sample usage per injection (Macias et al. 2023; Crittenden, Lanzillotti, and Chen 2023; Lanzillotti and Brodbelt 2024a). These on-line desalting methods arguably offer one of the most promising strategies for enhancing rapid analysis of larger nucleic acids and minimizing sample handling steps. For more complex and heterogeneous mixtures, separations using liquid chromatography or capillary electrophoresis are essential. A number of liquid chromatography modes have been explored for nucleic acids, including ones based on ion exchange, ion pairing reversed phase, size exclusion, and hydrophilic interaction liquid chromatography, and advances in these methods and capillary electrophoresis have been summarized in several recent reviews (Santos and Brodbelt 2021; Matos and Bülow 2018; Minkner et al. 2022; Wei, Goyon, and Zhang 2022; Guo 2024).

FIGURE 2 |.

Representation of nucleic acid desalting by size-exclusion chromatography. Small molecules (e.g., cations, anions from salts, shown as red circles) that frequently form adducts with nucleic acids interact with the pores of the stationary phase and are retained, whereas desalted nucleic acid molecules do not and elute with the solvent front.

2.2 |. MS/MS and Ion Activation Modes

Tandem mass spectrometry of oligonucleotides is well-established, and characteristic cleavages of the phosphodiester backbone result in an array of diagnostic fragment ions that facilitate sequencing (McLuckey, Van Berkel, and Glish 1992; McLuckey and Habibi-Goudarzi 1993). Fragment ions that retain the 5′ terminus (a,b,c,d) are complementary to those that retain the 3′ terminus (w, x,y,z) (Figure 3). While many modern mass spectrometry platforms can ionize and detect intact biomolecules greater than ~30 kDa at high resolution, additional considerations are necessary for effective activation and fragmentation of these large molecules. Developments in MS/MS methods have included widespread use of collisional activation, electron-based methods, and various photoactivation methods (Figure 4). While MS/MS of intact proteins has been broadly explored and optimized to yield improved sequence coverages and localization of modifications, (Macias, Santos, and Brodbelt 2020; Brown et al. 2020) these outcomes are not directly transferrable to fragmentation of nucleic acids owing to their differences in chemical structure and the types of bond cleavages induced (i.e., peptide backbone vs. phosphodiester backbone). For top-down analyses of nucleic acids, optimization of the activation parameters to balance production of abundant and informative backbone cleavages while mitigating excess fragmentation and conversion of diagnostic single-cleavage products into internal ions has been a key focus (Schürch 2016).

FIGURE 3 |.

Nucleic acid fragment ion nomenclature. Cleavages along the phosphodiester backbone are labeled based on which bond is cleaved and which terminus they contain.

FIGURE 4 |.

Schematic representation of the three major types of activation methods applied to nucleic acids (collision-, electron-, and photon-based) and the observed nucleic acid fragment ions produced by each method.

2.2.1 |. Collision-Based Dissociation (CAD) Methods

CAD is one of the most widely available and well-studied ion activation methods. CAD typically involves vibrational heating of the precursor ions via one or multiple collisions with neutral gas molecules in a collision cell (which may be a quadrupole, ion trap, or other module). In ion trap systems, CAD may be performed in a “resonant” manner in which the kinetic energies of selected precursor ions are increased via application of a frequency-selective waveform, or alternatively by accelerating ions through a separate collision cell (termed beam-type activation) in a process known as higher-energy collisional dissociation (HCD) (Olsen et al. 2009; Schuhmann et al. 2011). Stepwise deposition of energy on a 0.1–10 ms timescale leads to redistribution of vibrational energy throughout the molecule, resulting in cleavage of the most labile bonds. Collisional activation results in characteristic bond cleavages for nucleic acids anions and causes more nuanced dissociation dynamics based on nucleobase composition, functional groups present on the 2′ carbon, and charge state (McLuckey, Van Berkel, and Glish 1992; McLuckey and Habibi-Goudarzi 1993; Huang et al. 2008).

The earliest investigations into CAD of oligonucleotides coincide with some of the first analyses of nucleic acids by ESI-MS (McLuckey, Van Berkel, and Glish 1992; McLuckey and Habibi-Goudarzi 1993; Little et al. 1996; Greig and Griffey 1995). These initial studies focused on elucidation of the mechanisms by which small, synthetic deoxyribonucleic acids dissociated upon collisional activation. CAD of DNA strands containing 4–8 nt in a quadrupole ion trap exhibited high-abundance fragment ions arising from cleavage of the phosphodiester backbone, with secondary fragmentation occurring more frequently upon higher energy deposition (McLuckey, Van Berkel, and Glish 1992). Fragmentation of the 3′C-O bond was commonly preceded by loss of a nucleobase anion, particularly in adenine-containing nucleic acids, often resulting in complementary a-B/w ion pairs (McLuckey, Van Berkel, and Glish 1992). One of the first foundational studies confirmed that fragmentation in this manner proceeded via 1,2 elimination dependent on hydrogens from the deoxyribose sugar (McLuckey and Habibi-Goudarzi 1993). In addition, MS/MS spectra revealed preference for charged nucleobase loss upon CAD favoring adenine and thymine, resulting in limited fragmentation 3′ terminal to cytidine and guanine (McLuckey and Habibi-Goudarzi 1993). These fragmentation behaviors were also shown to depend on the charge state of the precursor ion, with higher charge states exhibiting increased Coulombic repulsion and thus more readily fragmenting, and disfavoring loss of the 3′ nucleobase and fragmentation at that nucleotide (McLuckey and Habibi-Goudarzi 1993).

A different pathway for formation of a/w fragment ions was observed for deprotonated oligodeoxynucleotides bound to metal cations (Wang, Taylor, and Gross 2001). CAD of these metal-adducted nucleic acids resulted in generation of a/w fragment ions at all nucleobases, with slightly higher abundances observed for fragments 3′ terminal to pyrimidines (Wang, Taylor, and Gross 2001). Here, metal cations not only stabilized the nucleic acids by minimizing Coulombic repulsion but also revealed another potential dissociation pathway for 3′ C-O bond cleavage, a pathway associated with thymine-containing oligodeoxynucleotides (Wang et al. Taylor, and Gross 2001, 1998). A charge-remote fragmentation mechanism was proposed, where nucleobase protonation from a mobile proton on its 5’ phosphodiester backbone precedes base loss and backbone cleavage in deoxyribonucleotides (Wang et al. 1998). This mechanistic interpretation was based on observation of site-specific base loss (a-B) fragments dependent on the gas-phase proton affinity of the nucleobase, where base loss followed the preference: G > A ~ C > T (Wang et al. 1998). Suppression of a-B fragment ions in T rich oligonucleotide anions and a lack of preference of backbone cleavage in oligonucleotide ions where all phosphodiester sites were bound to metal ions support this charge-remote mechanism (Wang et al. Taylor, and Gross 2001, 1998).

Examination of the dependence of CAD fragmentation on precursor ion charge state and nucleobase composition revealed both that phosphate groups adjacent to thymine nucleobases exhibited higher acidities, and thus higher average precursor charge states, but also that fragmentation trends appeared to follow the predicted delocalization of electron density in each nucleobase (Pan, Verhoeven, and Lee 2005). Experimentally, this appeared as preferential anionic base loss following the trend: A⁻ > G⁻ ≈ T⁻ > C⁻, where adenine anions were theorized to be the most stable via electron delocalization (Pan, Verhoeven, and Lee 2005). In addition, adenine-containing oligonucleotides exhibited formation of typical a-B/w fragment ion pairs, whereas thymine-containing sequences exhibited preference for a-B/w fragmentation at the 5’ terminus (Monn and Schürch 2007). Furthermore, the formation of a-B ions was suppressed for oligonucleotides with methylated phosphate functional groups which lack a mobile proton (Monn and Schürch 2007). Instead, d/z ions were observed in addition to typical a/w fragment ions (Monn and Schürch 2007).

CAD of multiply deprotonated RNAs resulted in production of c/y, a/w, and base loss products, (Huang et al. 2008) affording more streamlined fragmentation patterns than the corresponding deoxyribose nucleic acids. Fragmentation of oligonucleotides containing a mixture of ribose and deoxyribose sugars exhibited a strong dependence on the hydroxyl groups present at the 2’ carbon in ribose sugars (Schürch, Bernal-Méndez, and Leumann 2002). In these CID spectra, a mix of a/w products typical of nucleobase-driven pathways observed at deoxyribose sugars was observed, in addition to c/y fragments. Observation of these ions was indicative of fragmentation across the P-5’O bond along the phosphodiester backbone, which was attributed to abstraction of the 2’ hydroxyl proton to the 3’ phosphate. Schürch, Bernal-Méndez and Leumann (2002) The importance of the 2’ hydroxyl group on RNA fragmentation by CID was further confirmed by fragmentation of RNAs with 2’ fluorinated and 2’ O-methylated sugars, both of which limited formation of c/y fragments and instead yielded production of a/w fragment ions at those positions (Gao and McLuckey 2012).

These investigations into the dissociation of DNA and RNA by CAD established a framework for understanding the fragmentation pathways of nucleic acids (Schürch 2016; Wu and McLuckey 2004). The original studies applied CAD to increasingly large oligonucleotides, resulting in successful analysis of large (100 nt, ~30 kDa) molecules (Taucher and Breuker 2010; Little et al. Thannhauser, and McLafferty 1995, 1996; Kenderdine et al. 2023; Macias et al. 2023; Crittenden, Lanzillotti, and Chen 2023). Careful optimization of CAD parameters has resulted in in-depth characterization of heavily modified tRNAs (Taucher and Breuker 2012; Lanzillotti and Brodbelt 2024a; Huang, Liu, and McLuckey 2010). Follow-up studies have expanded the applicability of CAD to increasingly large oligonucleotides, including a 364 nt (118 kDa) ribonucleotide by leveraging the generation of secondary fragments, which originate from dissociation of a primary terminal fragment and contain neither the 5’ or 3’ terminus. Kenderdine et al. (2023) Assignment of internal fragment ions was performed utilizing commercial software (Bruker Daltonics) and were only considered for a-B, w, c, and y type fragment ions typical of CAD, resulting in up to 44% sequence coverage of the 364 nt molecule (Kenderdine et al. 2023). This methodology further expanded the utility of MS/MS characterization of nucleic acids by CAD (Kenderdine et al. 2023).

2.2.2 |. Electron-Based Dissociation Methods

Activation methods employing electrons (via free electrons or electron-donating reagents) to impart energy to precursor ions via exothermic electron transfer processes have been developed as an alternative to CAD methods to generate other types of fragment ions and enhance characterization of nucleic acids (Schürch 2016). These methods include electron capture dissociation (ECD), (Håkansson et al. 2003a, 2003b; Schultz and Håkansson 2004; Cooper, Håkansson, and Marshall 2005; Adamson and Håkansson 2007) electron detachment dissociation (EDD), (Mo and Håkansson 2006) electron transfer dissociation (ETD), (Smith and Brodbelt 2011; Hari, Leumann, and Schürch 2017) and negative-electron transfer dissociation (NETD), (Peters-Clarke et al. 2020; Guzmán-Lorite et al. 2024; Peters-Clarke et al. 2024) each leveraging generation of odd-electron radical precursor ions via capture or loss of an electron to induce fragmentation complementary to CAD methods.

ECD involves the precursor ion capturing a free electron generally generated from a heated filament (Zubarev, Kelleher, and McLafferty 1998). Originally developed for peptides and proteins, the ECD process cleaved different backbone bonds (carbonyl-alpha carbon bonds) of proteins rather than those cleaved by CAD (nitrogen-carbonyl bonds) (Zubarev et al. Kelleher, and McLafferty 1998, 2000). ECD typically resulted in a broader array of backbone cleavage sites compared to CAD (Zubarev et al. Kelleher, and McLafferty 1998, 2000; Kelleher et al. 1999; Ge et al. 2002; Kruger et al. 1999). Use of ECD was subsequently extended to nucleic acids, yielding production of w and d ions alongside isomeric a or z-B and c or x-B ions from synthetic DNA cations (Schultz and Håkansson 2004). Complete sequence characterization was attained for oligodeoxynucleotides when MS/MS data employing ECD was combined with MS/MS data using infrared multiphoton dissociation (Håkansson et al. 2003a). The limitation of ECD in its application to nucleic acids, however, is its reliance on activation of precursor cations, which are more difficult to generate at high abundance without adducts compared to nucleic acid anions (Cooper, Håkansson, and Marshall 2005; Adamson and Håkansson 2007; Yang and Håkansson 2006).

EDD provides more extensive applicability for nucleic acids compared to ECD because it is adapted for analysis of anions (Budnik, Haselmann, and Zubarev 2001). EDD proceeds by interaction of precursor anions with electrons of higher kinetic energy (> 10 eV) than those used for ECD (< 3 eV), resulting in electron detachment followed by backbone dissociation for peptides and nucleic acids (Mo and Håkansson 2006; Budnik, Haselmann, and Zubarev 2001). Application of EDD to DNAs and RNAs revealed comparable fragmentation for both types of oligonucleotides, without the characteristic dependence of 2’ carbon chemistry observed for CAD (Yang and Håkansson 2006). EDD produced an array of isomeric d or w, and c or x fragment ions when applied to oligoribonucleotides of homogeneous sequences, (Yang and Håkansson 2006) in addition to internal fragments (Taucher, Rieder, and Breuker 2010). Furthermore, application of EDD to structured nucleic acids revealed nuanced differences of fragment ion abundances depending on the secondary hairpin structure of several 15-mer sequences, an outcome related to stabilization of the oligonucleotide based on intramolecular interaction (Mo and Håkansson 2006). Application of EDD to large oligonucleotides (up to 61 nt) additionally yielded near-complete sequence coverage based on d and w fragment ions (Taucher and Breuker 2010). Base loss and production of internal ions, both of which are considered less informative or are difficult to assign, was decreased by reducing the internal energy, either by collisional cooling of the fragment ions or targeting precursor ions in lower charge states (Taucher and Breuker 2010; Taucher, Rieder, and Breuker 2010).

Unlike ECD and EDD, ETD and NETD rely on reagent ions, not free electrons, to carry out electron transfer reactions in the gas phase (Schürch 2016; Cooper, Håkansson, and Marshall 2005). ETD using fluoranthene as the reagent ion yielded complete sequence coverage of nucleic acid cations based on production of abundant d, w, a, a-B, and z ions (Smith and Brodbelt 2011). Sequence coverages were higher when ETD was combined with supplemental collisional activation, an outcome attributed to disruption of noncovalent interactions that allowed separation and detection of additional fragment ions (Smith and Brodbelt 2011). Like ECD, however, traditional ETD approaches were designed for use with cationic precursors as opposed to anions (Hari, Leumann, and Schürch 2017). To apply ETD to precursor anions, NETD utilizes different reagent ions such as xenon (Coon et al. 2005) or proton-adducted fluoranthene (Peters-Clarke et al. 2020; Guzmán-Lorite et al. 2024; Peters-Clarke et al. 2024; Huzarska et al. 2010) for the gas-phase ion-ion reactions. NETD of phosphorothioate RNAs by NETD generated a and z fragments, culminating in high sequence coverage (Peters-Clarke et al. 2024). Systematic characterization of 2’-modified RNAs was achieved using NETD with or without supplemental collisional activation (Peters-Clarke et al. 2020). More recently NETD afforded comprehensive sequence coverage of synthetic miRNAs based on production of abundant d and w fragment ions (Guzmán-Lorite et al. 2024). NETD offers another potential avenue for characterization of increasingly large and heavily modified RNAs (Peters-Clarke et al. 2020).

2.2.3 |. Photon-Based Dissociation Methods

Alternative ion activation methods leveraging photons to induce fragmentation have risen in popularity owing to their limited dependence on precursor charge state and monomer composition (Macias, Santos, and Brodbelt 2020; Brodbelt 2016). Photodissociation is typically implemented on instruments outfitted with an ion trap and employs a laser to irradiate the ion population (Macias, Santos, and Brodbelt 2020; Brodbelt 2016; Klein, Holden, and Brodbelt 2016). Photodissociation has been developed using infrared, visible or ultraviolet wavelengths (Macias, Santos, and Brodbelt 2020; Brodbelt 2016). The resulting ion fragmentation depends on the energy per photon: infrared multiphoton dissociation (IRMPD) relies on absorption of dozens or hundreds of low-energy infrared photons (< 0.1 eV per photon), resulting in a “slow-heating” vibrational activation process analogous to CAD (Macias, Santos, and Brodbelt 2020; Brodbelt 2016). Ultraviolet photodissociation (UVPD) leverages high-energy photons (> 3 eV per photon) to induce bond cleavage upon absorption of a single or few photons (Macias, Santos, and Brodbelt 2020; Brodbelt 2016; R. Julian 2017).

IRMPD has often been found to yield fragmentation analogous to CAD for various classes of biomolecules (Ledvina et al. 2009; Gardner et al. 2010; Greisch, van der Laarse, and Heck 2020). IRMPD has been used for characterization of both negatively and positively charged silencing RNA (siRNA), and the pre-dominant products were complementary c/y fragment ions along with losses of neutral nucleobases from precursor anions or losses of positively charged nucleobases from precursor cations (Gardner et al. 2010). IRMPD enabled characterization of modified nucleobases via neutral nucleobase loss from fragment ions or loss of nucleobase cations from the precursor ion (Gardner et al. 2010). Reducing the laser power allowed the energy deposition to be modulated to control the prevalence of low-energy pathways (decreasing the prevalence of nucleobase losses) versus high-energy pathways (increased secondary dissociation) (Gardner et al. 2010).

Owing to the greater energy deposition afforded by ultraviolet photons, UVPD offers access to higher-energy dissociation pathways which are otherwise inaccessible by CAD (Macias, Santos, and Brodbelt 2020; Brodbelt 2016; R. Julian 2017). In nucleic acids, this manifests as a complex mixture of a/w, c/y, d/z, and x ions, alongside some contributions from nucleobase loss and secondary fragments (Smith and Brodbelt 2011; Xu, Shaw, and Brodbelt 2013; Santos et al. 2022). An advantage of UVPD for analysis of modified nucleic acids is a lack of dependence on the functional group at the 2’ carbon; unlike collision-based methods, UVPD does not rely on a mobile proton at this site to initiate fragmentation and can reliably characterize modifications at this site (Smith and Brodbelt 2011; Santos et al. 2022). Likewise, the wealth of fragment ions produced by UVPD enabled straightforward localization of ligands, such as cisplatin, bound to a DNA strand (Xu, Shaw, and Brodbelt 2013). Ultraviolet photoactivation of precursor anions also results in production of intact, charge-reduced precursor ions generated by electron photodetachment (EPD) (Gabelica et al. 2006, 2007). Re-isolation of these charge-reduced precursors and application of supplemental collisional activation resulted in informative fragmentation exhibiting fragmentation tendencies of both UVPD and CAD (Crittenden, Lanzillotti, and Chen 2023; Santos et al. 2022; Gabelica et al. 2006, 2007). An example of a EPD mass spectrum is shown in Figure 5 for a modified oligoribonucleotide, patirasan antisense, a small interfering RNA-based therapeutic (Santos et al. 2022).

FIGURE 5 |.

a-EPD mass spectrum of patisiran modified antisense (m/z 1108, six-charge state) acquired in a Thermo Scientific Orbitrap Lumos mass spectrometer and the resulting sequence coverage map. UVPD was performed using 1 pulse at 1 mJ using 193-nm photons from an excimer laser, and HCD was applied using NCE 15%. The precursor ion is labeled with a star. Adapted from Ref (Santos et al. 2022). Copyright 2022 American Chemical Society.

2.3 |. Comparison of MS/MS Methods

As briefly summarized above, there have been inroads in the development and application of a number of MS/MS methods, and no single method has emerged as the dominant frontrunner. CAD remains the most widely accessible; the electron- and photon-based activation methods offer compelling alternatives with certain attributes for top-down applications for larger nucleic acids. The performance of CAD generally is enhanced for nucleic acids in lower charge states, whereas EDD offers better outcomes for higher charge states. Confounding secondary fragmentation can be minimized by focusing on precursors in lower charge states or by auxiliary collisional cooling to minimize internal energy (Taucher and Breuker 2010). The overall dissociation efficiency (i.e. conversion of precursor ions into fragment ions) is low for electron-based methods like EDD or for UVPD, and strategies to improve dissociation efficiency remain an important goal for advancing these MS/MS methods. Methods that integrate information from multiple activation methods provide the most comprehensive characterization of large nucleic acids, as also noted for top-down analysis of proteins (Chen et al. 2018; Fornelli et al. 2018).

2.4 |. Data Processing

The heterogeneity evident in a single MS/MS spectrum for even a relatively small (< 20 nt) nucleic acid is substantial, making manual interpretation difficult (Schürch 2016; Wetzel and Limbach 2016; Limbach and Paulines 2017; Herbert et al. 2024; Little et al. 1996). Unlike their counterparts in protein MS workflows, the available software tools for annotation of nucleic acid MS spectra are less developed and less numerous. A number of software tools have been recently launched by major MS vendors, largely in response to the growing emergence of nucleic acid-based biotherapeutics. The Oligo Workflow within Byos (Protein Metrics) offers both deconvolution of charge states and sequence confirmation based on MS/MS data (Protein Metrics). The OligoQuest workflow in Bruker’s BioPharma Compass software matches fragment ions, including visualization of internal ions, from MS/MS spectra to user-input sequences, allowing validation and examination of sequence truncations (BioPharma Compass n.d.). Within the Waters_Connect application, there is the MAP Sequence App and CONFIRM Sequence tools developed to confirm and characterize nucleic acids based on MS and MS/MS data (Waters Corporation n.d.). Agilent offers MassHunter BioConfirm software to confirm sequences based on MS/MS data, annotate spectra, and visualize outputs (Chemical n.d.). BioPharma Finder from Thermo Scientific has an oligonucleotide sequence editor to build custom oligonucleotide sequences including multiple modifications, then identify sequences based on MS/MS spectra derived from digested nucleic acids (Life Technologies - US 2024).

There are few sophisticated data processing workflows tailored to analysis of top-down MS/MS spectra of nucleic acids, particularly those employing higher-energy activation methods. The available tools employ different approaches for automated identification of nucleic acid fragment ions, typically in bottom-up workflows. The tools can be grouped into four general categories: those based on de novo sequencing, candidate sequence searching, database searching, or spectral library matching. A schematic representation of these four informatics approaches used to annotate complex MS/MS spectra is shown in Figure 6. The array of data processing tools is described in the following sections, and a summary of the available software approaches is provided in Table 1.

FIGURE 6 |.

Representation of the four major annotation approaches showcased for identification of nucleic acid fragment ions including: (A) de novo sequencing, (B) candidate sequence searching, (C) database searching, and (D) spectral library matching.

TABLE 1 |.

Summarized features of nucleic acid fragment ion annotation software.

Software	Annotation approach	MS workflow	Nucleic acid size (MS/MS level)	Fragment ion types
SOS	ab initio sequencing	Direct infusion	Up to 20 nt	a, a-B, c, w, y
MS2Links	Candidate sequence search with data reduction	Direct infusion top-down	Up to 20 nt	a, a-B, c, d-H2O, w, y
Ariadne	Database search	Bottom-up	5, up to 15 nt	a, c, w, y
RoboOligo	Automated and manual de novo sequencing	Bottom-up	Up to 15 nt	a-B, c, w, y
RAMM	Database search	Bottom-up	5, up to 20 nt	a-B, c, w, y
NIST	Spectral matching	Bottom-up	Up to 20 nt	Defined by spectral database
NASE	Database search	Bottom-up	15, up to 20 nt	All
Pytheas	Database search	Bottom-up	10, up to 20 nt	All
MIND4OLIGOS	Deconvolution	N/A	N/A	N/A
Nucleo-SAFARI	Candidate sequence search	Top-down	75+ nt	All

2.4.1 |. Simple Oligonucleotide Sequencer (SOS)

The SOS was one of the earliest developments in characterization of MS/MS data of nucleic acids (Rozenski and McCloskey 2002). Its initial development and implementation were in conjunction with a web-based application named Mongo Oligo Mass Calculator (Mongo Oligo, http://mass.rega.kuleuven.be/mass/mongo.html), that enables the calculation of various oligonucleotide masses, including a subset of possible fragment ions (Rozenski and McCloskey 2002). Mongo Oligo provides useful in silico generation of CID fragments (a, a-B, w, c, and y), possible nucleobase losses, intact precursor m/z values, internal fragment masses, and masses of potential endonuclease digestion products with a wide array of possible sugar and nucleobase modifications included. The lists of potential fragment ions created by Mongo Oligo can be manually matched to fragment ions in an MS/MS spectrum.

SOS extends the capabilities of Mongo Oligo by enabling ab initio annotation of MS/MS data for oligonucleotides of 20 nucleotides or fewer. This identification approach outlines several important considerations for nucleic acid MS/MS fragmentation, including recognition of unusual fragmentation owing to an unknown modification or substitution, discernment of sequential C or U nucleobases (0.985 Da mass difference), ambiguity in sequence assignment with increasing sequence length, and the high likelihood of isomeric or isobaric fragment ions (Rozenski and McCloskey 2002). To identify a nucleic acid sequence, a centroided mass list is provided from which a nucleic acid can be constructed based on identification of fragment ions. Fragments are identified bidirectionally, from both the 3’ and 5’ terminus, and putative assignments of the sequence are constructed based on user inputs and observed mass error of fragments containing the next potential nucleotide in the sequence. In this way, a sequence ladder can be constructed for relatively short oligonucleotides with confidence gained from identification of additional fragment ion series (including c/y ions) and overlap in the sequence ladders, which is assessed automatically within the software. SOS was applied to MS/MS data of several small oligonucleotide anions, including an example of its utility in reconstructing a modified sequence containing a 2’-O-methylated ribose within a 10-mer (Rozenski and McCloskey 2002).

2.4.2 |. MS2Links

The next advancement in processing nucleic acid MS/MS data was the development of MS2Links, (Kellersberger et al. 2004) software evolved from the previously developed MS2Assign (Schilling et al. 2003). MS2Assign was specialized to interpret MS/MS data of cross-linked peptides based on an input peptide sequence and potential cross-link locations. MS2Links extended MS2Assign with an additional data reduction step (Zhang and Marshall 1998; Horn, Zubarev, and McLafferty 2000) to select likely fragment ion peaks before identification and offered the capability to assign fragment ions of both canonical and modified oligonucleotides. This software was specifically developed for interpretation of nucleic acids modified by structural probes (Peattie and Gilbert 1980; Walker et al. 1982; Christiansen and Garrett 1988). These probes induced covalent labeling of nucleobases dependent on the molecule’s secondary structure, where base pairing suppressed the labeling reaction (Peattie and Gilbert 1980). Analysis of molecules modified by solvent accessibility reagents without mass spectrometry relied on labor-intensive chemical modification of target nucleic acids with subsequent SDS-PAGE, positioning mass spectrometry as an attractive alternative strategy (Kellersberger et al. 2004).

An important step introduced into the MS2Links software was the data reduction step. Data collected for evaluation of the software was recorded on a Fourier transform ion cyclotron resonance mass spectrometer, which provided excellent resolving power but included a significant number of potential peaks within each spectrum owing to the nature of the image current recorded for each scan (Marshall 2000; Marshall and Chen 2015). To capture even low-abundance fragment ions, MS2Links first filtered the raw data to only include plausible isotopic distributions based on a user-defined intensity threshold, followed by assessment of M + 1 and M + 2 isotopic peaks based on the charge state of the isolated precursor ion and the corresponding spacing of the isotopes. These isotopic distributions were then determined to be plausible by comparison to predicted isotopic abundances from the “averagine” model; (Senko, Beu, and McLaffertycor 1995) the hypothetical “average” amino acid. Because nucleic acids do not share the same average chemical composition as amino acids – the high relative prevalence of phosphorus (100% ³¹P) in nucleic acids biases the isotopic distribution towards the monoisotopic peak – large intensity tolerances were employed to ensure no legitimate ions were filtered out (Kellersberger et al. 2004). These intensity tolerances were also adjusted to capture low-abundance fragment ions whose abundances might be distorted. Observation of the potential monoisotopic mass was also required to be observed, even at low abundance (S/N = 1) (Kellersberger et al. 2004; Schilling et al. 2003).

This approach explored the complex nature of nucleic acids tagged with solvent accessibility reagents, for which a single species can potentially contain multiple positional isomers, further complicating the spectrum (Kellersberger et al. 2004). MS2Links was used for top-down analysis of various oligonucleotides ranging in size from 3700 to 6900 Da in conjunction with multi-stage MS/MS events where fragment ions were reisolated and dissociated again, ultimately achieving in-depth characterization of nucleotides modified with solvent accessibility reagents (Kellersberger et al. 2004).

2.4.3 |. Ariadne

Ariadne was one of the first database search engines to be applied to nucleic acid data (Nakayama et al. 2009). Specifically designed to process data generated from bottom-up LC-MS experiments, database search approaches are widely used in bottom-up proteomics workflows (McCormack, Eng, and Yates III 1994; Perkins et al. 1999). In the first study that deployed Ariadne, (Nyakas et al. 2013) modified noncoding RNAs, including a tRNA which exhibited a high degree of posttranslational modification and noncanonical nucleobases were characterized (Roundtree et al. 2017; Suzuki 2021). Digestion with RNAse T1 (cleavage 3’-terminal to G nucleobases) and subsequent HPLC separation using a reversed phase resin and ion-pairing mobile phases allowed collection of high quality bottom-up data that was supplied to the search engine.

The search engine requires a user-supplied set of DNA or RNA sequences in the FASTA format which are considered for searching. Ariadne generates an in silico digest of the supplied nucleic acid sequences—either from RNA sequences directly or via translation of genomic DNA—and creates a database of small oligonucleotide digestion products whose sequences are dependent on the restriction enzyme used in digestion (Nakayama et al. 2009). From these sequences, fragment ions were generated and searched within the MS/MS data recorded during HPLC separation. Special attention was devoted to typical CID fragment ions including c/y fragment ion pairs as well as a, a-B, and w fragments (McLuckey, Van Berkel, and Glish 1992; McLuckey and Habibi-Goudarzi 1993; Huang et al. 2008). Additionally, Ariadne is reported to tentatively identify internal fragments produced by secondary fragmentation. Considerations were also made within Ariadne to automatically identify potential modifications in RNA. Permutations of methylation sites in RNA on the four canonical RNA nucleobases, alongside reduction of uridine to dihydrouridine were included. These modifications were limited to two per oligonucleotide to keep the search times reasonable and to maintain high specificity (Nakayama et al. 2009).

To assess the fragment identifications in each spectrum for each potential set of fragment ions corresponding to digestion products, calculation of two scores for statistical validation was undertaken. First, the ‘nucleotide score’ calculates the probability of a false identification, where the likelihood of a random oligonucleotide matching to a significant number of fragment ions within the experimental spectrum is expected to be low (Nakayama et al. 2009). This calculation relies on the number of peaks searched within each spectrum, the resulting number of matched peaks, and the probability of a matched peak defined as the ratio of the mass tolerance to the mass searched (Nakayama et al. 2009). The “nucleotide score” is thus independent of the size of the database being searched. To evaluate spectral matches, a second ‘score threshold’ was calculated based on the number of identified fragments within the defined mass tolerance and a 0.05 significance level. These two scores allowed concise and statistically significant assessment of in silico oligonucleotide digestion products matched to MS/MS spectra and the observed fragment ions within (Nakayama et al. 2009).

Based on spectral matches, a nucleotide sequence map can be constructed. An additional scoring algorithm was included to generate a “mapping score” to characterize how well a given nucleic acid sequence was calculated based on the oligonucleotides identified by MS/MS, the relative frequency at which that oligonucleotide was identified, and the total number of oligonucleotides searched (Nakayama et al. 2009). This score is used to determine the most confident nucleic acid matches from those provided in the sequence database. The authors utilized Ariadne to achieve characterization of a small, unknown RNA and determine the candidate chromosomal from which it may have arisen based on MS/MS data and a genome-wide database, along with characterization of a post-translationally modified phenylalanine tRNA isolated from yeast (Nakayama et al. 2009).

2.4.4 |. Oligonucleotide Mass Assembler (OMA) and Oligonucleotide Peak Analyzer (OPA)

The oligonucleotide mass assembler (OMA) and oligonucleotide peak analyzer (OPA) work in tandem to fulfill the niche of analyzing MS data of hybridized DNA or RNA strands alongside accurate consideration of bound ligands including metal ions and small molecule drugs (Xu, Shaw, and Brodbelt 2013; Nyakas et al. 2013; Egger et al. 2008; Groessl et al. 2010). Furthermore, OMA and OPA took advantage of Java-based tools for direct read-in of Thermo Scientific mass spectrum “.raw“files, simplifying the user experience by eliminating the step of exporting a peak list.

The general workflow for OMA and OPA analysis first involves definition of one or more nucleic acid strands, which can contain a mixture of deoxyribose or ribose sugars, and any adducts (predefined as Na⁺, K⁺, or Pt(NH₃)_0–2²⁺) in OMA. OMA then generates an intact mass for the defined species, and corresponding fragment ions for both single and double-stranded precursor ions at multiple charge states. OPA then imports the theoretical fragment list and identifies observed m/z values based on a user-defined part-per-million m/z tolerance and with consideration of the experimental and theoretical fragment ion charge states. OMA likewise supports addition of user-defined modifications of the sugar, nucleobase, or backbone by their chemical formulas, and actively considers additional fragment ion series, including d/z fragment ions (McLuckey, Van Berkel, and Glish 1992) which were shown to produce high-quality characterization of a 2’-methoxy-methylphosphonate oligonucleotide which was expected to preferentially yield those fragments (Nyakas et al. 2013).

2.4.5 |. RoboOligo

RoboOligo was one of the more recent software applications that focuses on support for de novo sequencing of RNAs with modifications based solely on MS/MS results (Sample et al. 2015). This sequencing approach has three separate implementations; a fully automated procedure, manual sequencing, and a “variable sequencing”‘ methodology that combines the manual and automated methods. Each technique draws from the 102 unique nucleotide masses predefined within RoboOligo and was able to support charge states up to z = 3−. Additional restraints on automated sequencing restrict the number of modifications allowed per sequence in part based on RNAse digestion results with restriction enzymes of different specificities (McKenzie et al. 2012). Those with lower specificity will tolerate cleavage at modified residues, reducing the potential number of modifications in a single digestion product, while more specific enzymes will increase the potential size and complexity of digestion products.

The first step outlined in automated de novo sequencing generates a predicted combination of nucleotides based on the precursor ion m/z and charge. From here, sequences are constructed one nucleotide at a time, where the confidence of the next nucleotide in the sequence is dependent on the observed mass error and abundance of c and y fragment ions. This algorithm proceeds efficiently by removing potential sequences early in the process of de novo sequencing. The initial sequence is constructed from 5′ to 3′ based on identified c ions for potential nucleotides and is subsequently confirmed by the expected y ions from the sequence as a step to eliminate false identifications. Identification of w and a-B ions are subsequently used for scoring to aid in sequence confirmation. This process is repeated for each of the predicted nucleotide combinations determined based on the precursor m/z and charge. Scoring these identifications mimics the approach from SOS, where the abundances of fragments identified for a potential sequence are summed and the correct sequence is assumed to exhibit the highest total abundance (Rozenski and McCloskey 2002). The manual and variable sequencing methods operate upon similar principles as the automated de novo sequencing technique but involve user confirmation of each nucleotide. These aspects of the application are useful for rapid assessment of RNA sequences and aid in characterization of longer (> 10 nt) oligomers, as shown for annotation of two modified 14-mers generated from RNAse T1 digest of a Gln-tRNA^UUG in LC-MS/MS data (Sample et al. 2015).

2.4.6 |. RNAModmapper

RNAModMapper represented another advancement in annotation of RNA bottom-up data, as demonstrated for characterization of E. coli tRNAs as well as S. griseus rRNAs (Yu et al. 2017; Lobue et al. 2019a). Like Ariadne, RNAModMapper (RAMM) functions as a database search engine, where a set of potential RNA sequences are digested in silico to generate theoretical precursor and fragment ion masses. Up to 120 chemical modifications can be readily considered by RAMM, though positional isomers of each modification cannot be differentiated by mass alone and is reflected in the characterization outputs. Furthermore, modifications can be considered at fixed locations (e.g., the wobble base of a tRNA) or at variable locations based on a specific sequence motif (Yu et al. 2017).

Theoretical oligonucleotide masses are first compared to the m/z of the isolated precursor in each MS/MS scan, and theoretical fragment ions are searched within the spectrum for those that match within a defined precursor mass tolerance. Fragment identifications are then scored using a normalized binomial distribution probability relating to the identified c, y, w, and a-B fragment ions and overall oligonucleotide length, and a dot product score based on the similarity of experimental and reconstructed spectra. Both scoring metrics (normalized binomial distribution probability and dot product) are also used frequently in proteomics data processing approaches (Beausoleil et al. 2006; Yen et al. 2011). The binomial distribution probability score, (Beausoleil et al. 2006) termed the P-score, was weighted based on expected production of higher-abundance c and y fragments produced by CAD of RNA molecules, resulting in higher confidences when abundant fragment ions within each spectrum matched to c/y theoretical fragments. The dot product score (Yen et al. 2011) was employed to differentiate multiple sequences that matched a single MS/MS spectrum via spectrum reconstruction. This process involved retention of intensity values for all matched fragments within a spectrum and normalization of the abundances of unmatched ions to the average abundance of the remaining fragment ion peaks. The resulting dot product of the observed and reconstructed abundances determined whether the most abundant fragment ions were matched within the spectrum, increasing confidence in identifications with high dot product scores.

Application of this software to an E. coli tRNA extract resulted in interpretation of 58 of the 73 unique theoretical digestion products, including those generated for tRNAs containing known modifications sites. Overall, limitations still exist in using a single RNAse to map modifications, where small oligonucleotides limit determination of a unique locus within one RNA sequence and incomplete sequence coverage may result in identification of a given precursor RNA but not full characterization. The extension of RAMM to specific rRNAs isolated from S. griseus enabled more straightforward analysis of modifications as these RNAs exhibited less diverse modifications than the isolated tRNAs. Several oligonucleotides were characterized with high confidence, including two 16S rRNA digestion products containing methylations either on a nucleobase or ribose sugar (Yu et al. 2017). In this case, nearly all potential fragment ions were identified within a single MS/MS spectrum (Yu et al. 2017).

2.4.7 |. NIST Spectral Matching

Another common approach for MS/MS spectral identification utilized in proteomics involves spectral matching (Lam 2011). For spectral matching, rather than constructing a sequence de novo or referring to a sequence database for theoretical fragments generated in silico, instead experimental MS/MS spectra are compared to a database of known MS/MS spectra. In principle, well-controlled activation/fragmentation of a given precursor is reproducible, meaning that an MS/MS spectrum of a particular molecule in a specific charge state yields a unique “fingerprint.” Most often, this approach has been applied to GC-MS data of small molecules, where fragmentation tends to be less complex than that observed for peptides or oligonucleotides. However, recent developments have extended the spectral matching method to some bottom-up proteomics data with success (Vinaixa et al. 2016; Shao and Lam 2017).

To apply a spectral matching strategy to MS/MS data of oligonucleotides, the existing NIST spectral matching suite was augmented with a database of reference spectra from previously annotated MS/MS spectra from an RNAse T1 digestion of various modified tRNAs, in vitro transcripts, and synthetic oligonucleotides (Paulines, Wetzel, and Limbach 2019). Within the NIST spectral matching framework, previously characterized oligonucleotide MS/MS spectra, typically with all or most theoretical c/y fragments identified, were utilized as “search spectra,” and annotated fragment ions from these references were searched in the experimental LC-MS/MS spectra, named the ‘library spectra’. Peaks were identified using a 0.8 m/z tolerance, and spectral identifications were assessed using a cosine similarity score calculated via a dot product algorithm (Lam 2011).

Repurposing this spectral matching approach yielded many spectral matches from MS/MS data of a tRNA digest; however, a substantial limitation to spectrum matching was the dependence on the quality of the search spectra. In manual annotation of modified RNAs, a comparable set of search spectra was a necessity for spectral matching owing to the complexity of RNA modifications, (Cappannini et al. 2024) particularly those observed in tRNAs, (Sajek et al. 2019) and the necessary consideration of those modifications in the search spectra. As such, modifications can be overlooked easily without a proper set of reference spectra for the specific oligonucleotide sample being analyzed. Steps were taken to establish cosine similarity score thresholds to aid in subsequent interpretation of oligonucleotide LC-MS/MS data utilizing spectrum matching in future applications (Paulines, Wetzel, and Limbach 2019).

2.4.8 |. Nucleic Acid Search Engine (NASE)

The NASE represents one of the most contemporary additions to the field of oligonucleotide data analysis workflows and expands upon the utility of previously developed sequence database search engines (Wein et al. 2020). The major additions that NASE provides compared to existing database search methodologies are open code access via the OpenMS platform, (Röst et al. 2016; Sturm and Kohlbacher 2009) inclusion of scrambled decoy sequence searches and evaluation of false discovery rates (FDRs), and active consideration of potential adduct species during analysis. In addition, NASE can “correct” precursor masses in MS/MS spectra if the monoisotopic peak is not observed, which is critical for extending the utility of the application to larger oligonucleotides.

NASE was first showcased on a 21 nt synthetic, mature let7 microRNA, (Lee et al. 2016; Reinhart et al. 2000; Hutvágner et al. 2001; Johnson et al. 2005; Pillai et al. 2005) and achieved complete sequence coverage when using optimal collisional activation conditions (Wein et al. 2020). NASE also expanded the search to consider all potential main series fragment ion types (a/w, b/x, c/y, and d/z) and a-B fragments (McLuckey, Van Berkel, and Glish 1992). Comparison of the performance of NASE to Ariadne (Nakayama et al. 2009) and RAMM (Yu et al. 2017) for annotation of a 340 nt long S. cerevisiae noncoding RNA (MNE1) RNaseT1 digest exhibited that NASE was able to capture more and longer oligonucleotides compared to either Ariadne or RAMM. In part, this was due to the ability of NASE to account for sodium adducts present within the MS/MS spectra. The performance of Ariadne was hampered by the absence of a commercial preprocessing tool recommended for use, as acknowledged in the comparison (Wein et al. 2020). In addition, the performance of NASE enabled much faster analysis times, with searches completed on the NME1 and let7 datasets taking only seconds, and ~30 min when considering modified nucleonucleotides (Wein et al. 2020). The authors noted that the search times for RAMM and Ariadne when considering modified nucleotides were reported at over 1 month.

Utilizing the high-confidence identifications generated by NASE, searches of two NME1 (Mccorkle et al. 2014) digestion samples were carried out to assess localization of methylated nucleotides arising from treatment with NCL1, (Pavlopoulou and Kossida 2009) an RNA methyltransferase. In this experiment, no methylated nucleotides were identified in the control sample, as expected, and several 5-methylcytidine modifications were localized in the NCL1-treated sample. Further experiments comparing sequence coverage of human rRNAs and a highly modified tRNA sample generated additional localization of new modifications, including an methylguanine in 28S rRNA and characterization of several methylations and a thiouridine in the tRNA samples. These results showcased the utility of NASE in expanding the repertoire of bottom-up nucleic acid analysis tools to larger and more complex oligonucleotides, enabling rapid assessment of RNAse digests and expansion of high-throughput analyses for investigating epitranscriptomics (Wein et al. 2020).

2.4.9 |. Pytheas

Pytheas is, like NASE, a modern sequence database search engine developed in Python and available as a standalone application (D’Ascenzo et al. 2022). Analogous to NASE, Pytheas utilizes in silico generation of digestion products and theoretical fragment ion spectra from a sequence database followed by identification of fragment ions in MS/MS spectra. A new scoring system adapted from SEQUEST, (Gillet, Leitner, and Aebersold 2016) a well-established bottom-up proteomics analysis tool, was also implemented to assess oligonucleotide-spectrum matches (OSMs) and discriminate true and false identifications. Pytheas also includes a decoy search strategy to calculate a FDR for the experiment and introduces the capability of custom isotopic labeling for nucleosides which was utilized in conjunction with metabolic deuterium labeling to identify pseudouridine nucleobases within a yeast rRNA sample (D’Ascenzo et al. 2022).

Analysis of a purified 16S rRNA from E. coli grown in ¹⁴N (light) or ¹⁵N (heavy) growth media was examined on three mass spectrometry platforms to assess the applicability of Pytheas (Yamaki et al. 2020). Overall, oligonucleotides from 4 to 12 nt in length were routinely identified across all three platforms, including identification of five of the six digestion products that contained known methylation sites in the 1550 nt rRNA. Fewer OSMs were found for the ¹⁵N rRNA sample, likely owing to preprocessing of theoretical oligonucleotide sequence masses where theoretical sequences that are close (< 50 ppm) in mass are consolidated. This database search method were extended to S. cerevisiae 18S rRNA with metabolically labeled pseudouridine nucleobases, resulting in a + 2 Da mass shift for uridines and a + 1 Da shift for pseudouridines (Yamaki et al. 2020). Pytheas identified nine of the thirteen known pseudouridine sites based on these mass shifts, expanding the potential for mass spectrometry to characterize these previously undifferentiable modifications in a high-throughput manner (D’Ascenzo et al. 2022).

Pytheas was further applied to analysis of modified tRNAs, where 19 posttranslational modifications were identified in various tRNA families (Cappannini et al. 2024). Importantly, some chemically complex modifications, such as thiouridine, 5-carbamoylmethyluridine, N6-isopentyladenosine, and N4-acetylcytidine, were well-characterized using a targeted LC-MS/MS method. Modifications that are known to influence fragmentation by CAD, including 7-methylguanine, wybutosine, and N6-threonylcarbamoyladenosine, exhibited worse scores compared to modified nucleotides that exhibit standard fragmentation behavior. Last, application of Pytheas to an mRNA mimic, including a species labeled with ¹⁵N, showcased its applicability to quantitation workflows with a labeled internal standard, a critical factor in characterizing nucleic acid drugs (D’Ascenzo et al. 2022).

2.4.10 |. MIND4OLIGOS

MIND4OLIGOS is another recent algorithmic approach aimed at generating more easily identifiable monoisotopic masses of oligonucleotide ions and their fragments from m/z domain data (Prostko et al. 2024). Analogous to algorithmic deconvolution approaches often used to pre-process MS/MS data of protein ions, this application reduces the complexity of the m/z domain data (Senko, Beu, and McLaffertycor 1995). MIND4OLIGOS repurposes an existing approach designed for determination of the monoisotopic masses of protein ions (Lermyte et al. 2019) for use with nucleic acids. The MIND algorithm determines the monoisotopic mass based on the most abundant isotopic peak in a given isotopic distribution and its corresponding charge state and neutral mass (Lermyte et al. 2019). Recalibration of the deconvolution algorithm on existing MS data and in silico data including both ribo- and deoxyribooligonucleotides showcased reliable determination of the monoisotopic peak from a variety of isotopic distributions (Prostko et al. 2024). Furthermore, for ions of 11 kDa or above, the monoisotopic peak abundance tended to be less than 5% of the most abundant isotopic peak (Prostko et al. 2024). The MIND4OLIGOS approach was also relatively robust in regard to the presence of various posttranscriptional modifications (Prostko et al. 2024). In oligonucleotides containing over three sulfur atoms, errors in monoisotopic mass determination were observed, however, the algorithm was able to accurately assign many other common types of covalent modification (Prostko et al. 2024). This approach was also implemented in an R Shiny application, increasing its accessibility and ease-of-use (Prostko et al. 2024).

2.4.11 |. Nucleo-SAFARI

Nucleo-SAFARI is a recently developed R Shiny application, like MIND4OLIGOS, designed specifically for identification of fragment ions in top-down MS/MS spectra of nucleic acids based on their isotopic distributions (Lanzillotti and Brodbelt 2024b). This application is specifically compatible with mass spectral data recorded by Orbitrap instruments, where averaged spectra exported to a single-scan “. raw” file can be read into R directly using the rawrr package. Identification of fragment ions first proceeds through in silico generation of precursor and fragment ion chemical formulas, monoisotopic masses, and isotopic distributions. Isotope distributions are calculated via the IsoSpecR package, which generates fine isotopic structures for each chemical formula that are collated based on their nominal mass. These theoretical isotopic distributions are identified in the m/z domain based on a user-defined ppm error tolerance (default of 10 ppm) against the peak centroid values stored within the “. raw” file. Fragment ions are considered candidates for validation if a majority (default 70%) of the predicted probability of the isotopic distribution is matched to experimental centroids within this m/z tolerance. To validate identified fragment ions, abundances are predicted for each isotope based on their predicted probability and the sum of the peak centroid abundances corresponding to the matched peak centroids. If the mean absolute error between the predicted abundance and the observed peak centroid abundance for each isotope in the distribution is below a user-defined threshold (default 25%), the fragment ion is considered a match. Nucleo-SAFARI also includes a nonnegative least squares approach for identification of fragment ions exhibiting gain or loss of a hydrogen atom during dissociation, a common process observed upon UVPD.

This approach was applied to characterization of synthetic let7 miRNA precursors (each 78 nt, 25 kDa) exhibiting differential patterns of methylation and deoxidation (Lanzillotti and Brodbelt 2024a). Top-down MS/MS of these oligonucleotides using UVPD generated informative spectra, and annotation with Nucleo-SAFARI allowed identification of specific w fragment ions near the 3’ terminus that differentiated the modification states of the three let7 miRNA precursor variants (Lanzillotti and Brodbelt 2024a). Nucleo-SAFARI also facilitated the identification of isoforms of tRNAs introduced via an on-line desalting method combined with gas-phase fractionation that allowed sequential analysis of dozens of tRNAs (Lanzillotti and Brodbelt 2024c).

3 |. Conclusions and Outlook

The expansion of innovative tandem mass spectrometry approaches and new data processing methods has advanced characterization of myriad biological and synthetic nucleic acids. With these developments, along with ongoing improvements to the throughput and sensitivity of MS-based methods, mass-spectrometry is well-positioned to play a critical role in analysis of the posttranscriptional modification landscape and decipher how these nuanced regulatory mechanisms impact the biological functions of RNAs within the cell.

Two of the most critical challenges limiting analysis of biological nucleic acids, such as RNAs, by mass spectrometry are related to sample requirements. Decreasing the amount and concentration of nucleic acids required for top-down analysis is imperative to match the amount that can be effectively purified from cells. RNA purification yields are highly variable based on cell type and extraction method. Mammalian cells on average contain 10–30 pg of total RNA, the majority of which are tRNAs and rRNAs (El-Khoury et al. 2016). In some RNA classes, like mRNA, there may be fewer than 100 molecules of a single species per cell, making MS analysis difficult (Berry and Pelkmans 2022). Thus, isolation of a single RNA species for mass analysis can be prohibitive, either limiting MS analysis to high-abundance RNA species or requiring a substantial number of cells for purification.

Utilization of more sensitive mass spectrometry methods, including contemporary MS platforms and techniques like charge detection mass spectrometry (CDMS) can begin to address the discrepancy between the amount of an RNA that can be purified from cells and what can be effectively analyzed by MS. CDMS is an emerging technique that leverages the high sensitivity of MS detectors to measure fewer than 100 ions at a time, enabling simultaneous determination of the ion m/z and charge (Jarrold 2022). Using CDMS techniques enables mass analysis of low concentration (< 50 nM) samples, enhancing sensitivity (Kafader et al. 2020a). The recent development of the selected temporal overview of resonant ions (STORI) technique on Orbitrap platforms monitors the integral of induced current over time of a single ion at a given m/z value, where the slope of these plots is dictated by the single ion’s charge (Kafader et al. 2020a, 2019a, 2019b; Wörner et al. 2020; Kafader et al. 2020b). This technique leverages the high sensitivity for multiply charged ions, even when analyzing low concentration (1–0.1 nM) samples, potentially providing an avenue for analysis of low-abundance nucleic acids (Kafader et al. 2020a, 2019a).

The high degree of heterogeneity apparent in biological RNAs also presents substantial analytical challenges in MS workflows. Without an effective separation method such as chromatography or ion mobility, gleaning additional characterization information—most notably from tRNA isoforms with low relative abundances—will be difficult. This high relative homogeneity of the tRNAs in length and chemical characteristics make separation by HPLC prohibitive with the reversed phase chemistries commonly employed for analysis of smaller oligonucleotides. Development of additional separation strategies that leverage different chemical characteristics of large RNAs to resolve their isoforms is critical to more effective top-down MS analyses with less extensive sample preparation. An attractive candidate for investigation is hydrophilic interaction chromatography (HILIC), which operates using solvents compatible with nucleic acids and achieves separation based on the hydrophilicity of the analyte (Buszewski and Noga 2012; Lobue et al. 2019b; Huang et al. 2021; Lardeux, D’Atri, and Guillarme 2024). This type of separation may prove to be more effective than traditional reversed phase separations given the overall hydrophilicity of nucleic acids. Capillary electrophoresis (CE) separation is also a potential avenue for on-line separation of nucleic acids. Separation of nucleic acid based on electrophoretic mobility has been a successful method for characterizing small (< 20 nt) oligonucleotides and large (> 1000 nt) nucleic acids (Santos and Brodbelt 2021; Wei, Goyon, and Zhang 2022). The high separation efficiency afforded by CE along with low sample requirements provide an attractive alternative to HPLC-based methods (Wei, Goyon, and Zhang 2022). Additionally, gas-phase methods such as ion mobility can provide an additional dimension of separation based on the shape of the ion after introduction to the mass spectrometer (Santos and Brodbelt 2021; Fabris 2021). Ion mobility-based methods can provide additional detail about the structure of nucleic acids while potentially resolving samples that are highly heterogeneous in the m/z domain (Santos and Brodbelt 2021; Fabris 2021).

Finally, additional developments to fragment identification software approaches will enable more widespread adoption of top-down MS methods for analysis of nucleic acids. Although the approaches described earlier are effective, the algorithmic assessment of matched fragment ions, the speed at which spectra can be effectively analyzed, and statistical validation approaches can be further improved. Additional studies examining the fragmentation pathways and trends of large (> 50 nt) nucleic acids can provide more clarity about the effects of precursor charge state, nucleotide composition, and covalent modifications on the resulting MS/MS fragmentation patterns. Scoring based on the expected abundances of specific fragment ions may offer increased confidence in the identification of a given nucleic acid species or in localization of a specific modification, particularly in the case of nucleobases like 7-methylguanosine. Additionally, implementation of machine learning methods for interpretation of MS/MS fragment results may provide additional benefits in speed and accuracy of nucleic acid characterization by MS/MS. Furthermore, extension of previously developed de novo strategies for sequencing nucleic acids directly from an MS/MS spectrum to more complex nucleotide components and to larger potential molecules can be a powerful tool for discovering new posttranslational modifications (Rozenski and McCloskey 2002; Sample et al. 2015; Yu et al. 2017). Recent focuses on development of database search algorithms have largely moved away from de novo interpretation of larger (> 10 nt) oligonucleotides; however, development of optimized MS/MS strategies for top-down MS methods are providing fragmentation data amenable to this type of analysis. Incorporation of internal fragments into characterization approaches also presents an intriguing, if challenging, approach to gleaning additional sequence information from each MS/MS spectrum (Lyon et al. 2018; Kenderdine et al. 2023). The limited number of nucleobases in canonical nucleic acids and the symmetrical nature of the canonical phosphodiester backbone (O-PO₂-O) results in a high probability that small (< 10 nt) internal fragment ions cannot be uniquely localized within a candidate sequence. Additionally, the high likelihood of isomeric fragments further complicates identification of internal fragments. To successfully implement internal fragment identification into MS/MS workflows, acquisition of high-resolution (> 240,000) with high mass accuracy (< 2 ppm) is necessary, alongside rigorous statistical validation to ensure false identifications are mitigated.