Characterizations of G-Quadruplex RNA-Protein Interactions in Living Cells

Abstract

RNA guanine quadruplexes (rG4s) are noncanonical nucleic acid structures that contribute to diverse cellular functions and disease mechanisms. Defining the proteins that interact with rG4s (rG4IPs) is essential for elucidating their biological roles. Here, we build on the RNA–protein interaction detection (RaPID) platform to develop G4-RaPID, a tailored chemoproteomic strategy for the unbiased profiling of rG4IPs in living cells. Using G4-RaPID, we identified 105 candidate rG4IPs that were commonly enriched across three distinct rG4 sequences. Biochemical analyses confirmed that recombinant hnRNPA0, CHD4, and IGF2BP1 proteins directly bind rG4 structures in vitro. In addition, CLIP-seq experiments revealed significant enrichment of hnRNPA0 binding at endogenous rG4 loci. Luciferase reporter assays further demonstrated that hnRNPA0 engages the rG4 in the 5′ UTR of NRAS mRNA to negatively regulate its translation. Together, these results establish G4-RaPID as a robust approach for mapping rG4–protein interactions in living cells and document hnRNPA0–rG4 recognition as a regulatory mechanism controlling NRAS mRNA translation.

Graphical Abstract

INTRODUCTION

G4s are four-stranded nucleic acid structures formed in guanine-rich sequences.^1,2 Various G4 detection technologies, including G4 imaging and sequencing methods, e.g., rG4-seq, G4 ChIP-seq, and G4P-ChIP, revealed the prevalence of G4 structures in the human genome and transcriptome.^3,4 Putative G4-forming sequences (PQSs) are particularly enriched in gene regulatory regions, e.g., promoters, untranslated regions (UTRs), and splicing sites.^5-7 Emerging studies have revealed the roles of endogenous G4s in gene regulation and the association of aberrant G4 accumulation with pathological conditions, including cancer, immune disorders, and neurodegenerative diseases.^8-11 In addition, rG4 structures have been shown to contribute to cold adaptation in plants.¹² Despite significant progress in G4 structure mapping, the molecular mechanisms governing G4 function remain poorly understood. Identifying G4-interacting proteins (G4IPs) is key to elucidating the biological functions of rG4s and the underlying molecular mechanisms.

Existing methods for identifying G4IPs include biotinylated G4 pulldown assays with streptavidin-conjugated beads,^13,14 photoreactive pyridostatin (PDS) derivatives for cross-linking G4-binding proteins,^15,16 and motif-based identification of proteins with G4-binding domains (e.g., OB-fold-like subdomain, RGG motifs).^17,18 In addition, Lu et al.¹⁹ profiled G4IPs through proximity-mediated biotinylation by fusing an engineered biotin ligase with the G4-binding domain of DHX36. These methods have led to the identification of G4IPs with important cellular functions, including helicases, telomere-associated proteins, and epigenetic regulators.^15,16,19-23 While some of the aforementioned methods allowed for the identification of G4IPs in live cells,^15,16,19 to date, no methods have been specifically developed for comprehensive profiling of rG4IPs in live cells.

Proximity labeling (PL) has emerged as a powerful tool for studying molecular interactions in living cells.²⁴ This method utilizes promiscuous biotin ligases or peroxidases to covalently tag proteins near the target molecule with biotin, allowing for their subsequent enrichment and LC-MS/MS identification. Key advantages of these proximity labeling enzymes include high catalytic efficiency, a compact size, and customizable editing capabilities.²⁵ Recently, we developed a powerful RNA-protein interaction detection (RaPID) method, which leverages proximity-dependent protein labeling based on the BASU biotin ligase for identifying RNA-interacting proteins in live cells.²⁶ RaPID has been applied in various contexts, including assessing protein interactions with mutant RNA motifs in human genetic disorders, uncovering post-transcriptional networks in breast cancer, and identifying essential host proteins interacting with Zika virus RNA.²⁶

Here, we employed G4-RaPID and identified 105 candidate rG4IPs commonly enriched with three distinct rG4 sequences. We also validated the abilities of three of these proteins in binding directly with rG4 structures in vitro and demonstrated the enrichment of one of these proteins, i.e., hnRNPA0, at rG4 loci in cells. Moreover, we revealed the functions of the hnRNPA0-rG4 interaction in modulating the translation of NRAS mRNA.

EXPERIMENTAL SECTION

Cell Lines

HEK293T cells were purchased from ATCC. All cell lines used in this study were tested to be free of mycoplasma contamination using a LookOut Mycoplasma PCR Kit (Sigma, MP0035). Cells were maintained in DMEM (Life Technologies) supplemented with 10% FBS (Invitrogen) and 1% penicillin/streptomycin (v/v) at 37 °C in a humidified incubator with 5% CO₂. RaPID labeling was conducted according to previously published procedures with minor modifications.²⁶ Five million cells were seeded into T75 flasks the day before transfection. Plasmids for expressing motif RNA (10 μg) and 2 μg of RaPID construct were cotransfected into HEK293T cells using Mirus TransIT-X2. The day after transfection, biotin was added to the culture media until its final concentration reached 200 μM, and after a 30 min incubation, the cells were harvested.

Plasmids

Motif RNA and RaPID plasmids were constructed following previously published procedures (Addgene plasmids #107250, 107251, 107252, 107253). To generate the motif RNA plasmid, pairs of single-stranded DNA oligonucleotides encoding the motif RNA were annealed together to yield short double-stranded DNA fragments with 4 bp overhangs. These fragments were ligated into the BsmBI-digested motif RNA plasmid. Primers for constructing RNA motif plasmids are listed in Table S1.

RaPID-MS

Cells were collected and lysed in CelLytic M cell lysis reagent (Sigma) with a 1× protease inhibitor cocktail (Sigma). Biotinylated proteins were enriched using MyOne C1 streptavidin beads (Thermo Fisher) following our previously reported procedures with minor modifications.²⁷ One milligram of whole-cell lysate was incubated with 30 μL of beads in a rotator at 4 °C overnight. The beads were subsequently washed seven times with a series of buffers (1 mL each) to remove nonspecific binders: twice with RIPA lysis buffer, once with 1 M KCl, once with 0.1 M Na₂CO₃, once with 2 M urea in 10 mM Tris-HCl (pH 8.0), and twice with RIPA lysis buffer. The biotinylated proteins were eluted from the beads by boiling each sample in 30 μL of a 3× protein loading buffer supplemented with 4 mM biotin and 20 mM DTT for 10 min. The resulting mixture was centrifuged, and the supernatant was loaded onto a 15% SDS-PAGE gel, which was run at 120 V for a short duration (~10 min). Gel slices containing the proteins were excised and cut into 1 mm³ cubes.

The proteins were digested in-gel following a previously described protocol.²⁷ Briefly, excess SDS in the gel was removed by incubating with 500 μL of 100 mM NH4HCO3/CH3CN (1:1, v/v) in a thermomixer at 37 °C with interval mixing at 1,200 rpm. The supernatant was removed, and the gel pieces were dehydrated with acetonitrile for 5 min. Gel pieces were further dehydrated in a vacuum centrifuge for 1–2 min. Proteins were then reduced with 10 mM dithiothreitol (DTT, Sigma) in 25 mM NH₄HCO₃ at 55 °C for 1 h, and subsequently alkylated by incubating with 55 mM iodoacetamide (IAA, Sigma) in the dark for 1 h. Gel pieces were dehydrated with acetonitrile and washed three times with 1 mL of 25 mM NH₄HCO₃. Proteins were then digested with trypsin at 37 °C overnight. The peptides were subsequently eluted from the gel by incubating, with vigorous shaking at 37 °C for 15 min, first in 200 μL of 5% formic acid (v/v) in 25 mM NH₄HCO₃, then in 200 μL of 5% formic acid in 25 mM NH₄HCO₃ and 50% acetonitrile (v/v), and finally in 200 μL of 5% acetic acid in 25 mM NH₄HCO₃ and 70% acetonitrile (v/v). The eluted peptide fractions were pooled, evaporated to dryness, desalted using OMIX C18 Tips (Agilent), and analyzed by LC-MS/MS.

LC-MS/MS and Data Processing

LC-MS/MS experiments were performed as previously described with minor modifications.²⁸ Briefly, the affinity pull-down samples were analyzed using an EASY-nLC 1200 system coupled with a Q Exactive Plus quadrupole-Orbitrap mass spectrometer (Thermo Fisher Scientific). The HPLC separation was conducted using a trapping column followed by a separation column, both packed in-house with ReproSil-Pur C18-AQ resin (3 μm, Dr. Maisch HPLC GmbH, Germany). The peptides were separated using a 200 min linear gradient of 2–40% acetonitrile in 0.1% formic acid at a flow rate of 300 nL/min. The mass spectrometer was set up in positive-ion mode, and the spray voltage was 1.8 kV. MS/MS were recorded in a data-dependent acquisition mode, in which one full-scan MS was followed with 25 MS/MS scans.

LC-MS/MS data were processed using MaxQuant version 1.6.14 with the default parameters unless otherwise specified.²⁹ Database search was performed using the Andromeda search engine included with MaxQuant and the UniProt human sequence database (UP000005640). The mass tolerances for precursor and fragment ions were 4.5 and 20 ppm, respectively. Digestion enzyme specificity was set to trypsin, allowing a maximum of 2 missed cleavages. A minimum peptide length of 7 amino acid residues was imposed for protein identification. N-terminal acetylation and methionine oxidation were set as variable modifications, and cysteine carbamidomethylation was set as a fixed modification. “Match between runs” based on accurate m/z and retention time was enabled with a 5 min alignment time window. Label-free quantitation (LFQ) was performed using the MaxLFQ algorithm in MaxQuant.³⁰ Protein LFQ intensities were calculated from the median of pairwise intensity ratios of peptides identified in two or more samples and adjusted to the cumulative intensity across the samples. Quantification was performed using razor and unique peptides, including those modified by acetylation (protein N-terminus) and oxidation (Met). A minimum peptide ratio of 1 was imposed for protein intensity normalization.

Data were processed using Perseus version 1.6.13.0 (http://www.perseus-framework.org). Protein group LFQ intensities were log₂-transformed to reduce the effect of the outliers. For statistical comparisons between proteomes, protein groups with missing LFQ values were assigned values using imputation. Missing values were assumed to be biased toward low-abundance proteins that were below the MS detection limit, referred to as “missing not at random”, an assumption that is frequently made in proteomics studies. The missing values were replaced with random values taken from a median-downshifted Gaussian distribution to simulate low-abundance LFQ values. Imputation was performed separately for each sample from a distribution with a width of 0.3 and a downshift of 1.8. Differences in log₂(LFQ intensity) were calculated between the experimental and control groups. A two-tailed, unpaired Student’s t-test was employed to identify differentially enriched proteins. Visualization of the results was performed with volcano plots and Venn diagrams using the R libraries ggplot2 and VennDiagram. The mass spectrometry proteomics data were deposited to the ProteomeXchange Consortium via the PRIDE³¹ partner repository with the data set identifier PXD070623.

Expression and Purification of Recombinant hnRNPA0 Protein

The plasmid for expressing recombinant GST-hnRNPA0 protein was constructed by first amplifying the coding sequence of the HNRNPA0 gene from a cDNA library with primers containing BamHI and XhoI restriction recognition sites. The expression and purification ofhnRNPA0 were conducted following our previously published procedures with minor modifications.³² DE3 E. coli cells were transformed with the GST-hnRNPA0 plasmid and cultured in a 5 mL LB medium containing 100 μg/mL ampicillin at 37 °C overnight and then diluted to a 500 mL LB medium. After the optical density (OD600) reached approximately 0.6, the culture was cooled to 16 °C, and protein expression was induced with 1 mM isopropyl-β-D-1-thiogalactopyranoside (IPTG, Sigma) at 16 °C for 16 h. The cells were subsequently harvested by centrifugation and lysed by sonication in 20 mL of ice-cold PBS containing 10% (v/v) glycerol and 1 mM phenylmethylsulfonyl fluoride (PMSF, Sigma) for 15 min. The cell lysate was then centrifuged at 10000 × g for 15 min, and the supernatant was collected and filtered using a 0.45 μm syringe filter. The GST-tagged protein was purified from the supernatant by using a GSTrap column (Cytiva), following the manufacturer’s recommended procedures. Protein purity was verified by SDS-PAGE analysis (Figure S1) and stored in PBS containing 20% (v/v) glycerol at −80 °C until use.

In-Vitro Binding Assay

Fluorescently labeled RNA probes (500 nM, Integrated DNA Technologies, Table S1) were dissolved in an RNase-free buffer (10 mM Tris-HCl, pH 7.5, 100 mM KCl, and 0.1 mM EDTA) and annealed by heating the solution to 95 °C for 5 min, followed by cooling slowly to room temperature over 3 h.

Fluorescence anisotropy-based binding assay was performed following previously described procedures.³³ The aforementioned RNA probes (5 nM) were incubated, for 30 min on ice, with the indicated concentrations of recombinant proteins, including hnRNPA0, CHD4 (Active Motif), and IGF2BP1 (RayBiotech), in a 20 μL binding buffer containing 10 mM Tris-HCl (pH 7.5), 1 mM EDTA, 100 mM KCl, 0.1 mM DTT, and 10 μg/mL bovine serum albumin (BSA). Fluorescence anisotropy was subsequently recorded on a BioTek Synergy H1 Multimode Reader (Agilent Technologies, La Jolla, CA), with the excitation and emission wavelengths being 530 and 590 nm, respectively. The K_d values were calculated with GraphPad Prism 8 using nonlinear regression for curve fitting with the one-binding-site model.

CLIP-Sequencing of hnRNPA0

Plasmid for expressing HA-tagged hnRNPA0 (750 ng) was transfected into HEK293T cells in a 15 cm plate, followed by UV crosslinking, and harvested the next day. CLIP-seq on hnRNPA0 was performed according to the published easyCLIP protocol.³⁴ Cells were washed once with ice-cold PBS and cross-linked with 254 nm UVC light in a Stratalinker at 0.3 J/cm². Cross-linked cells were harvested in ~1 mL of 4 °C CLIP lysis buffer (50 mM Tris-HCl, pH 7.5, 200 mM NaCl, 1 mM EDTA, 10% glycerol, 0.1% NP40, 1% Triton X-100, 0.5% N-lauroylsarcosine) per plate. Approximately 2 mg of clarified lysate per replicate was combined with 22 μL Anti-FLAG M2 Magnetic Beads (Sigma, M8823) for 1 h, washed, treated with 0.02 U/μL Ambion RNase I (Thermo Fisher, #AM2294) at 30 °C for ~3 min, 3′-dephosphorylated with PNK and ligated overnight to L3 adapters. The L3 adapters had the sequences of /5′Phos/rNrNrNrNrNBBBBBAGATCGGAAGAGCACACGT-CAAAAAAAAAAAAAAAAAAAAAAAA/3′AzideN/, where rN is a random UMI RNA nucleotide, B is a barcode, and /3′AzideN/ is an azide group for dye conjugation and blocking 3′ ligation. L3 adapters were conjugated to DBCO-dyes before usage, and L3 barcodes were used to distinguish replicates. Following L3 adapter ligation, replicates were pooled, 5′ phosphorylated, and ligated to L5 adapters. The L5 adapters comprised the sequences of /5′AzideN/TCGGCAGCGT-CAGATGTGTATAAGAGA CAGGTATAGrNrNrNrNrNrNrN, where /5′AzideN/ is a 5′ azide group for dye conjugation and blocking ligation, B is a barcode, and rN is random RNA bases to serve as a UMI. RNA-protein complexes were resolved by SDS-PAGE, transferred to a nitrocellulose membrane, and RNA recovered by proteinase K digestion. RNA was purified by using oligo(dT) beads to capture the poly(A) tail on L3 adapters, reverse transcribed and amplifie by PCR following previously described protocol.³⁵ Libraries were sequenced on a NovaSeq X plus (paired-end, 150 bp; ~30 M reads per sample).

High-throughput sequencing reads were processed using our previously published easy-CLIP pipeline,³⁴ including trimming adapter, identifying reads uniquely mapping to the human genome (hg38), and removing PCR duplicates. Scripts used for CLIP analysis are available at github.com/dfporter/easyCLIP. For assessing peak distribution, RNA-protein binding sites were identified with CLIPper using default parameters (https://github.com/YeoLab/clipper/wiki/CLIPper-Home). The Integrative Genomics Viewer³⁶ was used to visualize the mapping results. The intersection between bed files was performed using BEDTools.³⁷ Overlapping percentage was calculated as (# of overlapped peaks)/(total # of peaks for the target protein) × 100%. The CLIP-Seq data have been deposited into the NCBI Gene Expression Omnibus (GEO) with the accession number GSE 312808.

Cell Culture and shRNA Knockdown

The sequences for shRNAs against HNRNPA0 are listed in Table S1. All shRNAs and control nontargeting shRNAs were cloned into the AgeI/EcoRI site of the pLKO.1 vector (Addgene, plasmid #10878) and confirmed by Sanger sequencing. Cells were transfected with pLKO.1/puro-shRNAs together with pLTR-G (Addgene plasmid no. 17532) envelope plasmid and pCMV-dR8.2 dvpr (Addgene plasmid no. 8455) package plasmid using PolyFect transfection reagent (QIAGEN). Viral particles were collected 48 h later and filtered through a 0.45-μm sterile filter. After that, the cells were transfected with lentiviral constructs expressing shRNA for 24 h and selected with puromycin for 7 days.

Western Blot

Protein samples were separated on a 15% SDS-PAGE gel and transferred onto a nitrocellulose membrane (Bio-Rad). After blocking with blotting-grade blocker (Bio-Rad), the membrane was incubated with PBS-T (PBS buffer with 0.05% Tween 20) containing primary antibody and 5% BSA for 2 h and then incubated with the HRP-conjugated secondary antibody in a 5% blotting-grade blocker. Following thorough washing with PBS-T, the immunoblots were detected using an ECL Western blotting detection reagent (Amersham). Primary antibodies used in this study included hnRNPA0 (Proteintech, 10848-1-AP; 1:1000), PITX1 (Proteintech, 10873-1-AP; 1:1000), NRAS (Proteintech, 10724-1-AP; 1:1000), APP (ABclonal, A17911; 1:2000), α-Tubulin Polyclonal antibody (Proteintech, 14555-1-AP; 1:5000), and GAPDH (Santa Cruz, sc-32233; 1:5000).

Real-Time Quantitative PCR (RT-qPCR)

Total RNA was extracted using an Omega Total RNA Kit I (Omega) and quantified. Reverse transcription was conducted using M-MLV reverse transcription enzyme (Promega) to obtain the cDNA library. RT-qPCR was performed using Luna Universal qPCR Master Mix (NEB) on a CFX96 RT-qPCR detection system (Bio-Rad). The primers used for RT-qPCR were previously reported³⁸ and are listed in Table S1.

Dual-Luciferase Reporter Assay

Cells stably expressing shHNRNPA0 and shControl were seeded in 12-well plates at a density of 2 × 1 × 10⁵ cells per well. After 24 h, the cells (at ~50% confluency) were cotransfected with 20 ng Renilla luciferase plasmid (pRL-CMV, Promega) and 1 μg firefly luciferase plasmid (NRAS-UTR-G4, Addgene #110490, or NRAS-UTR-ΔG4, Addgene #110493). After another 12 h, the cells were lysed in 1× passive lysis buffer and vortexed to obtain a homogeneous cell lysate. The firefly and Renilla luciferase activities of the cell lysates were measured, with a 10 s read time, using the dual-luciferase reporter assay system and a luminometer (Promega), following the manufacturer’s instructions.

RESULTS AND DISCUSSION

Identification of rG4IPs in Living Cells Using G4-RaPID

We set out to develop a proximity labeling-based chemo-proteomic method, i.e., G4-RaPID, to identify systematically rG4IPs in living cells. G4-RaPID consists of two key components, i.e., a G4 RNA element and a RaPID protein. The G4 RNA component includes BoxB stem loops flanking the rG4 motif (Figure 1A). Bacteriophage λ BoxB stem loops exhibit high-affinity binding to the λN peptide, with a dissociation constant of 200 ± 56 pM.³⁹ The RaPID protein component is a fusion of a 22-amino-acid λN peptide to the N-terminus of an engineered mutant BirA* from Bacillus subtilis, termed BASU (λN-BASU). In this system, the BoxB stem loops recruit the RaPID protein, which subsequently biotinylates proteins bound to neighboring rG4 (Figure 1A). The ensuing biotinylated rG4IPs are enriched by streptavidin affinity pull-down and analyzed by Western blotting and mass spectrometry (MS). To enable highly confident identification of general rG4IPs, we employed three distinct rG4 sequences, i.e., rG4-NRAS, rG4-APP, and rG4-C9ORF72, which are derived from the 5′ untranslated region (5′ UTR), 3′ UTR, and intron region of the NRAS, APP, and C9ORF72 genes, respectively (Figure 1B).^13,40,41

Figure 1.

G4-RaPID enables robust profiling rG4IPs in living cells. (A) A schematic diagram illustrating the G4-RaPID workflow in HEK293T cells. “BirA” and “B” represent engineered Bacillus subtilis biotin ligase and biotin, respectively. “BoxB” denotes bacteriophage λ stem loops, which exhibit high-affinity binding to the λN peptide. (B) RNA sequences used in G4-RaPID: rG4-NRAS, rG4-APP, and rG4-C9ORF72. (C–E) Volcano plot showing proteins enriched (highlighted in red) for rG4-NRAS (C), rG4-C9ORF72 (D), and rG4-APP (E), compared with control. Proteins were considered enriched when exhibiting a >2-fold signal over the control and a false discovery rate of <0.05. (F) Overlap of enriched proteins among the data sets obtained from the three rG4 probes. (G) Overlap between the enriched proteins in panels (C–E) and previously reported or predicted G4-interacting proteins.

We employed G4-RaPID to interrogate rG4IPs in living cells (Figure 1A, B). HEK293T cells were transiently transfected with plasmids for expressing the G4-RNA probe and RaPID, while a control RNA probe lacking G4 motif was introduced to label nonspecific RNA-binding proteins. Biotinylated proteins were then enriched using streptavidin beads and subjected to tryptic digestion and LC-MS/MS analysis in a data-dependent acquisition (DDA) mode for identifying the enriched proteins.

To minimize false positives, we performed rG4-RaPID experiments in three biological replicates and considered only those proteins consistently detected in all three replicates as positive hits. Proteins were deemed significantly enriched based on a threshold fold change (FC) of >2 and a false discovery rate (FDR) of <0.05. With this approach, we identified 236, 162, and 301 candidate proteins interacting with rG4-NRAS, rG4-APP, and rG4-C9ORF72, respectively, in HEK293T cells (Figure 1C-E, representative MS/MS images for several identified candidate rG4IPs are shown in Figure S2). Among these candidate rG4IPs, 105 were commonly enriched for all three rG4 probes (Figure 1F, Table S2). These high-confidence candidates were further compared to known G4-associated proteins from the literature (Figure 1F, Table S2). The proteins identified via G4-RaPID exhibited substantial overlap with existing entries of G4IPs previously reported (Figure 1F). Notably, 65% of these candidate rG4IPs were identified from previous pull-down assays, providing independent validation of our findings (Table S2). Additionally, 15 of these proteins were confirmed to be G4IPs based on prior in vitro assays (Figure 1F). Gene Ontology (GO) analysis revealed that the identified candidate rG4IPs are involved in transcription, RNA processing, and translation (Figure S3), which is consistent with the emerging roles of rG4s. In particular, G4-RaPID successfully enriched previously reported G4IPs, including DHX3,¹³ DDX5,⁵ FMR1,⁴² NCL,⁴³ and SERBP1.¹⁶

hnRNPA0, CHD4, and IGF2BP1 Can Bind Directly to rG4 Structures

Because proximity labeling-based method may also result in the identifications of proteins interacting directly with rG4 through protein–protein interactions, we next examined the abilities of several identified candidate rG4IPs in binding directly with rG4 in vitro. In particular, we measured the binding affinities of three selected candidate rG4IPs toward three previously characterized rG4 probes, one derived from the 5′ UTR of NRAS⁴⁴ and another two derived from the 3′ UTRs of APP⁴⁰ and PITX1.⁴⁵ For comparison, we also employed the corresponding mutated probes (rM4) unable to fold into the G4 structure. Fluorescence anisotropy results revealed that hnRNPA0 exhibited strong binding to all three rG4 probes, with the dissociation constants (K_d) being 4 ± 1, 3.9 ± 0.6, and 2.6 ± 0.9 nM toward NRAS, APP, and PITX1 rG4 probes, respectively (Figure 2C-D). In contrast, its binding to the rM4 probes was significantly weaker, with K_d values of 1700 ± 300, 70 ± 30, and 151 ± 40 nM, respectively (Figure 2). CHD4 did not bind PITX1 rG4 but displayed strong interactions with NRAS and APP rG4, with K_d values of 8 ± 2 and 2 ± 1 nM, respectively (Figure 2D and Figure S4). IGF2BP1, on the other hand, did not bind NRAS rG4 but exhibited a strong affinity for APP and PITX1 rG4 probes, with K_d values being 10 ± 3 and 2.7 ± 0.9 nM, respectively (Figure 2D and Figure S4). Its binding to the corresponding rM4 probes was substantially weaker, with only the PITX1 M4 probe showing moderate binding (K_d = 480 ± 50 nM, Figure 2D and Figure S4). Together, these findings establish hnRNPA0, CHD4, and IGF2BP1 as novel rG4IPs, underscoring the robustness and reliability of the G4-RaPID method in identifying previously uncharacterized rG4IPs.

Figure 2.

Fluorescence anisotropy revealed the abilities of hnRNPA0, IGF2BP1 and CHD4 in binding directly and selectively with rG4 structures in vitro. (A–C) Binding curves and dissociation constants (K_d) of full-length recombinant human hnRNPA0 with NRAS G4 and M4 (A), APP G4 and M4 (B), and PITXI G4 and M4. (D) A table summarizing the dissociation constants (K_d) for the interactions of CHD4, hnRNPA0, and IGF2BP1 with the three pairs of G4 and M4 sequences. “ND” indicates no detectable binding.

hnRNPA0 Binds to rG4 Structures in Cells

Next, we employed cross-linking immunoprecipitation (CLIP) sequencing³⁴ to examine the ability of hnRNPA0 in binding to rG4 structures in cells. hnRNPA0 is an RNA-binding protein playing a crucial role in RNA metabolism, including transcription, splicing, and translation.⁴⁶ After consolidating data acquired from three biological replicates (Figure S5), we identified 4,249 high-confidence hnRNPA0-binding loci in the transcriptome (Table S3). Notably, 9.9% of these sites overlapped with rG4 sites revealed by rG4-seq (Figure 3A, GSE77282),⁴ a technique that maps G4 structures in the human transcriptome. In contrast, only 0.1% of random sequences displayed overlap (P < 10⁻⁶). Furthermore, rG4-seq signals were highly enriched and centered at hnRNPA0-binding sites and vice versa (Figure 3B, C), indicating a direct association of hnRNPA0 with rG4 structures in cells. Integrative Genomics Viewer (IGV) plots further confirmed a high degree of colocalization between hnRNPA0-binding sites with rG4 loci (Figure 3D, E). Consistent with our proteomic data, hnRNPA0 is highly enriched at the rG4 sites located in the 3′ UTR of APP and the 5′ UTR of NRAS, where we detected strong hnRNPA0 peaks across all three replicates (Figure 3D, E). Collectively, these results establish that hnRNPA0 preferentially binds rG4 structures in cells.

Figure 3.

CLIP-seq reveals the hnRNPA0 binding landscape in human cells. (A) A comparison of the overlap rate between rG4 regions and hnRNPA0-binding sites versus randomly generated sequences of equal length (n = 1,000,000 trials). (B) Average signal of hnRNPA0 CLIP-Seq against the center of the rG4 peaks and (C) vice versa. (D, E) Integrative Genomics Viewer (IGV) plots showing colocalization of hnRNPA0 occupancy and rG4 structure loci in the 3′ UTR of APP (D) and the 5′ UTR of NRAS (E).

hnRNPA0 Modulates NRAS Translation through an rG4-Dependent Mechanism

Having established the ability of hnRNPA0 in interacting with rG4 structures in vitro and in cells, we next examined the biological functions of this interaction in cells. To this end, we first asked whether the loss of hnRNPA0 impacts the expression levels of NRAS, PITX1, and APP proteins in cells (Figures 4, S7 and S8). Our Western blot results showed that the level of NRAS protein was significantly increased upon genetic depletion of HNRNPA0 in HEK293T cells with two separate sequences of shRNAs, suggesting a role of hnRNPA0 in modulating the translation efficiency of NRAS mRNA (Figure 4A, B). However, no apparent changes in expression were detected for PITX1 or APP proteins in HNRNPA0 knockdown cells (Figure S8). These results suggest that the translation of the mRNAs of these two genes may be mainly modulated by other cellular factors (e.g., other G4IPs).

Figure 4.

hnRNPA0 depletion led to increased translation of NRAS mRNA and its dependence on rG4 in the 5′ UTR of NRAS. (A, B) Western blot analysis showing that genetic depletion of hnRNPA0 led to elevated NRAS protein levels (n = 3). (C–D) Relative firefly luciferase activity (normalized to Renilla luciferase) in control (shControl) and hnRNPA0 knockdown cells expressing NRAS-UTR or NRAS-UTR-ΔG4 constructs (n = 3). The data represent mean ± SD. The p values were determined using one-way ANOVA with Tukey’s multiple comparisons test. ns, p > 0.05; *, 0.01 < p < 0.05; **, 0.001 < p < 0.01; ****, p < 0.0001.

To further substantiate the above findings, we performed a dual-luciferase reporter assay using the 5′ UTR of NRAS mRNA with an rG4 structure (NRAS-UTR-G4) as well as its variant with a deletion of the first 29-nucleotide rG4-forming sequence of the NRAS 5′ UTR (NRAS-UTR-ΔG4, Figure 4C).⁴⁴ Compared with control cells, knockdown of HNRNPA0 significantly enhanced the luciferase activity driven by the NRAS 5′ UTR. However, when the rG4 motif was deleted (ΔG4), the differences observed between control and hnRNPA0-depleted cells were markedly diminished (Figure 4D). Notably, luciferase activity was pronouncedly higher in cells transfected with the NRAS-UTR-ΔG4 construct than those the NRAS-UTR-G4 construct, which is in keeping with the previous report that the rG4 structure in the 5′ UTR of NRAS mRNA impedes its translation.⁴⁴ These results support our hypothesis that the hnRNPA0-rG4 interaction plays a role in the translational regulation of NRAS mRNA.

Growing lines of evidence highlighted the critical roles of rG4 structures in regulating pre-mRNA processing, mRNA stability, and translation.^6,47 These regulatory functions often depend on rG4IPs, which can modulate rG4 conformation or act as scaffolds to recruit additional regulatory factors. While methods have been developed for profiling G4IPs in live cells^15,16,19 and for identifying rG4IPs through in vitro affinity pull-down,^13,47-49 no methods had been specifically developed for profiling rG4IPs in live cells. In the present study, we addressed this knowledge gap by developing a G4-RaPID method to enrich and systematically identify rG4IPs in living cells. With this approach, we identified 105 candidate rG4IPs that are commonly enriched with all three of the rG4 probes. Many of these proteins were previously identified as G4IPs through other methods, further supporting the robustness and reliability of our method (Figure 1). By analyzing the properties of G4-RaPID-identified proteins, we gained a comprehensive understanding of the biological processes associated with rG4 and its potential regulatory functions in cells. Given the strong association between G4 structures and various diseases, including Werner syndrome,⁵⁰ cancer,⁵¹ and Bloom’s syndrome,⁵² our findings provide an important knowledge base for the identification of new therapeutic targets and strategies for treating these conditions.

We further validated the abilities of three newly identified rG4IPs, i.e., CHD4, hnRNPA0, and IGF2BP1, to directly bind to rG4s in vitro. We also demonstrated the enrichment of hnRNPA0 at rG4 structure loci in living cells, suggesting that hnRNPA0 is capable of interacting with rG4 structures (Figures 2-3). Moreover, we found that hnRNPA0 suppresses the translational efficiency of NRAS mRNA, which harbors rG4 structures within the 5′ UTR (Figure 4). This notion finds further support from dual-luciferase reporter assay results, showing that the loss of the hnRNPA0 or rG4 structure led to elevated translation of luciferase mRNA containing NRAS 5′ UTR (Figure 4).

At the molecular level, proteins are known to interact with G4 structures through engaging the G-quartet grooves, loops, or external G-quartet surfaces via electrostatic, π-π, or hydrophobic interactions. For instance, a hydrophobic core formed by residues Ile65, Trp68, Tyr69, and Ala70 in the α-helical DHX36-specific motif in DHX36 produces a flat nonpolar surface that stacks on the top quartet of the bound G4 DNA.¹⁸ Nucleolin, on the other hand, exhibits little interactions with the G quartets, which are buried within the G4 core, and instead recognizes the loop and flanking regions of the globular G4 folding through hydrogen bonding, π-π stacking, and electrostatic interactions.⁵³ It will be important to determine, in the future, how CHD4, hnRNPA0, and IGF2BP1 recognize rG4 structures at the molecular level.

CONCLUSIONS

In summary, we established an innovative G4-RaPID method for detecting rG4IPs in living cells, and our work significantly expanded the repertoire of rG4IPs. Our approach is, in principle, broadly applicable to various cell types and states (e.g., under oxidative stress, upon exposure to environmental agents, and under pathological conditions), offering opportunities to uncover distinct features of rG4IPs and their biological functions. The ability to identify rG4IPs in their native cellular environment improves our understanding of rG4-mediated regulatory mechanisms and broadens the scope for future rG4-targeted studies and applications. A previous analysis of 15000 whole genome sequences suggests that variants affect G4 formation within UTRs may contribute to phenotypic variation and disease development.⁵⁴ The rG4IPs identified from the present study build a strong foundation for a better understanding about how rG4s modulate gene expression and how aberrant rG4-protein interactions may contribute to human diseases.

Supplementary Material

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.analchem.6c00902.

• Table S1: sequences of oligonucleotides are employed in this study; Table S2: summary for rG4-interacting proteins identified by G4-RaPID; Figure S1: SDS-PAGE gel image for monitoring the purification of recombinant GST-hnRNPA0 protein; Figure S2: representative MS/MS of tryptic peptides supporting the identification of CHD4, hnRNPA0, and IGF2BP1 as candidate rG4IPs; Figure S3: GO analysis for biological processes and molecular functions of the identified candidate rG4-interacting proteins; Figure S4: binding curves (the indicated KD values were obtained from fluorescence anisotropy assay) for interactions of CHD4 and IGF2BP1 proteins with rG4 and rM4; Figure S5: peak distribution analysis of hnRNPA0 obtained three replicates of CLIP-seq experiments; Figure S6: correlation analysis of the three biological replicates of CLIP-seq data; Figure S7: western blot for monitoring the knockdown efficiencies of hnRNPA0; Figure S8: western blot for monitoring the protein level of PITX1 and APP in hnRNPA0 knock-down cells (PDF)

• Table S3: hnRNPA0-binding loci identified by iCLIP-seq (XLSX)