Endogenous promoter G-quadruplexes scaffold apurinic/apyrimidinic endonuclease (APE1) to drive gene expression

Abstract

G-quadruplexes (G4s) are non-canonical DNA secondary structures enriched at promoters, yet their regulatory role in transcription remains elusive. While G4-ligand-based studies suggest transcriptional repression, their prevalence at oncogene promoters and correlation with high expression suggest a positive regulatory role. Here, we provide direct genetic evidence that promoter G4s function as positive activators of gene expression through a novel mechanism. By selectively mutating endogenous promoter G4 motifs, we demonstrate that G4 loss significantly impairs oncogene expression. Using the endogenous CXCL1 promoter G4 as an example, we revealed that loss of a single promoter G4 motif led to a marked down-regulation of CXCL1 expression as well as inhibition of cellular functions such as cell migration and invasion. Mechanistically, we identified apurinic/apyrimidinic endonuclease (APE1), a multifunctional DNA repair and redox factor, as a G4-binding protein which was recruited to promoters via its unique N-terminus. Subsequently, the redox activity of APE1 enhances transcription factor binding at G4-containing promoters, driving a pro-metastatic gene expression program. Disruption of the G4–APE1 interaction, either genetically or pharmacologically, suppresses gene expression and impairs tumor cell malignant traits. Our findings establish a direct genetic link and mechanistic basis for promoter G4s as crucial drivers of oncogene expression and tumor progression.

Graphical Abstract

Graphical Abstract.

Introduction

G-quadruplexes (G4s) are non-canonical four-stranded DNA secondary structures formed by the stacking of guanine tetrads in guanine-rich sequences [1, 2]. Genome-wide chromatin immunoprecipitation sequencing- (ChIP-seq) related high-throughput analyses have revealed that G4s are enriched in promoters, enhancers, and 3′-untranslated regions (3′-UTRs) of actively transcribed genes [3–5]. Notably, G4s are preferentially located in the promoters of oncogenes including MYC, KRAS, VEGF, PDGF, c-KIT, and hTERT, but are largely absent from tumor suppressors and housekeeping genes, suggesting a potential regulatory role in transcriptional activation of genes to promote cancer [6–8]. Accumulating genome-wide evidence supports a model in which G4s may enhance transcription by stabilizing R-loop formation [9], facilitating transcription factor (TF) binding [10–13] and promoter enhancer looping [14]. However, some G4-stabilizing small molecule studies have yielded conflicting interpretations. For example, G4 ligands such as TMPyP4 and pyridostatin (PDS) have been reported to suppress c-MYC and KRAS expression, leading to the idea that stabilizing G4s down-regulate gene expression, therefore, G4s function as transcriptional repressors [15, 16]. Recent studies have elegantly demonstrated that selective deletion of the potential G4-forming sequence (PQS) within the c-MYC and β-globin gene promoters resulted in a marked loss of transcriptional activity, pointing to an activator function for G4 [17, 18]. The inhibitory effect of G4-stabilizing ligands on transcription is likely to be attributable to displacing endogenous G4-binding proteins by competing for the overlapping or adjacent binding sites. Indeed, a study has demonstrated a decreased G4-specific BG4 antibody accessibility at the MYC promoter but increased accessibility at the PVT1 promoter following targeted PDS delivery, while another G4 ligand, PhenDC3, exerted opposite effects at PVT1, suggesting that transcriptional outcomes of G4 targeting are highly ligand and promoter context dependent [19]. Therefore, the molecular mechanisms by which an individual G4 functions and responds to G4-targeting ligands within its native chromatin context for gene expression are not fully understood.

We have recently demonstrated that apurinic/apyrimidinic endonuclease (APE1), a key enzyme in the base excision repair (BER) pathway, binds G4 DNA with high affinity [20–23]. Beyond its DNA repair function, APE1 also serves as a redox effector (Ref-1), enhancing the DNA binding affinity of numerous TFs such as AP-1 (c-Jun/c-Fos), p53, NF-κB, HIF1-α, and STAT3 by facilitating thiol exchange reactions involving conserved cysteine residues (Cys65 and Cys93), thereby activating target gene transcription [24–27]. G4 DNA levels are elevated under multiple pathophysiological conditions including in cancer [23, 28]. Notably, genome-wide profiling of G4s in patient-derived xenograft models revealed a high density of G4 structures in promoter and enhancer regions of signature genes associated with triple-negative breast cancer (TNBC), the most aggressive and metastatic breast cancer subtype [29], where the APE1 level is also highly elevated [30–32]. Nevertheless, the mechanistic relationship between promoter G4s, APE1, and transcriptional regulation in TNBC has not been defined.

Here, we use CRISPR/Cas9 [clustered regularly interspaced palindromic repeats (CRISPR)/CRISPR-associated protein 9] to disrupt G4 motifs in the endogenous gene promoters, and demonstrate that promoter G4 loss in the CXCL1 gene, encoding a chemokine crucial for TNBC cell growth and metastasis [33], abolishes APE1 recruitment, suppresses CXCL1 expression, and impairs growth and migration of TNBC cells. Transcriptome profiling revealed that APE1 regulates pro-survival and pro-metastatic gene programs. Mechanistically, we identified that promoter G4s act as scaffolds to recruit APE1 to gene promoters, which in turn enhances TFs occupancy at G4-enriched promoters via its Ref-1 function. We further identify the intrinsically disordered N-terminal domain of APE1, unique to mammalian APE1, as essential for direct G4 binding. Many G4-binding ligands, including TMPyP4, PDS, and PhenDC3, disrupt APE1–G4 interaction in vitro and in cells, and suppress gene expression, tumor growth, and metastatic dissemination in vivo. Together, our study provides direct genetic evidence and a new mechanism by which endogenous promoter G4 acts as a positive regulator to activate gene expression, and underscores its roles in driving tumor growth and metastasis in vivo.

Materials and methods

Cell lines

MDA-MB-231 (HTB-26) and BT-549 (HTB-122) TNBC cell lines were purchased from the ATCC. Cell line authentication was confirmed via short tandem repeat DNA profiling by Genetica DNA Laboratories (Burlington, NC, USA) prior to use. MDA-MB-231 cells were cultured in α-minimal essential medium (MEM; Gibco) supplemented with 10% fetal bovine serum (FBS; Sigma) and 1% (v/v) penicillin–streptomycin (Sigma). BT-549 cells were maintained in α-MEM containing 10% FBS, 1% (v/v) penicillin–streptomycin, 12.5 ng ml⁻¹ epidermal growth factor (EGF; PeproTech), and 1 μg ml⁻¹ hydrocortisone (Sigma). All cell cultures were maintained at 37°C in a humidified incubator with 5% CO₂ and were routinely tested for mycoplasma contamination.

CRISPR/Cas9 gene editing

Guide RNAs (gRNAs) targeting APE1, C-X-C motif chemokine ligand 1 (CXCL1), and vascular endothelial growth factor A (VEGFA) were designed using Benchling software and cloned into the pLenti-Cas9-GFP plasmid (Addgene #86 145). For APE1 knockout, plasmids expressing APE1-targeting gRNAs and Cas9 were transfected into TNBC cells using Lipofectamine 3000 (Invitrogen). Green fluorescent protein (GFP)-positive cells were sorted by fluorescence-activated cell sorting (FACS) at the UNMC Flow Cytometry Research Facility. Single-cell colonies were expanded, and APE1 knockout was confirmed via western blotting.

To mutate the G4 sequence in the CXCL1 promoter, CRISPR plasmids containing CXCL1 G4-targeting gRNAs and Cas9 were co-transfected with a pUC19 donor template plasmid carrying the desired G4 mutations. GFP-positive cells were single-cell cloned, and genomic DNA was isolated. Sanger sequencing confirmed the presence of the intended mutations. The same strategy was used to generate VEGFA G4-mutated cell lines using VEGFA-specific gRNAs and donor templates. APE1-gRNA, 5′-TAACGGGAATGCCGAAGCGT-3′; CXCL1-G4-gRNA, 5′-ACCCCAAGCGCTCCACCCTG-3′; VEGFA-G4-#1-gRNA, 5′-TCGCCTGTCCCCGCCCCCCG-3′; and VEGFA-G4-#2-gRNA, 5′-AGAGCCGGACAGGGACGGGT-3′.

Ectopic APE1 expression

Coding sequences of wild-type (WT) APE1 and its mutants were cloned into the doxycycline (Dox)-inducible pCW57 lentiviral vector (Addgene #80 921). APE1-knockout TNBC cells were transduced with the lentiviruses and selected with 40 μg ml⁻¹ blasticidin (AG Scientific). Dox (TaKaRa) was used to induce APE1 expression.

Western blotting

Cell fractionation and western blotting were performed as described previously [34]. A 20 μg aliquot of total protein extract per sample was loaded. Primary antibodies used included: APE1 (Novus, Cat# NB100-101), Snail (Cell Signaling Technology, Cat# 3879), c-Jun (Cell Signaling Technology, Cat# 9165), and HSC70 (Santa Cruz Biotechnology, Cat# sc-7298). Blots were developed using horseradish peroxidase (HRP)-conjugated secondary antibodies (Jackson ImmunoResearch) and the Luminate Crescendo detection system (Thermo Scientific).

In vitro migration and invasion assays

Cells were serum-starved in medium containing 0.5% (v/v) FBS 24 h prior to the assay. For the migration assay, 20 000 cells per well in low-FBS medium were seeded into 8.0 μm transparent PET membrane inserts (Falcon) placed in 24-well plates. The lower chamber contained 750 μl of complete medium supplemented with 10% FBS as a chemoattractant. For the invasion assay, 60 000 cells per well were seeded into Matrigel-coated invasion chambers (Corning) under the same serum-starved conditions. After 16–24 h of incubation, non-migrated cells were removed, and cells that had migrated or invaded through the membrane were fixed and stained with 0.5% crystal violet (Sigma) prepared in 100% methanol (Fisher Chemicals). Images were captured using an EVOS FL Auto microscope.

Tumorsphere formation assay

A total of 3000 cells per well were seeded in 96-well ultra-low attachment plates (Corning) and cultured in Dulbecco’s modified Eagle’s medium (DMEM)/F12 medium (Gibco) supplemented with 20 ng ml⁻¹ recombinant human EGF, 20 ng ml⁻¹ CXCL1, recombinant human basic fibroblast growth factor (bFGF; R&D Systems), 1× B27 supplement (Gibco), 10215 N2 supplement (Gibco), 4 μg ml⁻¹ heparin, and 5% growth factor-reduced Matrigel (Corning). On Day 7, tumor spheres with diameters >70 μm were counted per condition. Images were taken using an EVOS FL Auto microscope.

Cell viability assay

Cells (2000 per well) were seeded in 96-well plates (Thermo Scientific). At 24, 48, 72, and 96 h post-seeding, cell viability was measured using the CCK-8 assay (Dojindo). Cells were incubated with CCK-8 reagent at 37°C for 1 h, and absorbance was measured at 450 nm using a microplate reader (SpectraMax M5).

Quantitative real-time PCR

Total RNA was extracted using the RNeasy Mini Kit (QIAGEN). A 1 μg aliquot of RNA was reverse transcribed into cDNA using the Superscript VILO cDNA Synthesis Kit (Invitrogen). For quantitative PCR (qPCR), 2 μl of cDNA was used in a 10 μl reaction with SYBR Green Master Mix (Applied Biosystems) and run in triplicate. Gene expression was normalized to glyceraldehyde phosphate dehydrogenase (GAPDH) using the ΔΔCt method relative to the control sample. CXCL1-Fwd, 5′-TCCTGCATCCCCCATAGTTA-3′; CXCL1-Rev, 5′-CTTCAGGAACAGCCACCAGT-3′; VEGFA-Fwd, 5′-GAGGGCAGAATCATCACGAAG-3′; VEGFA-Rev, 5′-TGTGCTGTAGGAAGCTCATCTCTC-3′; phosphoinositide-3-kinase regulatory subunit 3 (PIK3R3)-Fwd, 5′-GAGAGGGGAATGAAAAGGAGA-3′; PIK3R3-Rev, 5′-ATCATGAATCTCACCCAGACG-3′; BMI1-Fwd, 5′-CTGGTTGCCCATTGACAGC-3′; BMI1-Rev, 5′-CAGAAAATGAATGCGAGCCA-3′; APE1-Fwd, 5′-TGGAATGTGGATGGGCTTCGAGCC-3′; APE1-Rev, 5′-AAGGAGCTGACCAGTATTGATGA-3′; GAPDH-Fwd, 5′-TGGGCTACACTGGAGCACCAG-3′; and GAPDH-Rev, 5′-GGGTGTCGCTGTTGAAGTCA-3′.

Immunofluorescence

Cells grown on coverslips (Fisherbrand) were fixed in 4% (v/v) paraformaldehyde (Sigma) for 20 min and permeabilized with 0.5% Triton X-100 (Sigma) for 20 min at room temperature. For G4 structure staining, cells were treated with 40 μg ml⁻¹ RNase A (New England BioLabs) for 20 min at 37°C, followed by blocking with 10% goat serum for 1 h. Primary anti-G4 antibody (Sigma-Aldrich, Cat# MABE1126) was applied and incubated overnight at 4°C. Other primary antibodies include APE1 (Novus, Cat# NB100-101). The following day, Alexa Fluor 594-conjugated goat anti-mouse secondary antibody (Invitrogen, Cat# A11005) was added for 1 h at room temperature in the dark. Coverslips were mounted using 4′,6-diamidino-2-phenylindole (DAPI)-containing mounting medium (Vector Laboratories), and images were acquired using a Zeiss confocal microscope.

Proximity ligation assay

Cells grown on coverslips were fixed with 4% (v/v) formaldehyde and permeabilized using 0.2% Triton X-100. Following permeabilization, cells were treated with 0.04 μg μl⁻¹ RNase A and blocked using the Duolink blocking solution. Proximity ligation assay (PLA) was performed using the Duolink® In Situ Red Starter Kit (Mouse/Rabbit; Sigma-Aldrich, DUO92101), following the manufacturer’s instructions. Primary antibodies included anti-G4 (Sigma, Cat# MABE1126), anti-APE1 [35], and anti-Bcl-2 (Santa Cruz Biotechnology, Cat# sc-7382). Images were captured using a Zeiss confocal microscope and processed with ZEN 2.3 SP1 software.

Chromatin immunoprecipitation and qPCR

Chromatin immunoprecipitation (ChIP) was performed as previously described [35]. Briefly, cross-linked chromatin was sonicated for 20 cycles (30 s on/ 60 s off) using a sonicator (Bioruptor, Diagenode). After pre-clearing, immunoprecipitations were conducted overnight at 4°C with antibodies against APE1 (Novus, Cat# NB100-101) or acetylated APE1 [35], Snail (Cell Signaling Technology, Cat# 3879), or c-Jun (Cell Signaling Technology, Cat# 9165). G4 ChIP was conducted using a well-established protocol with the BG4 antibody (Sigma, Cat# MABE917) and anti-FLAG M2 magnetic beads (Sigma, M8823).

For qPCR analysis, Ct values from 10% input were normalized by subtracting the dilution factor (log₂10), and enrichment was calculated by comparing IP versus input. IgG controls were included for all experiments. VEGFA-G4-Fwd, 5′-AGAAGGCCAGGGGTCACTC-3′; VEGFA-G4-Rev, 5′-TGAGAGCCGTTCCCTCTTTG-3′; CXCL1-G4-Fwd, 5′-ATCTGGGGCAGAAGGCGAAT-3′; CXCL1-G4-Rev, 5′-GGCCAGGGAAATTCCCGGAG-3′; Snail1-Fwd, 5′-AGAAGGAAGATAAGCAGGATATG-3′; Snail1-Rev, 5′-TGGTGGCAACAGATTATCAG-3′; c-Jun-Fwd, 5′-TTCGAGAGTGAGGACGTGT-3′; and c-Jun-Rev, 5′-ACACAGATCTATTGGAATCCTGG-3′.

Circular dichroism spectroscopy

Circular dichroism (CD) spectroscopy was conducted as previously described [23]. G4 oligonucleotides corresponding to putative G4-forming sequences from the promoters of CXCL1, VEGFA, and PIK3R3 were synthesized by Integrated DNA Technologies (IDT). CXCL1-G4 oligo, 5′-TTGGGGTGCGGGTGGGGTGTGGGGGTCC-3′; CXCL1-G4-Short Loops, 5′-GGGCGGGTGGGTGGGT-3′; VEGFA-G4-1# oligo: 5′-CCGGGGCGGGCCGGGGGCGGGGTCC-3′; VEGFA-G4-1#-Short Loops: 5′-GGGCGGGCGGGCGGGT-3′; PIK3R3-G4 oligo, 5′-GCGGGTGGGCAGCATGGGGCGGGGAGGGTGT-3′; and Telomere G4 oligo, 5′-TTAGGGTTAGGGTTAGGGTTAGGGTT-3′.

Enzyme-linked immunosorbent assay

Biotin-labeled DNA oligonucleotides were diluted in buffer [50 mM Tris–HCl pH 7.5, 2 mM MgCl₂, 1 mM dithiothreitol (DTT), 0.2 mM EDTA, 50 mM KCl] and 50 nM of each oligo was added to streptavidin-coated 96-well plates (Thermo Scientific). Plates were incubated for 30 min at room temperature and washed three times with wash buffer (the same buffer with 0.1% NP-40 and 10% glycerol). Plates were then blocked with 3% bovine serum albumin (BSA). Recombinant APE1 protein was diluted to varying concentrations and incubated in the wells for 1 h at room temperature. After washing, primary APE1 antibody (Novus, Cat# NB100-101) was added and incubated for 1 h, followed by a peroxidase-conjugated secondary antibody. After a final wash, TMB substrate (Invitrogen) and H₃PO₃ (Sigma) were added sequentially, and absorbance was measured at 450 nm using a microplate reader. CXCL1-G4-Biotin, 5′-TGGGGTGCGGGTGGGGTGTGGGGGTTTTT-biotin-3′; CXCL1-G4 complement strand-Biotin, 5′-GGACCCCCACACCCCACCCGCACCCCAATTTTT-biotin-3′; CXCL1-G4-Short Loops-Biotin, 5′-GGGCGGGTGGGTGGGT-biotin-3′; VEGFA-G4-#1-Biotin, 5′-CGGGGCGGGCCGGGGGCGGGGTTTTT-Biotin-3′; VEGFA-G4-#1-Short Loops-Biotin, 5′-GGGCGGGCGGGCGGGT-biotin-3′; VEGFA-G4-#2-Biotin, 5′-AGGGACGGGTGGGGAGAGGGTTTTT-biotin-3′; PIK3R3-G4-Biotin, 5′-CGGGTGGGCAGCATGGGGCGGGGAGGGTTTTT-biotin-3′; Telomere-G4-Biotin, 5′-TTAGGGTTAGGGTTAGGGTTAGGGTT-biotin-3′; and Non-G4-Biotin, 5′-GTCAGTTTCACTACTGCACCGCATG-biotin-3′.

Immunohistochemistry

Primary tumors and lung tissues from mice were collected, fixed in 10% formalin (Fisherbrand), and processed by the UNMC Tissue Science Facility. Tissue sections underwent standard deparaffinization, rehydration, antigen retrieval, peroxidase blocking, and goat serum blocking. Sections were incubated overnight at 4°C with primary antibodies: anti-APE1 (Novus, Cat# NB100-101), anti-Ki67 (Abcam, Cat# ab15580), anti-mouse IgG (Santa Cruz, Cat# sc-2025), and anti-rabbit IgG (Cell Signaling Technology, Cat# 2729). After washing, appropriate secondary antibodies (Dako) were applied. 3,3′-Diaminobenzidine (DAB; Vector Laboratories) was used as a chromogen and incubated for 2–5 min, depending on antibody reactivity. Slides were counterstained by the UNMC Tissue Science Facility. Quantification of positive staining was measured using ImageJ.

Orthotopic TNBC mouse model

All animal procedures were approved by the University of Nebraska Medical Center Institutional Animal Care and Use Committee (IACUC). MDA-MB-231 cells expressing luciferase and red fluorescent protein (RFP) were generated by lentiviral transduction and sorted via FACS for RFP expression. A total of 2 × 10⁶ Luc–RFP cells in a 1:1 mixture with Matrigel (Corning, Cat#354 234) were injected into the fourth mammary fat pad of 8-week-old female athymic nude mice (Charles River Laboratories) as described previously [36, 37]. Tumor growth was monitored weekly using caliper measurements and the in vivo imaging system (IVIS). When tumors reached 5–6 mm in diameter, mice received intraperitoneal injections of vehicle [phosphate-buffered saline (PBS)] or TMPyP4 (10 mg kg⁻¹; Sigma) twice weekly. Once tumors reached 2 cm, mice were administered luciferin and euthanized. Lung tissues were harvested and imaged using IVIS to assess metastasis.

In vivo imaging

Mice were injected intraperitoneally with 150 mg kg⁻¹  d-luciferin (GoldBio), anesthetized with isoflurane, and placed into the IVIS imaging chamber. Bioluminescence signals were measured as photons s⁻¹ in defined regions of interest. Image analysis was performed using Living Image Software.

RNA-seq

Total RNA was extracted using the RNeasy Mini Kit (QIAGEN, 74 104). RNA concentration and purity were measured using a NanoDrop spectrophotometer (Thermo Fisher). Samples were subjected to quality control and library preparation at the UNMC Sequencing Core Facility, followed by next-generation sequencing.

The sequencing reads were aligned to the UCSC hg19 human genome assembly using STAR (v2.7.3a) [38]. Differentially expressed genes (DEGs) were identified using DESeq2 (v1.44.0) [39] and normalized transcripts per million (TPM) values were derived using STRINGTIE (v2.1.1) [40]. Significantly changed genes were required to have a Benjamini–Hochberg-adjusted P-value of < 0.05 and a 2-fold change in expression. Function enrichment/Gene Ontology (GO) over-representation analysis was performed using ENRICHR [41].

Chromatin immunoprecipitation-seq

We mapped genome-wide occupancy of APE1 in chromatin using an APE1-sepcific antibody [35] by ChIP as described above. ChIP DNA samples were quantified using a Qubit fluorometer (Thermo Fisher) prior to library construction with the NEB Next Ultra II DNA Library Prep Kit (New England Biolabs). Libraries were sequenced at the UNMC Sequencing Core. ChIP-seq data were analyzed as previously described [35]. Peaks were called using MACS2 with the –broad option for APE1. Genome-wide peak distribution and annotation were performed using the ChIPseeker package in R.

Statistical analysis

GraphPad Prism (Version 10.2.3) was used for statistical analysis. All in vitro experiments were conducted in biological triplicates unless otherwise noted. Statistical comparisons between two independent groups were performed using a two-tailed unpaired Student’s t-test, and results in bar graphs are shown as means with standard deviation (SD). For in vitro binding assays involving recombinant APE1 and DNA oligonucleotides, non-linear regression analysis was used to determine binding affinity.

Results

APE1 regulates gene expression in TNBC cells

To elucidate the potential mechanistic connection between APE1 and G4 in regulating gene expression, we generated APE1 knockout (KO) cells in two TNBC cell lines, MDA-MB-231 and BT-549, using the CRISPR/Cas9 system (Supplementary Fig. S1A). We performed RNA-seq analysis on APE1-KO and wild-type (WT) MDA-MB-231 cells to explore gene expression changes resulting from APE1 depletion. Using a corrected P-value threshold <0.05 and fold change >2, we identified 2693 DEGs, of which 1717 were significantly down-regulated in APE1-KO cells compared with the WT (Fig. 1A). GO analysis revealed anti-apoptosis and cell migration among the top two biological processes significantly affected by APE1 deletion (Supplementary Fig. S1B). qRT-PCR validation in both MDA-MB-231 and BT-549 cells confirmed decreased expression of well-known regulators, CXCL1 [33, 42], VEGFA [43], PIK3R3 [44], and BMI1 [45] (Fig. 1B) involved in breast cancer cell survival and migration.

Figure 1.

APE1 and G4 are present at many gene promoters and regulate metastasis-associated genes in TNBC cells. (A) Volcano plot of RNA-seq data showing significantly down-regulated or up-regulated genes upon APE1 KO in MDA-MB-231 cells. Red and blue points indicate differentially increased and decreased genes, respectively, under thresholds of adjusted P-value <0.05 and log2 fold change > |1|. (B) Quantification of gene expression by qRT-PCR in WT and APE1-KO TNBC cells. (C) Analysis of genome-wide APE1 binding relative to genomic features in MDA-MB-231 cells. (D) Heatmaps of APE1 and H3K27ac ChIP-seq signal at peaks (summits) sorted by H3K27ac occupancy in MDA-MB-231 cells. (E) Genome browser screenshot illustrating the overlap between APE1 and H3K27ac peaks in the CXCL1 gene promoter in MDA-MB-231 cells (upper panel). Potential G4-forming sequence in the CXCL1 promoter region (−145 bp to −167 bp upstream of the TSS) that overlaps with APE1’s peak is shown (lower panel). (F) Genome browser screenshot illustrating the overlap between APE1 and H3K27ac peaks in the VEGFA gene promoter region in MDA-MB-231 cells (upper panel). Two potential G4-forming sequences VEGFA G4-#1 (−68 bp to −88 bp upstream of the TSS) and VEGFA G4-#2 (−387 bp to −406 bp upstream of the TSS) in theVEGFA promoter are shown (lower panel). (G) CD spectroscopy of CXCL1 G4-forming oligo showing formation of a hybrid (signatures characteristic maxima ∼265 nm and ∼295 nm, minima ∼245 nm) G4 structure in vitro in the presence of 50 mM KCl. CD spectroscopy of VEGFA promoter G4-forming oligo showing formation of parallel (signatures characteristic maxima ∼265 nm, minima ∼245 nm) G4 structure in vitro in the presence of 50 mM KCl. The y-axis indicates the ellipticity signal expressed in millidegrees. (H) Promoter-directed ChIP assay using G4-specific antibody BG4 in TNBC cells shows enrichment of folded G4 structures at CXCL1 and VEGFA promoter regions. Fold enrichment with G4-targeting antibody versus control IgG was calculated after input normalization. (I) Promoter-directed ChIP assay using APE1 antibody in TNBC cells shows enrichment of APE1 at CXCL1 promoter and VEGFA promoter G4 regions. Fold enrichment with anti-APE1 antibody versus control IgG was calculated after input normalization. *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001, Student’s t-test.

Notably, while APE1 deletion moderately impacted cell proliferation in two-dimensional cultures (Supplementary Fig. S1C), it significantly reduced tumor sphere formation (Supplementary Fig. S1D), cell migration (Supplementary Fig. S1E), and invasion (Supplementary Fig. S1F).

Genome-wide mapping reveals APE1 co-occupancy with G4-forming sequences

To elucidate how APE1 regulates gene expression, we performed genome-wide APE1 ChIP-seq using an AcAPE1 antibody [35] in the MDA-MB-231 cell line. APE1 was predominantly enriched at promoter regions [1 kb upstream of transcription start sites (TSSs)], with 20–25% of peaks located within this region (P < 0.001) compared with other regions (Fig. 1C). APE1-enriched regions showed a strong enrichment with H3K27ac, a mark of active regulatory elements (Fig. 1D). Many DEGs, including VEGFA, CXCL1, PIK3R3, and BMI1, had APE1-binding peaks at their promoters or enhancers (Fig. 1E, F, upper panels, Supplementary Fig. S1G, upper panel, Supplementary Fig. S1H, left panel). Intriguingly, these promoter regions contained PQSs overlapping with APE1 peaks (Fig. 1E, F, lower panels; Supplementary Fig. S1G, lower panel; Supplementary Fig. S1H, right panel). We synthesized DNA oligos containing the PQS of these gene promoters. CD spectroscopy confirmed distinct G4 folding patterns in CXCL1 (hybrid structure), VEGFA (parallel structure), and PIK3R3 (parallel structure) promoter PQS oligonucleotides in vitro (Fig. 1G; Supplementary SFig. 1I), suggesting that these PQS have the potential to fold into G4 structures in the promoters. We then validated the formation of folded G4 structures within CXCL1 and VEGFA promoters via promoter-directed ChIP using G4-targeting antibody, BG4 (Fig. 1H). We also confirmed the APE1 enrichment at the CXCL1 and VEGFA promoters using promoter-directed ChIP (Fig. 1I).

APE1 binds G4 structures in gene promoters

To explore the interaction between APE1 and G4, we performed enzyme-linked immunosorbent assay (ELISA) with CXCL1, VEGFA, and PIK3R3 promoter G4 oligos. ELISAs revealed strong binding of recombinant APE1 to these G4s as compared with non-G4-forming single-stranded DNA oligo or duplex DNA fragments (Fig. 2A, B; Supplementary Fig. S2A). APE1’s occupancy at the promoters in cells could be due to binding to the folded G4 structure, the G4 complementary single-stranded DNA (C-rich), or the G-rich double-stranded DNA. To tease this out, we compared the binding affinity of APE1 for CXCL1 G4-forming single-stranded DNA oligo, CXCL1 G4-complementary single-stranded DNA oligo, and CXCL1 PQS double-stranded DNA oligo. Intriguingly, the ELISA showed a higher binding affinity of APE1 for CXCL1 G4-forming single-stranded DNA oligo as compared with other DNA oligos (Fig. 2C), suggesting that APE1 preferentially binds to G4 structures compared with single- or double-stranded DNA in cells.

Figure 2.

APE1 binds to G4 structures in vitro and overlaps with G4 in cells. (A) CXCL1 G4-forming single-stranded oligo (CXCL1-G4), random double-stranded DNA oligo (DSD), and non-G4-forming random single-stranded oligo (Non-G4-random) labeled with 50 nM biotin were incubated with increasing concentrations of recombinant APE1 in the presence of 50 mM KCl, and ELISAs were performed with APE1 antibody. Non-linear regression was used to analyze the data. (B) ELISAs with 50 nM biotin-labeled VEGFA G4 #1- or G4 #2-forming oligo, non-G4-forming single-stranded (Non-G4-random), or random DSD were performed with increasing concentrations of recombinant APE1. Data were analyzed by non-linear regression. (C) CXCL1 G4-forming single-stranded oligo (CXCL1-G4), complementary C-rich single-stranded DNA oligo (CXCL1-G4 complement C-rich SSD), CXCL1 G4 double-stranded DNA oligo (CXCL1-G4 DSD), or Non-G4-random labeled with 50 nM biotin were incubated with increasing concentrations of recombinant APE1 in the presence of 50 mM KCl, and ELISAs were performed using APE1 antibody. Non-linear regression was used to analyze the data. (D) Representative images of tghe PLA show APE1 and G4 proximity (distance ≤ 40nm) in the nucleus in WT and APE1-KO TNBC cells. G4–APE1 PLA foci were visualized by confocal microscopy imaging (magnification: ×63). No antibody, anti-APE1 alone, G4 antibody alone, or BCL2 antibody, served as negative controls.

Next, we validated APE1–G4 interactions using a cellular approach. To visualize the interaction of APE1 and G4 within the cell nucleus, we performed a PLA, in which a focus is visible only when two proteins/antibodies are in proximity (<40 nm). A considerable number of APE1–G4 PLA foci were observed in TNBC cells (Fig. 2D), while no foci were observed between APE1 and BCL2 antibodies (Fig. 2D), demonstrating the specific interaction between APE1 and G4 structures.

The N-terminus of APE1 and G4 loops are involved in APE1–G4 interaction

G4-forming sequences can adopt three major topologies: parallel, antiparallel, and hybrid [46]. The G4s formed within the VEGFA and CXCL1 promoter sequences adopt parallel and hybrid topologies, respectively (Fig. 1G). In contrast, the well-characterized telomeric G-rich sequence forms an antiparallel G4 structure when folded in vitro in sodium-rich buffer (Supplementary Fig. S3A). To determine whether APE1 binding is influenced by G4 topology, we compared APE1 binding across G4s representing each topology. APE1 exhibited minimal binding to the antiparallel telomeric G4, whereas it showed robust binding to both parallel and hybrid G4s (Fig. 3A). These results indicate that APE1–G4 interactions are strongly topology dependent, with a clear preference for parallel and hybrid conformations over antiparallel structures.

Figure 3.

G4 loops and APE1’s N-terminus are crucial for APE1–G4 interaction. (A) ELISAs with 50 nM biotin-labeled VEGFA G4 #1 oligo (forming parallel G4), CXCL1 G4 oligo (forming hybrid G4), or telomeric G4 oligo (forming antiparallel G4 in Na⁺ buffer) were performed with increasing concentrations of recombinant APE1. Data were analyzed by non-linear regression. (B) VEGFA G4-#1 WT (VEGFA-G4-WT) or loop-shortened G4 oligos (VEGFA-G4-Short loops) labeled with 50 nM biotin were incubated with increasing concentrations of recombinant APE1 in the presence of 50 mM KCl, and ELISAs were performed with APE1 antibody. Non-linear regression was used to analyze the data. (C) CXCL1 WT (CXCL1-G4-WT) or loop-shortened G4 oligos (CXCL1-G4-Short loops) labeled with 50 nM biotin were incubated with increasing concentrations of recombinant APE1 in the presence of 50 mM KCl, and ELISAs were performed with APE1 antibody. Non-linear regression was used to analyze the data. (D and E) ELISAs with 50 nM biotin-labeled CXCL1-G4 oligo or VEGFA G4 #1 oligo were performed with increasing concentrations of recombinant WT APE1 or APE1 with the N-terminal 1–42 amino acids truncated (DeltaN42), respectively. Data were analyzed by non-linear regression.

G4 topology is dictated by the loop regions that connect adjacent G-tracts, including loop length (number of nucleotides) and loop type (propeller, lateral, or diagonal) [46]. G4-interacting proteins are thought to engage loop regions or grooves of the G4 structure. Given prior evidence that the size of the loop region affects APE1 binding to single-standard DNA structures having a stem and loop [47], we next asked whether G4 loop length modulates APE1 binding. Computational modeling predicts that the VEGFA promoter G4 adopts a parallel topology characterized by one longer propeller loop (3–4 nucleotides) and two short single-nucleotide loops, a common feature of parallel G4s [48]. To directly test the role of loop size, we engineered a modified VEGFA G4 sequence in which all three loops were reduced to single-nucleotide loops. CD spectroscopy confirmed that this modified oligonucleotide still formed a parallel G4 topology (Supplementary Fig. S3B). However, APE1 binding to this loop-shortened G4 showed a reproducible but moderate reduction compared with the WT VEGFA G4 containing a longer loop (Fig. 3B), indicating that loop length contributes to APE1–G4 interactions.

We extended this analysis to the CXCL1 promoter G4, which is predicted to form a hybrid (3 + 1) topology with two longer loops and one short loop. Reducing the length of the two longer loops to generate single-nucleotide loops altered the topology from hybrid to parallel (Supplementary Fig. S3C), and APE1 binding tended to be lower than that observed for the WT CXCL1 G4 (Fig. 3C). These findings demonstrate that G4 loop length plays a role in APE1 binding to G4.

Moreover, we found that deletion of APE1’s positively charged N-terminal 1–42 amino acids abolished APE1 binding to G4, suggesting that the APE1 N-terminus is required for G4 binding (Fig. 3D, E: Supplementary Fig. S3D).

G4 structures facilitate APE1 recruitment to the gene promoters

Strong binding affinity of APE1 for CXCL1 G4 (Fig. 2C) in vitro and enrichment of APE1 in the CXCL1 promoter PQS region in cells (Fig. 1I) raise the possibility that APE1 is recruited to the CXCL1 promoter through G4 binding. To test whether G4 is essential to recruit APE1 to promoters, we utilized the CRISPR/Cas9 knock-in technique to introduce mutations to disrupt the G4 sequence at the endogenous CXCL1 promoter without disturbing nearby DNA sequence in TNBC cells (Fig. 4A). The homozygous CXCL1 G4 mutant clone (CXCL1-G4 Mut) was confirmed by Sanger sequencing (Fig. 4A, right panel). Promoter-directed G4 ChIP assay revealed a significant decrease of G4 signal in the CXCL1-G4 Mut region, suggesting successful disruption of G4 folding at the CXCL1 promoter (Fig. 4B). Intriguingly, this abrogation of G4 formation significantly reduced APE1 occupancy at the CXCL1 promoter without altering APE1 protein level in cells (Fig. 4C; Supplementary Fig. S4A), which strongly supports our hypothesis that this G4 structure serves as a docking site for APE1 recruitment to the CXCL1 promoter. We next tested the impact of loss of promoter G4 on CXCL1 expression. Significantly, loss of CXCL1 promoter G4 structure reduced CXCL1 expression (Fig. 4D). Consistent with a previous study showing that down-regulation of CXCL1 in metastatic MDA-MB-231 cells reduced tumor growth and metastasis in the lungs [33], CXCL1-G4 Mut TNBC cells exhibited significantly reduced migration, invasion, and tumor sphere formation capacity in vitro in comparison with CXCL1-G4 WT cells (Supplementary Fig. S4B–D). Taken together, these results demonstrate that CXCL1 promoter G4 structure is functionally active and facilitates APE1 recruitment and transcriptional activation of this gene, and thus may play a crucial role in TNBC tumor growth and metastasis regulation.

Figure 4.

G4 is crucial to recruit APE1 to the CXCL1 gene promoter. (A) Schematic overview of CRISPR/Cas9-mediated generation of knock-in mutations in the CXCL1 promoter G4 sequence; the WT CXCL1 G4 sequence and mutated G4 sequences are shown (left panel). Sanger sequencing confirmed the homologous mutations (in both alleles) in CXCL1 G4 sequence (CXCL1-G4 Mut) compared with the CXCL1-G4 WT sequence (right panel). (B) Enrichment of folded G4 structure in the CXCL1 mutated G4 (CXCL1-G4 Mut) promoter versus the CXCL1 WT G4 promoter region in MDA-MB-231 and BT-549 cells was examined by promoter-directed ChIP assay with G4-specific antibody. (C) Promoter-directed ChIP assay with APE1 antibody in TNBC cells shows enrichment of APE1 in the CXCL1-G4 WT promoter region and the CXCL1-G4 Mut promoter. (D) Quantitation of CXCL1 expression in TNBC cells containing the CXCL1-G4 WT and CXCL1-G4 Mut promoter by qRT-PCR. *P < 0.05; **P < 0.01; ***P < 0.001, Student’s t-test.

Unlike the CXCL1 gene promoter, the VEGFA gene promoter contains several PQSs in the promoter region. First, we made an attempt to mutate two canonical G4-forming sequences VEGFA G4 #1 and VEGFA G4 #2 in the promoter (Supplementary Fig. S4E). While we were able to successfully mutate the VEGFA G4 #2 sequence in both alleles (homozygous mutant), despite our multiple attempts, we could not mutate the VEGFA G4 #1 sequence in both alleles (Supplementary Fig. S4E). Mutations in these G4 sequences in theVEGFA promoter had a minimal effect on APE1 occupancy and VEGFA expression (Supplementary Fig. S4F, G), suggesting redundancy with additional non-canonical PQS sites in this region, which can form G4s and recruit APE1.

G4–APE1 axis facilitates transcription factor binding

Our findings that G4 loss reduces APE1 occupancy and loss of either G4 or APE1 diminishes CXCL1 expression led us to investigate the molecular mechanism underpinning this regulation. Given that APE1 can promote gene expression directly by enhancing binding of TFs through its Ref-1 function [24, 25] or indirectly by promoting the damage repair in the transcribed gene region which might also affect TF binding, we first assessed whether APE1 deletion affects occupancy of all known TFs in CXCL1 and VEGF promoters. Promoter-directed ChIP revealed that deletion of APE1 significantly impaired Snail1 binding at the CXCL1 promoter (Fig. 5A). APE1 deletion markedly reduced c-Jun binding at the VEGFA promoter (Fig. 5B). Importantly, total levels of Snail1 or c-Jun were unchanged by APE1 deletion (Supplementary Fig. S5A). Moreover, promoter-directed ChIP demonstrated reduced Snail1 binding to the CXCL1 promoter in CXCL1-G4 Mut cells (Fig. 5C). Since the CXCL1 promoter G4 sequence itself does not contain any known TF-binding sites, and the G4 mutation did not alter Snail1 levels (Supplementary Fig. S5B), we hypothesized that promoter G4 structures recruit APE1, thereby facilitating TF DNA binding activity through its Ref-1 function or DNA repair function.

Figure 5.

The APE1–G4 axis promotes TF binding at promoters. (A and B) Promoter-directed ChIP assay shows enrichment of Snail1 at the CXCL1 promoter(A) and of c-Jun at the VEGFA promoter (B) in WT and APE1-KO cells. (C) Promoter-directed ChIP assay shows enrichment of Snail1 in CXCL1-G4 WT and CXCL1-G4 Mut promoter-containing cells. (D) MDA-MB-231 APE1-KO cells expressing control vector (KO-Vec), WT-APE1 (KO-WT), APE1 Ref-1 function-defective mutant (KO-CS), or APE1 endonuclease-defective mutant (KO-ED) under a Dox-inducible promoter were treated with 2 µg ml⁻¹ Dox to induce the expression of APE1. Then the expression levels of CXCL1 and VEGFA were quantitated in each group of cells by qRT-PCR. (E) Representative images of PLAs of APE1 and G4 co-localization in APE1-KO MDA-MB-231 cells expressing WT-APE1 (WT) or the APE1 N-terminal deletion mutant (DeltaN42) under treatment with 2 µg ml⁻¹ Dox (left panel); average numbers of PLA foci of 30 cells were quantitated (right panel). (F) MDA-MB-231 APE1-KO cells expressing control vector (KO-Vec), WT-APE1 (KO-WT), or the APE1 N-terminal deletion mutant (KO-DeltaN42) under treatment with 2 µg ml⁻¹ Dox. The expression levels of CXCL1 and VEGFA were quantitated in each group of cells by qRT-PCR. ns, not significant; *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001, Student’s t-test.

Both APE1’s N-terminal domain and Ref-1 function are essential for gene expression

Given that APE1 has well-characterized DNA repair activity (mediated by a globular C-terminal domain) and Ref-1 activity (mediated by 60–127 amino acids), we generated Dox-inducible expression plasmids containing the WT, the Ref-1-mutant (Cys65A99A) [24], and the endonuclease-deficient mutant (E96Q&D210N) [49]. Dox was administrated to induce the expression of APE1 mutants in APE1-KO cells to a similar level as APE1 in parent cells (Supplementary Fig. S5C). The qRT-PCR analysis revealed that the WT and endonuclease mutant APE1, but not the Ref-1 mutant, were able to activate CXCL1 and VEGFA expression (Fig. 5D), indicating that APE1’s Ref-1 activity is required for transcriptional activation of these two genes. Treatment with E3330, a small molecule specifically inhibiting APE1’s Ref-1 function [50, 51], inhibited CXCL1 and VEGFA expression (Supplementary Fig. S5D), further confirming the effect of APE1’s Ref-1 function on regulating the expression of these genes.

Moreover, to examine whether APE1’s N-terminal loss, which impaired APE1’s G4 binding in vitro (Fig. 3D, E; Supplementary Fig. S3D), affects APE1’s function in cells, we generated Dox-inducible plasmids expressing N-terminal deleted (1–42 amino acids) truncated APE1 (DeltaN42) fused with a c-Myc nuclear localization signal (NLS), which helped DeltaN42 translocate into the nucleus in the absence of its own N-terminal NLS (Supplementary Fig. S5E, F). We observed that deletion of the APE1 N-terminus significantly impaired APE1–G4 co-localization in cells (Fig. 5E). The qRT-PCR analysis further revealed that APE1 with its N-terminus deleted failed to activate CXCL1 and VEGFA expression (Fig. 5F), indicating that the N-terminus-mediated G4 binding is important for regulating gene expression. Collectively, these data suggest that G4s recruit APE1 to promoters via APE1’s N-terminal interaction, and APE1 further promotes TF binding via its Ref-1 function. Therefore, both APE1’s G4 binding and its Ref-1 function are essential to activate gene expression.

G4 ligands block G4–APE1 interaction and inhibit gene expression

Our findings imply that the G4–APE1 interaction is a crucial regulator to activate gene expression. Thus, disrupting this interaction could negatively affect gene expression. Several G4-binding small molecules are known to stabilize G4 structures and could potentially displace endogenous G4-binding proteins in cells [13, 52]. To test this hypothesis, we examined the effect of TMPyP4, a porphyrin-based G4 ligand that stabilizes G4 structures in cells by stacking on terminal G4s and interacting with external loop regions (Fig. 6A) [53, 54]. ELISA-based binding assays revealed that TMPyP4 inhibited APE1 binding to CXCL1 and VEGFA promoter G4s in a dose-dependent manner (Fig. 6B; Supplementary S6A). Consistently, PLAs demonstrated that treatment with these ligands significantly disrupted endogenous APE1–G4 interactions in cells (Fig. 6C). ChIP assays showed that TMPyP4 treatment significantly reduced APE1 recruitment to G4-containing promoter regions, without altering total cellular APE1 protein levels (Fig. 6D; Supplementary Fig. S6B). Correspondingly, qRT-PCR analysis revealed that TMPyP4 treatment significantly decreased CXCL1 and VEGFA mRNA expression (Fig. 6E), supporting the conclusion that blocking APE1–G4 interactions impairs transcriptional activation. Functionally, TMPyP4 treatment led to significant reductions in cell migration (Supplementary Fig. S6C), invasion (Supplementary Fig. S6D), and tumor sphere formation (Supplementary Fig. S6E), while exerting only a modest effect on cell proliferation in vitro (Supplementary Fig. S6F). Collectively, these findings suggest that disruption of APE1–G4 interactions by a G4-stabilizing ligand suppresses oncogenic gene expression and attenuates tumorigenic traits in TNBC cells.

Figure 6.

G4 ligands blocks G4–APE1 interaction. (A) MDA-MB-231 cells treated with either vehicle or the G4-stabilizing ligand TMPyP4 (150 µM) for 24 h and then immunostained with G4-specific antibody and visualized by confocal microscopy. (B) CXCL1-G4 or VEGFA-G4 oligos labeled with 50 nM biotin, which were attached to a streptavidin-conjugated plate, were incubated with a saturating dose (64 nM) of APE1 protein and 50 mM KCl, titrated with increasing concentrations of TMPyP4, and then an ELISA was performed with APE1 antibody. Non-linear regression was used to analyze the data. (C) Representative images of PLAs of APE1 and G4 co-localization in TNBC cells after treatment with vehicle or TMPyP4 (150 µM) (left panel); average numbers of PLA foci of ~25 cells were quantitated (right panel). (D) Promoter-directed ChIP assay shows enrichment of APE1 versus IgG at CXCL1 G4 and VEGFA G4 promoter regions following 150 µM TMPyP4 treatment. (E) Expression levels of genes involved in migration in TNBC cells after treatment with vehicle or 150 µM TMPyP4 for 24 h, determined by qRT-PCR. (F) CXCL1-G4 or VEGFA-G4-#1 oligos labeled with 50 nM biotin were incubated with 64 nM APE1 protein, titrated with increasing concentrations of PDS or PhenDC3, respectively, and then an ELISA was performed with APE1 antibody. Non-linear regression was used to analyze the data. (G) Representative images of PLAs of APE1 and G4 co-localization in TNBC cells after treatment with vehicle, PDS (10 μM), or PhenDC3 (20 μM) (left panel); PLA foci of ~30 cells were quantitated in each group (right panel). *P < 0.05; **P < 0.01; ***P < 0.001, Student’s t-test.

Since recent studies highlight that transcriptional outcomes of G4 targeting are highly ligand and context dependent [19], we examined whether other well-characterized G4 ligands such as PDS and PhenDC3 can also disrupt APE1–G4 interactions and modulate CXCL1 and VEGFA expression. ELISA-based binding assays revealed that PDS and PhenDC3 each inhibited APE1 binding to CXCL1 and VEGFA G4 oligos in a dose-dependent manner (Fig. 6F) in vitro. Consistently, PLAs demonstrated that treatment with these ligands significantly disrupted endogenous APE1–G4 interactions in cells (Fig. 6G). Moreover, treatment with either PDS or PhenDc3 decreased CXCL1 and VEGFA mRNA expression (Supplementary Fig. S6G) and led to significant reductions in cell migration (Supplementary Fig. S6H), supporting the conclusion that many G4 ligands block APE1–G4 interactions in cells and lead to reduced gene expression.

Disruption of the G4–APE1 axis suppresses TNBC cells growth and metastasis in vivo

First, to evaluate APE1’s role in tumor growth in vivo, we orthotopically implanted luciferase-expressing APE1-WT and APE1-KO MDA-MB-231 cells into the fourth mammary fat pads of mice. APE1 KO markedly suppressed primary tumor growth (Fig. 7A, B) and reduced lung metastasis (Fig. 7C, D). Immunohistochemistry confirmed the absence of APE1 expression and reduced proliferation (Ki67 staining) in KO tumors (Supplementary Fig. S7A). Since down-regulation of CXCL1 in metastatic MDA-MB-231 cells was shown to reduce TNBC growth and metastasis in the lungs [33], we examined the impact of CXCL1 promoter G4 mutation on tumor growth. We found that G4 mutation in the CXCL1 promoter significantly reduced both primary tumor growth and lung metastasis (Fig. 7E–H), underscoring the critical role of the CXCL1 promoter G4 in regulating tumor cell growth in vivo. Because TMPyP4 has been shown to affect cell adhesion and migration [55], and our study demonstrated that it interfered with APE1–G4 binding, altered gene expression, and suppressed TNBC cell migration and invasion in vitro, we next examined whether TMPyP4 treatment could prevent TNBC metastasis in vivo. TMPyP4 treatment of MDA-MB-231 tumor-bearing mice modestly affected primary tumor growth (Fig. 7E, F) but significantly suppressed lung metastasis (Fig. 7G, H). Together, these results highlight crucial regulatory roles of the G4–APE1 axis in TNBC cell growth and metastasis in vivo.

Figure 7.

Disruption of the APE1–G4 axis inhibits TNBC tumor growth and lung metastasis in vivo. (A) Representative bioluminescence IVIS images of a nude mouse showing tumor burdens at days 21, 35, 52, and 63 after orthotopic implantation of MDA-MB-231 APE1-WT (mice n = 5) and APE1-KO cells (mice n = 4) in the mammary fat pad of nude mice. A different bioluminescence intensity scale was used to show the tumor burden on the indicated days. (B) Line chart of luciferase bioluminescence intensity showing tumor burden over time in mice bearing APE1-WT (mice n = 5) or APE1-KO (mice n = 4) tumors shown in (A). (C) Bioluminescence imaging of lungs harvested from WT and APE1-KO TNBC tumor-bearing mice as shown in(A), exhibiting tumor metastases to the lungs. (D) Quantitation of bioluminescence fluxes of lung metastatic tumors in each WT and APE1-KO group as shown in (C). (E) Representative bioluminescence imaging of a nude mouse tumor burden at days 21, 36, 47, and 62 after orthotopic implantation of MDA-MB-231 CXCL1-G4 WT- (n = 5) and CXCL1-G4 Mut- (n = 5) containing cells. Mice (n = 5) bearing MDA-MB-231 CXCL1-G4 WT tumors were treated with 10 mg kg⁻¹ TMPyP4 twice weekly. The IVIS images were taken on the indicated days. (F) Line chart of luciferase bioluminescence intensity showing tumor burden over time in mice bearing CXCL1-G4 Mut (n = 5) or CXCL1-G4 WT (n = 5) tumors treated with or without TMPyP4, as shown in (E).(G) CXCL1-G4 WT, CXCL1-G4 Mut, and TMPyP4-treated primary tumors were allowed to grow for different time periods until they reached 1.5 cm in diameter, mice were then euthanized, and lung tissues were harvested. Bioluminescence imaging of lungs harvested from CXCL1-G4 WT, CXCL1-G4 Mut, and TMPyP4-treated TNBC tumor-bearing mice as shown in(E). (H) Quantification of bioluminescence intensity of lung tissues in MDA-MB-231 CXCL1-G4 WT, CXCL1-G4 Mut, and TMPyP4-treated groups as shown in (G).

Discussion

The presence of potential G4-forming sequences in promoter regions of almost 50% of human genes suggests a role in regulating their expression through the ability to form G4 structures [56]. There is now increasing correlative evidence showing that G4s act as transcriptional regulators by shaping the structure of chromatin or by influencing the formation and stability of transcriptional loops including R-loops and long-range enhancer–promoter loops [13, 14, 57, 58]. However, these works were limited to correlative study. Although recent work has demonstrated that selective deletion of promoter G4s alters transcription [17, 18], the roles of G4s in regulating cellular functions and cell fate remain poorly understood. Here, we have provided direct evidence that loss of G4 formation suppresses CXCL1 expression and G4 plays a crucial role in regulating cellular functions such as cell migration and invasion both in vitro and in vivo. Moreover, this study expands our understanding by demonstrating that G4 structures recruit the multifunctional protein APE1, subsequently enhancing TF binding via APE1’s Ref-1 function, thus providing a new mechanism by which G4 acts as a positive regulator of gene expression (Graphical abstract). Multiple lines of evidence support the model: (i) genome-wide mapping shows APE1 enrichment at promoter regions containing G4 structures in TNBC cells, with confirmed co-localization; (ii) recombinant APE1 robustly binds CXCL1 and VEGFA promoter G4 structures in vitro; (iii) loss of G4 formation at the CXCL1 promoter reduces APE1 recruitment; (iv) disruption of either G4 formation or APE1 expression decreases Snail1 TF binding at the CXCL1 promoter, leading to reduced CXCL1 expression; (v) G4 ligands, including TMPyP4, PDS, and PhenDC3, which inhibit APE1–G4 interactions in cells decrease CXCL1 and VEGFA expression; and (vi) ectopic expression of a Ref-1-function-deficient APE1 mutant failed to restore reduced CXCL1 expression in APE1-KO cells, confirming the role of APE1’s Ref-1 activity in transcriptional activation.

APE1’s dual functions in BER and gene expression regulation have long raised questions about its mechanisms of recruitment to promoter regions. Unlike its bacterial counterpart, human APE1 contains an intrinsically disordered N-terminal domain, which we show is essential for G4 binding. This provides a mechanistic explanation for how APE1 is targeted to specific regulatory regions despite lacking intrinsic sequence specificity. Although studies suggested that guanine oxidation in gene promoters containing G4-forming sequences, which induces AP site formation and activates APE1’s DNA repair function, may mediate APE1’s binding to promoter G4s and promote gene activation [59, 60], while we have provided evidence in the ELISA that APE1 binds to G4 independently of DNA damage, and the endonuclease-defective APE1 does not impair gene activation. Instead, our data suggest that promoter G4s serve as structural docking sites for APE1 via its N-terminal interaction, and its Ref-1 activity is required for transcriptional activation. Thus, our work integrates the regulatory roles of promoter G4 structures and APE1 into a unified mechanism of transcriptional control. Elevated APE1 expression in TNBC correlates with poor metastasis-free survival (Supplementary Fig. S7B), and APE1 deletion significantly reduces TNBC tumor growth and metastasis. Collectively, our data propose that G4-bound APE1 enhances TF binding to critical gene promoters, facilitating transcriptional activation essential for cell migration, invasion, and TNBC progression.

Our unbiased approach identified CXCL1 among several genes whose expression is significantly altered upon APE1 deletion. Notably, previous studies demonstrated that down-regulating CXCL1 in highly metastatic MDA-MB-231 cells suppressed mammary tumor growth and reduced lung metastasis in vivo, confirming CXCL1’s pivotal role in breast cancer progression [33]. Consistent with these findings, we demonstrated that mutations disrupting G4 formation at the CXCL1 promoter significantly decreased its expression and attenuated TNBC tumor growth and lung metastasis in vivo, highlighting the crucial role of G4-mediated CXCL1 activation. Nevertheless, our findings clearly delineate the mechanism by which the APE1–G4 axis promotes CXCL1 and confirms the critical role of promoter G4 structures in TNBC tumor growth and lung metastasis in vivo. Although we have provided substantial evidence showing that G4 in the CXLCL1 promoter acts as an activator for CXCL1, we cannot argue that G4s act solely as transcriptional activators for all genes, as there may be individual cases where G4s are repressive in nature. For example, the KRAS promoter G4 recruits MAZ, PAR P1, and hnRNP A1 to regulate transcription, yet unfolding of this G4 by hnRNP A1 or its derivative UP1 enhances MAZ binding and activates gene expression [61]. Together, these findings underscore that G4s act as dynamic regulatory platforms whose transcriptional output is dictated by the identity and mode of engagement of associated proteins.

Several small-molecule ligands capable of selectively stabilizing G4 structures have emerged as promising therapeutic agents for cancer [54, 62]. These ligands stabilize dynamic G4 structures in vitro and in cells, and, in many cases, displace endogenous G4-binding proteins. Here, we have shown that treatment with multiple G4 ligands, including TMPyP4, PDS, and PhenDC3, disrupts APE1–G4 interactions both in vitro and in cells, concomitant with reduced expression of the APE1-regulated genes CXCL1 and VEGFA. These findings support a model in which G4 ligands can suppress transcription by perturbing productive protein–G4 interactions at gene promoters. Importantly, recent studies highlight that the transcriptional consequences of G4 ligand binding are highly context dependent. An elegant targeting strategy selectively directing PDS to the MYC promoter G4 revealed decreased BG4 antibody accessibility at MYC but increased accessibility at the PVT1 promoter G4, whereas PhenDC3 exerted the opposite effect at PVT1 [19]. These observations suggest that distinct ligands may dock differently on individual promoter G4s, enabling either displacemebnt of TFs by competing with the binding sites or enhancement of TF–G4 interaction by stabilizing G4 structures. Therefore, the transcriptional outcome is largely dependent on a TF, ligand, and G4 topology. Uncovering how ligands and proteins interact with G4s in a certain context can provide insights to predict the ligand-induced transcriptional outcome.

Furthermore, our data demonstrate that G4 loop length is a determinant of APE1 binding. This observation aligns with prior biochemical studies showing that loop size influences APE1 binding to structured single-stranded DNA substrates [47]. We propose that APE1 recognizes loop regions of promoter G4s and that TMPyP4 disrupts this interaction by competing for overlapping binding surfaces, thereby preventing APE1 recruitment to chromatin in cells. Future structural studies using nuclear magnetic resonance (NMR) spectroscopy or crystallography will be essential to define the precise molecular interfaces governing APE1–G4 recognition. In addition, genome-wide analyses examining how G4 ligands, such as TMPyP4, PDS, and PhenDC3, alter APE1 chromatin occupancy and transcriptional programs will provide critical biological insight.

In conclusion, our comprehensive study provides direct genetic evidence that endogenous promoter G4 structures can function as transcriptional activators by recruiting APE1, which enhances TF binding through its Ref-1 activity. We reveal a critical role for this G4–APE1 axis in regulating CXCL1 expression and driving TNBC tumor growth and metastasis. These findings not only demonstrate that G4s function as transcriptional activators for many genes in cancer cells but also identify the G4–APE1 interaction as a potential therapeutic vulnerability in aggressive cancers such as TNBC.

Supplementary data

Supplementary data are available at NAR online.

Conflict of interest

The authors declare that they have no conflicts of interest with the contents of this article.

Funding

The Department of Défense (DOD) [HT9425-23-1-0051, BC220592, HT9425-23-1-0052, and BC220592P1 to K.K.B. and V.B. (MPI), and R35GM147467 to M.J.R.].

Data availability

All raw and processed sequencing data (FASTQ and BigWig files) have been deposited in the Gene Expression Omnibus (GEO), GSE297827.