A Novel Adductomics Workflow Incorporating FeatureHunter Software: Rapid Detection of Nucleic Acid Modifications for Studying the Exposome

Abstract

Exposure to the physicochemical agents that interact with nucleic acids (NA) may lead to modification of DNA and RNA (i.e., NA modifications), which have been associated with various diseases, including cancer. The emerging field of NA adductomics aims to identify both known and unknown NA modifications, some of which may also be associated with proteins. One of the main challenges for adductomics is the processing of massive and complex data generated by high-resolution tandem mass spectrometry (HR-MS/MS). To address this, we have developed a software called “FeatureHunter”, which provides the automated extraction, annotation, and classification of different types of key NA modifications based on the MS and MS/MS spectra acquired by HR-MS/MS, using a user-defined feature list. The capability and effectiveness of FeatureHunter was demonstrated by analysing various NA modifications induced by formaldehyde or chlorambucil in mixtures of calf thymus DNA, yeast RNA and proteins, and by analyzing the NA modifications present in the pooled urines of smokers and non-smokers. The incorporation of FeatureHunter into the NA adductomics workflow offers a powerful tool for the identification and classification of various types of NA modifications induced by reactive chemicals in complex biological samples, providing a valuable resource for studying the exposome.

Graphic abstract

1. INTRODUCTION

Exposure to certain endogenous and exogenous agents and processes, such as cellular metabolism, radiation, various chemicals, and xenobiotics, can cause changes to the nucleic acids (NA) in organisms, leading to the formation of DNA and RNA modifications a.k.a. adducts. The totality of these modifications is known as the NA adductome. DNA modifications are complex and various types of alterations have been identified, including nucleobase modifications, DNA-DNA crosslinks (DDCL), and DNA-protein crosslinks (DPCL) ^1–3. These DNA modifications may interfere with, or otherwise participate in, DNA replication, transcription, and epigenetic processes ^{4, 5}. If undesired DNA modifications are not properly repaired or reverted, downstream consequences such as genetic mutations can occur and increase the risk of developing diverse diseases, including cancer ⁶. Compared to DNA, RNA is more vulnerable to modification ⁷, which affects its coding properties and obstructs the synthesis of crucial cellular proteins. Over 170 RNA modifications have been identified and associated with the epitranscriptome (e.g., N⁶-methyl-adenosine) ⁸, and these modifications play important roles in various cellular processes such as RNA stability, localization, transcription and translation initiation ^{9, 10}. Like DNA, new types of RNA modifications and damage are frequently being discovered. These include RNA-RNA crosslinks (RRCL) caused by UV exposure, for example ¹¹, RNA-protein crosslinks (RPCL) which can be caused by UV or formaldehyde exposure ¹², and the recently observed DNA-RNA crosslinks (DRCL) induced by formaldehyde in mice treated with 4-(methylnitrosamino)-1-(3-pyridyl)-1-butanone ¹³. RNA modifications can significantly affect RNA processing and metabolism at multiple levels ⁹ and the dysregulation of RNA functionality has been linked to various diseases, including cancer, neurodegenerative, and metabolic disorders ^14–18. Because NA modifications can reflect the effect of agents which reach the body’s cells, where they may then lead to disease ^19–21, the analysis of the NA adductome can reveal how environmental exposures alter DNA and RNA, improving our understanding of how they participate in pathogenesis, which is critical to exposome research.

Targeted tandem mass spectrometry (MS/MS) is currently the most widely adopted approach for identifying DNA and RNA adducts ^{6, 22}. However, this approach may overlook some significant, but unexpected, DNA and RNA adducts. Non-targeted analytical methods have recently been demonstrated to be a promising strategy for comprehensively detecting a broad range of NA modifications though these approaches are currently mostly limited to the detection of DNA modifications (i.e., DNA adductomics ^{23, 24}). Several DNA adductomics approaches are based on liquid chromatography-high-resolution tandem mass spectrometry (LC-HR-MS/MS) operated in data-dependent acquisition (DDA) or data-independent acquisition (DIA) mode ^{25, 26}. For DNA adductomics analysis using LC-HR-MS/MS, DNA adducts are featured by a precise neutral loss (NL) mass due to the dissociation of 2-deoxyribose (dR) moiety from the adduct because the N-glycosidic bond between dR and nucleobase (nB) in 2’-deoxyribonucleosides (2’-dN) is labile. To facilitate the identification of such DNA adducts (signals) from the massive MS/MS spectra acquired by DDA- or DIA-HR-MS/MS, several post-data analysis software have been developed, such as the wSIM-City ²⁷ and nLossFinder ²⁸ for DIA, and DFBuilder ²⁵ for DDA within the MZmine platform. Recently, we introduced the concept of NA adductomics using HR-MS/MS and demonstrated its feasibility using human urine. Our results reveal the existence of at least six types of NA modifications, including mono-2’-dN adducts and -ribonucleosides (rN) adducts, nB adducts, DRCL, RRCL and RPCL, and we show, for the first time, that examples of all of these forms of adducts are present in urine ²⁹. To date, the identification of NA modifications from complex NA adductomics datasets relies heavily on the careful manual inspection of acquired MS spectra, which is time-consuming and may lead to inconsistencies or inaccuracies in the results.

In the present study, we report the development of a software named “FeatureHunter”, which provides an automated and comprehensive MS data analysis workflow for NA adductomics. FeatureHunter enables automated extraction, annotation, and classification of various NA modifications based on HR-MS/MS data. This software allows for the identification of, not only typical MS/MS features of DNA adducts but also, additional MS/MS features to mark typical and non-typical (e.g., NA adducts associated with proteins) adducts derived from both DNA and RNA by deeply mining the signal features in HR-MS and HR-MS/MS spectra.

Additionally, FeatureHunter facilitates the detection of peak pairs with a fixed mass difference in MS¹ and effectively excludes false positive results arising from in-source fragmentation. The reliability and performance of FeatureHunter for NA adductomics were demonstrated through the analysis of: (i) 92 modified NA standards, (ii) NA modifications induced by formaldehyde and chlorambucil (CLB, an alkylating agent) in a mixture of calf thymus DNA (CT-DNA), yeast RNA, and protein (bovine serum albumin, BSA), and (iii) NA modifications present in the pooled urines of smokers and non-smokers. This study demonstrated the successful development and application of FeatureHunter which will facilitate NA adductomics analysis through the high-throughput and high-accuracy profiling of NA modifications in biological samples.

2. EXPERIMENTAL SECTION

2.1. Chemicals.

Native 2’-dN, rN, unlabelled CLB (d₀-CLB), unlabelled paraformaldehyde, CT-DNA, yeast RNA, BSA, lysine (Lys) and cysteine (Cys) were from Sigma-Aldrich (St. Louis, MO, USA). d₈-CLB and ¹³C,d₂-paraformaldehyde were from Cayman chemical (Ann Arbor, MI, USA) and Cambridge Isotope Laboratories (Tewksbury, MA, USA), respectively. A complete description of the NA modifications used is provided in the Supporting Information, including commercially available 73 unlabeled NA modifications (23 modified 2’-dN, 21 modified rN, 29 modified nB) and 19 stable isotope-labeled (SIL) NA modifications (Table S1), and 13 in-house synthesized CLB-induced DDCL and nucleobase-nucleobase crosslink (nBCL) products (Table S2) ³⁰. Other potential types of NA modifications induced by CLB [i.e., DRCL, DPCL, RRCL RPCL, and nucleobase-protein crosslinks (nBPCL)] were synthesized using the native 2’-dN, rN and two amino acids (AA; i.e., Lys and Cys). Full details, including a list of abbreviations, are provided in the Supporting Information.

2.2. Induction of multiple NA modifications by formaldehyde and CLB.

To comprehensively demonstrate the capability and performance of NA adductomics using FeatureHunter, formaldehyde and CLB were employed as representative toxic agents to induce a broad range of NA modifications in a mixture of DNA, RNA and proteins ^{30, 31}. This study also applied the stable isotope labeling and mass spectrometry (SILMS) ³ technique, which involves the co-exposure of both unlabeled and labeled toxicants to the biological system to facilitate the confirmation of toxicant-induced NA modifications. A protocol combined from three earlier methods ^{30, 32, 33}, involving nuclease P1, alkaline phosphatase, snake venom phosphodiesterase I, RNase A and protease was used to release the potential crosslinked products from the mixtures of DNA, RNA and proteins. Full experimental details are given in the Supporting Information.

2.3. Human urine samples collection and purification.

Two pooled urine samples were prepared to represent non-smokers and heavy smokers by combining equal aliquots from 11 non-smokers or 12 smokers. Details of urine collection and purification are given in the Supporting Information. To confirm the self-reported smoking status of each participant, the urine samples were also analyzed for cotinine and total 4-(methylnitrosamino)-1-(3-pyridyl)-1-butanol (NNAL, a well-known tobacco-specific nitrosamine metabolite) using our validated online SPE LC-MS/MS methods ^{34, 35}. To correct for urine concentration, urinary creatinine was also measured using a HPLC-UV method, as previously described ³⁶.

2.4. LC-HR-MS/MS analysis with data-dependent (DD)-MS² acquisition.

All the NA modifications were separated by a reversed-phase HPLC system (Thermo Fisher Scientific, Vanquish UHPLC, Waltham, MA, USA) using an Inertsil ODS-3 C18 column (150 × 2.1 mm i.d., 5 μm, GL Sciences, Tokyo, Japan). The LC conditions applied has been described in detail previously ³⁷. A hybrid quadrupole-linear ion trap-orbitrap MS was used (Q-LIT-OT-MS; Orbitrap Fusion Lumos Tribrid MS from Thermo Fisher Scientific, MA, USA) equipped with a heated electrospray ionization (HESI) source and operated in positive ESI mode. The source voltage was 3.5 kV. The gas setting of sheath gas, aux gas and sweep gas were 35, 7 and 0 arbitrary units, respectively. The vaporizer and ion transfer tube temperatures were optimized to be 300 and 250 °C, respectively. The RF lens was set at 45%. The detailed optimization of conditions is described in section 2.5. Solvent blanks were analyzed before and after acquisition to evaluate potential contamination and sample carryover between injections. Data acquisition and processing were conducted by Xcalibur software 4.3 (Thermo Fisher Scientific).

Non-targeted DD-MS² analysis was performed with full-scan detection ranged from m/z 125 to 1000, followed by MS² fragmentation with a cycle time of 2.5 s. Full-scan detection was conducted using the Orbitrap detection at a resolution of 60,000 with automatic gain control (AGC) at standard mode and an auto maximum ion injection time setting. Ion signals with an ion intensity greater than 10,000 from the MS¹ full-scan spectra, were automatically selected by a quadrupole (isolation width of 1.6 amu) and fragmented in the linear ion trap (MS² fragmentation) with collision-induced dissociation (CID) at a normalized collision energy (NCE) of 40% or in the ion-routing multipole with higher energy collisional dissociation (HCD) at a fixed NCE of 30% or in the ion-routing multipole with wide higher energy collisional dissociation (wide-HCD) at a stepped NCE of 20, 60 and 100%. With the incorporation of three distinct collision modes governing the mechanism of fragmentation and the utilization of varied collision energies ^{23, 38}, the aim was to effectively capture representative and reproducible features for detecting a diverse range of modifications. The MS² resolution was 15,000. Dynamic exclusion was enabled to prevent repeated acquisition of spectra from the most abundant ions during the analysis (exclude after one time triggered; exclusion duration: 6 s). To improve the sensitivity, an exclusion list was applied in MS¹ precursors selected for MS/MS fragmentation, including the unmodified 2’-dNs, rN, nB, the corresponding dimers and ion adducts (e.g., [2M+H]⁺, [M+Na]⁺, [M+K]⁺ and [M+NH₄]⁺), as well as the background ions from solvent blank, as recommended previously ³⁹.

2.5. Optimization of MS parameters for analyzing various NA modifications.

To achieve sufficient sensitivity for simultaneous determination of various types of NA modifications, the MS parameters, including vaporizer temperature (150-350 °C), ion transfer tube temperature (150-350 °C) and RF lens (10-80%), were optimized and determined to be optimized at 300 °C, 250 °C and 45%, respectively. The results of optimization are summarized and visualized as a heat map in Supporting Information Figure S1.

3. RESULTS

3.1. Establishment of characteristic features for FeatureHunter by analysis of multiple NA modifications under CID, HCD and wide-HCD fragmentation.

FeatureHunter was developed to search and tag specific features of NA modifications. Therefore, at the core of FeatureHunter is the establishment of specific features of different types of NA modifications that arise through collision in MS. Both the commercially available NA modifications and in-house synthesized CLB-induced NA crosslinked products were used to obtain their dominant features/fragments in CID, HCD and wide-HCD. Figure 1 shows an example of the representative features of seven types of NA modifications under CID, HCD and wide-HCD (including modified 2’-dN, rN, nB, DDCL, RRCL, DRCL and nBCL products). Figure S2 shows the representative features of three other types of NA modifications (i.e., DPCL, RPCL and nBPCL), following CID, HCD and wide-HCD. The major fragmentation patterns of these ten different types of NA modifications are summarized and illustrated in Figure 2 (modified 2’-dN, rN, DDCL, RRCL and DRCL), Figure S3 (nB adducts and nBCL) and Figure S4 (DPCL, RPCL and nBPCL). In general, the dominant features in MS/MS for modified 2’-dN, rN, nB, DDCL and DRCL were consistent with previous work ^{13, 30, 37, 39–42}. As shown in Figure 1, the modified 2’-dNs tend to lose the dR, resulting in the product ion of [M+H-dR]⁺ and [dR+H]⁺. Modified rNs tend to lose ribose (R) or the methylated ribose moiety (MeR), resulting in the product ion of [M+H-R]⁺ or [M+H-MeR]⁺. For the nB adducts, these produce specific product ions, such as [nB]^+•, [nB+H]⁺ and [nB+H-NH₃]⁺. Additionally, nB adducts can lose the nB itself, resulting in a product ion [M+H-nB]⁺ (Figure S2D). DDCL gives product ions of [M+H-dR]⁺ and [M+H-2dR]⁺. Furthermore, RRCL has product ions of [M+H-(R or MeR)]⁺ and [M+H-(2R or 2MeR)]⁺ or [M+H-(R+MeR)]⁺ while DRCL has product ions of [M+H-dR]⁺ and [M+H-(R or MeR)]⁺ and [M+H-(dR+R) or M+H-(dR+MeR)]⁺. For nBCL, these produce specific product ions including [M+H-nB1]⁺, [M+H-nB2]⁺ and the modification itself (i.e., the CLB moiety, see the final panel in Figure 1). As shown in Figures S2 and S4, DPCL has the specific product ions [M+H-dR]⁺, [M+H-dN]⁺, [M+H-AA]⁺ and [AA]⁺, while RPCL produces the specific product ions [M+H-(R or MeR)]⁺, [M+H-rN]⁺, [M+H-AA]⁺ and [AA]⁺. For nBPCL, it gives the products ions of [M+H-nB]⁺, [M+H-AA]⁺ and the fragment of modification itself (the CLB moiety, see Figure S2C).

Figure 1.

Representative fragmentation spectra of seven types of NA modifications (modified 2’-dN, rN, nB, DDCL, RRCL, DRCL and nBCL) under CID (40%), HCD (30%) and wide-HCD (20, 60, 100%).

Figure 2.

Proposed fragmentation patterns for various NA modifications, including (A) 2’-dN adducts, (B) rN adducts, (C) DDCL, (D) RRCL and (E) DRCL.

The results demonstrated that most of the NA modifications containing dR or R moieties exhibited similar and consistent MS/MS fragmentation features under CID, HCD and wide-HCD. Nevertheless, it was noted that for the DDCL, RRCL and DRCL the most common features of “NL of dR” for DNA adducts or “NL of R” for RNA adducts were often absent in MS² under HCD and wide-HCD (see Figure 1). Interestingly, the informative product ions of modified 2’-dN (i.e., [dR+H]⁺) ⁴², modified nB (i.e., [nB]^+• and [nB+H-NH₃]⁺, see Figure 1) and the crosslinking products associated with AA (i.e., [AA]⁺, see Figure S2A and S2B), were mainly observed in HCD or wide-HCD. To detect more diverse types of NA modifications, we therefore performed CID and wide-HCD fragmentation MS/MS for each of the precursor ion. A complete feature list for tagging the NA adducts in FeatureHunter is provided in Table S3, and a ‘ready-to-use’ tag list of each type of NA modifications is also provided in Table S4.

3.2. . The design and application workflow of FeatureHunter for characterizing NA modifications.

FeatureHunter was developed using both the C and C# programming languages. The full pseudocode and a visualized flowchart of the pseudocode of FeatureHunter are provided in the Supporting Information (Experimental section and Figure S5). Within FeatureHunter, each specific feature of the NA modifications is defined as a “Tag”. Based on the above observations of fragmentation pattern for multiple types of NA modifications , the tags (features) within FeatureHunter are categorized into three major groups: (i) NL of a certain functional group [e.g., NL of dR (−116.047344 Da) for DNA modifications ²⁹ and NL of R (−132.042258 Da) ⁴³ or 2’-O-methylated R (−146.057908 Da) for RNA modifications], (ii) specific product ions in MS² (e.g., m/z 152.056686 as a diagnostic product ion of guanine-associated adducts ²⁵) and (iii) a fixed delta mass in MS¹ (e.g., a paired peak with a delta mass of 8.050216 Da due to eight deuterium labeling). These tags can be used alone or in combination (in intersection or union), to accurately group/identify each type of NA modifications (including the modified 2’-dN, rN, nB, DDCL, RRCL, DRCL, nBCL, DPCL, RPCL and nBPCL).

As shown in Figure 3, raw data files from LC-HR-MS/MS sample analysis were converted into mzML format using MSConvert (ProteoWizard) and imported into FeatureHunter. Initially, the specific features of NA modifications were extracted according to the NA adduct tags present. Based on the LC-HR-MS/MS data quality acquired in this study, the FeatureHunter parameters were set to mass tolerance of 5 ppm, intensity threshold of 10,000, retention time (RT) tolerance of 60 s, top 20 most intense ions in MS², minimum number of consecutive MS¹ scan of 5, an intensity ratio of 1.0 ± 0.3 in MS¹ for paired peaks. It is worth mentioning that these parameters could be customized according to the specific requirements or preferences for each feature. All the adduct ions formed artificially from mobile phase additives (e.g., [M+Na]⁺) and ESI processes were grouped with a 5 ppm mass error tolerance. Chromatography peak for each precursor ion was characterized within a m/z window of 5 ppm and a ± 1 min window of its ion selection in MS/MS analysis. Precise chromatography RTs and approximate ion abundances were ascertained using the centroid time and the height at 80% of the peak’s maximum, respectively ⁴⁴. Duplicate peaks and isotopes were removed with a mass tolerance of 5 ppm and RT tolerance of 60 s. Subsequently, all the extracted signals were grouped and annotated according to the tags used for ten different types of NA modifications. Peak alignment was achieved by using the pooled QC samples (i.e., a mixture by combining equal volume aliquots from each sample within a batch) as the first injection. The RTs and m/z values of potential signals obtained in the pooled QC samples were used to align the subsequent samples (or runs). The final results (including the information on the precursor ion m/z, RT, peak height and the tags matched) were exported as a TSV file for further analysis.

Figure 3.

Workflow of post-data analysis by FeatureHunter for NA adductomics.

3.3. Detection of authentic standards using NA adductomics with FeatureHunter.

To validate FeatureHunter, we analyzed a standard mixture of 23 2’-dN adduct, 21 rN adducts, 29 nB adducts, and 19 corresponding SIL adducts (Table S1) at three different concentrations (0.5, 5 and 500 ng/mL of each NA modification). The representative NA adductome maps obtained from a standard mixture solution of each NA modification at 500 ng/mL, using the optimized LC-Q-LIT-OT-MS method in DD-MS² scan mode followed by FeatureHunter processing, are provided in Figure S6. Table 1 summarizes the detection rates of the NA modifications spiked at different concentrations (0.5, 5 and 500 ng/mL). At the concentration of 500 ng/mL, features of 2’-dN adducts, rN adducts, and nB adducts were successfully detected, grouped, and tagged (extracted) by FeatureHunter, with a detection rate of 100%, except for pseudo-uridine (Ψ). It should be noted that Ψ failed to be tagged by FeatureHunter because it lacks a common feature of NL of ribose in CID or wide-HCD ⁴⁵. The detection rates for 2’-dN adducts, rN adducts, and nB adducts decreased as the spiked concentrations decreased, particularly for nB adducts. It is also worth noting that FeatureHunter successfully tagged the pair-peaks (e.g., d₀- vs. d₃-5-hm-dCyd, d₀- vs. d₄-N⁶-HE-dAdo and ¹⁴N₅ vs. ¹⁵N₅-8-oxo-Gua) at all three tested concentrations.

Table 1.

Detection rates of spiked NA modifications at three concentrations in deionized water using different NA adductomic analysis strategies.

Concentration of multiple types of NA modifications (number of adducts or pairs)	Numbers of the spiked NA modifications detected
	Manual inspection		DD-NL-MS3 acquisition	DD-MS2 acquisition
	MS1	features in MS2	by MS3 triggering eventa	With DFBuilder in MZmineb	With FeatureHunter
Level 1 (500 ng/mL each)
2’-dN (23)	23c (100%)d	23 (100%)	23 (100%)	23 (100%)	23 (100%)
rN (21)	21 (100%)	20 (95%)	20 (95%)	20 (95%)	20 (95%)
nB (29)	29 (100%)	29 (100%)	－e	29 (100%); 70*	29 (100%)
d0 - d3 peak pair (9)	9 (100%)	－	－	－	9 (100%)
d0 - d4 peak pair (2)	2 (100%)	－	－	－	2 (100%)
14N5 - 15N5 peak pair (4)	4 (100%)	－	－	－	4 (100%)
Level 2 (5 ng/mL each)
2’-dN (23)	23 (100%)	23 (100%)	22 (96%)	23 (100%)	23 (100%)
rN (21)	21 (100%)	20 (95%)	18 (86%)	20 (95%)	20 (95%)
nB (29)	28 (97%)	20 (69%)	－	20 (69%); 50*	20 (69%)
d0 - d3 peak pair (9)	9 (100%)	－	－	－	9 (100%)
d0 - d4 peak pair (2)	2 (100%)	－	－	－	2 (100%)
14N5 - 15N5 peak pair (4)	4 (100%)	－	－	－	4 (100%)
Level 3 (0.5 ng/mL each)
2’-dN (23)	23 (100%)	19 (83%)	18 (78%)	16 (70%)	19 (83%)
rN (21)	20 (95%)	14 (67%)	10 (48%)	13 (62%)	14 (67%)
nB (29)	18 (62%)	2 (7%)	－	2 (7%); 14*	2 (7%)
d0 - d3 peak pair (9)	9 (100%)	－	－	－	9 (100%)
d0 - d4 peak pair (2)	2 (100%)	－	－	－	2 (100%)
14N5 - 15N5 peak pair (4)	4 (100%)	－	－	－	4 (100%)

^aThe DD-NL-MS³ acquisition method used was previously reported by Chang et al. (2021)³⁷;

^bModule was developed by Murray et al. (2021)²⁵ within MZmine software;

^cThe number of spiked adducts that were successfully extracted;

^dDetection rate is calculated as: (total adducts extracted/spiked adducts) × 100;

^eNot available or not applicable;

^*The total number of extracted adducts, which included both the spiked adducts and the false positive results (derived from the in-source fragmentation of 2’-dN or rN adducts).

Using the same mixtures of standards, the performance of FeatureHunter was compared with DFBuilder module ²⁵ in MZmine (the parameters applied are detailed in Table S5) and the traditional instrument-dependent data analysis (i.e., DD-NL-MS³ acquisition method) ^{37, 41}. DD-NL-MS³ acquisition uses the appearance of an MS³ event (upon observing the criteria NLs of dR or R) to search potential DNA or RNA adducts and therefore is limited to the detection of only 2’-dN and rN adducts. As shown in Table 1, with the DFBuilder module in MZmine, the detection rates of 2’-dN adduct, rN adducts, and nB adducts are 100% as high as those achieved by FeatureHunter at 500 ng/mL. Notably for the DFBuilder module, while a total of 70 nB adducts were identified, these comprised 29 spiked nB adducts and 41 false positive identifications due to in-source fragmentation. Such false positive results were consistently observed in the low concentrations of 5 and 0.5 ng/mL. Moreover, as the concentration of the spiked NA adducts decreased to 0.5 ng/mL, the detection rates of 2’-dN adducts and rN adducts in DFBuilder module dropped below those in FeatureHunter (i.e., 70% vs. 83 % for 2’-dN adducts and 62% vs. 67% for rN adducts). Table 1 also shows that use of the traditional DD-NL-MS³ acquisition method ³⁷ generally yielded lower detection rates for 2’-dN adducts and rN adducts compared to those of the DFBuilder module and FeatureHunter. The performance of each method was evaluated further by analyzing a blank sample (deionized water without spiking standards). The results are provided in Table S6, demonstrating that the DFBuilder module and FeatureHunter did not produce any false positive results. It would appear that the performance of both software-based methods is better than that of the traditional approach using the DD-NL-MS³ acquisition method.

3.4. Detection of formaldehyde-induced NA modifications using NA adductomics with FeatureHunter.

Figure 4 shows the NA adductome maps obtained from the mixture of CT-DNA, RNA and BSA following co-treatment with ¹²C,d₀-formaldehyde and ¹³C,d₂-formaldehyde using the optimized Q-LIT-OT-MS method in DD-MS² scan mode. A total of eight types of NA modifications were detected, apart from nBCL and nBPCL. The known types of formaldehyde-induced NA modifications, such as the mono-2’-dN adducts, nB adducts, DDCL, DRCL, RPCL and DPCL ^{12, 13, 30, 46} were all successfully detected. Using FeatureHunter, we further demonstrated, for the first time, novel types of rN adducts and RRCL (Figure 4B and 4E). Among these NA modifications, some were already present in the mixture of DNA, RNA, and BSA prior to formaldehyde treatment, but we determined with high confidence that others were directly induced by formaldehyde treatment, based on the observation of ¹²C,d₀-¹³C,d₂ paired peaks (with a fixed mass difference of 3.015909, peak intensity ratio of 1.0 ± 0.3 and RT difference within 0.1 min) in MS¹ by FeatureHunter. The formaldehyde-induced NA modifications with their paired peaks are listed in Table S7. These ¹²C,d₀-¹³C,d₂ paired adducts induced by ¹²C,d₀-formaldehyde and ¹³C,d₂-formaldehyde all have the same features in MS² that were successfully extracted by FeatureHunter. A representative MS² spectra of ¹²C,d₀-¹³C,d₂ paired RRCL is shown in Supporting Information Figure S7.

Figure 4.

NA adductome maps obtained from a mixture of CT-DNA, yeast RNA and BSA following co-treatment with ¹²C,d₀- and ¹³C,d₂-formaldehyde each at 55.5 mM, at 37 °C for three days, determined by LC-Q-LIT-OT-MS with DD-MS² acquisition and FeatureHunter post-data analysis. Data are reported as accurate mass-to-charge ratios (m/z; Y axis, left) with retention times (RT; X axis) and associated peak intensities (spot color; Y axis, right), using OriginPro. (A)-(H) show eight different types of NA modifications identified. Ions labeled with numbers indicated that the ¹²C,d₀-¹³C,d₂ paired peaks (with a fixed mass difference of 3.015909, peak intensity ratio of 1.0 ± 0.3 and RT difference within 0.1 min) were detected in MS¹ by FeatureHunter. Ions labeled with both a number and lowercase letter indicated the isomers.

3.5. Detection of CLB-induced NA modifications using NA adductomics with FeatureHunter.

NA adductome maps obtained from the mixture of CT-DNA, RNA and BSA following the d₀- and d₈-CLB co-treatment using the optimized Q-LIT-OT-MS method in DD-MS² scan mode, are shown in Figure S8. Ten types of NA modifications were clearly induced by CLB. In addition to the conventional types of CLB-induced NA modifications (i.e., mono 2’-dN adducts, nB adducts and DDCL) reported previously ^{47, 48}, our present study revealed several new types of NA modifications induced by CLB, including rN adducts, RRCL, DRCL, nBCL, DPCL, RPCL and nBPCL. Some of these novel NA modifications have d₀-d₈ paired peaks (with fixed mass difference of 8.050216, peak intensity ratio of 1.0 ± 0.3 and RT difference within 0.1 min) detected in MS¹ and tagged by FeatureHunter, confirming that they were induced specifically by CLB. The CLB-induced NA modifications with their paired peaks and the tagged features by FeatureHunter, are listed in Table S8. A representative MS² spectra of d₀-d₈ paired DRCL is shown in Figure S9.

3.6. Detection of NA modifications in the urines of non-smokers and heavy smokers using NA adductomics with FeatureHunter.

The characteristics of the human volunteers are shown in Table S9. Heavy smokers had a mean urinary cotinine concentration of 1302 ng/mg creatinine and a mean total NNAL of 0.30 ng/mg creatinine. Non-smokers had non-detectable cotinine levels (i.e., < 0.03 ng/mL, the assay’s limit of detection ³⁴) and non-detectable levels of total NNAL (i.e., < 4 pg/mL, the assay’s limit of detection ³⁵). Figure 5 shows the NA adductome maps obtained from the pooled urine of twelve heavy smokers and the NA adductome maps obtained from the pooled urine of eleven non-smokers are shown in Figure S10. The peak intensities of the identified NA modifications were adjusted for urinary creatinine before being illustrated via the NA adductome maps. The NA adductome maps revealed that both smokers and non-smokers had at least ten types of NA modifications present in their urine, indicating the diversity of NA modifications present in human urine. Heavy smokers demonstrated a significantly greater number of most types of NA modifications compared to non-smokers (e.g., 62 ions vs. 35 ions for 2’-dN adducts). The identity of the NA modifications which could be closely linked to tobacco use were indicated by numbers (Figure 5), and are summarized in Table S10. The results revealed that some NA modifications were detected exclusively in smokers; example nB adducts include (4-(3-pyridyl)-4-oxobut-1-yl)-cytosine and (4-(3-pyridyl)-4-oxobut-1-yl)-adenine, which are reported to be directly associated with tobacco-specific carcinogen exposure ⁴⁹. Meanwhile, certain NA modifications (indicated by numbers with a prime symbol) were detected in both non-smokers and smokers, but smokers exhibited dramatically higher levels (2.3 - 66-fold) compared to non-smokers. These included various nB adducts (e.g., N3-methyl-adenine, N7-methyl-guanine and N7-ethyl-guanine) which have been previously identified as smoking-related adducts ⁵⁰. Tobacco-induced oxidative stress also led to significantly elevated (at least two-fold higher than in non-smokers) levels of the well-known urinary biomarkers of oxidative stress 8-oxo-7,8-dihydro-2’-deoxyguanosine, 8-oxo-7,8-dihydroguanosine and 8-oxo-7,8-dihydroguanine in smokers, together with ethenocytosine and ethenoadenine which were detected only in smokers ^{50, 51}.

Figure 5.

NA adductome maps obtained from the pooled urines of heavy smokers (n = 12), as measured by LC-Q-LIT-OT-MS with DD-MS² acquisition with FeatureHunter post-data analysis. Ions labeled with numbers were exclusively detected in smokers. Ions labeled with numbers and a prime showed at least two-fold higher levels in smokers compared to non-smokers.

4. DISCUSSION

Adductomics is a rapidly evolving field with great potential to provide valuable insights into the mechanisms of chemical toxicity and disease. By analyzing the chemical modifications or adducts that are formed on biological molecules such as proteins and DNA, researchers can gain a deeper understanding of the effects of environmental and endogenous agents, together with endogenous processes. The use of HR-MS/MS has revolutionized the field of adductomics and various novel classes of NA-associated adducts have recently discovered in mice and humans, such as DRCL ¹³, RRCL and RPCL ²⁹.

NA adductomics data analysis can be challenging due to the complexity (multiple classes of NA adducts) and large amount of data generated by LC-HR-MS/MS. In the present study, we developed FeatureHunter to apply a diagnostic fragment filtering/grouping algorithm to achieve fully automated data processing as part of the NA adductomics approach. Previously, such rule-based query approaches have been formalized and mainly used in lipidomics (e.g., molecular fragmentation query language ⁵²).

As shown in Figure 6, FeatureHunter will facilitate the proposal of a “top-down” NA adductomics approach ²⁹, by which patterns of adducts may be used to trace, and identify the originating exposure source, showing great promise as a valuable tool for unraveling the complexities of the exposome. FeatureHunter not only screens for user-defined features (i.e., NLs, product ions and pair-peaks) in MS¹ and MS² spectra in input raw data files and builds extracted ion chromatograms (EICs) for precursors of interest, but also classifies and characterizes NA adducts. FeatureHunter brings a number of breakthrough innovations for post-data processing:

Figure 6.

A proposed role for FeatureHunter in facilitating and expediting a top-down NA adductomics approach to investigate the health effects of the exposome. The exposome (e.g., exogenous and endogenous exposures) leads to a wide range of NA modifications, which can be efficiently and simultaneously identified and categorized using FeatureHunter. Subsequently, NA adductome maps are generated which enable the identification of precursor reactive species/metabolites based on the profiles of adducts formed, and hence the potential nature of the exposure source.

Firstly, FeatureHunter permits fully automated feature extraction, together with adduct annotation and classification. MZmine is an open-source software tool, which is well-established and widely used in the field of metabolomics ⁵³ and more recently, adductomics ^{37, 54}. For the discovery of DNA adducts (or RNA adducts), MZmine is often used with the traditional analysis strategy DD-NL-MS³ acquisition ^{55, 56}, which uses the appearance of an MS³ event as an indicator of a putative DNA (or RNA) adduct. Due to its reliance on the instrument’s ability to trigger MS³ data acquisition upon observing the criteria for NL of dR (or R), this approach has limited flexibility being only capable of detecting 2’-dN adducts and rN adducts (as shown in Table 1). Furthermore, it is worth noting that only MS instruments with ion trap capabilities are capable of triggering MS³ data acquisition. Additionally, under a fixed cycle time, performing MS³ data acquisition can lead to a reduction in the number of MS² spectra collected which, in turn, can decrease the overall sensitivity of the analysis. Based on our experience, conducting MS³ data acquisition resulted in a reduction of up to 40% in the number of MS² spectra (data not shown). In contrast to the traditional DD-NL-MS³ analysis strategy ^{37, 55, 56}, when using FeatureHunter the detection of DNA or RNA modifications no longer depends on the instrument’s settings and capabilities. This is because FeatureHunter can extract the specific features of NA modifications if the NA modifications possess their own distinctive characteristics in MS². Unfortunately, use of MZmine alone is unable to filter the abundant informative features present in MS² spectra, which is a drawback, especially when many substantial informative features (e.g., alternative NLs and product ions) are produced in the MS² spectra.

The DFBuilder module was built using the MZmine platform and enables automated filtering the NL and nB-associated product ions in MS² spectra of DNA adducts, using a user-specified features list ²⁵. However, the classification of putative DNA adducts must still be made by manual review. Like the DFBuilder module within MZmine, the FeatureHunter algorithm also uses the user-defined tags (characteristic features of different adduct types) to search for the presence of a diagnostic pattern in all MS² spectra, but it has more advanced capabilities because it automates the annotation and classification of different types of putative NA adducts based on the tags applied (a total of 132 tags were established in this study, as a means to classify ten types of NA modifications, see Table S3). For example, the “NL of dR (Tag 1)” and the NL of R (Tag 2) are used for grouping DNA and RNA adducts, respectively. In fact, FeatureHunter generates a “classified peak list” as the final exported result, depending on the applied tags. This classified peak list contains the precursor (m/z) value, RT, and MS² scan information (tags matched) of each potential NA adduct within the group of interest. A representative output of FeatureHunter is provided in Figure S11.

Secondly, FeatureHunter increase the specificity of the NA adductomic approach. This is mainly because the tags in FeatureHunter can be used in various combinations [i.e., in union (∪), intersection (∩) or difference (−)] to achieve a better classification of the putative NA adducts. For example, it can be challenging to classify the DNA-associated modifications (including the mono 2’-dN adducts, DDCL, DRCL and DPCL) from the complex MS² data, because all these DNA modifications have the common dominant feature of NL of dR. To accurately classify these four types of modifications, FeatureHunter relies on the use of tags in intersection and difference. When the tags are used in intersection [e.g., “NL of dR (Tag 1)”∩“NL of 2 dR (Tag 4)” for DDCL; “NL of dR (Tag 1)”∩“NL of R (Tag 2)”∩“NL of dR+R” (Tag 7)” for DRCL], only the DNA-associated crosslinked products are identified. When the tags are used in difference [e.g., “NL of dR (Tag 1)” – (“NL of 2dR (Tag 4)”∪“NL of dR+R (Tag 7)”∪“NL of dR+MeR (Tag 8)”∪“NL of dR+AA (Tags 46-49)”], only the mono 2’-dN adducts are identified. Importantly, the tags used in difference are powerful in eliminating false positive results generated during the ionization process. As shown in Table 1, the DFBuilder module of MZmine extracted both the spiked nB adducts and false positive results due to in-source thermolysis and CID during the ionization process ⁵⁷. In contrast, these in-source fragments were successfully excluded in FeatureHunter by using the tags in difference [Tags 78-92 (nB-associated product ions) - Tags 115-123 (ion-source CID induced ions), see Table S3], resulting in a satisfactory detection rate and accuracy for nB adducts. For example, to accurately extract the nB adducts associated with Gua by FeatureHunter, Tags 78-80 (Gua-associated products ions) were initially used in union and then Tags 115-117 (Gua-associated products ions from the in-source CID fragmentation) were used in difference to exclude the false positive results [i.e., (Tag 78∪Tag 79∪Tag 80) – (Tag 115∪Tag 116∪Tag 117)]. In this way the more tags that are applied, the more accurate the classification of NA adducts by FeatureHunter.

Thirdly, FeatureHunter allows for an automated detection of “pair-peaks” with a fixed delta mass in MS¹. SILMS is a powerful technique and has been frequently adopted in metabolomics ⁵⁸ and DNA adductomics ³. SILMS involves introducing a 1:1 mole ratio mixture of both unlabeled and labeled toxicants to an organism or cell, to study their metabolism, distribution, and interactions with biomolecules, such as DNA. This technique is crucial for identifying the DNA adducts induced directly by a specific chemical agent. The “pair-peak” detection in SILMS plays a vital role in the confident identification and profiling of biomolecular modifications induced by chemical agents. As show in Table 1, FeatureHunter with the tags applied (Tags 105-114, see Table S3) successfully extracted various pair-peaks (m/z) in MS¹ with a fixed delta mass (e.g., Δ mass: 3.018830 for labeling with three deuterium, 4.025107 for labeling with four deuterium, and 4.985175 for labeling with five ¹⁵N). The Adduct search module in MZmine and similar software are also capable of detecting SIL pair-peaks. However, FeatureHunter provides a more advanced solution that enables users to process and manipulate data (i.e., using ∪, ∩ and –) for multiple purposes simultaneously.

The fourth strength of FeatureHunter is the availability of customizable parameter settings for individual tags (features). Proper parameter settings for mass tolerance, signal thresholds and repeat count are critical for minimizing the risk of false positive or negative results, which can be a significant issue in the analysis of complex data such as adductomics data. FeatureHunter provides a customized parameter setting for each individual tag, which means that the user can adjust (i.e., optimize) the mass tolerance, signal thresholds, repeat count and pair-peak intensity ratio for individual tags applied. This is practical and useful because different types of HR-MS/MS have different collision mechanisms and energy, which can substantially affect the fragmentation efficiency and result in different dominant features, ultimately leading to significant variation in sensitivity and selectivity of the analysis. As illustrated in Figure 1 and Figure S2, the results demonstrate that even with the use of the same HR-MS/MS, the dominant fragments or features of NA modifications can differ depending on the collision mode. For example, for the 2’-dN (or rN) adducts, the dominant feature in CID was the NL of dR (or R), however NL of dR (or R) was often minor or even absent in HCD. The current available software commonly used for adductomics (e.g., MZmine with DFBuilder module) lacks the ability to customize parameters for individual features, as it utilizes the same parameter criteria for all features (e.g., the same mass tolerance and signal thresholds for each feature). This inability can considerably impact the sensitivity of adductomic analysis, particularly when the NA adduct concentration is low. As indicated in Table 1, at a low concentration of 0.5 ng/mL each, FeatureHunter demonstrates an equivalent satisfactory detection rate of 83% for 2’-dN adducts when compared to the performance of manual inspection utilizing features observed in MS². In contrast, the DFBuilder module achieved a detection rate of only 70%. This decreased detection rate with the DFBuilder module is because the distinctive features of 2’-dN and rN adducts (i.e., NLs of dR or R) were too low to be extracted. Therefore, the ability of FeatureHunter to customize parameter settings for individual features is particularly useful for detecting modifications that occur at low concentrations, given that these modifications exhibit distinctive features.

Additionally, FeatureHunter enables the processing of data with the universal format “mzML” and offers the capability to customize parameter settings for individual features. This allows FeatureHunter to process massive amounts of data easily and comprehensively from various mass analyzer types, including the Orbitrap, Time-of-Flight (TOF), and other hybrid MS instruments. These capabilities further enhance the utility of FeatureHunter in identifying NA modifications from complex datasets and will help facilitate the integration of datasets from different laboratories worldwide, for achieving such goals as forming a comprehensive database of NA modifications. Figure S12 provides an example of the NA adductome maps obtained from the same CT-DNA sample following CLB treatment, as measured by both the LC-Orbitrap Fusion Lumos Tribrid MS and LC-Q-TOF MS (Waters Synapt G2 QTOF) instruments with post-data processing by FeatureHunter. Interestingly, despite the use of different LC-HR-MS/MS instruments, the NA modifications identified were highly consistent.

The current version of FeatureHunter (Ver. 1.3) presents some limitations that may require improvement. Despite its usefulness, false positive results may still occur due to difficulties in accurately filtering noise. The surrounding noise levels may significantly obscure detected peaks, creating challenges in differentiating true signals from the background noise. Such challenges ultimately increase the risk of false positives, which can lead to incorrect conclusions. Nevertheless, an effective approach to decrease the rate of false positives in FeatureHunter is to apply appropriate tags. With the use of more specific and accurate tags, it is less likely that the noise will match all the tags applied, thereby enabling a more effective discrimination of true signals from background noise. Consequently, the application of additional appropriate tags may help improve the accuracy of FeatureHunter and hence reliability of the data analysis. Alternatively, users can utilize an optional function called “Peak Confirmation via MZmine (optional)” within FeatureHunter to validate the quality of MS¹ peaks, as ensured by MZmine 3. A workflow incorporating MZmine 3 is provided in Supporting Information Figure S13. This integration was achieved by intersecting both datasets. The final output includes the matched ion signal with its informative feature (from FeatureHunter), along with the qualified peak and peak area data for MS¹ (from MZmine 3).

It is important to acknowledge that the current focus of FeatureHunter is exclusively on the processing of DDA data. However, DIA has numerous potential benefits for NA adductomics research because it has the capability to perform an all-inclusive analysis of all modifications present in a sample, regardless of their abundance, by comprehensively fragmenting all analytes without precursor selection based on ion intensity criterion ²⁷. Considering this, we are currently exploring options to expand the capabilities of FeatureHunter to include the analysis of DIA data. The current version of FeatureHunter (Ver. 1.3) lacks an interface for statistical analysis and database search, which may limit its ability to provide a comprehensive analysis in NA adductomics research. However, there are several free software options available (e.g., MetaboAnalyst) for statistical analysis, and efforts are underway to develop DNA adductome databases ^{59, 60}. The current version of FeatureHunter does not include a built-in browser function for visually displaying raw LC-MS data (i.e., chromatograms and spectra). However, this can be achieved by using other data browsers in parallel, such as Xcalibur and MZmine.

In conclusion, FeatureHunter represents a significant advance in the post-data processing for NA adductomics, providing automated feature extraction, adduct annotation, and classification. A novel strength of FeatureHunter is that it increases the sensitivity and specificity of the NA adductomic approach, and supports the exploration of the exposome by facilitating the annotation and classification of different putative NA adduct types based on user-defined tags. Using FeatureHunter, novel types of NA modifications induced by genotoxins such as formaldehyde or CLB were successfully identified and classified; this was also achieved for the urine of human smokers and non-smokers, although the exact chemical structures of some of the modifications remain unknown and require further elucidation. While our results focused on detecting NA modifications using LC-Q-LIT-OT-MS, FeatureHunter’s filtering of informative features in MS² spectra and its ability to classify certain compounds based on the user-defined tags makes it versatile for use with other mass analyzers. Furthermore, numerous classes of toxicologically and pharmacologically relevant compounds are also revealed through neutral loss or fragment ion detection, making FeatureHunter a valuable asset across diverse research domains (e.g., the analysis of protein phosphorylation, glutathione conjugates and lipid species ^{52, 61, 62}). The incorporation of FeatureHunter into the NA adductomics workflow is likely to be an invaluable tool to the successful processing of the extensive and intricate datasets produced by LC-HR-MS/MS. Hence our novel software will empower researchers to gain a more profound understanding of the potential health effects of the exposome.

Synopsis:

The global software, FeatureHunter, automates nucleic acid modification detection, facilitating exploration of the exposome. FeatureHunter successfully identifies and classifies novel types of nucleic acid modifications induced by endogenous/exogenous agents.