Identifying G-quadruplex-interacting proteins in cancer-related gene promoters

Introduction

In the last decades, G-quadruplexes (G4s) have emerged as one of the most attractive and targetable noncanonical DNA/RNA secondary structures. G4s can fold from guanine-rich sequences through stacked G-tetrads, a square planar arrangement stabilized by monovalent cations (such as potassium or sodium), connected by loops of varying lengths. A monomeric G4 is typically described by a G_≥3N_1-7G_≥3N_1-7G_≥3N_1-7G_≥3N_1-7 motif, where G_≥3 stands for a G-tract and N represents any nucleotide in the loop^1,2.

G4 structures are highly polymorphic and exhibit significant versatility, stability, and functional diversity. Their folding and stability are greatly influenced by the length and composition of the G4-forming sequence, the number of G-tetrads and strands involved, the size of the loops, and the type of the stabilizing cation. Depending on the number of strands involved, G4s can be classified as intramolecular (formed by a single G-rich strand) or intermolecular (formed by two or four interacting strands) (Supplementary Fig. S1). The strands’ orientation defines the topology of the secondary structure, which can be: i) parallel, with all strands oriented in the same direction; ii) antiparallel, with two strands oriented in opposite directions; or iii) hybrid (3 + 1), with three strands oriented in one direction and the fourth in the opposite direction (Supplementary Fig. S1)³. Generally, intramolecular constructs give rise to monomeric G4s. However, multimeric structures are also possible when a single strand contains multiple G-tracts⁴.

G4-forming sequences are abundant in various functional regions of higher eukaryotic genomes, including telomeres, oncogene promoters, and 5’ or 3’ untranslated regions of mRNAs. This suggests they play a variety of biological roles, such as regulation of transcription, translation, DNA replication, and RNA localization^5,6. G4 structures typically exert their cellular functions by interacting with or recruiting several proteins, also known as G4-related proteins (G4RPs), such as helicases, telomere-binding proteins, epigenetic modulators, and transcription factors^7,8,9,10.

Several transcription factors have been identified as G4RPs. For instance, the Myc-associated zinc finger protein (MAZ) and transcription factor Sp1 bind to KRAS and c-KIT* G4s, respectively. Poly(ADP-ribose) polymerase 1 (PARP1) specifically binds the KRAS G4^11,12, which can also be recognized and unfolded by the ribonucleoprotein HNRNP A1^13,14. Yin Yang 1 (YY1), another zinc finger protein, binds two close-spaced DNA G4s and dimerize, bringing the two DNA fragments closer¹⁵. Nucleolin (NCL), a nuclear phosphoprotein, preferentially binds and induces folding of G4s with long loops¹⁶. Notably, NCL and nucleoside diphosphate kinase (NM23-H2) act as a repressor and activator of the c-MYC oncogene by inducing the G4 formation or promoting unfolding, respectively^17,18. NM23-H1 and NM23-H2 proteins have also been shown to repress the transcription of the PDGF-A gene via functional interactions with promoter elements¹⁹. Additionally, nucleophosmin (NPM) and tumor suppressor protein p53 can bind to c-MYC G4^20,21. HMGB1, an architectural nuclear protein that bind to DNA and regulate multiple genomic processes²², can interact with G4-forming sequences and stabilize the noncanonical conformation^23,24,25. More recently, vimentin (VIM), an intermediate filament protein, was identified for the first time as a binder of multimeric G4s from both telomeres and oncogene promoters^26,27.

Various methods have been developed to identify DNA G4RPs from nuclear/cytoplasmic extracts, as well as native chromatin^{17,26,28,29,30,31}. A widely used proteomic approach employs biotin-labeled G4-forming sequences as baits in solution. These sequences are incubated with cellular extracts, allowing specific protein interactions to occur. The G4 structures, along with their bound proteins, are subsequently captured using streptavidin-coated magnetic beads. After isolation, the complexes are eluted under denaturing conditions, digested into peptides, and analyzed by LC-MS/MS tandem analysis to identify the associated proteins^17,28,29. While this method does not account for the native chromatin context, it has successfully identified numerous G4RPs using specific G4 sequences. Chromatin immunoprecipitation (ChIP) combined with mass spectrometry-based proteomics has been instrumental in characterizing chromatin-associated protein complexes³². However, this method relies on high-affinity, high-specificity antibodies, which are not always available, posing significant limitations in certain experimental setups. To overcome these limitations, Balasubramanian et al. developed a co-binding-mediated protein profiling (CMPP) strategy to identify G4RPs in living cells³⁰. This method involves incubating cells with a G4-ligand probe functionalized with a photo-crosslinking group. Upon irradiation, the probe forms a covalent complex with nearby G4RPs, which can then be isolated, digested, and analyzed by LC-MS/MS³⁰. Similarly, the G4 ligand-mediated crosslinking and pull-down strategy has been proposed for labeling and isolating G4RPs within cells, thought it often gives distinct results compared to CMPP³³. A more recent approach introduced by Lu et al. is the G4-interacting proteins-specific biotin-ligation procedure, which employs a G4-targeting biotin ligase to profile G4RPs in cells³⁴. However, methods based on ligands or ligation procedures have inherent limitations, as they may stabilize G4 structures, induce transient G4 formation, or alter G4 topology, potentially biasing the resulting G4 interactome.

While all these approaches offer valuable insights into the identification of G4RPs, their focus on specific, stable, and single G4 structures, may exclude interactions with proteins that preferentially bind to dynamic or transient G4s. Such interactions are important for understanding the biological role of G4s, especially in complex cellular contexts where G4 polymorphism is crucial. Notably, many proteins that play key roles in cancer development and progression are encoded by genes with multiple G-/C-rich tracts in the complementary strands of their promoter regions. Key examples include death-associated protein 1 (DAP1), hypoxia-inducible factor 1α subunit (HIF-1α), juxtaposed with another zinc finger protein 1 (JAZF-1), and platelet-derived growth factor (PDGF-A)^{35,36,37,38,39,40}. These genes are involved in critical pathways related to cell growth, survival, and angiogenesis. While extensive research has investigated the C-rich strands of these genes and their ability to form i-motifs⁴¹ (another noncanonical DNA structure^40,41,42), the complementary G-rich, G4-forming strands, despite their potential biological importance, have received less attention, with the exception of HIF-1α and PDGF-A^{36,39,43,44,45}.

HIF-1α is one of the two subunits of the HIF-1 transcription factor, which regulates over 60 genes involved in many processes³⁶. In tumors, the activation of hypoxia-responsive genes promotes angiogenesis, metabolic adaptation, resistance to apoptosis, and the expression of genes associated with local invasion and metastasis³⁷. The polypurine/polypyrimidine tract in its promoter is adjacent to several putative transcription factor binding sites, and its G-rich strand can fold into several G4s in vitro in the presence of potassium. Mutagenesis of this region results in lower basal HIF-1α expression⁴³.

The PDGF-A signaling pathway leads to proliferation, migration, angiogenesis, and metastasis, particularly in pancreatic cancer progression^37,46. Its gene proximal 5’-flanking region contains a nuclease hypersensitive element critical for transcription. This region includes a G4-forming sequence with five G-tracts in the -82 to -47 region, which can fold into two stable intramolecular G4 structures in dynamic equilibrium under physiological conditions. These structures are involved in transcriptional regulation, and ligand-mediated stabilization of the major G4 structure can silence PDGF-A expression³⁹.

DAP1 is a highly conserved phosphoprotein with pro-apoptotic functions³⁵. It is correlated with disease progression and long-term survival of patients with colorectal cancer, while low DAP1 expression is associated with clinicopathological parameters of breast cancer^35,47. The G-rich strand of DAP1 promoter contains five G-tracts and can potentially form multiple G4 structures.

JAZF-1 is a zinc finger protein associated with tumor progression and type 2 diabetes^38,48,49. Its overexpression is correlated with enhanced prostate cancer cell proliferation, migration, and invasion by regulating JNK/Slug signaling, although the underlying molecular mechanism remain unclear³⁸. The G-rich strand of JAZF1 promoter also contains multiple consecutive G-tracts, suggesting a high degree of structural polymorphism.

In this study, the G-rich sequences from the DAP, HIF-1α, JAZF-1, and PDGF-A gene promoters (Table 1) were first examined in vitro to evaluate their ability to form G4 structures and their inherent structural polymorphism using a combination of biophysical techniques. Biotin-labeled derivatives of these sequences were then synthesized and used as baits in solution to identify G4RPs from the nuclear extracts of osteosarcoma cancer cells (U2OS) using a classical “fishing–for–partners” proteomic approach^28,29,50. To ensure selectivity for G4 structures over non-G4-forming G-rich sequences, the enrichment protocol was optimized using an unfolded biotinylated G-rich oligonucleotide. By targeting a mixture of G4 conformations, our strategy aimed to provide a more physiologically relevant representation of the G4 cellular landscape. This approach also addresses a key limitation of conformation-specific fishing techniques, which often fail to capture the full diversity of G4-protein interactions occurring in vivo. Notably, 86 putative G4RPs were identified, 14 of which were able to interact with all four G4-forming sequences. Importantly, to the best of our knowledge, 7 of these 14 proteins are new G4 interactors, uncovering potential new players in G4-mediated gene regulation.

Table 1 List of G-rich sequences investigated in this study

Full size table

Results and discussion

Biophysical analysis of the selected G4-forming sequences

To demonstrate that the G-rich sequences from HIF-1α, PDGF-A, JAZF-1, and DAP fold into G4 structures under the experimental conditions used in the pull-down assay (i.e., 20 mM HEPES buffer at pH 6.6 containing 150 mM KCl, and 1 mM EDTA), a combination of spectroscopic techniques, including fluorescence, circular dichroism (CD), nuclear magnetic resonance (NMR), and non-denaturing polyacrylamide gel electrophoresis (PAGE) was employed. A GT-repeated sequence (GT15), not expected to form G4 structures, was used as a negative control (Table 1 and Supplementary Table S1).

First, fluorescence assays were performed using selective light-up probes for G4 recognition, namely thioflavin T (ThT)⁵¹ and N-methylmesoporphyrin IX (NMM)⁵². These small molecule probes selectively bind to G4 structures, causing significant fluorescence enhancement (FI) at 487 (ThT) and 609 nm (NMM) compared to their unbound forms (FI₀)^51,53,54. This fluorescence enhancement serves as a reliable indicator of G4 presence in solution. Notably, the response of these probes is highly sensitive to G4 topology and interaction stoichiometry −for example, NMM exhibits a preference for parallel over hybrid G4s^54,55. To avoid ambiguity, four well-known G4-forming sequences (AS1411, c-KIT2, c-MYC, and mTEL₂₄) were used as positive controls, representative of different G4 topologies and degree of polymorphism. Specifically, AS1411 is known for its high structural G4 polymorphism, forming a mixture of different conformations predominantly parallel⁵⁶, c-KIT2 and c-MYC form single monomolecular G4s with parallel topology^57,58, while mTEL₂₄ predominantly folds into a monomolecular hybrid-type G4⁵⁹. In addition, a hairpin duplex-forming sequence (Hrp₂₀, Supplementary Table S1) and the GT15 sequence were included as negative controls. Results of the ThT assay, performed at a ThT/DNA ratio of 1:2, revealed distinct fluorescence enhancements for the G4-forming sequences (Fig. 1A and Supplementary Fig. S2). AS1411, c-KIT2, c-MYC, and mTEL₂₄ G4s induced 60-, 145-, 180-, and 200-fold fluorescence enhancement of ThT, respectively. In contrast, GT15 and Hrp₂₀ showed less than a 15-fold increase, confirming their inability to form G4 structures. When the G-rich sequences from DAP, HIF-1α, JAZF-1, and PDGF-A were tested, ThT fluorescence increased by 72-, 197-, 131-, and 94-fold, respectively. These results indicate that all four investigated G-rich sequences effectively fold into G4 structures under the experimental conditions used. Fluorescence assays with NMM, performed at a 1:10 NMM/DNA ratio, further supported these observations (Fig. 1B and Supplementary Fig. S3). Parallel G4s formed by AS1411, c-KIT2, and c-MYC sequences induced over a 40-fold increase in NMM fluorescence at 609 nm. The mTEL₂₄ hybrid G4 generated a moderate 22-fold increase, consistent with a discrimination ratio of 2.0 between parallel and hybrid G4s. For the G-rich sequences from DAP, HIF-1α, JAZF-1, and PDGF-A, NMM fluorescence increased more than 30-fold, suggesting predominant folding into parallel G4s. On the other hand, the GT15 sequence and the Hrp₂₀ hairpin-duplex structure, both unable to form G4s, displayed minimal fluorescence enhancements of less than 5-fold with NMM (Fig. 1B).

**Fig. 1: DNA sequences from the *DAP*, *HIF-1α*, *JAZF-1*, and *PDGF-A* gene promoters form G4s.**

To further corroborate these results and gain insights into G4 topologies, CD spectra of the DAP, HIF-1α, JAZF-1, and PDGF-A sequences were acquired (Fig. 1C). Different G4 topologies exhibit characteristic CD signal patterns: i) parallel, with a maximum band at around 264 nm and a minimum band at around 240 nm; ii) antiparallel, with a maximum at 290 nm and a minimum at 260 nm; iii) hybrid, combining features of parallel and antiparallel orientations, with maxima at around 290 and 264 nm and a minimum at 240 nm⁶⁰. The CD spectra of all four G4-forming sequences recorded at 20 °C indicated parallel G4 conformations, in agreement with the NMM assay results. However, the spectra of HIF-1α and DAP exhibited a less intense maximum and a bump at 295 nm, slightly more pronounced for DAP, suggesting the possible contribution of other minor G4 topologies with antiparallel strands or higher-order G4s characterized by additional base stacking interactions. As expected, the CD spectrum of GT15 showed features typical of an unstructured G-rich oligonucleotide, with a weak positive band at 280 nm (Fig. 1C)⁶¹. Additionally, the thermal stability of these G4-forming sequences was evaluated through CD melting experiments. All DNA structures exhibited high thermal stability in solution in the presence of 150 mM potassium ions, with melting temperatures above 80 °C (Supplementary Fig. S4). The multimeric nature of DAP, HIF-1α, JAZF-1, and PDGF-A G4s was examined using PAGE. All samples displayed smeared bands, characteristic of high-order DNA structures (Fig. 1D and Supplementary Fig. S5). The electrophoretic pattern of HIF-1α was consistent with previously reported findings, confirming the formation of mono-, bi-, and tetramolecular G4 structures under the experimental conditions used⁴³. For DAP and JAZF-1, only smeared bands were observed, suggesting structural heterogeneity and the formation of higher-order structures for these sequences. Similarly, the results for PDGF-A were in agreement with literature data, showing a predominance of intermolecular G4 structures with some intramolecular conformations³⁹.

To further investigate the structural features of these G4s, 1D ¹H NMR spectra were acquired. The imino protons of guanine bases involved in G-tetrads through Hoogsteen hydrogen bonding, typically resonate in the 12.0–10.0 ppm range of the proton spectrum⁶². For HIF-1α and PDGF-A G4s, the presence of humped and unresolved signals in this region, along with signals in the aromatic proton region, indicated the existence of multiple G4 conformations (Supplementary Fig. S6). In contrast, DAP and JAZF-1 showed no detectable signals in the imino proton region, along with humped, unresolved signals in the aromatic one. This suggests highly heterogeneous folding (Supplementary Fig. S6), as also indicated by PAGE experiments. Such structural heterogeneity likely results in rapid conformational exchange, with significant broadening or disappearance of NMR signals. To reduce this dynamic exchange, 1D ¹H NMR experiments were conducted under reduced ionic strength conditions (20 mM HEPES buffer containing 5 mM KCl) and at varying temperatures (Supplementary Fig. S7). Ionic strength and temperature influence G4 polymorphism, with lower ionic strength and higher temperatures reducing heterogeneity by destabilizing intermolecular and higher-order structures⁶³. Notably, for the HIF-1α G4, distinct imino proton peaks were observed between 11.8 and 11.0 ppm already at 25 °C, which further resolution achieved at higher temperatures (Supplementary Fig. S7). In contrast, DAP, JAZF-1, and PDGF-A sequences maintained significant structural heterogeneity even under reduced ionic strength, although the bump becomes more resolved. Upon heating, these sequences showed an enrichment of specific higher-order structures over the others, as indicated by up-field-shifted humped imino signals.

Profiling of DNA G4-interacting proteins

With the aim of identifying nuclear G4RPs taking into account the polymorphic nature of G4-forming sequences, biotinylated DAP, HIF-1α, JAZF-1, and PDGF-A G-rich sequences were prepared and used to pull-down proteins from U2OS cancer cell nuclear lysates. This was combined with a classical proteomic approach integrating G4RP affinity enrichment and proteomic analysis. This method was chosen for its ability to preserve the endogenous G4 landscape, avoid limitations associated with antibody dependence, and provide a more comprehensive view of G4-protein interactions. Unlike other methods that artificially stabilize a single G4 conformation through mutations or chemical modifications, our strategy maintained the natural conformational diversity of the sequences, reflecting their physiological conditions. Therefore, this strategy should provide a more accurate representation of G4 structures as they are expected to exist under physiological conditions, better capturing their dynamic and biologically relevant conformations.

To verify that the 5’-biotinylated sequences retained their G4 structures, CD spectra were acquired under the same conditions as their unmodified counterparts. The resulting spectra showed no significant differences from those of the unmodified oligonucleotides (Supplementary Fig. S8). To minimize nonspecific binding and enrich for G4-specific interactors, nuclear lysates were preincubated with unstructured biotinylated GT15 (biotin-GT15, Fig. 2) before each pull-down experiment. After this step, oligonucleotide probes were recovered using streptavidin-coated magnetic beads, exploiting the strong and specific interaction between streptavidin and the biotin moiety. The unbound fraction (i.e., the washed fraction) was subsequently incubated with biotinylated G4-forming sequences. Protein-loaded G4s were then recovered using magnetic beads again. Proteins bound to the G4 DNA, representing potential G4-specific interactors, as well as those bound to GT15, were eluted under denaturing conditions (Fig. 2). This protocol was carried out separately for each G4-forming sequence investigated.

**Fig. 2: Schematic overview of the workflow for MS-based identification of G4-interacting proteins.**

The bound fractions (from both GT15 and G4 probes) were enzymatically digested in solution with trypsin, and the resulting peptide mixtures were analyzed by LC-MS/MS. All proteomic data obtained for biotinylated DAP, HIF-1α, JAZF-1, and PDGF-A samples, as well as GT15, are provided in Supplementary Data 1. UniProt ID, protein name, number of peptides, sequence coverage (%), molecular weight (kDa), score, and label-free quantification (LFQ) intensity were reported for each identified protein. Aiming to define the G4RPs dataset, we excluded proteins binding exclusively to the biotinylated GT15 sequence, as well as those identified as cytosolic or environmental impurities. After the filtering step, a dataset of 86 putative G4RPs was obtained (Fig. 3A and Supplementary Table S2). These proteins were classified according to biological activities and molecular functions using the DAVID database (Fig. 3B, C)^64,65. The analysis of the annotated biological processes revealed that the identified G4RPs are implicated in various nuclear processes, mainly mRNA processing and splicing (Fig. 3B). Notably, over 25% of these proteins are involved in transcriptional regulation. In addition, a significant fraction of proteins is already recognized as nucleic acid-binding proteins, with 51.3% classified as DNA-binding and 39.8% as RNA-binding proteins (Fig. 3C). Furthermore, 19 of the 86 G4RPs have been previously reported as RNA and/or DNA G4 interactors.

**Fig. 3: Profiling of the 86 proteins identified as putative G4RPs.**

Since our protocol did not allow for a quantitative analysis of binding fractions, proteins were classified based on the number of times they were identified as binders of DAP, HIF-1α, JAZF-1, or PDGF-A G4s compared to GT15. This comparison, expressed as the G4s/GT15 ratio, produced three groups: i) 19 proteins bound more frequently to G4s than to GT15 (G4s/GT15 > 1); ii) 47 proteins interacted equally with G4s and GT15 (G4s/GT15 = 1); iii) 20 proteins bound more frequently to GT15 than to G4s (G4s/GT15 < 1) (Supplementary Table S2).

Among the 66 proteins with G4s/GT15 ≥ 1, 13 are known G4RPs. Four of these—nucleolin (NCL), far upstream element-binding protein 2 (FUBP2, also known as KHSRP), lamin B1 (LMNB1), and heterogeneous nuclear ribonucleoprotein K (HNRNP K)—exhibited a clear preference for G4 structures. These proteins are well-established G4-interactors, underscoring the reliability of our protocol^{16,17,28,66,67,68,69}. For instance, NCL is widely recognized as a highly selective G4RP strongly binding both DNA and RNA G4s^{16,17,70,71,72}, while KHSRP and LMNB1 have been previously identified as interactors of human telomeric G4²⁸, although these and other findings also suggest roles beyond telomeres⁷³. The other nine proteins, equally bound to G4s and GT15, include five known to unwind G4 structures and/or bind to unstructured G-rich sequences. These are DHX9, known to unwind both DNA and RNA G4s contributing to transcriptional activation and genomic stability^67,68, HNRNP A1 and HNRNP A2B1, two ribonucleoproteins with helicase activity^69,74,75,76, HNRNP D and HNRNP H1 which regulate telomere overhangs and interact with G-rich sequences in both structured and unstructured states^77,78. Another notable example is PARP1, first identified as a c-KIT1 G4 binder and later shown to regulate KRAS and c-MYC transcription through G4 interactions^79,80.

Additionally, among proteins with G4s/GT15 < 1 already known to bind to G4s, 6 were previously identified as RNA G4 interactors. For instance, DDX17 acts on RNA G4s in the 5’-UTR, while nucleophosmin (NPM) and non-POU domain-containing octamer-binding protein (NONO) interact with the lncRNA MALAT1 G4⁷². Other examples include HNRNP R, ILF3, and SFPQ, previously identified with HNRNP K and HNRNP L as interactors of the RNA G4 of the mutant C9orf72 gene⁸¹.

Notably, 46 of the 86 identified proteins interacted exclusively with a single G4-forming sequence: 30 with JAZF-1, 7 with HIF-1α, and 9 with DAP, while no potentially specific interactors were identified for PDGF-A. Among the G4-forming sequences investigated, JAZF-1 and HIF-1α attracted the largest number of interactors, with 65 and 44 proteins, respectively, with 32 proteins shared between them (Fig. 3A). DAP and PDGF-A yielded 33 and 25 interactors, respectively, with 16 in common (Fig. 3A). It is worth noting that these fishing-for-partners experiments were conducted using U2OS cancer cells, where HIF-1α is overexpressed—a feature associated with increased tumor aggressiveness and likely contributing to the recruitment of several G4RPs⁶⁶.

Interestingly, 14 proteins were shared among all four G4-forming sequences. Among them, several known G4 binders, including the aforementioned LMNB1 and vimentin (VIM), a protein known for its binding to multiple G4 structures^26,27,28. Additionally, 8 of the 17 ribonucleoproteins here identified acted as G4RPs across all four G4-forming sequences. Among them, to the best of our knowledge, HNRNP AB, HNRNP DL, and HNRNP M were newly identified as G4RPs in this study. Other newly identified interactors include prelamin A/C (LMNA), crucial for nuclear assembly⁸², chromatin organization⁸³, and telomere dynamics⁸⁴; neuroblast differentiation-associated protein (AHNAK), essential for neuronal differentiation; peptidyl-prolyl cis-trans isomerase A (PPIA), known to catalyze the cis-trans isomerization of proline imidic peptide bonds; and glyceraldehyde-3-phosphate dehydrogenase (GAPDH), known for its involvement in various nuclear functions including transcription. In summary, we identified seven proteins with potential G4-binding capabilities across the promoter regions of the four cancer-related genes studied: AHNAK, GAPDH, HNRNP AB, HNRNP DL, HNRNP M, LMNA, and PPIA.

Validation of protein interaction with G4-forming sequences in vitro

To validate the direct interaction between selected proteins and the G4-forming sequences from DAP, HIF-1α, JAZF-1, and PDGF-A, surface plasmon resonance (SPR) analysis was performed. SPR experiments, performed using the single-cycle kinetics method, indicated specific interactions, as shown by the representative SPR sensorgrams obtained for the interaction of the G4-forming sequences with the immobilized proteins (Fig. 4 and Supplementary Fig. S9). Equilibrium dissociation constants (K_d) obtained from curve fitting revealed affinities in the nanomolar range (Table 2). Notably, CD analysis confirmed that these interactions take place without consequent unwinding of the G4 structures (Supplementary Fig. S10).

**Fig. 4: SPR sensorgrams for protein interaction with the G4-forming DNA sequences.**

Table 2 Dissociation constants (K_d) for the interaction of proteins with the investigated G4-forming sequences, determined by SPR experiments

Full size table

AHNAK and PPIA —two proteins not previously associated with nucleic acids interactions due to their lack of specific RNA/DNA recognition motifs— were found to bind directly to DNA G4s, with K_d values ranging from 8 to 37 nM. Both proteins showed the highest binding affinity for PDGF-A, with K_d values of 8 ( ± 4) for AHNAK and 15 ( ± 8) nM for PPIA. AHNAK also exhibited strong binding to HIF-1α, followed by JAZF-1, with K_d values comparable to PPIA for these sequences. These findings are particularly significant in light of prior studies linking PPIA to the assembly of TAR DNA-binding protein 43 (TADBP) in HNRNP complexes⁸⁵. Notably, TADBP was identified in this study as a putative specific interactor of the JAZF-1 G4 DNA.

HNRNPs are a group of proteins characterized by similar structural domains that play essential roles in various cellular functions, including DNA repair, telomere biogenesis, cell signaling, and the regulation of gene expression at both transcriptional and translational levels. Additionally, they are critical for processes such as angiogenesis, cell invasion, and inhibition of apoptosis, all of which are implicated in tumor development and progression⁸⁶. Among these, HNRNP AB, HNRNP DL, and HNRNP M identified here, each contains at least two RNA recognition motifs (RRMs). Here, we showed that the RRM1 motif of HNRNP M interacts with DNA G4-forming sequences. Interestingly, RRM1 displayed high affinity for PDGF-A and HIF-1α [K_d = 10 ( ± 2) and 26 ( ± 16) nM, respectively] but significantly lower affinity for DAP and JAZF-1 [K_d = 179 ( ± 110) and 342 ( ± 58) nM, respectively], which are the sequences with the highest levels of structural heterogeneity.

As for GAPDH protein, its presence in the nucleus suggests a potential role as a DNA-binding protein, a hypothesis further supported by evidence of its strong binding to nuclear DNA⁸⁷. Interestingly, in this study, GAPDH exhibited the strongest average affinity for G4s among the proteins investigated, with K_d values ranging from 10 nM for PDGF-A to 26 nM for JAZF-1.

Lamins, key components of the nuclear lamina, interact with specific genomic regions known as lamina-associated domains (LADs). LMNB1 is localized at the nuclear periphery where it associates with heterochromatin and plays a pivotal role in chromatin restructuring during the epithelial-to-mesenchymal transition (EMT). It binds G-rich promoters of EMT-related genes, facilitating their transcriptional activation⁸⁸. In contrast, LMNA is distributed throughout the nucleus, interacts with chromatin-binding protein lamina-associated polypeptide (LAP)2α and plays roles in chromatin organization and gene expression⁸⁹. Prior studies showed that LMNB1 binds the G4-forming human telomeric sequence²⁸, which predominantly folds into a parallel-stranded G4 under molecular crowding conditions. Here, SPR analysis revealed that LMNA strongly interacts with all investigated G4-forming sequences, with K_d values ranging from 9 to 44 nM, and HIF-1α emerging as the preferred target [K_d = 9 ( ± 3) nM]. Finally, to evaluate the proteins’ selective recognition of G4 secondary structures, we assessed their interactions with the unstructured G-rich GT15 strand and a non-G-rich DNA sequence (T₃₀) (Supplementary Fig. S11). Notably, none of the investigated proteins exhibited significant binding to T₃₀. As for GT15, the only relevant interaction was observed for PPIA, with a K_d value of around 31 nM, comparable to its affinity for HIF-1α. This suggests that PPIA may selectively recognize G-rich DNA strands in both folded and unfolded states.

Conclusions

G4 DNA structures play pivotal roles in regulating various biological processes, including gene expression. In this context, the recognition of G4s by nucleic acid-directed proteins represents a crucial event in modulating physiological and pathological pathways. Identifying these G4-binding proteins is essential to clarify their actual role in such processes.

Motivated by the significance of these interactions, in this study, we employed a classical fishing-for-partners proteomic approach to identify putative interactors of G4-forming DNA sequences from the promoter regions of the DAP, HIF-1α, JAZF-1, and PDGF-A genes. After confirming that these DNA molecules fold into G4 structures, forming a mixture of conformations representative of natural G4 conformational diversity, biotinylated G4s were synthesized and used as baits in solution to identify putative G4RPs in U2OS cancer cells. Noteworthy, 86 G4RPs were identified, 19 of which were already reported as RNA and/or DNA G4 interactors. Among these, LMNB1 and KHSRP proteins previously identified as telomeric G4 interactors, some helicases, and several heterogeneous nuclear ribonucleoproteins. Interestingly, 14 proteins (8 of which were ribonucleoproteins) potentially interact with all four investigated G4-forming sequences. Among these, 7 were identified as G4RPs for the first time: AHNAK, GAPDH, HNRNP AB, HNRNP DL, HNRNP M, LMNA, and PPIA.

The direct interaction of five of them (AHNAK, GAPDH, HNRNP M, LMNA, and PPIA) with the investigated G4s was validated in vitro by performing SPR experiments. The results of these experiments indicated high-affinity interactions between the proteins and the G4-forming sequences. Notably, AHNAK and PPIA, which were not previously associated with nucleic acid binding, demonstrated strong affinity for G4s, particularly PDGF-A. HNRNP M’s RRM1 motif and LMNA also emerged as G4 interactors, with preferences for PDGF-A and HIF-1α, respectively. Finally, GAPDH exhibited the strongest overall binding to G4s, underscoring its potential nuclear role. The selective recognition of G4s over non-G-rich or unstructured DNA further supported the specificity of these interactions, with PPIA uniquely recognizing both folded and unfolded G-rich strands.

Overall, this study lays the groundwork for further research aimed at elucidating the biological significance of these interactions and deepen our understanding of the mechanisms by which G4s regulate gene transcription.

Methods

Oligonucleotide synthesis and sample preparation

DNA sequences were synthesized on an ABI 394 DNA/RNA synthesizer (Applied Biosystem, Foster City, CA, USA) using standard β-cyanoethyl phosphoramidite solid phase chemistry at 1 μmol synthesis scale. Deprotection and detachment were performed by using a concentrated NH₄OH aqueous solution at 55 °C for 12 h. DNA filtrates and washings were combined and concentrated under reduced pressure, solubilized in water, and then purified by high-performance liquid chromatography using an anionic exchange column (Nucleogel SAX, Macherey-Nagel, 1000-8/46) as previously reported⁹⁰. The purified fractions of the oligomers were then desalted by using Sep-Pak cartridges (C-18). The biotinylated oligonucleotides were purchased from Biomers.net (Ulm, Germany). All DNA samples were prepared in the appropriate buffer solutions, and their concentration measured by UV adsorption at 90 °C using the appropriate molar extinction coefficient values, ε (λ = 260 nm), calculated by the nearest-neighbor model⁹¹. Samples were annealed by heating at 90 °C per 5 min followed by slow cooling to room temperature and then stored at 4 °C for at least 24 h before performing experiments.

Circular dichroism (CD) experiments

CD experiments were carried out on a Jasco J-815 spectropolarimeter (JASCO Inc., Tokyo, Japan) equipped with a PTC-423S/15 Peltier temperature controller using a quartz cuvette with a path length of 1.0 cm. Biotin-labeled and unlabeled oligonucleotides were prepared at 50 µM as described above in 20 mM HEPES buffer (pH 6.6) containing 150 mM KCl, and 1 mM EDTA. CD spectra were recorded between 230-360 nm at 100 nm min^-1 scan speed, 1.0 nm bandwidth, 0.5 s response, and 3 accumulations. CD spectra for the proteins and corresponding mixtures with HIF-1α in 20 mM Tris buffer (pH 6.0) containing 150 mM KCl, and 1 mM EDTA were recorded between 200-320 nm at 50 nm min^-1, 2.0 nm bandwidth, 2.0 s response, and 10 accumulations. CD melting experiments were performed on diluted samples at 3 µM, carried out in the 20-100 °C temperature range with a heating rate of 1 °C min^-1, by following changes of the CD signal at the wavelengths of maximum CD intensity (i.e., 264 nm). All measurements were performed at least in duplicate.

Thioflavin T (ThT) and N-methylmesoporphyrin IX (NMM) fluorescence assays

Experiments were conducted in a 1 cm path-length cell at 20 °C on a Jasco FP-8300 spectrofluorometer (Jasco, Easton, MD, USA) equipped with a PCT-818 Peltier cell holder. For the ThT experiments, a stock solution (ca. 300 μM) of ThT was prepared in water and the concentration was determined using the molar extinction coefficient of 36 000 M^-1 cm^-1 at 412 nm⁵¹. A solution of DNA/ThT in a 2:1 ratio was prepared in 20 mM HEPES buffer (pH 6.6) containing 150 mM KCl, and 1 mM EDTA and allowed to equilibrate in the dark for 30 min at room temperature. Fluorescence emission spectra were recorded in the 440-650 nm range at 100 nm min^-1 scan speed, using the excitation wavelength of 420 nm and setting both the excitation and emission slits at 5 nm. As for the NMM assays, stock solution (ca. 60 μM) of the light-up probe was prepared in water and the concentration was determined using the molar extinction coefficient of 145 000 M^-1 cm^-1 at 379 nm⁵⁴. A solution of DNA/NMM in a 10:1 ratio was prepared in 20 mM HEPES buffer (pH 6.6) containing 150 mM KCl, and 1 mM EDTA and allowed to equilibrate in the dark for 5 min at room temperature. Fluorescence emission spectra were recorded in the 550-700 nm range at 100 nm min^-1 scan speed, using the excitation wavelength of 399 nm and setting both the excitation and emission slits at 5 nm. The results were reported as fluorescence intensity enhancement (FI/FI₀) of ThT or NMM at 487 nm and 609 nm, respectively, where FI is the fluorescence of the light-up probe in the presence of DNA and FI₀ is the background fluorescence of the probe alone, after subtraction in both cases of the buffer fluorescence. All measurements were performed in triplicate from distinct samples. Data shown as Mean ± SD.

Non-denaturing polyacrylamide gel electrophoresis

Non-denaturing 15% PAGE gels were prepared with 29:1 acrylamide/bisacrylamide solution and Tris-Borate-EDTA (TBE 1×), pH 8.4. Gels were run at 4 °C and 100 V for 90 min. Oligonucleotide samples annealed at 50 µM concentration in 20 mM HEPES buffer at pH 6.6 containing 150 mM KCl, and 1 mM EDTA were loaded. A solution of glycerol/TBE was added (10% final) to facilitate sample loading into the wells. Bands were visualized by UV shadowing at 254 nm.

Nuclear magnetic resonance (NMR) experiments

1D ¹H NMR spectra were recorded on a Bruker Advance NEO NMR spectrometer (Bruker BioSpin, Rheinstetten, Germany), operating at 600 MHz (¹H Larmor frequency) and equipped with a 5-mm QCI cryo-probe set and a cooled SampleJet autosampler. DNA samples (50 μM in 20 mM HEPES buffer at pH 6.6 containing 150 mM KCl, and 1 mM EDTA, or 5 mM KCl) were transferred into 5-mm NMR tubes, and 10% D₂O was added. Spectra were acquired from 5 to 90 °C, using excitation sculpting with gradients for water suppression⁹². All experiments were performed using 512 scans per spectrum, with a recovery delay of 1.5 s. The free induction decays were multiplied by an exponential function equivalent to a line-broadening factor of 1.0 Hz before Fourier transformation. The transformed spectra were phase adjusted, baseline corrected and calibrated against the sodium trimethylsilylpropanesulfonate (DSS) signal as an external reference. Spectra were processed with the Bruker TopSpin 4.3.0 software package and analyzed with MestReNova software.

Cell culture

U2OS (osteosarcoma, ATCC) cells were grown in Dulbecco’s Modified Eagle Medium (DMEM) supplemented with 10% fetal bovine serum, 100 U mL^-1 penicillin, and 100 μg mL^-1 streptomycin. Cells were sub-cultured at 90% confluence in the split ratio 1:4, every 3–5 days, and maintained at 37 °C in a humidified atmosphere containing 5% CO₂.

Nuclear proteins extraction

U2OS cells were grown in 10 cm Petri dishes, at 37 °C, in a 5% CO₂—95% air atmosphere. Once 90% confluence was reached, cells were washed and scraped into PBS to be then pelleted by centrifugation (2000 rpm, 3 min). After removing the supernatant, the cellular pellet was re-suspended in a cold hypotonic buffer (10 mM HEPES pH 7.9, 10 mM KCl, 0.1 mM EDTA pH 8.0, 0.1 mM EGTA pH 8.0, 0.5% NP-40, 0.5 mM phenyl-methyl-sulfonyl fluoride (PMSF), 1 mM dithiothreitol (DTT), 1× protease inhibitor cocktail) and left on ice for 15 min. Afterwards, nuclei were pelleted by centrifugation (2000 rpm, 5 min, 4 °C), re-suspended in a cold high-salt buffer (20 mM HEPES pH 7.9, 150 mM NaCl, 1 mM EDTA pH 8.0, 1 mM EGTA pH 8.0, 0.5 mM PMSF, 1 mM DTT, 1x protease inhibitor cocktail), left on ice for 15 min, and then disrupted by sonication using the Bioruptor® Plus device (Diagenode) with high power mode, for 5 cycles, at 4 °C (sonication cycle: 30 s ON, 30 s OFF). Insoluble cell matter was removed by centrifugation (13,000 rpm, 30 min, 4 °C). Released nuclear proteins were quantified according to the Bradford method.

G4RPs fishing protocol

Nuclear lysates (ca. 200 µg) of U2OS cancer cells were stored at -80 °C in 20 mM HEPES, 150 mM NaCl, 1 mM EDTA, and 1 mM EGTA (pH 7.9). DNA solutions were prepared as described above in 20 mM HEPES buffer (pH 6.6) containing 150 mM KCl, and 1 mM EDTA at 0.2 mg mL^-1. Before use, the pH of nuclear lysates was adjusted to 6.6 value, followed by incubation with biotin-GT15 for 20 min at 4 °C. Next, 1 mg of Dynabeads™ M-280 Streptavidin (10 mg mL^-1, Thermo Fisher) was added to the mixtures and incubated for 30 min at 4 °C. The supernatant, enriched with putative G4 selective proteins, was washed away and further incubated first with the biotin-G4 structure for up to 30 min at 4 °C, and then with 1 mg of beads in the same conditions. The supernatant, containing the unbound proteins, was washed away. Both G-rich strand and G4s bound fractions were eluted in denaturing conditions in 6 M urea, 300 mM Tris, and 10 mM EDTA buffer for 5 min at 70 °C.

In solution digestion

For the proteomics analysis, eluates recovered from G4s and GT15 fishing experiments were subjected to an in-solution digestion protocol. First, the cysteines were reduced using a 20 mM dithiothreitol (DTT) solution in 25 ammonium bicarbonate (AMBIC), followed by incubation at 60 °C for 1 h. Samples were then cooled, and the reduced cysteines were alkylated with a 40 mM iodoacetamide (IAM) solution in 25 mM AMBIC for 45 min at room temperature in the dark. A solution of 100 mM DTT was then added to the reaction to quench the alkylation process, and the mixture was left for 1 h at room temperature. The protein mixture was diluted with 50 mM AMBIC to a final concentration of 1 M urea prior to enzymatic digestion. An aliquot of 10 µL of trypsin 0.1 µg µL^-1 was added to each sample in a 1:50 enzyme:substrate ratio. The enzymatic hydrolysis was performed overnight at 37 °C in a thermostatic bath. Each sample was treated with the same desalting procedure using stage tips containing 3 layers of 3 M Empore C18 membrane. Stage tips were washed with 100 µL of 0.1% formic acid (HCOOH) and peptides were eluted with 50 µL of 50% acetonitrile (ACN) and, subsequently, with 80% ACN, both acidified with 0.2% HCOOH. The peptide mixture was dried in a vacuum Speed-Vac and re-suspended in 50 µL of a solution containing 5% ACN and 0.2% HCOOH.

LC-MS/MS analysis

The peptide mixture was analyzed by LC-MS/MS on an LTQ Orbitrap XL system (Thermo Fisher) coupled with the nanoACQUITY UPLC system (Waters). Peptides were fractionated onto a C18 reverse-phase capillary column (250 mm, 75 μm, 1.8 μm, Waters Bioseparation Technology) at a 250 nL min^-1 flow rate. A linear gradient from 10% to 60% of eluent B (0.2% HCOOH, 95% ACN, LC-MS Grade) was used over 80 min. Parameters of electrospray ion sources were: electrospray voltage at 3500 V, capillary voltage at 45 V, capillary temperature at 300 °C. Mass spectrometric analyses were carried out in Data Dependent Acquisition mode (DDA) spanning from 300 to 1800 m/z in positive ion mode. From each MS scan, the five most abundant ions were selected and fragmented in the collision-induced dissociation (CID) modality in the linear ion trap: normalized collision energy 25-35%.

Proteomics data analysis

Raw data files were processed using MaxQuant software (1.6.8.0 version)⁹³. The following parameters were applied for raw data processing: trypsin enzyme specificity, 3 missed tryptic cleavages, oxidation of methionine, formation of pyroGlu from N-terminal glutamine (Q) as variable modifications, and carbamidomethylation of cysteine (C) as a fixed modification. Identification criteria included a minimum peptide length of 6 amino acids and a minimum of 1 peptide (both razor and unique peptide), with a peptide tolerance of 10 ppm, and a fragment mass tolerance of ± 0.2 Da. All proteins were filtered based on a false discovery rate (FDR) of 0.01%, applied at both peptide and protein levels, and a maximum peptide posterior error probability (PEP) of 1. The peak list derived from Quant.exe (the initial step of MaxQuant) was searched using the Andromeda search engine integrated into the MaxQuant against the Homo sapiens FASTA file downloaded from the UniProt website. MaxQuant output files were subsequently processed using Perseus (version 1.6.8.0) software platforms⁹⁴. An experimental design template was used to merge replicate experiments (each data set contained two technical replicates) into a single column containing all the proteins in each sample. Contaminants, reverse, and peptides identified only by site hits were filtered out. Expression values of LFQ intensity were then log2 transformed, and only the protein rows containing a minimum of 1 valid value, were maintained within the Perseus matrix.

Surface plasmon resonance (SPR) experiments

SPR experiments were performed at 25 °C using a Biacore X100 (GE Healthcare) equipped with a research-grade CM5 sensor chip. AHNAK (TP315337, OriGene Technologies GmbH, Germany), GAPDH (TP302309, OriGene Technologies GmbH), HNRNP M (orb1784775, Biorbyt, UK), LMNA protein (TP304970, OriGene Technologies GmbH), and PPIA (orb706916, Biorbyt) were immobilized using amine-coupling chemistry and HBS-EP as running buffer (10 mM HEPES buffer at pH 7.4, containing 150 mM NaCl, 3 mM EDTA, and 0.005% Surfactant P20). The surfaces of flow cells were activated with a 1:1 mixture of 0.1 M N-hydroxysuccinimide (NHS) and 0.1 M 3-(N,N-dimethylamino)propyl-N-ethylcarbodiimide (EDC) at a flow rate of 10 μL min^-1. The proteins diluted to a 50 μg mL^-1 concentration with 10 mM sodium acetate at pH between 4.0 and 4.7, were immobilized at a density up to 3,000 response unit (RU) on the sample flow cell, leaving the reference cell blank. Unreacted activated groups were blocked by the injection of 1.0 M ethanolamine at 10 μL min^-1 over the chip surface. Samples were prepared at 50 μM single-strand concentration in 20 mM TRIS buffer at pH 6.0 containing 150 mM KCl, and 1 mM EDTA. Kinetic binding data were collected by using the single-cycle kinetics approach⁹⁵. Oligonucleotides were injected sequentially in the same cycle from a low to high concentration (from 0.06 to 1 µM, except for LMNA protein from 0.3 to 5 μM) with an association time of 30 s and a dissociation time of 600 s at the end. Injections were performed at a flow rate of 30 μL min^-1 using 20 mM TRIS buffer at pH 6.0 containing 150 mM KCl, and 1 mM EDTA as running solution. No regeneration after each sample was required. Data were fitted to a simple 1:1 interaction model, using the global data analysis option available within the BIAevaluation software provided with the device. All measurements were performed at least in duplicate from distinct samples. Data shown as mean ± SD.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.