Mass Spectrometry for Assessing Protein–Nucleic Acid Interactions

Protein–nucleic acid interactions, including the interactions of proteins with DNA, RNA, and DNA/RNA hybrid, often confer important biological consequences to targeted nucleic acids.¹ It has been extensively explored because protein–nucleic acid interactions play essential roles in regulating numerous biological processes in human cells, including DNA replication and transcription, DNA damage response, epigenetic modifications, RNA processing, and RNA translation, among others.^2,3 Aberrant protein–nucleic acid interactions have deleterious effects on human health and result in many human diseases.⁴ Thus, a comprehensive study of protein–nucleic acid interactions is vital for understanding the molecular mechanisms underlying these diseases and for providing an important knowledge basis for disease diagnosis and therapies.

A variety of analytical techniques have been developed for assessing protein–nucleic acid interactions, and these methods can be protein-centric or nucleic acid-centric.^3,5,6 The aim of protein-centric methods is to interrogate the DNA/RNA-binding sites of a specific protein at the genome-wide and transcriptome-wide scales. Protein-centric methods for mapping protein–nucleic acid interactions are usually coupled with high-throughput sequencing techniques, such as chromatin immunoprecipitation-sequencing (ChIP-seq) and cross-linking immunoprecipitation-sequencing (CLIP-seq) for studying the interactions of proteins with DNA and RNA, respectively.^7,8 On the other hand, nucleic acid-centric methods are developed to identify the binding proteins of a specific DNA, RNA, or RNA/DNA hybrid. In nucleic acid-centric methods, affinity purification (AP) is often required to pull down protein–nucleic acid complexes from cell lysates, followed by immunoassays or mass spectrometry (MS)-based analysis of interacting proteins.^3,9 Compared to immunoassays, MS-based proteomic methods exhibit much higher throughput and higher efficiency and do not require the availability of a specific antibody. Moreover, the sensitivity of MS-based methods has improved rapidly, which is attributed to continuous advances in MS instrumentation and sample preparation methods.¹⁰ Owing to these advantages, MS-based methods have been extensively employed to study protein–nucleic acid interactions since the end of the last century.¹¹

In this Review, we will provide a conceptual overview of protein–DNA, protein–RNA, and protein–RNA/DNA hybrid interactions. We will also summarize and discuss representative MS-based methods for investigating protein–nucleic acid interactions in the last three years. We will place our emphasis on MS-based proteomic methods, including in vitro pulldown methods, and in cellulo methods based on cross-linking, proximity labeling, and clustered regularly interspaced short palindromic repeats (CRISPR) targeting. Because of the widespread applications of MS in structural biology, we will also review MS-based methods for the structural characterizations of protein–nucleic acid interactions.

OVERVIEW OF PROTEIN–NUCLEIC ACID INTERACTIONS

Protein–DNA Interactions.

Protein–DNA interactions are fundamental for a wide range of cellular processes, including DNA replication, transcriptional regulation, DNA repair, etc. (Figure 1A).¹² The interactions between protein and DNA can occur through direct or indirect recognition, which are also known as “base readout” and “shape readout”, respectively.¹³ “Base readout” involves direct interactions between protein residues and nucleobases in DNA through hydrogen bonds, hydrophobic interactions, electrostatic interactions, and π–π interactions. “Shape readout” refers to local and global DNA shape properties in protein–DNA recognition, such as DNA sequence-dependent recognition.¹³ The contributions of these two readout mechanisms vary across DNA-binding protein families. Based on the current knowledge about the mechanisms of protein–DNA interactions, many computational methods have been developed to predict protein–DNA binding sites through sequence- or structure-based prediction.^14–17

Figure 1.

Protein–nucleic acid interactions and their biological functions. (A) DNA–transcription factor interactions regulate transcription. (B) RNA–splicing factor interactions modulate alternative splicing of RNA. (C) The R-loop–GADD45A interaction at TCF21 promoter recruits ten-eleven translocation (TET) DNA demethylase to demethylate the promoter DNA and activate its transcription. Created with BioRender.com.

Over the past few decades, protein–DNA interactions have been extensively studied to uncover the mechanisms of how regulatory proteins recognize specific DNA sites and their resulting biological consequences. In general, DNA-binding proteins can be divided into several types, such as histones for DNA packaging, transcription factors (TFs) for transcriptional regulation, DNA polymerases for DNA replication, nucleases for DNA cleavage, etc.^18–20 Interactions between histones and DNA are among the most crucial ones, as these interactions are assumed to regulate chromatin structures and gene expression in cells.²¹ TFs are another group of essential proteins that account for fundamental gene regulation when binding to corresponding sites in the genome to activate or repress RNA synthesis.²²

Protein–RNA Interactions.

Aside from protein–DNA interactions, protein–RNA interactions also assume essential roles in the vast majority of biological processes (Figure 1B). Physical interactions between proteins and RNA promote a variety of cellular activities, including RNA splicing, localization, stability, and post-transcriptional regulations.^3,23 Previous studies showed that the perturbation of protein–RNA interactions can result in loss of cellular homeostasis and other cellular dysfunctions that could possibly lead to diseases.^24,25 Owing to the importance of protein–RNA interactions, numerous studies have been conducted to unveil the complex network of RNA and RNA-binding proteins (RBPs).

More than 2000 RNA-binding proteins have been found to have RNA-dependent biological functions.²⁶ RBPs harbor diverse motif structures for RNA recognition, such as RNA-recognition motif, S1 domains, zinc fingers, and RNA-binding domain (RBD), among others, enabling high specificity and affinity in interactions between RBPs and RNA.²⁷ The bindings between RBPs and RNA are facilitated by various types of chemical interactions, including hydrogen bonding, van der Waals forces, and π–π stacking between aromatic amino acids in proteins and nucleobases in RNA, resulting in dynamic rearrangement of RNA and protein structures.²⁸ Recent development and advancement in experimental strategies have substantially expanded the number of RNA–RBP interactions, which will be discussed in detail in the following sections of this Review.

Protein–RNA/DNA Hybrid Interactions.

In the past decades, accumulating evidence has suggested that RNA/DNA hybrids are widely distributed across the genome of bacteria, yeast, and higher eukaryotes.^29,30 They are formed when the RNA strand hybridizes with its template DNA, displacing the non-template strand and forming a Watson–Crick RNA/DNA hybrid called R-loop.³¹ The structure of RNA/DNA is more stable compared to traditional double-stranded DNA (dsDNA) and RNA.³² R-loops play essential roles in several biological processes such as epigenome regulation, transcription modulation, and DNA repair and can elicit DNA damage in some circumstances (Figure 1C).³⁰ Dysregulation of R-loops is implicated in several genetic diseases, including neurodegenerative disorders and cancer-prone syndromes.³³ Over the past decades, advances in high-throughput sequencing methods such as DNA–RNA immunoprecipitation-sequencing (DRIP-seq) and DNA–RNA in vitro enrichment coupled to sequencing (DRIVE-seq) have facilitated the genome-wide mapping of R-loops.³⁴

Despite the breadth of functions of R-loops in fundamental cellular processes, many aspects of them remain unknown. There are potentially a significant number of proteins that act as R-loop “writers”, “readers”, and “erasers”, which regulate R-loop formation, functions, and resolution, respectively. To assess the physical interaction between R-loop-binding proteins (RLBPs) and R-loops, approaches such as immunoprecipitation with S9.6 monoclonal antibody coupled with mass spectrometry (IP-MS) and proximity labeling coupled with mass spectrometry (Prox-MS) have been implemented, which enabled the detection of hundreds of RLBP candidates.^35,36 In the future, more integrated approaches for RLBP detection could help expand our knowledge on how R-loops impact cellular homeostasis.

MASS SPECTROMETRY-BASED QUANTITATIVE PROTEOMIC METHODS FOR ASSESSING PROTEIN–NUCLEIC ACID INTERACTIONS

Prior to MS analysis, AP-based sample preparation methods are often required to isolate proteins of interest from cell lysates. Hence, a variety of AP methods coupled with MS have been developed for the investigation of protein–nucleic acid interactions. Depending on the environments where they capture these interactions, AP methods can be conducted in vitro and in cellulo.³ In vitro approaches are commonly used to examine protein–nucleic acid interactions outside the context of an intact cell. In vitro methods are relatively easier to implement and thus have been widely employed in protein–nucleic acid interaction studies. One particular advantage of in vitro methods is the ability to identify protein partners of a nucleic acid carrying a specific modification or adopting a unique secondary structure. However, many weak and transient interactions are often lost during cell lysis and in vitro purification procedures. More importantly, in vitro methods are unable to detect dynamic protein–nucleic acid interactions in living cells. To overcome the limitations of these in vitro assays, a number of in cellulo approaches have been developed, such as those based on cross-linking, proximity labeling, RNA aptamer, and more recently, CRISPR-targeting systems.

AP methods are usually used in combination with quantitative proteomics. Several quantitative proteomic strategies, including label-free and labeling techniques, are introduced to differentiate specific from nonspecific interactors. Label-free quantification (LFQ) strategies gain broad applications in quantitative proteomics due to their low cost, ease of use, and ability to compare an unlimited number of samples in parallel (Figure 2A).³⁷ However, they suffer from poor precision because any experimental variations in sample preparation and MS analysis can diminish quantification reproducibility and accuracy. Therefore, stable isotope-labeling strategies are considered the gold standard for quantitative proteomics owing to their precision and robustness. In labeling approaches, stable isotope labels can be introduced into proteins or peptides through metabolic or chemical labeling.³⁸ Stable isotope labeling by amino acids in cell culture (SILAC) is the most commonly used metabolic labeling strategy, where cells are cultured in a medium supplemented with “light” or “heavy” isotope-labeled amino acids (Figure 2B).³⁹ However, metabolic labeling methods are not amenable to tissue samples or primary cells that minimally divide. Chemical labeling is complementary to metabolic labeling. An attractive chemical labeling method is the tandem mass tag (TMT) labeling system, which is designed to enable multiplex (6- to 11-plex) quantitative proteomics for cells and tissues (Figure 2C).⁴⁰ One central aspect of the TMT strategy is that up to 11 samples with different isotopic labels can be combined and run simultaneously for high-throughput analysis to save instrument time.

Figure 2.

Workflows of common quantitative proteomic strategies. (A) Label-free quantification (LFQ). (B) Metabolic labeling: stable isotope labeling by amino acids in cell culture (SILAC). (C) Chemical labeling: tandem mass tag (TMT) labeling system. Created with BioRender.com.

Protein–DNA Interactions.

In Vitro Affinity Purification Methods for Assessing Protein–DNA Interactions.

Owing to their simplicity and unbiased nature, in vitro AP-based quantitative proteomic methods have gained widespread applications in the identification of novel cellular proteins that can bind to DNA carrying a structurally defined lesion^41,42 or adopting a unique secondary structure.^24,43,44 They usually employ a biotinylated DNA or RNA probe as “bait”, which can be immobilized on streptavidin beads to pull down prey proteins from cell lysates. After washing extensively to remove nonspecific interactors, the bound proteins are eluted from beads, digested with trypsin, and analyzed by liquid chromatography-tandem mass spectrometry (LC-MS/MS) (Figure 3A).¹

Figure 3.

Schematic diagrams showing the general procedures of in vitro affinity purification methods. (A) A biotinylated DNA or RNA probe (bait) is immobilized on streptavidin beads to pull down proteins from cell lysates. (B) An RNA of interest is tagged with an RNA aptamer and transfected into cells. After cell lysis, the aptamer-tagged RNA together with its binding proteins are isolated from cell lysates using an aptamer-specific ligand-containing bead. Created with BioRender.com.

Guanine quadruplexes (G4s), four-stranded DNA and RNA structures, exist ubiquitously in the human genome and transcriptome, respectively.^45,46 G4s play essential roles in DNA replication and transcription and RNA processing and translation as well as maintenance of genomic stability.^47–49 Wang and co-workers^43,44 employed three biotin-labeled G4 probes and their corresponding mutated probes as baits to perform affinity purification, together with a SILAC-based quantitative proteomic method, for screening of DNA G4-binding proteins. Utilizing this approach, three proteins were identified to selectively bind to all three G4 probes, and 78 proteins were identified to selectively bind to one or two G4 probes. Among these proteins, SLIRP,⁴³ YY1,⁵⁰ and VEZF1⁵¹ were shown to bind directly with G4 DNA, where YY1 and VEZF1 were found to assume important functions in G4-dependent transcriptional regulation and angiogenesis, respectively. Similarly, Yan et al.²⁴ performed a biotin-labeled rDNA G4 pulldown assay coupled with LFQ-based quantitative proteomics to identify G4-binding proteins in yeast cells and revealed that two helicases (i.e., Dbp2 and Ded1) bind selectively to G4 structures through their C-terminal RGG domain and modulate G4 destabilization. To increase pulldown efficiency and specificity, Zhang et al.⁵² designed a photoactivatable pyridostatin derivative that can bind to endogenous G4s and subsequently cross-link with proximal G4-binding proteins. After adding a biotin tag onto the G4 ligand, the G4 ligand–protein conjugates can be isolated for MS analysis. Since this small-molecule G4 ligand is cell-permeable, the approach enabled the identification of a number of putative G4-interacting proteins in living cells. A potential pitfall of this approach, however, resides in that the G4 ligand may displace G4-binding proteins from G4 sites, thereby preventing the discovery of some G4-binding proteins.

DNA damage can be induced by a variety of exogenous toxicants and/or endogenous metabolites, and the ensuing DNA lesions, if not repaired, may confer deleterious effects on human health.¹⁰ A comprehensive understanding about how cellular proteins recognize DNA lesions may provide important insights into how damaged DNA interacts with cellular proteins and offer a mechanistic basis for designing strategies for the prevention and treatment of human diseases arising from DNA damage. Du et al.⁴² synthesized a cisplatin-containing oligodeoxyribonucleotide (ODN) and a thiolated polyethylene glycol (SH-PEG)-containing complementary strand. After annealing into duplex DNA, the resulting thiolated lesion-containing dsDNA was immobilized onto gold nanoparticles and applied for pulling down interactors from light and heavy SILAC-labeled cell lysates, followed with quantitative proteomics analysis. Using this approach, they revealed a role of PC4 in DNA damage response through its selective interaction with platinated DNA. One concern about this method is that gold nanoparticles might have nonspecific interactions with cellular proteins.⁵³ A slightly different affinity pulldown method was designed by He et al.,⁴¹ in which they synthesized biotin-labeled DNA probes that can be immobilized onto streptavidin beads for affinity purification. They applied this method to identify O²- and O⁴-alkylated thymidine DNA-binding proteins and revealed that TFAM binds selectively to O⁴-alkylated thymidine DNA and promotes the transcriptional mutagenesis of these DNA lesions.

Abasic site and oxidized abasic site can form in DNA following the loss of nucleobases through hydrolysis of the N-glycosidic bond.⁵⁴ Since they are chemically reactive, (oxidized) abasic sites can form DNA–protein cross-links (DPCs) with their interacting proteins, which may inactivate DNA repair enzymes or induce histone modifications.⁵⁴ Recently, Jacinto et al.⁵⁵ developed a chemical reaction-based proteomic approach to interrogate histone post-translational modifications induced by a C₄′-oxidized abasic site (C4-AP). In particular, C4-AP can react with histone lysine residues to yield a 5-methylene-2-pyrrolone modification on lysine (K_MP). K_MP-modified histones contain an alkenyl group that can cross-link with thiol-containing biotin, which facilitates the affinity pulldown of K_MP-modified histones by neutravidin beads. Using LC-MS/MS, they uncovered 17 unique K_MP sites at 57 lysine residues throughout the four core histones isolated from HeLa cells treated with bleomycin. Moreover, Tang et al.⁵⁶ developed a MS-based method to identify cross-linking sites in abasic site-elicited DPC at single-amino acid resolution. They applied this method to examine TFAM–abasic site cross-links and revealed several lysine residues involved in TFAM–abasic site interactions.

Cross-Linking-Based Methods for Assessing Protein–DNA Interactions.

Formaldehyde cross-linking is the most widely used method to preserve DNA–protein contacts in the ChIP-seq experiment. To study chromatin-bound proteins, Ginno et al.⁵⁷ employed formaldehyde-induced DNA–protein cross-linking in cells. By utilizing the buoyant density centrifugation strategy, DNA–protein conjugates were enriched and analyzed using TMT-based quantitative proteomics. This method was applied to quantitatively analyze chromatin-associated proteins in different cell cycle phases, which provided comprehensive knowledge about chromatin changes throughout the cell cycle.

Formaldehyde is a well-established reagent for cross-linking proteins with DNA, RNA, and other proteins. Therefore, the formaldehyde-based method leaves ambiguity about whether the proteins identified interact directly with DNA or indirectly through protein–protein interactions. To address this issue, many recent studies employed UV cross-linking, which can largely reduce protein–protein cross-linking. Stützer et al.⁵⁸ reported a UV cross-linking together with LC-MS/MS method to investigate protein–DNA interactions and to identify the interaction sites. In this method, isolated nuclei were cross-linked with 254 nm UV light, followed by DNA extraction. By removing most RNAs and proteins by RNase and trypsin digestion, they could separate intact DNA–peptide conjugates from non-cross-linked peptides and RNA–peptide conjugates by size-exclusion chromatography. After DNase digestion, the cross-linked peptides were enriched and analyzed by LC-MS/MS. The MS data were analyzed by the RNA–protein cross-linking (RNP^xl) computational workflow⁵⁹ to reveal exact cross-linking sites, which facilitated the identification of the DNA binding pocket on target proteins and the differentiation of protein–DNA from protein–RNA cross-links.

To increase UV cross-linking efficiency between protein and DNA, Reim et al.⁶⁰ developed a femtosecond laser-induced cross-linking mass spectrometry (fliX-MS) pipeline to investigate protein–DNA interactions. The femtosecond UV laser can provide a high cross-linking yield while minimizing DNA damage.⁶¹ With an optimal purification protocol for cross-linked peptides, MS instrumentation, and RNP^xl computational workflow,⁵⁹ the proposed fliX-MS method was capable of mapping protein–DNA contacts at single-nucleotide and single-amino acid resolution, both in vitro and in cells.

Proximity Labeling-Based Methods for Assessing Protein–DNA Interactions.

In in vitro methods, protein–nucleic acid interactions are captured after cell lysis and thus may be different from their native environment in living cells.^5,62 To address this issue, proximity labeling (PL) followed by proteomic analysis was developed to provide a complementary method for studying protein–DNA interactions in living cells, from a protein- or DNA-centric perspective.^5,62 The basic principle of PL methods is to add a biotin tag to proteins close to the DNA or RNA of interest in living cells.⁵ To this end, many engineered enzymes, such as peroxidases, e.g., engineered ascorbate peroxidase 2 (APEX2),⁶³ and biotin ligases, e.g., a mutant biotin ligase BirA^R118G from E. coli (BioID),⁶⁴ BioID2,⁶⁵ birA* from Bacillus subtilis (BASU),⁶⁶ and TurboID,⁶⁷ are developed as PL enzymes (Figure 4). The labeling time and labeling sites on targeted proteins vary among these PL enzymes.⁶²

Figure 4.

Proximity labeling methods for the identification of nucleic acid-binding proteins. (A) Peroxidase catalyzes the biotinylation of proximal proteins using biotin phenol as a substrate, followed by pulling down biotinylated proteins for LC-MS/MS analysis. (B) Biotin ligase-based proximity labeling method using biotin as a substrate. Created with BioRender.com.

In the DNA-centric method, proteins proximal to a specific genomic locus or chromatin complex are biotinylated by PL enzymes, followed by affinity pull-down and LC-MS/MS analysis.⁵ For instance, ChromID was developed by fusing well-established chromatin readers (eCRs) to the biotin ligase BASU, in which each eCR protein can target a distinct chromatin modification, and BASU can catalyze biotinylation of proximal proteins for identifying protein interactomes of DNA methylation (i.e., CpG) and histone trimethylations (i.e., H3K27me3, H3K9me3, and H3K4me3).⁶⁸ Moreover, eCRs can be used individually or combinatorially to detect proteins associated with individual and multiple modifications, respectively.⁶⁸

Recently, the development of CRISPR-based DNA targeting systems offers excellent opportunities to target specific genomic loci.^69,70 These methods rely on catalytically inactive nucleases (e.g., dCas9 and dCas12), which can specifically bind to targeted DNAs in the presence of guide RNA (gRNA) without cleaving them.⁷¹ Therefore, many recent AP methods adopted CRISPR-targeting systems to direct PL enzymes to endogenous DNA of interest. Gao et al.⁷² reported a restricted spatial tagging (C-BERST) method through fusing APEX2 with catalytically inactive dCas9 to target genomic elements (e.g., telomeres and centromeres) with a small guide RNA (sgRNA), which allows for the biotinylation, enrichment, and MS identification of telomere- or centromere-associated proteins. The proposed C-BERST method is complementary to the BioID method⁷³ in studying telomere-associated proteins due to their different labeling kinetics and specificities.

To increase targeting specificity, Myers et al.⁷⁴ designed five sgRNAs that can be stably expressed in a single-colony 293T cell line to target genomic loci of interest individually. In this genomic locus proteomics (GLoPro) method, the cross-validation of proteomic results generated from these five cell lines can greatly improve the confidence in identifying telomere-associated proteins.⁷⁴ Moreover, this method was also applied to identify interacting proteins of nonrepetitive genomic locus (i.e., MYC gene). It is worth noting that, in the CRISPR-based method, an inducible dCas9–fusion construct is usually adopted to minimize the time that the dCas9–fusion protein occupies and its excess accumulation at the targeted genomic locus, which leads to high background biotinylation.

These PL methods, however, still suffer from low specificity, especially in assessing protein–DNA interactions, because of low copy number of DNA “bait” (two alleles per cell) in cells for protein targeting.⁷⁵ This problem also leads to low resolution of these approaches in detecting alternations arising from single nucleotide variants (SNVs) in DNA.⁷⁵ To address this issue, Mondal et al.⁷⁵ developed a proximal biotinylation by episomal recruitment (PROBER) method, which increased DNA targets per cell using high-copy episomes, and applied the method to identify DNA-binding proteins in living cells. Notably, this approach is conducive toward investigating the function of regulatory SNVs, disease-associated mutations, and novel DNA motifs, which should have widespread applications in the future.

Protein–RNA Interactions.

In Vitro Affinity Purification Methods for Assessing Protein–RNA Interactions.

Traditional in vitro AP approaches usually use a synthetic or in vitro-transcribed RNA carrying an affinity tag (e.g., biotin, desthiobiotin, digoxigenin) as “bait”, which can be immobilized onto solid supports (e.g., streptavidin beads) for convenient separation.^8,76 After incubation with cell lysates, the beads are thoroughly washed and the bound fraction is subsequently eluted for MS analysis. In vitro AP methods have been extensively employed to study protein partners of modified RNA,^77,78 RNA G4s,^79,80 and long-noncoding RNAs (lncRNAs).^81–83

RNA contains over 100 types of post-transcriptional modifications, which can influence its processing, stability, and translation.⁸⁴ To explore cellular proteins involved in N1-methyladenosine (m¹A) and N⁶-methyladenosine (m⁶A) recognition, Dai et al.⁷⁷ employed biotin-labeled m¹A- and m⁶A-harboring RNA probes to pull down m¹A- and m⁶A-binding proteins, respectively, from whole-cell lysates using a SILAC-based quantitative proteomic workflow. They revealed that several YTH domain-containing proteins can bind directly to both m¹A- and m⁶A-harboring RNAs, suggesting a potential functional similarity between m¹A and m⁶A modifications in epitranscriptomic regulations. Using a similar AP approach, they identified YTHDF2 protein as a reader protein for 5-methylcytidine (m⁵C)-modified RNA, and this interaction can regulate pre-rRNA processing in cells.⁷⁸

To study lncRNA–protein interactions, Haemmig et al.⁸¹ reported a lncRNA pulldown combined with quantitative proteomics method to identify SNHG12 lncRNA-binding proteins. They were able to uncover an interaction between SNHG12 lncRNA and DNA-PK protein, which can mediate DNA repair and vascular senescence. To explore the mechanism of NKILA lncRNA nuclear export, Khan et al.⁸² employed a biotinylated NKILA lncRNA probe to isolate proteins from nuclear extract, followed by MS analysis. They identified SRSF1 and SRSF7 proteins as NKILA lncRNA-interacting proteins, and these interactions are required for lncRNA export. Instead of using RNA substrate as “bait”, Zhang et al.⁸³ performed a CLFR circular RNA (circCLFR) pulldown assay by using its biotin-labeled sense DNA probe as “bait” and a biotin-labeled antisense DNA probe as the negative control. This approach led to a discovery that circCLFR interacts with MSH2 protein to positively modulate the chemoresistance of bladder cancer.

Another commonly used RNA pulldown approach is to tag an RNA of interest with aptamers in cells (Figure 3B),⁷⁶ such as short RNA hairpin-type aptamers (e.g., MS2,^85,86 PP7,⁸⁷ and CRISPR-Cys4⁸⁸), S1 and S1m aptamers,^89,90 and tobramycin aptamer.⁹¹ After cell lysis, the tagged RNA–protein complex can be enriched by specific ligand-containing beads for LC-MS/MS analysis. It is noteworthy that the binding affinities toward ligands vary among these aptamers, where high affinity leads to a higher enrichment efficiency but lower elution efficiency and vice versa.⁸ For instance, Zhang et al.⁸⁶ employed a tagged RNA affinity purification (TRAP) method to identify circRNA-binding proteins through co-overexpression of circRNA-MS2 RNA and GST-MS2 protein, which can form the GST-MS2-circRNA complex in cells. This complex, together with circRNA-binding proteins, can be pulled down by GST beads and subjected to MS analysis. Moreover, the RNA aptamer-based method is often integrated with the PL strategy, which will be discussed in the following sections.

Cross-Linking-Based Method for Assessing Protein–RNA Interactions.

Tedious cell lysis and RNA fragmentation steps may disrupt weak RNA–protein interactions, and postlysis reassociation may result in new ribonucleoprotein particles (RNPs).⁹² To address these issues, researchers introduced many cross-linking-based methods to maintain the integrity of RNPs during and after cell lysis, such as chemical cross-linking (e.g., formaldehyde or paraformaldehyde) and UV cross-linking with or without photoreactive nucleosides, e.g., 4-thiouridine (4-SU) and 6-thioguanosine (6-SG) (Figure 5).⁹³

Figure 5.

Workflow of cross-linking-based method for assessing protein–RNA interactions. First, protein–RNA cross-linking is induced by UV or formaldehyde in cells. After cell lysis and RNA fragmentation, RNAs of interest are hybridized with biotinylated-RNA antisense probes, and targeted RNA–protein conjugates can be purified by streptavidin (SA) beads for proteomics analysis. Created with BioRender.com.

In 2019, severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) caused coronavirus disease 2019 (COVID-19).⁹⁴ A variety of cross-linking-based quantitative proteomic methods have been developed to investigate RNA virus–host protein interactions.^95–100 Flynn et al.⁹⁷ conducted a comprehensive identification of RNA-binding proteins using mass spectrometry (ChIRP-MS) to profile cellular proteins that bind to SARS-CoV-2 RNA during active infection. In ChIRP-MS, the cross-linking was induced by formaldehyde, which can recover entire protein complexes associated with cellular RNAs. However, formaldehyde treatment can also induce extensive protein–protein cross-linking, which may result in false-positive identification of RNA–protein interactions.¹⁰¹

To cope with this issue, UV cross-linking has emerged as an alternative to chemical cross-linking in studying protein–RNA interactions.¹⁰¹ For instance, Kamel et al.⁹⁵ developed a viral RNA interactome capture (vRIC) method to identify cellular and viral proteins interacting with SARS-CoV-2 virus. The vRIC method was based on 254 nm UV light-induced RNA–protein cross-linking, followed by polyadenylated (polyA) RNA isolation from cell lysates using oligo(dT) beads. After on-beads RNase digestion, the released proteins were tryptic digested and analyzed by LC-MS/MS. By comparing RNA interactomes in mock-treated and virus-infected cells through quantitative proteomics, they identified 139 cellular proteins and 6 viral proteins that directly interact with viral RNA. They also revealed a new role of tRNA ligase complex as a key regulator of SARS-CoV-2. Moreover, Lee et al.⁹⁶ developed an RNA-antisense purification mass spectrometry (RAP-MS) method using a pool of biotinylated antisense 90-mer probes to capture cross-linked RNPs.

To capture incoming virus–host interactions, Kim et al.¹⁰² employed a 4-SU-labeled virus to infect unlabeled host cells and specifically induced 4-SU-labeled viral RNA–protein cross-linking through irradiation with 365 nm UV light. When added to the culture medium, 4-SU can be metabolically activated to the corresponding nucleoside triphosphate and incorporated into nascent RNA.⁹³ The optimal UV wavelength for inducing 4-SU-protein cross-linking is 365 nm, which can improve cross-linking efficiency between RNA and protein as well as avoid nonspecific cross-linking compared to 254 nm irradiation.⁹³ This method is named viral cross-linking and solid-phase purification (VIR-CLASP), where the protein–RNA complexes were isolated by SPRI beads in a sequence-independent manner mainly through the RNA–beads interaction. Thus, compared to vRIC⁹⁵ and RAP-MS⁹⁶ methods, VIR-CLASP can enable the investigation about sequence-independent interactions between exogenous RNA virus and host proteins.

The RAP-MS approach is also widely used for the identification of lncRNA-binding proteins in recent years.^103–105 There are some slightly different variants of traditional RAP-MS. To improve cross-linking efficiency, Ninomiya et al.¹⁰⁴ combined the RAP method with the ChIRP-MS method to identify HSATIII lncRNA-binding proteins, in which they employed paraformaldehyde-induced cross-linking before RNA antisense pulldown; Hu et al.¹⁰⁵ combined 4-SU labeling with RAP-MS to study BGL3 lncRNA-binding proteins.

Small-molecule affinity tags can also be introduced into RNA for RNP isolation. Fan et al.¹⁰⁶ introduced a biotin tag into cellular RNA through metabolic ethynyluridine (EU) incorporation and “click-chemistry”-based labeling. After UV cross-linking, biotin-labeled RNA–protein complexes were enriched by streptavidin beads and subjected to LC-MS/MS analysis.

Proximity Labeling-Based Methods for Assessing Protein–RNA Interactions.

A variety of RNA-centric PL methods have been developed for screening protein binding partners of a specific RNA.⁵ Moreover, this technique is particularly handy for capturing transient interactions and studying RNPs that are poorly soluble.⁸ Qin et al.¹⁰⁷ developed a method, by combining ascorbate peroxidase-catalyzed PL with organic-aqueous phase separation of cross-linked protein–RNA complexes (APEX-PS), to enrich subcompartment-specific RBPs. In this method, the authors conducted APEX-catalyzed biotinylation and formaldehyde-induced RBP cross-linking. The cells were then lysed; the cross-linked protein–RNA complexes were enriched by orthogonal organic phase separation, where protein–RNA complexes partition to the interface, and the biotinylated protein–RNA can be further enriched by streptavidin beads. Using this method, they generated proteomic data sets of nuclear, nucleolar, and outer mitochondrial membrane RBPs, which may be useful for future functional explorations.

PL also can be conducted in vitro with recombinant PL enzyme, like the Hypro-MS method,¹⁰⁸ which works especially for cell lines with poor transfection efficiencies, organisms, and clinical samples. Hypro-MS is a hybridization-proximity (HyPro) labeling approach for profiling proximity proteome associated with RNAs of interest in fixed cells.¹⁰⁸ In particular, a recombinant bifunctional protein, where APEX2 is fused with a DIG10.3 domain that specifically binds to digoxigenin, is used in HyPro labeling. After fixation and permeabilization, the fixed cells were hybridized with digoxigenin-labeled antisense probes against RNAs of interest. The digoxigenin tag can direct HyPro protein to targeted transcripts through the digoxigenin–DIG10.3 interaction to achieve biotinylation of the proximity proteome.

Except for digoxigenin, RNA aptamers, including BoxB and MS2, are also adopted to develop aptamer-directed PL methods, such as RNA–protein interaction detection (RadID) utilizing BoxB aptamer and BASU ligase,⁶⁶ MS2-APEX2,¹⁰⁹ and MS2-BioID¹¹⁰ methods. A recent study utilized the MS2 tagging approach to deliver MS2-coated protein-fused APEX2 (MCP-APEX2) to the MS2-tagged human telomerase RNA (MS2-hTR) site for biotinylation of proximal protein partners (Figure 6).¹⁰⁹ In this respect, a 4×MS2-hTR plasmid and an MCP-APEX2 plasmid were cotransfected into cells for in cellulo tagging and proximity labeling. The method led to the discovery of a list of candidate hTR-binding proteins, including m⁶A demethylase ALKBH5. It was also found that the selection of MS2 stem-loop fusion site on targeted RNA and the number of stem loops are important for MCP-APEX2 targeting.

Figure 6.

A schematic diagram showing APEX2-mediated proximity labeling of human telomerase RNA (hTR)-interacting proteins. (Left) APEX2 is targeted to MS2 aptamer-tagged hTR via fusion to MS2 coat protein (MCP). (Right) APEX2 is targeted to hTR via fusion to dCas13 and gRNA. Adapted with permission from Han, S.; Zhao, B. S.; Myers, S. A.; Carr, S. A.; He, C.; Ting, A. Y. RNA–protein interaction mapping via MS2- or Cas13-based APEX targeting, Proc. Natl. Acad. Sci. USA 2020, 117, 22068–22079.¹⁰⁹

PL has also been coupled with a CRISPR-based targeting technique, where a number of combinational methods have been developed, such as the CRISPR-based RNA-united interacting system (CRUIS)¹¹¹ and CRISPR-assisted RNA–protein interaction detection method (CARPID).¹¹² These methods rely on catalytically inactive nucleases (e.g., dCas13 and dCasRx), which can bind specifically to, but are unable to cleave target RNAs in the presence of gRNA.⁷¹ Therefore, gRNA in the CRISPR-based method directs the PL enzyme to a specific RNA, which is more accessible than the aptamer-directed method because it obviates the needs of introducing an exogenous aptamer-tagged RNA “bait” into cells (Figure 6A).¹⁰⁹ However, the CRISPR-based method may suffer from poor protein identification compared to the aptamer-based approach,¹⁰⁹ which could be due to the lower copy number of endogenous RNA for CRISPR targeting than exogenously introduced aptamer-tagged RNA. Nevertheless, the CRISPR-based method is more likely to reflect the cellular environment in which the RNA resides.

In the CRUIS method, the CRISPR-dCas13a-based RNA targeting system¹¹³ was combined with a pupylation-based interaction tagging (PUP-IT) labeling system¹¹⁴ to discover NORAD ncRNA-binding proteins.¹¹¹ In the CARPID method, the CRISPR-dCasRx-based RNA targeting system was combined with the BASU-based PL method to identify protein binding partners of specific lncRNAs.¹¹² Different from dCas13,¹¹³ this dCasRx protein was able to process gRNA into two or more sgRNAs without cutting targeted RNA transcripts. In particular, the authors designed a gRNA array composed of two sgRNAs to target two adjacent loci on the same transcript, which may increase the targeting specificity. This CARPID method was able to capture a list of known interacting proteins and a set of novel interactors of XIST lncRNA.

In contrast to the enzyme-based method, the specific pupylation as identity reporter (SPIDER) method utilized a small-molecule substrate as a reporter to identify interacting proteins of various types of RNAs, such as m⁶A-modified RNA and SARS-CoV-2 Omicron.¹¹⁵ The SPIDER method comprises three components, including Pup^E-fused streptavidin (SA-Pup^E), a biotinylated RNA “bait”, and PafA ligase. Among them, SA-Pup^E and biotinylated RNA can form a complex prior to PL, and PafA ligase can catalyze the formation of a covalent linkage between Pup^E and proximal proteins of RNA. Compared to enzyme-based PL methods, this method does not require a large dCas protein that may sterically interfere with protein–RNA interactions, and it is capable of identifying protein partners of specific modified RNAs.

Protein–RNA/DNA Hybrid Interactions.

R-loops are three-stranded nucleic acid structures consisting of an RNA/DNA hybrid and a displaced nontemplate DNA strand.¹¹⁶ Since R-loops play important roles in many cellular processes, several proteomic methods have been developed for the identification of R-loop-binding proteins, including in vitro affinity purification using a biotinylated RNA/DNA probe¹¹⁷ or S9.6 antibody^118,119 and an in cellulo PL-based method.³⁶ S9.6 is a commonly used R-loop antibody, which can specifically recognize RNA/DNA hybrids at a subnanomolar binding affinity.¹²⁰

Cristini et al.¹¹⁸ employed S9.6 antibody to pull down R-loops together with R-loop-associated proteins from non-cross-linked HeLa nuclear extracts. To minimize the coimmunoprecipitation (co-IP) of nonspecific interactors and to increase the IP efficiency of RNA/DNA hybrids, the authors sonicated nuclear extracts to fragment DNA prior to IP. By using IgG as a negative control, they identified a series of potential R-loop-binding proteins, including helicases, splicing factors, rRNA processing factors, and so on. They also validated DHX9 helicase as a novel R-loop-associated protein and revealed its role in preventing R-loop-associated DNA damage. Based on R-loop interactome data sets,^117,118 a recent study confirmed that DDX18 is another R-loop helicase that regulates DNA replication, DNA damage response, and genome stability through preventing R-loop accrual.¹²¹ Similarly, Li et al.¹²² conducted S9.6 co-IP followed by MS analysis in mouse pluripotent stem cells and identified Sox2, Ddx5, and Ddx9 as R-loop-associated factors in mouse cells. They demonstrated that Sox2 prevents Ddx5/Ddx9 from resolving R-loops, which maintains the balance of R-loop levels during reprogramming.

Given that R-loop encompasses hundreds of base pairs even after DNA fragmentation, chromatin-associated proteins in close proximity of R-loops can also be copurified with R-loops, which may affect the identification specificity. To cope with this issue, Wu et al.¹¹⁹ used a high-salt washing buffer to remove most chromatin-associated proteins. They conducted S9.6 co-IP in nuclear extracts with or without UV-induced RNA–protein cross-linking followed by proteomic analysis. Only proteins enriched by both methods were considered putative R-loop-binding proteins. It is worth noting that cross-linking reagents could induce R-loops, which may prevent the identification of bona fide R-loop-binding proteins.¹¹⁸

More recently, Mosler et al.³⁶ developed an RNA/DNA proximity proteomics (RDProx) method by combining RNA/DNA proximity labeling with quantitative proteomics to investigate R-loop proximal proteome in human cells. In the RDProx method, the hybrid-binding domain (HBD) of ribonuclease H1 (RNase H1) was fused with APEX2 to target RNA/DNA hybrids and to biotinylate the proximal proteins of R-loop structures in cells. RNase H1 is a conserved endonuclease that can specifically recognize RNA/DNA hybrids and hydrolyze the phosphodiester backbone of the RNA moiety.¹²³ A mutated version of HBD (HBD-WKK), which lost its binding affinity toward the RNA/DNA hybrid, was used as a negative control. Using this method, they generated a list of R-loop-binding proteins, confirmed DDX41 as a novel R-loop-binding protein, and revealed the role of DDX41 in minimizing R-loop formation and DNA double-strand break accumulation in promoters.

STRUCTURAL MASS SPECTROMETRY

Except for the aforementioned proteomic studies, MS is a powerful method for the structural characterizations of proteins and complexes, and a series of structural MS techniques have been developed, such as native mass spectrometry (nMS), hydrogen–deuterium exchange mass spectrometry (HDX-MS), cross-linking mass spectrometry (CX-MS), and covalent labeling mass spectrometry (CL-MS).¹²⁴ Structural MS is an emerging alternative technique that can complement traditional tools such as nuclear magnetic resonance (NMR) spectroscopy and X-ray crystallography in structural biology studies, especially for protein complexes that display unfavorable conformational dynamics in NMR analysis or fail to crystallize properly.^125,126

nMS, native top-down mass spectrometry (nTDMS) in particular, has been widely used to study the structure, binding affinity, and stoichiometric ratios of noncovalent protein–nucleic acid complexes, including the transcription factor–DNA complex,¹²⁷ CRISPR-associated complex,¹²⁸ G4–protein complex,¹²⁹ protein–RNA virus complex,¹³⁰ and RNase H1-RNA/DNA hybrid complex.¹³¹ For instance, Schneeberger et al.¹³⁰ employed nMS and MS/MSto study the assembly of an arginine-rich motif (ARM) peptide derived from HIV rev protein and rev response element (RRE) stem II RNA from HIV-1. nMS provided time-resolved, stoichiometric information on the RRE stem II–rev complex, while MS/MS revealed RNA-binding sites on the protein at single-nucleotide resolution. The nMS and MS/MS data uncovered a mechanism that rev protein binds to the upper stem of RRE IIB RNA and is relayed to binding sites involved in rev dimerization.¹³⁰

UV cross-linking in conjunction with MS analysis not only can be used to identify DNA- or RNA-binding proteins based on proteomic analysis but also is a valuable technique in determining DNA- or RNA-binding sites on proteins.^{58,60,132–134} One approach focused on the MS detection of cross-linked peptides after extensive DNase or RNase treatment, which showed a mass shift induced by a nucleic acid remnant in the mass spectrum.^58,60 Data analysis was performed using RNP^xl,⁵⁹ a computational software allowing for identification of exact cross-linking sites. To avoid the assignment difficulty arising from covalent cross-links on peptides, RBDmap was developed to assign nucleic acid-binding sites through neighboring non-cross-linked peptides.¹³³ In RBDmap, further digestion with a protease was introduced to achieve cleavages in every 17 amino acids. RBS-ID is another approach to addressing the challenge in identifying the cross-linked peptides.¹³⁵ In RBS-ID, RNA on the protein–RNA complex was fully cleaved into mononucleosides by hydrofluoric acid in pyridine, thereby minimizing the search space to enhance identification coverage and to achieve identification of the cross-linking site at single-amino acid resolution.

HDX-MS is another emerging technique for studying the structural and dynamic properties of proteins and complexes.¹³⁶ In a typical HDX-MS experiment, protein is diluted into a deuterated buffer, where backbone amide protons exchange with deuterons in a manner depending on solvent accessibility and exchange time, and the deuteration of peptides can be measured by MS.¹³⁷ In a protein–nucleic acid mixture, DNA or RNA bound to the targeted protein can alter the HDX rate of the protein with the surrounding deuterated solvent at the DNA- or RNA-binding sites, thereby providing information about the binding dynamics and interface of protein–nucleic acid interactions.¹³⁸ A series of studies have been conducted on utilizing HDX-MS to structurally characterize protein–nucleic acid interactions, such as the RNA–helicase interaction,¹³⁹ ssDNA–replication factor interaction,¹⁴⁰ DNA–ATPase interaction,¹⁴¹ and DNA–transcription initiation factor interaction.¹⁴²

CONCLUSIONS AND FUTURE PERSPECTIVES

In this Review, we summarized recent MS-based methods for studying protein–nucleic acid interactions and discussed the advantages and limitations of these methods. Overall, the field witnesses continuous advances in technologies and methodologies for studying protein–nucleic acid interactions, and recent studies have expanded the number of nucleic acid-centric methods, especially in cellulo methods. These advances have shed light on understanding more complex and critical protein–nucleic acid interactions in cells.

From a technological perspective, the analytical performances of mass spectrometers, including sensitivity, resolution, and scan speed, have increased rapidly and will keep on improving because of continuous innovation in instrumentation. Technological advances in instrumentation are pushing the limits of protein characterization and quantification capacities of MS techniques. In this respect, MS has been and will likely continue to be an essential technology for current and upcoming studies on protein–nucleic acid interactions. Therefore, we expect further developments of MS-based methods for the investigations of novel nucleic acid-binding proteins, the involvement of protein post-translational modifications in protein–nucleic acid recognition, and the identification of the binding interfaces between protein and nucleic acids. Given that sample preparation is critical to MS-based methods, future efforts should be made to improve sample preparation methods. For instance, peptide-level affinity purification is emerging as an attractive alternative to traditional protein-level pulldown due to its improved specificity and ability to identify protein–nucleic acid interfaces.

From a methodological perspective, in cellulo methods relying on cross-linking, proximity labeling, and CRISPR targeting provide complementary approaches to traditional in vitro affinity purification methods. Each currently available method has its own strengths and weaknesses; thus, an optimal method should be tailored to address a particular research question. For instance, in vitro methods are easy, cost-effective, and capable of studying the interactome of modified nucleic acids, but they suffer from limitations associated with in vitro assays; cross-linking-based methods can capture intrinsic protein–nucleic acid interactions in high efficiency and in a sequence-independent manner, but nonspecific cross-linking events lead to high background. The PL method can specifically biotinylate the endogenous proximal proteins of a specific nucleic acid in living cells, but it has limitations when working with poorly transfectable cell lines and its labeling efficiency can be affected by low copy number of DNA in cells.

As a newly developed technique, PL has been proven to be a powerful and irreplaceable technique for identifying nucleic acid-binding proteins in cells. Since biotin ligases (e.g., BioID and TurboID) have been successfully applied in organisms for in vivo proteomic studies, further developments of PL methods for exploiting protein–nucleic acid interactions in organisms may offer insightful information. Furthermore, the CRISPR-based targeting system is a powerful, promising, and revolutionary technique. Future efforts should be made to continually expand the applications of current CRISPR-targeting systems in identifying interactomes of specific nucleic acids as well as employ novel CRISPR-targeting systems to develop optimal MS-based proteomic methods. In terms of PL- and CRISPR-based approaches, improvements should be made to address the issues about poor transfection efficiency and low copy number of DNA or RNA targets in cells.

Biographies

Xiaomei He received her B.S. degree in Chemistry (2012) and Ph.D. degree in Analytical Chemistry (2018) from Xiamen University (China) and Wuhan University (China), respectively. She was also an exchange student at Hong Kong Baptist University in 2018. Currently, she is a postdoctoral fellow in Prof. Yinsheng Wang’s laboratory at the University of California Riverside. Her research interest involves the identification and functional characterizations of nucleic acid-binding proteins, including G-quadruplex-binding proteins, DNA damage recognition proteins, R-loop-binding proteins, etc., through mass spectrometry-based method and bioinformatic analysis.

Xingyuan Chen obtained her B.S. degree in Biology at the University of California Irvine in 2020. She worked in Prof. Aileen Anderson’s laboratory in the Department of Physical Medicine and Rehabilitation from 2018 to 2020. She is currently pursuing her Ph.D. degree in Environmental Toxicology at the University of California Riverside. Since 2020, she has been working in Prof. Yinsheng Wang’s laboratory in the Department of Chemistry with her major focus being placed on the application of mass spectrometry in studying interactomes of proteins carrying amyotrophic lateral sclerosis (ALS)-associated mutations.

Yinsheng Wang received his Ph.D. degree from Washington University in St. Louis in 2001 after earning his B.S. and M.S. degrees from Shandong University and Dalian Institute of Chemical Physics, Chinese Academy of Sciences, respectively. He is now a Distinguished Professor in Chemistry at the University of California Riverside. Yinsheng’s current research interest encompasses the investigation of the occurrence and biological consequences of DNA damage as well as the characterizations of functions of nucleic acid- and nucleotide-binding proteins. His research team employs a highly interdisciplinary approach, including mass spectrometry, synthetic organic chemistry, and molecular biology, as well as genetic and genomic tools, in their research.