Comprehensive identification of proteins binding to RNA G-quadruplex motifs in the 5' UTR of tumor-associated mRNAs

Abstract

G-quadruplex structures in the 5' UTR of mRNAs are widely considered to suppress translation without affecting transcription. The current study describes the comprehensive analysis of proteins binding to four different G-quadruplex motifs located in mRNAs of the cancer-related genes Bcl-2, NRAS, MMP16, and ARPC2. Following metabolic labeling (Stable Isotope Labeling with Amino acids in Cell culture, SILAC¹) of proteins in the human cell line HEK293, G-quadruplex binding proteins were enriched by pull-down assays and identified by LC-orbitrap mass spectrometry. We found different patterns of interactions for the G-quadruplex motifs under investigation. While the G-quadruplexes in the mRNAs of NRAS and MMP16 specifically interacted with a small number of proteins, the Bcl-2 and ARPC2 G-quadruplexes exhibited a broad range of proteinaceous interaction partners with 99 and 82 candidate proteins identified in at least two replicates, respectively. The use of a control composed of samples from all G-quadruplex-forming sequences and their mutated controls ensured that the identified proteins are specific for RNA G-quadruplex structures and are not general RNA-binding proteins. Independent validation experiments based on pull-down assays and Western blotting confirmed the MS data. Among the interaction partners were many proteins known to bind to RNA, including multiple heterogenous nuclear ribonucleoproteins (hnRNPs). Several of the candidate proteins are likely to reflect stalling of the ribosome by RNA G-quadruplex structures. Interestingly, additional proteins were identified that have not previously been described to interact with RNA. Gene ontology analysis of the candidate proteins revealed that many interaction partners are known to be tumor related. The majority of the identified RNA G-quadruplex interacting proteins are thought to be involved in post-transcriptional processes, particularly in splicing. These findings indicate that protein-G-quadruplex interactions are not only important for the fine-tuning of translation but are also relevant to the regulation of mRNA maturation and may play an important role in tumor biology. Proteomic data are available via ProteomeXchange with identifier PXD005761.

1. Introduction

Guanine quadruplexes (G-quadruplexes) are non-canonical nucleic acid structures formed by guanine-rich sequences. In these structures, four guanines assemble to build a coplanar G-quartet through Hoogsteen hydrogen bonding. The complete G-quadruplex structure is formed by stacking of two or more G-quartets that are linked together by three loops [1]. G-quadruplex motifs are stabilized by monovalent cations, of which potassium confers the highest stability [2]. Computational analysis predicted the existence of more than 300,000 putative G-quadruplex forming sequences in the human genome [3, 4]. A more recent high-throughput sequencing approach even identified more than 700,000 putative G-quadruplex structures, approximately 450,000 of which were not predicted by the previously performed in silico analyses [5]. Furthermore, G-quadruplexes have been reported to reveal a higher prevalence in transcribed regions [6] including untranslated regions (UTRs) [7].

While DNA G-quadruplexes have been investigated for decades, such structures formed by RNA came into the focus of research with the finding that G-quadruplexes in the 5' UTR of mRNAs repress translation in vitro [8]. In subsequent experiments, this inhibitory effect was also observed in vivo [9, 10]. In the meantime, repression of translation has been confirmed for many G-quadruplex motifs located in the 5' UTR region of mRNAs, including those under investigation in the present study, namely the G-quadruplexes of the matrix metalloproteinase-16 (MMP16), also known as MT3-MMP or MT-MMP3 [11], the actin-related protein 2/3 complex subunit 2 (ARPC2) [12], the neuroblastoma RAS viral oncogene homolog (NRAS) [8], and the apoptosis inhibitor B-cell CLL/lymphoma 2 (Bcl-2) [13]. While RNA G-quadruplexes in the 5' UTR of mRNAs have usually been reported to repress translation, some examples demonstrate augmentation of translation by G-quadruplexes, e.g. in the 5' UTR of the Transforming Growth Factor β2 [14] or in the internal ribosome entry site of the human vascular endothelial growth factor (VEGF) mRNA [15]. In addition, an RNA G-quadruplex in the 3' UTR of the proto-oncogene PIM1 was reported to suppress translation [16].

The four above mentioned factors addressed in the present study, NRAS, Bcl-2, MMP16, and ARPC2 are well-known to have important functions in cancer biology. The NRAS gene encodes a membrane-bound GTPase that is involved in signal transduction pathways controlling cell proliferation and differentiation as well as oncogenesis [17–19]. More recently, four non-canonical NRAS isoforms were identified to regulate differentially downstream pathways, cell growth, and cell transformation [20]. The Bcl-2 family regulates cell survival and programmed cell death in malignant diseases, with the Bcl-2 protein being one of the best studied oncogenic factors [21, 22]. Cell-cell adhesion, cell migration and metastasis are important factors in malignant transformation. The MMP16 produces four transcripts, two of which are protein-coding isoforms encoding the endopeptidase MMP16. This endopeptidase is relevant for important steps in cancer progression, such as migration and invasion [23]. ARPC2 is one of the seven subunits of the human Arp2/3 complex which is involved in actin polymerization, nucleation and filament branching [24, 25]. Silencing of Arp2/3 subunits, including ARPC2, was shown to cause reduced cell migration and invasion [26].

Although G-quadruplexes have been intensively studied in vitro, their presence in vivo remained elusive for many years. In 2014, the formation of RNA G-quadruplex structures could be confirmed in vivo using structure-specific antibodies [27]. More recently, the NRAS RNA G-quadruplex structures could be visualized in cells with an engineered fluorogenic hybridization probe [28]. In addition, the stability of G-quadruplexes has successfully been modulated by small molecule compounds. For example, the cationic porphyrin TmPyP4 was demonstrated to unfold the G-quadruplex in the MMP16 mRNA and thereby alleviate its repressive effect on translation in eukaryotic cells [29]. Consequently, G-quadruplex motives have been considered as potential novel therapeutic targets for diseases such as cancer [30]. However, the presence of RNA G-quadruplex structures in eukaryotic cells was recently questioned by intracellular structural probing which found that the G-quadruplex motifs are globally unfolded in eukaryotic cells [31]. Still, the authors allowed that G-quadruplex structures might transiently fold and remain undetected by their steady-state measurements. In addition, they suggested that interactions of the G-quadruplex regions with proteins might be functionally relevant.

Although RNA G-quadruplexes are now widely accepted to play an important role in translational regulation, mRNA processing and maintenance of chromosomal end integrity [32], the mode of action of these unusual nucleic acid secondary structure elements still remains unclear. To approach an understanding of the intracellular processes associated with G-quadruplexes, we recently identified proteins binding to the G-quadruplexes in the MMP16 and ARPC2 mRNAs by a focused approach based on pull-down assays and MALDI-ToF mass spectrometry [12]. We found heterogenous nuclear ribonucleoproteins (hnRNPs), ribosomal proteins, splicing factors, and several proteins that have not previously been described to interact with nucleic acids. Surface plasmon resonance measurements revealed dissociation constants in the low nanomolar range typical for strong biological interactions. In another study, the fragile X mental retardation protein (FMRP) was found to bind to a G-quadruplex in the coding region of its own mRNA and thereby control its own translation by a negative feedback loop [33]. In addition, the FMRP may interact with G-quadruplexes in other mRNAs to repress their translation by recruiting the translation repressor, inducing the microRNA pathway and interacting directly with the translating ribosome [34]. The FRAXE-associated mental retardation protein (FMR2) was reported to interact with G-quadruplex structures in mRNAs and to be involved in alternative splicing [35, 36]. Among the proteins interacting with the G-rich sequence of the telomeric repeat-containing RNA (TERRA) were multiple members of the hnRNP family [37]. A comprehensive review revealed that due to the structural similarities between DNA and RNA G-quadruplexes, their binding proteins overlap significantly [38]. Another important transient interaction partner is the RNA helicase associated with AU-rich element (RHAU) that was reported to resolve RNA G-quadruplex structures [39, 40].

The present study describes the global analysis of cellular proteins binding to G-quadruplexes located in the 5' UTRs of four different mRNAs. To obtain a more comprehensive picture of the proteinaceous interaction partners of the G-quadruplexes than in our previous study, we carried out stable isotope labeling with amino acids in cell culture (SILAC), followed by pull-down assays and LC-orbitrap mass spectrometry. Furthermore, we included two additional G-quadruplex motifs in genes of utmost importance as oncogenic factors, Bcl-2 and NRAS, in addition to the G-quadruplexes in the UTRs of the MMP16 and ARPC2 mRNAs. Although the four G-quadruplex motifs are structurally diverse, as they differ in the lengths of the loops connecting the G-tracts, they all repress translation. The use of mutated control sequences ensured exclusion of general RNA-binding proteins and enrichment of proteins that interact with RNA G-quadruplexes. Interestingly, we observed two different patterns of protein binding for the four G-quadruplexes. While the G-quadruplex motifs with shorter loops and shorter guanine stretches (in NRAS and MMP16) have a comparatively specific binding pattern, those with longer loops and longer guanine stretches (in Bcl-2 and ARPC2) bind a broader range of proteins. Candidate proteins binding selectively to the RNA G-quadruplex motifs of NRAS and MMP16 are mainly classified as involved in post-transcriptional events, translation and cancer biology. For Bcl-2 and ARPC2, a substantially larger number of proteins were captured, of which the strongest candidate proteins were splicing-related. According to these results, the interactions between the candidate proteins and their respective RNA G-quadruplexes can be assumed to play crucial roles in post-transcriptional processing and translational control. The identified proteins can be considered as targets to modulate the function of the RNA G-quadruplex structures.

2. Experimental Procedures

2.1. Cell culture

Human Embryonic Kidney 293 cells (HEK 293 cells) were cultured at 37°C in a humidified atmosphere containing 5% CO₂ in low glucose Dulbecco’s modified Eagle’s medium (PAA Laboratories GmbH, Pasching, Austria) containing 10% fetal bovine serum, 2 mM glutamine, MEM non-essential amino acids and the antibiotics penicillin and streptomycin. The culture medium was additionally supplemented with 4.5 mg/ml glucose.

2.2. RNA oligonucleotides

All the RNA oligonucleotides used in the present study were purchased from Purimex GmbH (Grebenstein, Germany). Mutated sequences (mt) with substitutions in the consecutive guanine stretches (underlined) served to allow differentiation between specific interactions of proteins and RNA G-quadruplex motifs and unspecific RNA-binding capacity of proteins. The following, PAGE purified RNA oligonucleotides carrying biotin at their 3' ends were used:

Bcl-2 GQwt	5' GGGGGCCGUGGGGUGGGAGCUGGGG-biotin 3'
Bcl-2 GQmt	5' GGAGGCCGUGAAGUGAGAGCUGAAG-biotin 3'
NRAS GQwt	5' GGGAGGGGCGGGUCUGGGUG-biotin 3'
NRAS GQmt	5' GAGAGGAGCGAGUCUGAGUG-biotin 3'
MMP16 GQwt	5' GGGAGGGAGGGAGAGGG-biotin 3'
MMP16 GQmt	5' GAGAGAGAGAGAGAGAG-biotin 3'
ARPC2 GQwt	5' GGGGGCUGGGCGGGGACCGGG-biotin 3'
ARPC2 GQmt	5' GUAGACUGAGCGAAGACCGAG-biotin 3'

2.3. Circular Dichroism and UV-melting analysis

To confirm the RNA G-quadruplex formation, spectroscopic studies were performed. For Circular Dichroism (CD) analysis, 5 μM RNA samples were prepared in a buffer containing 10 mM Tris-HCl, pH 7.5 and 100 mM KCl. Formation of secondary structures was achieved by heating the RNA to 95°C for 5 min in a thermocycler and cooling it down to 4°C in 2°C per minute steps prior to the CD experiment. CD spectra of the folded RNA samples were monitored at 20°C in 1 mm quartz cuvettes from 220 to 320 nm with four accumulations using JASCO J-815 (JASCO, Gross-Umstadt, Germany) with a Peltier temperature controller and their average was calculated.

UV melting analysis was carried out with RNA samples at a final concentration of 1 μM in 10 mM Tris-HCl pH 7.5 either 1 mM KCl, which strongly stabilizes G-quadruplex structures, or LiCl, which does not support the formation of G-quadruplex structures [41]. Experiments were performed with a JASCO V-650 UV-visible spectrophotometer with a Peltier temperature controller. Samples were heated to 95°C and cooled down to 20°C with a temperature gradient of 0.2°C min^-1. Absorption data were recorded at 295 nm and were collected every 0.5 min.

2.4. Dual luciferase reporter assays

Dual luciferase reporter assays were carried out with the psiCHECK-2 vector (Promega, Wisconsin, USA). The G-quadruplex sequences and their respective mutated controls were inserted upstream of the coding region of the Renilla luciferase. Pairs of oligonucleotides encoding the G-quadruplexes were annealed and cloned into the NheI restriction site of the vector. The following oligonucleotides were used:

Bcl-2 GQwt	Sense	5' CTAGCGGGGGCCGTGGGGTGGGAGCTGGGGG 3'
	Antisense	5' CTAGCCCCCAGCTCCCACCCCACGGCCCCCG 3'
Bcl-2 GQmt	Sense	5' CTAGCGGAGGCCGTGAAGTGAGAGCTGAAGG 3'
	Antisense	5' CTAGCCTTCAGCTCTCACTTCACGGCCTCCG 3'
NRAS GQwt	Sense	5' CTAGCGGGAGGGGCGGGTCTGGGTGG 3'
	Antisense	5' CTAGCCACCCAGACCCGCCCCTCCCG 3'
NRAS GQmt	Sense	5' CTAGCGAGAGGAGCGAGTCTGAGTGG 3'
	Antisense	5' CTAGCCACTCAGACTCGCTCCTCTCG 3'

For the dual luciferase reporter assay, subconfluent HEK293 cells grown in 24-well plates were transfected with 0.8 μg of the respective psiCHECK-2 reporter plasmid using Lipofectamine 2000 (Life Technologies GmbH, Darmstadt, Germany) according to the manufacturer’s instructions. Activities of firefly and Renilla luciferase were measured 24 h after transfection using the Dual-Luciferase Reporter Assay Kit (Promega) on a TriStar² Multimode Reader LB 942 luminometer (Berthold Technologie GmbH & Co. KG, Bad Wildbad, Germany). The ratio of Renilla to firefly luciferase activity (R/FF) was normalized to the value of the empty psiCHECK-2 vector. The data represent mean values from three independent measurements, each performed in triplicate.

2.5. Quantitative RT PCR assays

For quantitative reverse transcription PCR (qRT-PCR) assays, total RNA was isolated from HEK293 cells 24 hours after transfection using the RNeasy Mini Kit (Qiagen, Hilden, Germany). The isolated RNA was treated with RQ1 RNase-Free DNase (Promega) according to the manufacturer’s instructions. After reverse transcription using the RevertAid H Minus First Strand cDNA Synthesis Kit (Thermo Fisher Scientific, Waltham, MA, USA), relative transcript levels of the two luciferases were quantified by real-time PCR using the Maxima SYBR Green qPCR Master Mix (2X) (Thermo Fisher Scientific). Experiments were performed with a CFX96™ Real-Time System on a C1000™ Thermal Cycler (Bio-Rad Laboratories, Hercules, CA, USA) using the following primer pairs:

Renilla luciferase	forward	5' GCTGGACTCCTTCATCAACTACTA 3'
	reverse	5' GACTTACCCATTCCGATCAGATCA 3'
firefly luciferase	forward	5' CTATTTTCGGCAACCAGATCATCC 3'
	reverse	5' ACTGAATTTTGTAGTCTTGCAGGC 3'

The following amplification conditions were used: 39 cycles with denaturation at 95°C for 15 s, annealing at 60°C for 30 s, elongation at 72°C for 30 s. At the end of the amplification, a melting curve was recorded to confirm the formation of a single fragment. All samples were analyzed as three independent replicates, each measured in triplicate. Data analysis was performed with the Bio-Rad CFX Manager version 3.1. The mRNA level of the each luciferase was calculated using the following formula [42] which is based on the corresponding real-time PCR efficiency E (E=10^(-1/slope)) of one cycle in the exponential phase and the Cq deviation of the target (Renilla luciferase) versus the reference (firefly luciferase).

$$\text{ratio}=\frac{{\left(E_{\text{target}}\right)}^{{\text{ΔCq}}_{\text{target}}\left(\text{control-sample}\right)}}{{\left(E_{\text{ref}}\right)}^{{\text{ΔC}}_{{\text{q}}_{\text{ref}}\left(\text{control-sample}\right)}}}$$

The calculated values were normalized to the value of the empty psiCHECK-2.

2.6. SILAC and pull-down assay

To identify proteins that specifically bind to one of the RNA oligonucleotides (Bcl-2 GQwt, NRAS GQwt, MMP16 GQwt, ARPC2 GQwt and their respective mutants), three independent biological replicate experiments, including one label swapped experiment, were compared using SILAC, pull-down assays and LC-orbitrap mass spectrometry. This sample size will produce reliable datasets.

For SILAC, HEK 293 cells were cultured in SILAC DMEM high glucose (4.5 g/l) (PAA Laboratories GmbH) containing 10% dialyzed fetal bovine serum, 2 mM glutamine, MEM non-essential amino acids and the antibiotics penicillin and streptomycin. One population of HEK293 was grown in SILAC medium containing 0.2 mM L-Arginine-monohydrochloride (Carl Roth, GmbH & Co. KG, Karlsruhe, Germany) and 0.79 mM L-Lysine hydrochloride (Carl Roth). A second population was grown in SILAC medium containing 0.2 mM L-Argine HCl U-13C6 (EURISO-TOP GmbH, Saarbruecken, Germany) and 0.79 mM L-Lysine 2HCl U-13C6 U-15N2 (EURISO-TOP GmbH). HEK 293 cells were grown in the respective SILAC medium up to the seventh passage for complete labeling the amino acid isotopes. Cells were subsequently harvested and precipitated by centrifugation. The cell pellet was washed with 30 ml of cold PBS and resuspended in 1 ml of 10 mM Tris-Hcl, pH 7.5, 1.5 mM MgCl₂, 0.5 mM DTT and 10 mM KCl. After incubation for 15 min on ice and centrifugation, the cell pellet was resuspended in a mixture of 2 ml of the above mentioned buffer and 1 ml of a buffer containing 300 mM Tris-HCl, pH 7.5, 3 mM MgCl₂ and 1.4 M KCl.

Cell lysates containing the protease inhibitor cocktail cOmplete Mini (Roche Diagnostics, Rotkreuz, Switzerland) was prepared using an Ultrasonics cell disruptor connected to a Sonifier 250 (Branson, Danbury, CT, USA). The protein extract was subsequently dialyzed against 1 l of cold 20 mM Tris-HCl, pH 7.5, 20% (v/v) glycerin, 0.2 mM Na-EDTA, 0.5 mM DTT and 100 mM KCl under stirring overnight at 4°C. Concentrations of the extracted cellular proteins were determined with the Pierce BCA Protein Assay Kit (Pierce/Thermo Fisher Scientific).

For G-quadruplex formation, 1.4 nmol of each RNA oligonucleotide dissolved in G-quadruplex folding buffer containing 10 mM Tris-HCl, pH 7.5, 0.1 mM Na-EDTA and 100 mM KCl, was heated to 95°C for 5 min and cooled down to 4°C at a rate of 2°C per minute in a thermocycler. All buffers for the pull-down assays were prepared with RNase-free H₂O.

To capture the interaction partners of each of the G-quadruplex motifs, 100 μl magnetic beads (Dynabeads MyOne™ Streptavidin T1, Thermo Fisher Scientific, Massachusetts, USA) were initially prepared according to the manufacturer’s instructions. The folded RNA oligonucleotides were immobilized to the magnetic beads according to the manufacturer’s instruction using binding and washing buffer containing 10 mM Tris-HCl, pH 7.5, 0.1 mM Na-EDTA, 950 mM NaCl and 50 mM KCl. The RNA-magnetic beads were dissolved in 200 μl pull-down wash buffer containing 10 mM Tris-HCl, pH 7.5, 1 mM DTT, 0.1 mM Na-EDTA, 0.01% (v/v) Triton X-100 and 400 mM KCl. 2 mg protein extract treated with RNase inhibitor (RiboLock RNase Inhibitor (40 U/μl), Thermo Fisher Scientific,) was added to the RNA-magnetic beads. They were subsequently incubated under stirring for 1 h at 30°C. Subsequently, the beads were washed by a gradient of the pull-down wash buffer containing increasing concentrations of KCl (400 mM and 600 mM), and the proteins were finally eluted with the pull-down wash buffer containing 2,600 mM KCl. Amicon Ultra-0.5 ml Centrifugal Filters (Ultracel®-3K, Merck Chemicals GmbH, Darmstadt, Germany) were used to concentrate the proteins and to remove KCl. All eight eluted protein extracts labeled with light amino acids (L samples) were pooled and used as the standard. Equal amounts of the each of the protein extracts labeled with heavy amino acids (H sample) and the standard were mixed to identify proteins binding to the respective RNA G-quadruplex motifs. The protein samples were subsequently run on a polyacrylamide gel for a short time to concentrate the proteins and stained with Coomassie Blue.

2.7. LC-orbitrap mass spectrometry and data processing

Coomassie stained gel bands were excised and the proteins were digested with trypsin, as previously described [43]. In brief, proteins were reduced in 10 mM DTT (Sigma Aldrich, St. Louis, MO, USA) for 30 min at 37°C and alkylated in 55 mM iodoacetamide (Sigma Aldrich) for 20 min at ambient temperature in the dark. They were then digested overnight at 37°C with 12.5 ng/μl trypsin (Pierce/Thermo Fisher Scientific). Following digestion, samples were diluted with equal volume of 0.1% TFA and spun onto StageTips as previously described [44]. Peptides were eluted in 20 μl of 80% acetonitrile in 0.1% TFA and concentrated down to 4 μl by vacuum centrifugation (Concentrator 5301, Eppendorf, Hamburg, Germany). The peptide sample was then prepared for LC-MS/MS analysis by diluting it to 5 μl by 0.1% TFA. MS-analyses were performed on a QExactive mass spectrometer (Thermo Fisher Scientific), coupled on-line to Ultimate 3000 RSLCnano Systems (Thermo Fisher Scientific). The analytical column with a self-assembled particle frit [45] and C18 material (ReproSil-Pur C18-AQ 3 μm; Dr. Maisch GmbH, Ammerbuch-Entringen, Germany) was packed into a spray emitter (75-μm ID, 8-μm opening, 300-mm length; New Objective) using an air-pressure pump (Proxeon Biosystems). Mobile phase A consisted of water and 0.1% formic acid; mobile phase B consisted of 80% acetonitrile and 0.1% formic acid. The gradient used was 220 min. The peptides were loaded onto the column at a flow rate of 0.5 μl min^-1 and eluted at a flow rate of 0.2 μl min^-1 according to the following gradient: 2 to 40% buffer B in 180 min, then to 95% in 16 min. FTMS spectra were recorded at 70,000 resolution and the top 10 most abundant peaks with charge ≥ 2 and isolation window of 2.0 Thomson were selected and fragmented by higher-energy collisional dissociation [46] with normalized collision energy of 25. The maximum ion injection time for the MS and MS2 scans was set to 20 and 60 ms respectively and the AGC target was set to 1 E6 for the MS scan and to 5 E4 for the MS2 scan. Dynamic exclusion was set to 60 s. All biological replicates were subjected to the same experimental conditions.

The MaxQuant software platform [47] version 1.5.2.8 was used to process the raw files and a search was conducted against Human complete/reference proteome set of UniProt database (released on 12/05/2014), using the Andromeda search engine [48]. The first search peptide tolerance was set to 20 ppm while the main search peptide tolerance was set to 4.5 ppm. Isotope mass tolerance was 2 ppm and maximum charge was set to 7. The MS/MS match tolerance was set to 20 ppm. Maximum of two missed cleavages were allowed. Carbamidomethylation of cysteine was set as fixed modification. Oxidation of methionine and acetylation of the N-terminal were set as variable modifications. Multiplicity was set to 2 and for heavy labels Arginine 6 and Lysine 8 were selected. Peptide and protein identifications were filtered to 1% FDR. Unique and non-unique peptides were used for quantitation. Proteins with minimum of two quantified labeled peptide pairs/triplets were reported for quantitation and the isoforms with the highest peptide counts were considered for quantitation. Biological replicates of each RNA nucleotide (wild type and mutants) were analyzed together. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE [49] partner repository with the dataset identifier PXD005761.

2.8. Evaluation of RNA G-quadruplex-binding proteins

Statistical analyses were performed on Perseus version 1.5.2.6. Proteins mapped to potential contaminants, reverse protein hits and proteins only identified by site were excluded prior to statistical analysis. The used static test was significance B calculating intensity-dependent significant outliers relative to a certain population from p-values [47]. Classification related to molecular function as well as to biological process were performed with the PANTHER classification system [50] version 10.0 (www.pantherdb.org) and UniProt database (http://www.uniprot.org). Because of the different binding features of the four wild type G-quadruplex samples, proteins were chosen using following criteria; significance B test (p < 0.05) for the NRAS and MMP16 G-quadruplexes. Results for the Bcl-2 and ARPC2 G-quadruplexes were calculated by choosing proteins from a population showing higher ratio H/L values than the unspecific population (with the exception of one replica which had to be evaluated according to the above described procedure). In addition, proteins identified in at least two replicates were considered as candidate proteins. Common candidates were represented in Venn diagrams [51].

2.9. Western Blotting for Validation of Interaction Partners

Western Blot experiments were carried out to validate interactions between RNA G-quadruplexes and proteins identified by the mass spectrometric analysis. To this end, either 80 μg of the cell lysate or 20% of the magnetic beads mixed with the proteins from the pull-down assays as described above, were loaded onto a 10% SDS-polyacrylamide gel. Following semi-dry transfer of proteins to PVDF membranes, immunodetection was carried out for four candidate proteins, NSUN5, RBM4, HNRNP U and HNRNP F. All antibodies (NSUN5 (H-10): sc-376147, 1:100; RBM4 (E-3): sc-373852, 1:100; hnRNP U (3G6): sc-32315, 1:200; hnRNP F (3H4): sc-32309, 1:100) were purchased from Santa Cruz Biotechnology (TX, USA). Membranes were developed by Pierce™ ECL Western Blotting Substrate (Thermo Fisher Scientific) and visualized using the ChemiDoc™ MP Imaging System (Bio-Rad Laboratories, CA, USA) followed by densitometric analysis using Image Lab Version 4.1 (Bio-Rad). The values were normalized to the value of the cell extract prior to the pull-down assays. Data are represented as mean values from three independent experiments. Statistical analysis (Student's t-test) was carried out to compare the amount of the isolated target proteins between the G-quadruplex sample and the respective mutated sample.

3. Results

3.1. Confirmation of RNA G-quadruplex formation

To confirm that the G-rich sequences located in the 5' UTRs of the mRNAs under investigation in the present study (Bcl-2, NRAS, MMP16 and ARPC2) form G-quadruplex structures, analyses by CD spectroscopy and thermal melting experiments using UV spectroscopy were carried out. CD spectra of the RNA quadruplexes of MMP16 and ARPC2 were already presented in our previous study [12]. Figure 1 shows the CD spectra of the G-rich sequences in the Bcl-2 (A) and NRAS (B) mRNAs in the presence of potassium ions. The spectra exhibit typical features of a parallel G-quadruplex structure, i.e. a negative peak near 240 nm and a positive peak around 260 nm [8, 52]. The maxima of the mutated control sequences were shifted and far less pronounced.

Figure 1. Spectroscopic characterization of RNA G-quadruplex motifs.

The RNA G-quadruplex structures of Bcl-2 and NRAS were characterized by CD spectroscopy and thermal melting experiments. CD experiments (A, B) were carried out with 5 μM of the respective RNA oligonucleotide in 10 mM Tris, pH 7.5 and 100 mM KCl. Thermal melting of the G-quadruplex structures of Bcl-2, NRAS, MMP16, and ARPC2 (C-F) was performed with 1 μM of the respective RNA oligonucleotide in 10 mM Tris, pH 7.5 and 1 mM KCl. Absorbance was normalized to the value at 90°C. Curves for melting and reannealing are shown.

Thermal melting experiments are widely used to characterize the stability of G-quadruplexes [53]. Melting curves were measured between 20 and 90°C at 295 nm (Figure 1C-F). All four G-quadruplexes under investigation in the present study showed a sigmoidal transition, which is typical for a G-quadruplex motif. Neither of the mutated controls exhibited a temperature dependent transition, confirming that they do not fold into a G-quadruplex structure.

To further validate the formation of the G-quadruplex structure, we carried out thermal melting experiments in the presence of lithium ions. This control maintains the sequence constant, but in contrast to potassium ions, lithium ions impair G-quadruplex formation. While a clear transition can be observed in the presence of potassium ions (Figure 1C-F) no such transition occurs in the presence of lithium ions (Supplementary Figure S1). This control experiment further strengthens the assumption that all four G-rich sequences under investigation have the potential to form a G-quadruplex structure. The melting temperature (Tm) of the G-quadruplex forming sequences is a measure for the stability of the respective structure. The G-quadruplex structures of MMP16 and ARPC2 were found to be slightly more stable (Tm of 60.9°C ± 0.9°C and 60.3°C ± 0.4°C, respectively) than those of Bcl-2 and NRAS (Tm of 55.0°C ± 0.9°C and 53.5°C ± 0.3°C, respectively).

3.2. RNA G-quadruplexes inhibit translation

The current hypothesis regarding the main function of RNA G-quadruplex motifs in the 5' UTR of mRNAs is to repress translation without affecting transcription. A common strategy to test this effect is to insert the G-quadruplex forming sequence and its mutated variant, respectively, upstream of a reporter gene. We cloned the G-rich sequences of Bcl-2 and NRAS into the unique NheI site upstream the start codon of Renilla luciferase in the psiCHECK-2 vector (Figure 2A). The vector expresses firefly luciferase as an internal control. Expression of the two luciferases was quantified by qRT-PCR and dual luciferase reporter assays (Figure 2B). The ratio of Renilla and firefly luciferase activity (R/FF) was normalized to the value of the empty psiCHECK-2 vector. While the G-quadruplex motifs of Bcl-2 and NRAS only had a negligible effect on transcription, they significantly reduced translation by approximately 60% and 40%, respectively. In contrast, the mutated controls did not influence transcription or translation. These results are consistent with previous reports on the respective G-quadruplex motifs [8, 13]. In addition, our previous study had shown that the G-quadruplex motifs of MMP16 and ARPC2 repress translation to approximately the same degree [12].

Figure 2. Influence of G-quadruplexes on transcription and translation.

A, The psiCHECK-2 vector used for qRT-PCR and dual luciferase reporter assay. The letters represent the G-quadruplex-forming sequences and the underlined letters represent mutations. The G-quadruplexes and their mutated sequences, respectively, were inserted into the NheI restriction site upstream of the Renilla luciferase coding region in the psiCHECK-2 vector (Promega). R Luc: Renilla luciferase. FF Luc: firefly luciferase. B, Quantitative RT-PCR and dual luciferase reporter assays to determine influence of the G-quadruplex motifs (closed bars) and their mutated sequences (hatched bars) on transcription and on translation. The ratios of Renilla and firefly luciferase activity were normalized to the value of the empty psiCHECK-2 vector. The data represent mean values from three independent measurements (***p < 0.001).

Taken together these data and previous publications [8, 12, 13], confirmed the initial characterization of the G-rich sequences in the 5' UTRs of Bcl-2, NRAS, MMP16 and ARPC2 as forming stable G-quadruplex structures that repress translation by 40-60% without substantial effects on transcription.

3.3. Identification of cellular RNA G-quadruplex-binding proteins

To further understand the biological role of the G-quadruplexes, we aimed at identifying cellular proteins which interact with the RNA G-quadruplex structures. While our previous study [12] only comprised the G-quadruplexes of MMP16 and ARPC2 and was restricted to those proteins that could be identified on Coomassie-stained polyacrylamide gels followed by MALDI-ToF analyses, the current study was designed to reveal a comprehensive picture of the G-quadruplex interacting proteins by SILAC, followed by pull-down assays and LC-orbitrap mass spectrometry. In addition, we included the G-quadruplexes in the mRNAs of Bcl-2 and NRAS, as these genes are of particular relevance in cancer biology.

The experimental design is shown in Figure 3. Protein extracts from cells grown in the presence of amino acids with light or heavy isotopes were used for pull-down assays which were carried out with either of the G-quadruplex motifs and their respective controls. Pools of all eight light samples were used as standard. Equal amounts of one of the heavy samples and the standard were mixed and analyzed by LC-orbitrap mass spectrometry. For each sample, three independent replicates were carried out as described in the Materials and Methods section.

Figure 3. Identification of RNA G-quadruplex-binding proteins.

Schematic workflow for identifying the G-quadruplex-binding cellular proteins. Cellular proteins binding to RNA G-quadruplexes (black) and their respective mutated variants (gray) were isolated by pull-down assay using protein extract labeled with heavy (H samples, hatched) and light (L samples, closed) amino acids, respectively. All the eluted eight L samples were pooled and used as the standard. Equal amounts of the standard and either of the H samples were mixed and analyzed by LC-orbitrap mass spectrometry.

The MS raw files of the three replicates of one G-quadruplex, together with the three replicates of its respective mutant control, were processed in one step by MaxQuant to identify the proteins that bind specifically to the G-quadruplex structure (Supplementary Table S1). Statistical analyses were performed with Perseus Framework. The statistical test used was the significance B test, which calculates intensity-dependent significant outliers relative to a certain population from p-values [47].

Surprisingly, we observed two different patterns for the proteins interacting with the G-quadruplex motifs under investigation. Proteins detected in the samples of the NRAS and MMP16 G-quadruplexes were clustered in one population within the outliers in the ratio H/L-intensity scatter plots (Figure 4A, Supplementary Figure S2). For these samples, candidate proteins were chosen using the significance B test for the calculation of the p-values (p < 0.05). We consider the G-quadruplexes of NRAS and MMP16 to be specific for binding of certain proteins. In total, 68 and 75 proteins were found to be significantly enriched in the samples of the NRAS and MMP16 G-quadruplexes, respectively, compared to the reference sample. None of the potential interaction partners of the G-quadruplex motifs was found to bind to the respective mutated control. Six candidate proteins binding to the NRAS G-quadruplex were detected in at least two replicates (Supplementary Figure S3, Supplementary Table S3D). However, it should be noted that two of the identified proteins belong to the myosin family whose members are known to be common contaminants in proteomics studies [54]. Three of the remaining four proteins are related to nucleic acid regulation. Eight of the 75 candidate proteins for the MMP16 G-quadruplex were detected in at least two replicates (Supplementary Figure S3, Supplementary Table S4D). According to the GO analysis, seven of the eight candidate proteins are known to bind nucleic acids and to regulate metabolic processes such as post-transcriptional regulation (Table 1A).

Figure 4. Protein binding patterns.

Comparison of the protein binding patterns of the four RNA G-quadruplex-forming motifs. Each dot in ratio H/L-intensity scatter plots represents one protein identified to interact with the given G-quadruplex. A, Significantly enriched candidate proteins (p < 0.05) for the specific binding G-quadruplexes (NRAS and MMP16) are marked in black. B, Candidate proteins interacting with one of the broad range binding G-quadruplexes (Bcl-2 and ARPC2) are displayed as a population and they are indicated in gray and black, respectively. In order to display the MS data clearly, the value of the ratio H/L as well as the intensity of each proteins were logarithmically (log2 and log10, respectively) transformed.

Table 1. Candidate proteins for the specific G-quadruplexes of NRAS and MMP16 and the strongest candidate proteins for the broad range binding G-quadruplexes of Bcl-2 and ARPC2.

A, The six candidate proteins for the NRAS G-quadruplex and the eight candidate proteins for MMP16 G-quadruplex which were detected in at least two replicates. The listed proteins were significantly enriched (p < 0.05) compared to the standard. B, The six strongest candidate proteins for the Bcl-2 G-quadruplex and the three strongest candidate proteins for ARPC2 G-quadruplexes detected in all three replicates.

A
Gene symbol	Ensembl Gene ID	Protein Name	Biological Process	Number of times
6 RBPs specific for the NRAS G-quadruplex
NSUN5	ENSG00000130305	Putative methyltransferase NSUN5	rRNA metabolic process	3
MYO1D	ENSG00000176658	Unconventional myosin-Id	cell signaling pathways	3
MYO1C	ENSG00000197879	Unconventional myosin-Ic	cell signaling pathways	2
TRIM26	ENSG00000234127	Tripartite motif-containing protein 26	tumor suppressor activity, innate immune response, transcriptional regulation	2
TLN1	ENSG00000137076	Talin-1	tumor progression, cell migration	2
RPS29	ENSG00000213741	40S ribosomal protein S29	Diamond-Blackfan anemia tumor suppressor activity, translation	2
8 RBPs specific for the MMP16 G-quadruplex
TUSC1	ENSG00000198680	Tumor suppressor candidate gene 1 protein	tumor suppressor activity	2
SCAF11	ENSG00000139218	Protein SCAF11	post-transcriptional regulation, mRNA splicing	2
HNRNPU	ENSG00000153187	Heterogeneous nuclear ribonucleoprotein U	post-transcriptional regulation, mRNA splicing	2
DUSP11	ENSG00000144048	RNA/RNP complex-1-interacting phosphatase	post-transcriptional regulation, mRNA splicing	2
CPSF1	ENSG00000071894	Cleavage and polyadenylation specificity factor subunit 1	3' processing of pre-mRNAs	2
MBD1	ENSG00000141644	Methyl-CpG-binding domain protein 1	negative regulation of transcription from methylated DNA, tumorigenesis	2
CHTOP	ENSG00000160679	Chromatin target of PRMT1 protein	mRNA export	2
RPS2	ENSG00000140988	40S ribosomal protein S2	tumor cell proliferation, translation	2

A different pattern was observed for the analysis of proteins interacting with the G-quadruplexes of Bcl-2 and ARPC2. In this case, two populations within the outliers in the H/L-intensity scatter plots were observed (Figure 4B), although this pattern was not as clear in one of the label swap experiments (Supplementary Figure S2A). We consider these G-quadruplexes to be less specific and to bind to a broad range of proteins. To adequately process these data, a different evaluation strategy was followed. As G-quadruplex-binding candidates, proteins were chosen from the whole population which shows high ratio H/L values that could be distinguished well from the unspecific population (Supplementary Figure S2A, Supplementary Tables S2 and S5). Only proteins detected in at least two replicates were considered for further analysis. Proteins interacting with the respective mutated sequences were identified by the same procedure. For classification of the candidate proteins, data from MaxQuant were loaded into the PANTHER classification system. Proteins which were not annotated in the PANTHER classification system were classified using the UniProt database.

In total, 534 and 407 proteins were enriched in the samples for the Bcl-2 and ARPC2 G-quadruplex motifs, respectively. Out of these groups, 99 and 82 candidates were identified in at least two replicates (Supplementary Figure S3). None of the proteins for either of the G-quadruplex motifs were present in the sample for the respective mutated sequence (Supplementary Tables S2D and S5D). The candidates were classified according to the GO domains Molecular Function and Biological Process and represented as pie charts using the PANTHER classification system (Figure 5).

Figure 5. Candidate proteins for the broad range binding G-quadruplexes of Bcl-2 and ARPC2.

A, Classification of 99 Bcl-2 G-quadruplex-binding proteins and 82 ARPC2 G-quadruplex-binding proteins according to the GO domain Molecular Function using the PANTHER classification system. B, C, and D, Classification of the Bcl-2 G-quadruplex-binding proteins and the ARPC2 G-quadruplex-binding proteins according to the GO domains Biological Process, metabolic process and primary metabolic process using the PANTHER classification system. One protein (RNA-binding protein 4B) could not be classified.

Interestingly, more than half of the candidates for the Bcl-2 and ARPC2 G-quadruplexes were identified in both cases (Supplementary Figure S4, and consequently the overall classification of the proteins for the two G-quadruplexes was very similar. In the GO domain Molecular Function “binding”, “catalytic activity” and “structural molecule activity” were the largest categories. Most of the proteins in the category “binding” were classified to interact with other proteins, i.e. they may interact indirectly with the G-quadruplex by binding a direct interaction partner. Many proteins with “catalytic activity” have hydrolase activity. Approximately 25% of the candidate proteins were categorized as structural constituents of the cytoskeleton or the ribosome. In the GO domain Biological Process, the main categories were “cellular process” (~23%) and “metabolic process” (~20%). Many Proteins in the class “cellular process” are involved in “cell communication”. More than 60% of the proteins categorized in “metabolic process” play important roles in “primary metabolic process” (Figure 5C). Of these, the majority of the proteins were further categorized into “protein metabolic process” and “nucleobase-containing compound metabolic processes” (Figure 5D) which include RNA metabolic processes.

Proteins detected in all three replicates were further analyzed (Table 1B, Supplementary Tables S2D and S5D). Five of the six strongest candidates for the Bcl-2 G-quadruplex were categorized as “nucleotide binding”. Their main function is in post-transcriptional regulation such as mRNA splicing. As before, myosin may be suspected as a contaminant. Three proteins were detected in all three replicates for the ARPC2 G-quadruplex. All of them are assigned to be involved in post-transcriptional regulation (mRNA splicing).

B.

Gene symbol	Ensembl Gene ID	Protein Name	Biological Process	Number of times
6 strongest RBPs specific for the Bcl-2 G-quadruplex
HNRNPF	ENSG00000169813	Heterogeneous nuclear ribonucleoprotein F	post-transcriptional regulation, mRNA splicing	3
HNRNPH1	ENSG00000169045	Heterogeneous nuclear ribonucleoprotein H	post-transcriptional regulation, mRNA splicing	3
HNRNPH3	ENSG00000096746	Heterogeneous nuclear ribonucleoprotein H3	post-transcriptional regulation, mRNA splicing	3
GRSF1	ENSG00000132463	G-rich RNA sequence binding factor 1	post-transcriptional regulation, translation	3
RBM4	ENSG00000173933	RNA-binding protein 4	post-transcriptional regulation, mRNA splicing, translation, tumor suppressor activity	3
MYO1D	ENSG00000176658	Unconventional myosin-Id	cell signaling pathways	3
3 strongest RBPs specific for the ARPC2 G-quadruplex
HNRNPF	ENSG00000169813	Heterogeneous nuclear ribonucleoprotein F	post-transcriptional regulation, mRNA splicing	3
HNRNPH3	ENSG00000096746	Heterogeneous nuclear ribonucleoprotein H3	post-transcriptional regulation, mRNA splicing	3
ACINI	ENSG00000100813	Apoptotic chromatin condensation inducer in the nucleus	apoptosis, post-transcriptional regulation, mRNA splicing	3

3.4. Validation of protein-RNA G-quadruplex interactions

For an independent validation of the results obtained by the MS approach, additional pull-down assays followed by Western blotting were carried out. Four candidates were randomly chosen: According to the MS analysis, NSUN5, hnRNP U, and RBM4 were identified to specifically bind to the the G-quadruplexes of NRAS, MMP16, and Bcl-2, respectively, while hnRNP F was among the strongest RBPs for the G-quadruplexes of Bcl-2 and ARPC2 (Table 1). As can be seen in Figure 6A, the proteins under investigation were substantially enriched in pull-down assays with the G-quadruplex motifs compared to the untreated lysates or the control experiments with the mutated sequences. Three independent experiments were carried out (Supplementary Figure S5) and densitometrically evaluated (Figure 6B).

Figure 6. Western blot analysis of protein-RNA G-quadruplex interactions.

Following pull-down assays with either of the G-quadruplex motifs or their mutated controls, enriched proteins were analyzed by Western blotting. A) One fifth of the magnetic beads mixed with proteins of the pull-down experiment, or untreated cell lysate as control were separated by SDS-PAGE and transferred to a PVDF membrane. Detection was carried out with either of the antibodies specific for NSUN5, RBM4, hnRNP U, and hnRNP F, respectively, as indicated. B) Three independent experiments (Supplementary Figure S5) were densitometrically evaluated using Image Lab Version 4.1 (Bio-Rad). Mean and standard of the three independent experiments are shown. *p < 0.05 and **p < 0.01 versus the respective mutated sequence.

Overall, the validation experiments confirmed the results obtained by the MS approach very well. NSUN5 was specifically enriched by the G-quadruplex of NRAS, but not by its mutated control or any of the other G-quadruplexes as predicted by the MS analysis (Table 1). Likewise, RBM4 was strongly enriched by the G-quadruplex of Bcl-2 (and, to a much lesser extent and without statistical significance, by that of ARPC2). The results of the MS approach were also precisely reproduced for hnRNP F, which was enriched by two G-quadruplexes (Bcl-2 and ARPC2), but not by either of the other G-quadruplexes or any of the mutated controls. This is in perfect agreement with the candidate list shown in Table 1. The situation is less clear for hnRNP U, which is highly expressed in HEK 293 cells, as can be seen in the lysate lane. The highest and statistically significant enrichment was found for the G-quadruplex of Bcl-2 for which hnRNP U was listed as one of the major interacting partners in Table 1. However, it was also found to interact with two other G-quadruplexes (NRAS and Bcl-2), though not with their mutated controls and with a lower statistical confidence. Thus, even in this case, the interaction partner identified by MS was also found to be the tightest binder in the validation experiments. Taken together, these experiments with exemplarily candidates clearly confirm the validity of the MS approach.

4. Discussion

RNA G-quadruplex structures have been demonstrated to fulfill a broad variety of cellular functions, including repression of translation, modulation of splicing, and subcellular targeting of mRNAs [32, 55, 56]. In addition to the characterization of the biological functions of RNA G-quadruplexes, their existence in cells could be visualized by a structure specific antibody [27] and an engineered fluorogenic hybridization probe [28]. These findings were challenged by a recent study, according to which RNA G-quadruplex structures are globally unfolded in eukaryotic cells [31]. However, these steady-state measurements might not detect transiently folded G-quadruplexes. In addition, the G-rich sequences might interact with specific proteins that might influence the translational efficiency.

In fact, we and other have started to investigate interactions between RNA G-quadruplexes and cellular proteins [12, 33, 35, 37, 38] as G-quadruplex-protein complexes may finally represent novel drug targets [57]. In our restricted previous study, we searched candidate proteins that bind to the G-quadruplexes in the 5' UTR of the MMP16 and ARPC2 mRNAs [12]. The interaction partners were isolated by pull-down assays and identified by gel electrophoresis and MALDI-ToF mass spectrometry. However, this study was restricted to only two G-quadruplex motifs and could only detect abundant proteins that were visible in the gel. The present study includes G-quadruplex motifs in the mRNAs of two additional genes of utmost importance in cancer biology, namely the apoptosis inhibitor Bcl-2 and the proto-oncogene NRAS. To obtain a comprehensive picture of the interaction partners of the four G-quadruplex motifs, we made use of the more sophisticated LC-orbitrap mass spectrometry which does not require separation of the proteins by gel electrophoresis.

In our study, we carried out pull-down assays with G-quadruplex-forming oligonucleotides and lysates from HEK293 cells. As CD and UV melting experiments confirmed that the sequences form very stable G-quadruplex structures, it is reasonable to assume that they maintain their conformation in the cell lysate. Furthermore, the use of mutated G-rich oligonucleotides that are not capable of forming G-quadruplex structures as controls whose interactions partners were subtracted in the evaluation procedure ensures that the identified candidates are in fact specific binders of the G-quadruplex-forming sequence rather general RNA-binding proteins.

Unexpectedly, we observed different patterns for the G-quadruplex-protein interactions for the four motifs under investigation. While the G-quadruplex motifs of NRAS and MMP16 specifically interacted with a small set of proteins, the G-quadruplex motifs of Bcl-2 and ARPC2 bound to a large number of proteins.

Six top candidates were identified to interact with the NRAS G-quadruplex: NSUN5, MYO1C, MYO1D, TRIM26, TLN1 and RPS29. The putative methyltransferase NSUN5 (also called NOP2/Sun RNA methyltransferase family member 5 and probable 28S rRNA (cytosine-C(5))-methyltransferase) is deleted in the genome of individuals with the Williams-Beuren syndrome [58, 59]. The protein encoded by this gene adds a single methyl group to rRNA and plays an important role in aging processes in which it acts as a pro-aging factor [60]. Two proteins, MYO1C (unconventional myosin-Ic) and MYO1D (unconventional myosin-Id), belonging to the myosin family were identified. Myosins are known to be actin-based motor proteins and to be involved in a number of cell signaling pathways [61]. However, they are not known for being involved in regulatory processes of gene expression and have frequently been identified as contaminants in proteomics studies [54]. The two myosins may therefore be false-positives. A further identified interaction partner of the NRAS G-quadruplex, the cytosolic protein TRIM26 (tripartite motif-containing protein 26), is involved in cancer biology. Several members of the TRIM family have been described as functioning as regulators for carcinogenesis [62]. TRIM26 acts as a tumor suppressor in hepatocellular carcinoma by regulating cell proliferation, colony forming ability, migration and invasion [63]. Its downregulation correlated with a poor prognosis in patients with liver cancer. However, the detailed mechanism by which TRIM26 modulates carcinogenesis is not yet fully understood. The cytoskeleton protein TLN1 (Talin-1) has been shown to play an important role in tumor biology and to serve as a potential therapeutic target for cancer [64]. Finally, the ribosomal protein RPS29 (40S ribosomal protein S29) was found to interact with the NRAS G-quadruplex. This interaction may result from stalled ribosomes at the site of the G-quadruplex structure, which would explain the reduced translation efficiency of mRNAs with a respective motif. In addition, whole-exome and whole-genome sequencing approaches have recently revealed a link between mutations in ribosomal proteins and carcinogenesis [65]. In one of these studies, RPS29 was identified as a novel gene mutated in multicase Diamond-Blackfan anemia families [66].

Eight proteins were detected in two replicates to interact with the MMP16 G-quadruplex. Most of the candidates are involved in processing of primary mRNA transcripts. The SCAF11 protein (SR-Related CTD Associated Factor 11, also called SC35-interacting protein 1) is an important factor for pre-mRNA processing known to interact with the splicing factor U2AF65 [67] which we identified as an interaction partner of the MMP16 G-quadruplex in our previous study [12]. Another interaction partner whose function in splicing has been well documented is the heterogeneous nuclear ribonucleoprotein U (hnRNP U) that binds RNA via its glycine-rich domain [68, 69]. The hnRNPs, together with additional proteins, form complexes which play important roles in pre-mRNA splicing [70]. In our previous study we also identified hnRNP U as a MMP16 G-quadruplex interaction partner [12].

Three further binders of the MMP16 G-quadruplex were determined to be factors involved in RNA processing. DUSP11 (RNA/RNP complex-1-interacting phosphatase, also known as PIR1 and dual specificity phosphatase 11) is a member of the dual specificity phosphatase family which negatively regulates MAPK signaling pathways. It binds RNA with high affinity and interacts with various splicing factors [71]. Moreover, the DUSP11 gene is a target gene of p53 and it interacts with the splicing factor SAM68 [72]. It has been described as contributing to p53-dependent inhibition of cell proliferation and as involved in the regulation of RNA splicing. CPSF1 (cleavage and polyadenylation specificity factor subunit 1) is a member of the cleavage and polyadenylation specificity factor (CPSF) complex that plays a crucial role in the 3' processing of pre-mRNAs [73]. The small nuclear protein CHTOP (chromatin target of PRMT1) belongs to the TREX complex that couples nuclear pre-mRNA processing with mRNA export [74].

The identified MMP16 G-quadruplex binding partner MBD1 (methyl-CpG binding domain protein 1) is also known to interact with nucleic acids, as it has been described to bind to methylated CpG islands in promoters and to act as a transcriptional repressor [75]. In addition, MBD1 was shown to be important for pancreatic cancer therapy [76]. Another interaction partner of the MMP16 G-quadruplex is the TUSC1 (tumor suppressor candidate gene 1 protein) whose gene TUSC1 has been shown to reduce tumor cell growth in vitro and in vivo [77]. Finally, RPS2 (40S ribosomal protein S2) is a component of the small subunit of the eukaryotic ribosome. As outlined above, this interaction is indicative for stalling of the ribosome by the G-quadruplex. The ribosomal proteins identified in the present study are components of the small ribosomal subunit, which is plausible as this subunit scans the 5' UTR for the AUG start codon. RPS2 was found to be overexpressed in liver tumors [78] and has also been considered a potential therapeutic target in prostate cancer [79].

For the other two G-quadruplexes in the mRNAs of Bcl-2 and ARPC2, we found a larger number of interaction partners. For the two G-quadruplexes, 99 and 82 proteins, respectively, were identified as potential interaction partners in at least two replicates. All of the G-quadruplex-binding proteins that showed up in all three replicates are involved in post-transcriptional regulation.

The four top interaction partners of the Bcl-2 G-quadruplex, hnRNP F, hnRNP H1, hnRNP H3, and GRSF1, belong to the hnRNP family that plays important roles in multiple aspects of RNA processing [70]. The hnRNPs are further classified into the hnRNP F/H family containing two or three quasi-RNA recognition motifs (qRRMs) [80]. The qRRMs recognize G-rich sequences, i.e. putative G-quadruplex-forming sequences, in pre-mRNAs and regulate post-transcriptional processing, including splicing [81–83]. The identification of proteins with domains that recognize G-rich sequences and the absence of related proteins possessing KH domains that bind to AU-rich sequences (e.g. hnRNP K) further validates our approach [84]. GRSF1 (G-rich RNA sequence binding factor 1), the fourth candidate binding to the G-quadruplex motif of Bcl-2, is mainly involved in translational regulation and is named according to its function of binding to G-rich RNA sequences [85, 86]. MYO1D was also detected for the Bcl-2 G-quadruplex, but as discussed for NRAS it might be a contaminant. The last candidate protein for the Bcl-2 G-quadruplex is RBM4 (RNA-binding protein 4). Similar to the hnRNP F/H family, RBM4 contains two RNA recognition motifs (RRMs) and an additional CCHC-type zinc finger [87, 88]. It is reasonable to assume that the RRMs associate with the G-quadruplex motifs. RBM4 is involved in the regulation of splicing [88] and was reported to act as a tumor suppressor by restricting proliferation and migration of various cancer cells by specifically controlling cancer-related splicing [89].

Two of the identified interaction partners for the ARPC2 G-quadruplex, hnRNP F and hnRNP H3, were also found to bind to the Bcl-2 G-quadruplex. In addition, hnRNP F was also detected as a candidate protein for the ARPC2 G-quadruplex in our previous study [12]. Thus hnRNPs may have a general potential to interact with G-quadruplex structures, or they may act selectively with certain G-quadruplex structures.

In addition to the hnRNPs, ACIN1 (apoptotic chromatin condensation inducer 1, also known as ACINUS) was identified as an interaction partner of the ARPC2 G-quadruplex. ACIN1 contains a RRM and was reported to be involved in the regulation of splicing [90].

Some of the interaction partners detected here were also found as G-quadruplex-binding proteins in our previous study [12], including hnRNP U for MMP16 and hnRNP F for ARPC2. It should be noted that both studies used different controls which may affect the identification of G-quadruplex-binding proteins. While our focused previous study [12] directly compared either of the two G-quadruplex motifs under investigation with its specific mutated control, we used a mixture of all four G-quadruplexes and their four controls as the reference sample here, to ensure that the internal standard contains all proteins seen in any of the eight samples without adding substantially more proteins than needed, which would have been the case if using SuperSILAC with whole cell extract [91].

To ensure the validity of our approach, we carried out independent experiments to study the protein-G-quadruplex interactions. Exemplary candidate proteins enriched by pull-down experiments were analyzed by Western blotting. The outcome of this analysis was in very good accordance with the results of the MS approach. The identified interaction partners were confirmed to specifically bind to their respective G-quadruplex motif, but not to the mutated control or any of the other G-quadruplexes. These validation experiments give good confidence in the reliability of the MS data.

Although it is not fully clear why two of the four G-quadruplexes under investigation specifically interact with a small set of proteins while the other two G-quadruplexes had a higher number of interaction partners, it is tempting to speculate that the different modes of interaction depend on the lengths of the loops in the G-quadruplex structures. The specific G-quadruplex motifs had a total length below 20 nucleotides (17 and 18 nucleotides, respectively), whereas the G-quadruplexes with a broad range of interaction partners were longer than 20 nucleotides (21 and 25 nucleotides, respectively).

In addition, not only the loop length differed between the two classes of G-quadruplexes, the lengths of the G-repeats also varied. While the specific G-quadruplexes were well defined, with three consecutive guanines separated by the loops (with one exception of four consecutive guanines), the G-quadruplexes binding to a broader range of proteins had longer stretches of guanines (up to five guanines in a row). This raises the possibility that they fold into a number of different G-quadruplex structures. This assumption was further strengthened by analyzing the G-quadruplexes with the QGRS mapper, a bioinformatic tool for the prediction of G-quadruplex motifs [92]. According to the QGRS mapper, the G-rich sequences of NRAS and MMP16 have the potential to fold into 38 and 17 G-quadruplex motifs, respectively. Only 1 and 2 of those, respectively, had a high G-score above 35, which is a measure for the likelihood of a G-rich sequence to form a stable G-quadruplex structure. In contrast, the G-rich sequences of Bcl-2 and ARPC2 had the potential to form 201 and 123 G-quadruplex structures, with 12 and 6 putative motifs, respectively, having a G-score above 35. Thus, the bioinformatic analysis further strengthens the hypothesis that the G-rich sequences of NRAS and MMP16 form well-defined G-quadruplex motifs that specifically interact with a limited set of proteins. The structures formed by the G-rich sequences of Bcl-2 and ARPC2 are less well defined and a dynamic equilibrium between different G-quadruplex structures is very likely the case which may explain why they interact with a broader range of proteins. However, the correlation between the composition of G-quadruplexes and their specificity for binding of cellular proteins needs to be analyzed in more detail in future studies.

In recent studies, the RNA-interactome was determined for HeLa [93], HEK293 [94], murine embryonic stem cells [95], RAW 264.7 macrophages [96], and beating cardiomyocytic HL-1 cells [97]. In our study with a whole cell extract from HEK293, we identified a total of 139 candidate proteins as interacting with one or more of the four RNA G-quadruplex structures under investigation. When we compared this set of proteins with the published RNA interactome of HEK293 [94], 35 proteins were detected in both studies, while 104 novel proteins were found in the present experiments (Supplementary Figure S6, Supplementary Table S6).

The Rhau helicase was described as the dominant helicase responsible for unwinding RNA G-quadruplexes [39]. We detected Rhau in our interaction study; however, it was not specifically enriched in the G-quadruplex-forming motifs. This may due to the fact that the helicase generally interacts with RNA, but does not specifically interact with G-quadruplex structures. The described unwinding process for G-quadruplex motifs may be a transient event that escapes detection in our assay. In addition, RNAi-mediated knockdown of Rhau revealed that the helicase unwinds only a subset of G-quadruplex motifs (data not shown). This finding is consistent with the observation that the global G-quadruplex folding pattern remained largely unchanged after deletion of Rhau by Cre-mediated recombination in mouse embryonic fibroblasts [31]. The authors of this study speculate that redundant functions with other helicases may be responsible for this observation.

Another finding was that only a small number of the strongest interaction partners were identified for more than one of the G-quadruplex motifs (hnRNP F and hnRNP H3 were identified as interaction partners for the G-quadruplexes of Bcl-2 and ARPC2; and the ambiguous candidate MYO1D was found for the G-quadruplexes of NRAS and Bcl-2). This validates our experimental approach to identify specific binders of the respective RNA G-quadruplexes. However, this observation also suggests that the structures of G-quadruplex motifs are diverse enough to require specific proteins to selectively modulate a given G-quadruplex motif [98].

Recent research indicates that regulation of the transient folding of G-rich sequences into G-quadruplex structures may be an important cellular process. Here, we carried out a comprehensive analysis of interactions between proteins and four G-quadruplex motifs. We found two G-quadruplex motifs to have a higher specificity in binding to proteins, whereas the other two G-quadruplex motifs bound to a broader range of proteins. Several ribosomal proteins were found as interaction partners, which supports the assumption the G-quadruplex structures actively stall the ribosome and thereby suppress translation. In addition, we identified numerous factors involved in post-transcriptional mRNA processing, e.g. hnRNPs. The enrichment of this class of proteins is indicative for a role of the G-quadruplex motifs in the regulation of splicing. Another curious finding of the present study is the large number of cancer-associated factors that were identified as G-quadruplex-binding proteins. This indicates that G-quadruplexes may in fact constitute attractive targets for anti-cancer therapeutics.