IP-SNPs-seq links noncoding risk alleles to lineage transcription factor programs in prostate cancer

Abstract

Most prostate cancer risk variants reside in noncoding DNA, but connecting germline alleles to lineage transcription factor (TF) programs has been challenging. We developed immunoprecipitation-coupled SNPs-seq (IP-SNPs-seq), enabling high-throughput, allele-specific, TF-resolved interrogation of candidate regulatory variants. Screening 903 prostate cancer–associated SNPs with androgen receptor (AR) immunoprecipitation, we identified multiple alleles with biased AR binding and convergent evidence from eQTL and ChIP-seq datasets. Among these, rs7600820 emerged as a functional enhancer variant: the risk G allele conferred stronger reporter activity, heightened AR responsiveness to dihydrotestosterone, and increased ODC1 expression; chromatin profiling and Hi-C revealed an active enhancer loop to the ODC1 promoter. ODC1 was consistently upregulated in primary and metastatic tumors across independent cohorts, associated with adverse clinicopathologic features, and required for prostate cancer cell proliferation. Gene-set enrichment analyses linked high ODC1 expression to MYC target signatures, positioning ODC1 as a clinically relevant, AR-regulated oncogenic node that integrates germline risk with core prostate cancer circuitry. IP-SNPs-seq thus provides a scalable route from association to mechanism, broadly applicable to diverse TFs and diseases, and nominates the AR–rs7600820–ODC1 axis as a potential biomarker and therapeutic vulnerability in androgen-driven prostate cancer.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00018-025-05964-7.

Introduction

Prostate cancer is the second most frequently diagnosed malignancy and the fifth leading cause of cancer mortality among men worldwide, accounting for ~ 1.5 million new cases and ~ 0.4 million deaths annually [1]. Although most patients with localized disease have excellent outcomes, survival drops sharply once metastases occur, underscoring the urgent need for deeper understanding of disease biology and novel therapeutic strategies [2].

Large-scale epidemiological and twin studies consistently demonstrate a strong inherited component to prostate cancer, with heritability estimated at nearly 60% [3, 4]. Genome-wide association studies (GWAS) have identified more than 450 risk single-nucleotide polymorphisms (SNPs), representing the largest known genetic architecture for any cancer [5, 6]. Strikingly, the majority of these variants map to noncoding regions, suggesting that dysregulation of gene expression, rather than protein-coding changes, plays a central role in prostate cancer susceptibility [7, 8]. Deciphering how these variants exert their effects remains a major challenge, yet it holds the key to connecting germline risk to somatic alterations and ultimately to clinical outcome.

Post-GWAS investigations have begun to link prostate cancer risk loci to biological processes such as DNA repair, cell cycle regulation, metabolism, and inflammation [9, 10]. However, a central obstacle lies in pinpointing causal variants, defining their regulatory mechanisms, and identifying their target genes in relevant cellular contexts. To bridge this gap, high-throughput functional genomics technologies have emerged. Massively parallel reporter assays (MPRAs) and self-transcribing active regulatory region sequencing (STARR-seq) allow systematic evaluation of enhancer activity across thousands of sequences [11–14], while CRISPR interference (CRISPRi) screens and pooled ChIP-based approaches have revealed regulatory elements and protein–DNA interactions underlying disease risk [15–18]. Our group previously developed single-nucleotide polymorphisms sequencing (SNPs-seq), a scalable assay to quantify allele-specific protein binding across thousands of loci [19, 20]. Yet, despite its power, conventional SNPs-seq lacks the resolution to identify which transcription factors (TFs) directly mediate risk-associated binding events.

In this study, we address this gap by advancing SNPs-seq into an immunoprecipitation-coupled platform, termed IP-SNPs-seq, which enables high-throughput, allele-specific mapping of TF binding. Applying IP-SNPs-seq to 903 candidate SNPs, we identified dozens of variants with preferential androgen receptor (AR) binding. Functional validation, including reporter assays and mechanistic studies, uncovered a previously unrecognized AR–rs7600820–ODC1 regulatory axis, which contributes to prostate cancer susceptibility and progression. These findings not only establish IP-SNPs-seq as a broadly applicable strategy for linking noncoding variation to TF binding but also reveal a clinically relevant regulatory pathway with therapeutic potential in prostate cancer.

Materials and methods

Cell culture

Human prostate cancer cell line LNCaP and 22Rv1 were sourced from the American Type Culture Collection (ATCC, USA) and the Cell Bank of the Chinese Academy of Sciences (China). Cells were maintained in RPMI 1640 medium supplemented with 10% fetal bovine serum (FBS) (FSP500, Genetimes Technology) and 1% penicillin/streptomycin (MA0110, MeilunBio) at 37 °C under 5% CO₂. All cell lines underwent regular mycoplasma testing and short tandem repeat (STR) authentication. For androgen treatment experiments, cells were switched to phenol red-free RPMI medium containing 10% charcoal-stripped FBS and treated with 10 nM dihydrotestosterone (DHT) or 0.1% ethanol (ETH) for 48 h.

IP-SNPs-seq

A total of 903 SNPs were selected from 51 regions that showed significant expression quantitative trait locus (eQTL) signals (p < 1.97E-7) within 2.2 Mb window of 88 genes, with further filtering for linkage disequilibrium (LD > 0.5) and significance (p < 3.02E-8) [21, 22] (the workflow of IP-SNPs-seq is illustrated in Fig. 1a, and oligo libraries are showed in Fig. 1b).

Fig. 1.

Workflow of IP-SNPs-seq. (a) Schematic overview of the IP-SNPs-seq method, comprising three main steps: binding of SNP-containing oligonucleotides to nuclear proteins, washing to remove unbound oligonucleotides, and elution of transcription factor–bound oligonucleotides followed by sequencing analysis. (b) Representation of the structural organization of initiation, input, and sequencing libraries at different stages of the IP-SNPs-seq workflow

• Step 1. For each candidate SNP, four single-stranded oligonucleotides (21 bp in length, covering both variant and reference alleles, +/- strand, with the SNP positioned centrally) were synthesized at 20 µM in 25 µl duplex buffer (Integrated DNA Technologies). For annealing, 10 µl of complementary strands were mixed, denatured at 95 °C for 3 min, and gradually cooled to 25 °C in 70 min. All double-stranded oligos (ds-oligos) were pooled by combining 2 µl of each.

• Step 2. Preparation of oligo pool initiation library by adding adaptor sequence to 21 bp SNP oligos using ThruPLEX DNA-seq 48S Kit (R400427). Then, the oligo pool input library was prepared by primers (5’ Adaptor half primer: 5’ TCTTTCCCTACACGACGCTCTTCCGATCT, 29nt; 3’ Adaptor half primer: 5’ GTGACTGGAGTTCAGACGTGTGCTCTTCCGATCT, 34nt) using Q5 hot start High fidelity DNA polymerase (M0493S, New England Biolabs). The input library was purified using DNA Clean & Concentrator kit (D4033, Zymo research).

• Step 3. We extracted the nuclear protein using Ne-Per nuclear and cytoplasmic extraction reagents (Thermo Fisher Scientific). Protein concentrations were determined using Pierce BCA protein assay kit (Thermo Fisher Scientific). Add the nuclear extract to the tube containing pre-washed Pierce Protein A/G Magnetic Beads and incubate at room temperature for 30 min with mixing. Collect the beads with a magnetic stand, pipette the supernatant to a new 1.5 ml tube.

• Step 4. For oligo–protein binding assay, we mixed 2 µg ds-oligos pool, 200 µg nuclear extract, and 15 µl 10X binding buffer (100 mM Tris, 500 mM KCl, 10 mM DTT) in ultrapure water at total 150 µl volume reaction. The binding reaction mixtures were incubated 30 min at room temperature. For immunoprecipitation, we combined nuclear extract-oligo mixtures with 10 µg of anti-AR antibody (ab74272) in a microcentrifuge tube. Dilute the antibody/lysate solution to 500 µL with IP Lysis/Wash Buffer. Incubate for overnight at 4 °C to form the immune complex.

• Step 5. Add the antibody-protein-oligo mixtures to the tube containing pre-washed magnetic beads and incubate at room temperature for 1 h with mixing. With magnetic stand, the mixture was washed three times using 500µL of IP Lysis/Wash Buffer. Then the mixture was eluted in 100 µl Elution Buffer. After protein digested by Proteinase K, the DNA oligos were further purified using Oligo Clean kit (D4060, Zymo research). Two repeats were used in each experimental condition to ensure reproducibility and to minimize technical variability.

• Step 6. We prepared sequencing libraries employing purified eluted oligos and paired primers (5’ P5 universal primer: AATGATACGGCGACCACCGAGATCTACACTCTTTCCCTACACGACGCTCTTCCGATCT, 58 nt; 3’ P7 index primer: CAAGCAGAAGACGGCATACGAGATNNNNNNGTGACTGGAGTTCAGACGTGTGCTCTTCCGATCT, 64 nt, NNNNNN is index sequence) by Q5 hot start High fidelity DNA polymerase (M0493S). After 15-cycle amplification, the libraries were purified using SPRIselect reagent (Beckman Coulter Life Sciences) and quantified by Qubit. The libraries were sequenced on Illumina HiSeq2500 with 50 bp single read.

FastQC (v.0.11.9) was for quality assessment of raw data. Adapters were removed using TrimGalore (v.0.6.7). The trimmed reads were mapped into the 1806 unique sequences template (a total of 903 SNP sites) using Bowtie2 (v.2.2.5) [23] with the default parameters. Only perfect match was allowed during mapping. To determine the allelic protein-binding difference, we calculated the biased allelic binding (BAB) score using this formula: BAB = log2 [IP sample (variant allele/reference allele read counts)/Input sample (variant allele/reference allele read counts)]. The sequences data of IP-SNPs-seq are available upon request.

Luciferase enhancer reporter assay

Allele-specific sequences for rs7600820 (ODC1), rs3800285 (FOXP4), and rs902774 (KRT8) were cloned into firefly luciferase pGL4.23-minP vectors (E8411, Promega). Constructs were transfected into 22Rv1 and LNCaP cells using Lipofectamine 3000 DNA Transfection Reagent (L3000015, Thermo Fisher Scientific). To normalize the results, we co-transfected cells with the renilla luciferase pGL4.75 plasmid (E6931, Promega) as an internal control. Luciferase activity was measured 48 h post-transfection using the Dual Luciferase Reporter Assay System (E1960, Promega) on a bioluminometer. Each construct was tested in at least three replicate wells. The results were then statistically analyzed using a two-tailed Student’s T-test. Details of the primer sequences and cloning methods are available in Supplementary Table 1.

SiRNA knockdown assay

Cells were transfected with control siRNA or gene-specific siRNAs targeting ODC1 using Lipofectamine RNAi MAX Transfection Reagent (13778150, Thermo Fisher Scientific). The medium was replaced 12 h post-transfection, and RNA and protein were extracted 48 h for analysis. The specific sequences of siRNAs used are detailed in Supplementary Table 2.

RNA isolation and quantitative PCR

Total RNA was isolated using the EZ-10 DNAaway RNA Mini-Preps Kit (B618133, Sangon Biotech). 1 µg total RNA was reverse transcribed using the HiScript III RT SuperMix for qPCR kit (R323-01, Vazyme) and the resulting cDNA was diluted 20 times. RNA expression was quantified using the ChamQ universal SYBR qPCR master mix (Q711-02, Vazyme) on the Light Cycler 480 (Roche). GAPDH or β-actin served as reference genes. Each sample was measured in triplicate to ensure the accuracy and reliability of the data. Relative gene expression was calculated via the ΔΔCT (ΔCT [sample] – ΔCT [control average]) method. The sequences of oligonucleotides are provided in Supplementary Table 3.

Tumor cell biology assays

For cell proliferation assay, cells were seeded in 96-well plates. Cell viability and proliferation were measured using MTT kit (SY316, Beyotime Biotechnology). Absorbance readings at 490 nm (MTT) were taken at specific time points. Data obtained from at least triplicate wells and analyzed using two-tailed Student’s T-test or two-way ANOVA.

For cell migration assay, cells were resuspended in serum-free medium, and placed into 8 μm transwell inserts (353097, BD). Lower chambers were filled with 600 µl of normal growth medium and cells were incubated for 36 h. Post-incubation, cells were fixed with 4% formaldehyde and stained with crystal violet.

For cell invasion assay, the transwell inserts were coated with 100 µl Matrigel (40183ES10, Yeasen) diluted in serum-free medium. Invasive cells on the bottom surface of the filters were counted in five microscopic fields per membrane. The data were statistically analyzed using two-tailed Student’s T-test or two-way ANOVA, with each assay performed in three replicates.

Chromatin Immunoprecipitation (ChIP) and ChIP-seq

For the tissue ChIP assay, samples were dissected into small pieces using fine scissors and fixed in 1.5% formaldehyde for 10 min at room temperature, followed by quenching with glycine. The tissues were mechanically disrupted through eight cycles of grinding in a tissue freezing grinder (Jingxin, China). Nuclei were isolated by resuspending the tissue pellet in hypotonic lysis buffer (supplemented with DTT and protease inhibitor cocktail) and incubating for 40 min at 4 °C. The tissue debris was removed by filtration through a sterile 100 μm filter. Chromatin was sheared to fragments of 200–500 bp using a high-power Bioruptor Plus sonicator or Covaris system. For each ChIP reaction, chromatin (1.5 µg for histone modification ChIP) was incubated with antibodies (2 µg for histone antibodies) overnight at 4 °C. The antibody-chromatin complexes were then captured by incubating with pre-washed Protein G Dynabeads overnight at 4 °C. The immunoprecipitated complexes were eluted in 100 µl and reverse- crosslinked by adding 6 µl of 5 M NaCl and 5 µl of Proteinase K, followed by incubation overnight at 65 °C. DNA from both immunoprecipitated and input samples was purified using the MinElute PCR Purification Kit (28006, Qiagen).

ChIP-seq libraries were constructed with the NEBNext Ultra II DNA Library Prep Kit (E7103L, NEB) following the manufacturer’s protocol. Sequencing was carried out by Annoroad Company. Libraries for histone modifications (H3K27ac, H3K4me1, and H3K4me3) were sequenced to generate 150 bp paired-end reads. Raw data quality was assessed with FastQC (v.0.11.9). Adapters trimming and removal of short reads were performed using TrimGalore (v.0.6.7, RRID: SCR_011847). Trimmed reads were aligned to the human reference genome Hg38 using Bowtie2 (v.2.2.5) [23] under default parameters. Low-quality alignments were filtered out with SAMtools (v.1.13) [24] using the parameters “-q 30 -F 3844”. Duplicate reads were marked and removed using Picard Toolkit (v.2.25.1, RRID: SCR_006525). Peak calling was performed with MACS2 (v.2.1.4) [25] under default settings. Peak visualization and analysis were conducted using the Integrated Genome Viewer (IGV, v.2.12.3).

Expression quantitative trait loci (eQTL) analysis

To evaluate the associations between genotypes of rs7600820, rs3800285, rs902774 and the expression levels of ODC1, FOXP4, KRT8, we performed an eQTL analysis using the R package “Matrix eQTL” (v.2.2) in the CPGEA cohort comprised of 134 paired normal prostate and prostate cancer samples, GTEx dataset of 218 normal prostate samples and TCGA-PRAD dataset of 490 prostate cancer samples. The eQTL analysis was applied by fitting a linear regression model (“useModel = modelLINEAR”), setting up other parameters as default (pvOutputThreshold = 0.05, errorCovariance = numeric ()”). The transcriptional profiling in CPGEA cohort was assessed by RNA-Seq and the CPGEA cohort was genotyped using whole genome sequencing (WGS) strategy.

Clinical association analysis

We performed the co-expression analysis to evaluate the expression correlation between ODC1, FOXP4, KRT8 and AR, from TCGA-PRAD cohorts. Spearman correlation method was applied in all linear expression correlation tests.

We used the Kruskal-Wallis test to compare gene expression between normal, prostate tumors, and metastasis from TCGA prostate cancer dataset [26]. We investigated the association of candidate gene expression with clinicopathological features such as clinical stages, Gleason score, prostate specific antigen (PSA) level.

The publicly available GWAS data in prostate cancer used in this study were obtained from the GWAS catalog. The publicly available RNA-seq or microarray data including CPGEA, TCGA, GTEx, Rld, Penney (GSE62872), GSE6811, Taylor (GSE21034), and FHCRC171 were retrieved from public databases including cBioPortal for Cancer Genomics, Oncomine database, and GEO database. The deposited data were listed in Supplementary Table 4.

Statistics analysis and data visualization

All statistical analyses were performed using R environment (v.4.0.5) or unless specified. The Kruskal-Wallis test was applied for gene expression in clinical cohorts. For the experimental part, data were presented as means ± SD using the GraphPad Prism 6 software. Differences between two groups were estimated using the two tailed student’s T test. The variables in three or more groups were compared using the two-way ANOVA test. Asterisks indicate the significance levels (*P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001). For comparative analyses, P < 0.05 was considered statistically significant. The software and algorithms were listed in Supplementary Table 5.

Results

Generation of high-quality IP-SNPs-seq libraries

To evaluate the performance of our IP-SNPs-seq approach, we first assessed the quality of the constructed sequencing libraries. The double-stranded oligonucleotides (21 bp) combined with adaptors of 58 bp and 64 bp yielded an expected library size of ~ 143 bp, which was confirmed by a sharp, well-defined band at the predicted size (Fig. 2a). This tight distribution indicates both accurate library construction and minimal by-products.

Fig. 2.

Quality control of IP-SNPs-seq. (a) Size distribution of IP-SNPs-seq libraries showing an average fragment length of ~ 140 bp. (b) Raw read counts and perfectly matched read counts in IP and input samples. (c–g) Correlation analysis of raw read counts (log₂-transformed) between technical replicates in dihydrotestosterone (DHT), ethanol (ETH), and input samples

Next, we quantified the proportion of sequence reads that perfectly matched one of the 1,806 designed oligonucleotide templates. Across all libraries, the mapping rates were consistently high, ranging from 78% to 84% (Fig. 2b), demonstrating the robustness and specificity of our strategy. Technical reproducibility was excellent, with replicate experiments yielding highly correlated read counts (R ≥ 0.99) under both dihydrotestosterone (DHT) and ethanol (ETH) conditions, as well as for input samples (Fig. 2c-g).

To further benchmark data quality, we compared our read counts with a previous SNPs-seq study [19] across 116 shared SNP sequences. This analysis revealed a strong cross-study correlation (R ≥ 0.76; Supplementary Fig. 1), confirming that IP-SNPs-seq not only achieves high reproducibility within experiments but also provides results that are consistent with independent datasets.

Collectively, these results establish IP-SNPs-seq as a highly reliable and reproducible platform for profiling SNP-associated sequences with high fidelity. This methodological rigor ensures that downstream analyses—particularly the interrogation of functional variants in cancer-relevant pathways—are based on robust, high-quality data.

Identification of candidate SNPs with allele-dependent AR binding

We next sought to systematically identify SNPs exhibiting allele-specific binding by the androgen receptor (AR), a central driver of prostate cancer. To this end, we calculated a biased allelic binding (BAB) score based on the ratio of sequencing reads between variant and reference alleles. BAB scores displayed a broad distribution (− 2.076 to 1.662) and were significantly correlated between DHT- and ethanol-treated conditions (R ≥ 0.46) (Fig. 3a, b and Supplementary Tables 6,7). Importantly, when BAB scores from ethanol-treated samples were ranked, those from DHT-treated samples followed the same trend (Fig. 3c, d), underscoring the robustness of the allele-specific signal.

Fig. 3.

Correlation and distribution of biased allelic binding (BAB) scores. (a, b) Correlation of BAB scores between DHT- and ETH-treated groups in LNCaP and 22Rv1 cells, respectively. (c, d) Distribution of BAB scores in DHT and ETH groups, ranked by BAB score from low to high in the ETH group. (e, f) Overlap of significant SNPs (|BAB score| ≥ 0.58) identified in DHT and ETH groups

Using an absolute BAB score ≥ 0.58 (reflecting a ≥ 1.5-fold allelic imbalance) as the threshold, we identified a substantial number of significant SNPs across two prostate cancer models: 16 (DHT) and 24 (ethanol) SNPs in LNCaP cells, and 17 (DHT) and 20 (ethanol) SNPs in 22Rv1 cells (Fig. 3e, f and Supplementary Tables 6,7). This categorization yielded three functional groups of SNPs: DHT-specific, ethanol-specific, and shared. Given that AR activation—particularly in response to DHT—is a defining hallmark of prostate cancer progression [27], these SNPs represent a set of compelling candidates for functional validation.

To prioritize biologically relevant variants, we integrated our significant SNP list with eQTL datasets from CPGEA [28], TCGA [29], and GTEx [30], as well as AR ChIP-seq profiles from CPGEA prostate samples. This rigorous multi-layered analysis converged on six high-confidence SNPs associated with allele-specific AR binding and gene regulation: rs10901849 (CTBP2), rs3800285 (FOXP4), rs5933764 (GPR143), rs6883630 (IRX4), rs7600820 (ODC1), and rs902774 (KRT8) (Fig. 4a).

Fig. 4.

Allele-specific AR binding at rs7600820, rs3800285, and rs902774. (a) ChIP–seq profiles showing histone modification enrichment (H3K4me1, H3K4me3, and H3K27ac) in cancerous prostate tissues, AR binding, and eQTL signals at the indicated loci. (b–d) AR binding preference for the A versus G alleles of rs7600820, rs3800285, and rs902774, as determined by IP–SNP–seq analysis. (e–g) Association of the risk allele—G at rs7600820 and rs902774, or A at rs3800285—with increased expression of ODC1, KRT8, and FOXP4 in the TCGA prostate cancer cohort. Boxplots indicate the interquartile range (IQR), with the median shown as the center line; whiskers extend to 1.5× IQR, with outliers plotted individually. (h–j) Correlation between AR mRNA expression and expression of ODC1, FOXP4, or KRT8 in the TCGA cohort

Notably, several of these loci are already implicated in prostate cancer susceptibility by GWAS, including CTBP2 (rs10901849) [31–33] and IRX4 (rs6883630) [34], and we previously identified CTBP2 as a functional prostate cancer locus [19, 22]. These genes carry important biological and clinical significance: CTBP2, a transcriptional co-regulator overexpressed in prostate tumors, promotes tumorigenesis and resistance to apoptosis [35–37]; IRX4, located at the 5p15 risk locus, functions as a tumor suppressor in prostate cancer but also gives rise to a micropeptide that enhances progression and chemoresistance via Wnt signaling [38]; and FOXP4, a transcription factor associated with multiple malignancies, promotes tumor growth and therapy resistance, including radioresistance through ferroptosis regulation [39]. Additionally, KRT8, a cytoskeletal intermediate filament protein, is prominently upregulated in prostate cancer and strongly associated with metastatic progression, apoptotic resistance, and adverse clinical outcomes [40–43]. Together, these data highlight the convergence of allele-specific AR binding with known cancer-relevant genes, providing mechanistic links between germline variation and prostate cancer biology.

To validate allele-specific AR occupancy, we visualized AR IP-SNPs-seq binding patterns for rs7600820, rs3800285, and rs902774, which confirmed preferential enrichment for specific alleles (Fig. 4b-d and Supplementary Table 8). eQTL analysis of TCGA-PRAD [28] further revealed that the G alleles of rs7600820, rs902774, rs10901849, and rs6883630 correlated with higher expression of ODC1, KRT8, CTBP2, and IRX4, respectively, while the A allele of rs3800285 was associated with elevated FOXP4 expression (Fig. 4e-g and Supplementary Fig. 2). Importantly, transcriptome analysis in TCGA-PRAD demonstrated strong positive correlations between AR levels and expression of ODC1 and FOXP4 (Fig. 4h-j), suggesting that germline variants at these loci influence AR-driven transcriptional programs linked to aggressive prostate cancer.

Collectively, these findings define a panel of candidate SNPs with allele-dependent AR binding that converge on established oncogenes and tumor suppressors. By functionally connecting germline variants with AR signaling and clinically relevant gene regulation, our study reveals new mechanisms by which inherited risk loci contribute to prostate cancer susceptibility and progression.

Functional analysis of candidate SNPs

To determine whether allele-dependent AR binding translates into functional transcriptional regulation, we performed luciferase reporter assays using constructs containing either the reference or variant allele of rs7600820, rs3800285, and rs902774. Among these, rs7600820 demonstrated the most consistent allele-specific effect. In both 22Rv1 and LNCaP cells, the risk-associated G allele drove significantly stronger luciferase activity than the A allele, with further enhancement upon DHT stimulation (Fig. 5a, b). This effect was recapitulated under AR overexpression, where the G allele again conferred markedly higher enhancer activity relative to the A allele (Fig. 5c, d). These results provide direct evidence that the rs7600820 locus functions as an AR-responsive regulatory element, with the G allele conferring greater transcriptional activation potential.

Fig. 5.

Functional validation of SNP rs7600820. (a, b) In 22Rv1 and LNCaP cells, the G allele of rs7600820 confers stronger luciferase enhancer activity than the A allele, with enhanced effect under DHT treatment. (c, d) G allele shows elevated enhancer activity upon AR overexpression. (e) DHT treatment increases ODC1 expression, paralleling KLK3 induction, in LNCaP cells. (f) ChIP-seq tracks show enrichment of active histone marks (H3K27ac, H3K4me1) at the rs7600820 locus in prostate tissues. (g) Hi-C analysis of LNCaP cells reveals chromatin looping between rs7600820 and the ODC1 promoter

To further investigate whether rs7600820 regulates ODC1, the nearest candidate gene, we treated LNCaP cells with DHT and monitored expression changes by RT-qPCR. Consistent with its role as an AR target, ODC1 expression increased in parallel with KLK3 induction following androgen stimulation (Fig. 5e). These findings link AR activity at rs7600820 to ODC1 transcriptional regulation.

We next examined the chromatin landscape at rs7600820 in clinical prostate cancer specimens. ChIP-seq analyses revealed strong enrichment of active enhancer marks (H3K27ac and H3K4me1) at the rs7600820 locus (Fig. 5f), supporting its functionality as an active regulatory element in vivo. Importantly, Hi-C data from LNCaP cells demonstrated chromatin looping interactions connecting the rs7600820 locus to the ODC1 promoter (Fig. 5g), establishing a direct three-dimensional regulatory relationship between the SNP-containing enhancer and ODC1 transcription.

Taken together, these results provide functional validation of rs7600820 as an AR-responsive regulatory SNP. The risk-associated G allele enhances AR-driven transcriptional activity, increases ODC1 expression, and physically engages the ODC1 promoter through chromatin looping. By mechanistically linking germline variation at rs7600820 to androgen-regulated oncogene activation, our findings highlight ODC1 as a functional effector of prostate cancer susceptibility.

Clinical association and implications of ODC1 in prostate cancer

To determine the clinical relevance of ODC1, we analyzed its expression across multiple independent cohorts, including CPGEA [28], TCGA-PRAD [29], GSE62872, GSE6919, GSE6811, and GSE21034. In every dataset, ODC1 expression was significantly upregulated in primary prostate tumors and further elevated in metastatic lesions compared to normal prostate tissues (Fig. 6a, b and Supplementary Fig. 3a-d). Elevated ODC1 expression was also consistently associated with aggressive clinicopathological features, including higher Gleason score (Fig. 6c), elevated PSA levels (Supplementary Fig. 3e), and advanced tumor stage (Fig. 6d and Supplementary Fig. 3f). These results establish ODC1 as a marker of tumor progression and poor prognosis in prostate cancer.

Fig. 6.

ODC1 functions as an oncogene in prostate cancer. (a, b) ODC1 expression is significantly higher in prostate tumors and metastases compared to normal tissues across multiple cohorts. (c, d) Elevated ODC1 expression correlates with higher Gleason score and advanced tumor stage. (e) RT-qPCR validation of ODC1 knockdown in 22Rv1 cells. (f) MTT assays show reduced cell proliferation after ODC1 silencing. (g–i) GSVA and GSEA reveal enrichment of MYC target gene programs in tumors with high ODC1 expression in CPGEA cohort.

To assess the functional role of ODC1, we performed siRNA-mediated knockdown experiments in 22Rv1 cells (Fig. 6e). Loss of ODC1 expression significantly impaired cell proliferation, as demonstrated by MTT assays (Fig. 6f), underscoring its requirement for prostate cancer cell growth. To explore the pathways through which ODC1 may promote malignancy, we carried out gene set variation analysis (GSVA) and gene set enrichment analysis (GSEA) in CPGEA and TCGA-PRAD cohorts. Both analyses revealed a strong enrichment of MYC target gene signatures in tumors with high ODC1 expression (Fig. 6g-i and Supplementary Fig. 4a-c), implicating ODC1 as an upstream regulator of MYC-driven oncogenic programs.

Together, these findings demonstrate that ODC1 is consistently overexpressed in prostate cancer and associated with clinically aggressive disease features. Functional studies confirm that ODC1 promotes tumor cell proliferation, at least in part through activation of MYC signaling pathways. These results not only position ODC1 as a promising biomarker for disease progression but also identify it as a potential therapeutic target in aggressive prostate cancer.

Discussion

In this study, we developed and applied an enhanced high-throughput platform, IP-SNPs-seq, to systematically interrogate allele-specific transcription factor binding across prostate cancer–associated risk loci. By integrating an immunoprecipitation step into our previous SNPs-seq framework, this approach enables the direct identification of transcription factor–specific binding events at SNP-containing oligonucleotides. Using an AR-specific antibody, we screened 903 prostate cancer–associated SNPs and uncovered a substantial subset exhibiting allelic differences in AR occupancy. Among these, functional validation pinpointed rs7600820 as a causal regulatory variant linked to the expression of ODC1, a key oncogenic driver that we demonstrate is directly regulated by AR and embedded within MYC-related transcriptional programs. These findings delineate a mechanistic connection between germline risk variation, androgen receptor signaling, and downstream oncogenic networks that accelerate prostate cancer progression.

Our work addresses one of the central challenges in the post-GWAS era: moving from statistical associations to molecular mechanisms. While thousands of disease-associated SNPs have been catalogued, functional annotation and mechanistic understanding remain limited, especially for noncoding variants [10, 44–47]. To address this, we previously developed SNPs-seq [19] for high-throughput detection of allele-specific protein binding differences. While effective, SNPs-seq does not identify the specific transcription factors bound to the SNP-containing oligos. The development of IP-SNPs-seq thus directly tackles this bottleneck by providing a scalable, factor-specific strategy to resolve allele-dependent regulatory interactions. By focusing on AR, the lineage-defining transcription factor in prostate cancer, we identified functional risk variants with direct regulatory impact, thereby advancing the framework for linking germline variation to transcriptional dysregulation and oncogenesis.

Our discovery that ODC1, regulated through the rs7600820 locus, acts as an AR target and oncogene in prostate cancer adds an important dimension to its emerging role in human malignancies. Previous studies have implicated ODC1 in gastric cancer [48, 49], colorectal cancer [50], hepatocellular carcinoma [51, 52], and neuroblastoma [53], where it drives proliferation, metabolic reprogramming, and aggressive phenotypes. Our findings extend this oncogenic portfolio to prostate cancer, positioning ODC1 as a clinically relevant node that integrates androgen signaling and MYC pathways—two central axes of prostate tumor biology. The observation that ODC1 expression associates with aggressive disease highlights its potential as both a prognostic biomarker and a therapeutic vulnerability, particularly given the availability of pharmacologic inhibitors targeting polyamine biosynthesis.

Beyond prostate cancer, the methodological innovation presented here has broad translational implications. The IP-SNPs-seq platform is adaptable to other transcription factors and disease contexts, providing a powerful tool to functionally annotate GWAS loci at scale. Its ability to reveal mechanistic links between germline variation, transcription factor binding, and oncogenic gene regulation creates opportunities for systematic dissection of risk alleles across diverse cancers and complex traits.

While our study demonstrates the utility of IP-SNPs-seq in identifying functional risk alleles, an important consideration remains regarding experimental design to further bolster the robustness of such high-throughput approaches. In the present work, we incorporated multiple layers of validation to guard against false discoveries, including the use of technical replicates with high reproducibility (R ≥ 0.99), cross-validation in independent prostate cancer cell lines (LNCaP and 22Rv1), and most importantly, convergence of IP-SNPs-seq signals with orthogonal datasets such as eQTLs and in vivo ChIP-seq profiles. This multi-tiered validation pipeline substantially strengthens the confidence in our high-priority candidates, such as rs7600820. Looking forward, the explicit inclusion of a set of negative control SNPs in the initial screening library would represent a valuable refinement. For instance, incorporating SNPs with weaker eQTL associations (e.g., p > 0.05) or those situated in genomic regions with low TF-binding potential (e.g., heterochromatic domains) would establish an empirical null distribution. Such an approach would enable a more data-driven and stringent estimation of false discovery rates (FDR) and facilitate the refinement of significance thresholds, such as the BAB score cutoff, thereby enhancing the robustness of the IP-SNPs-seq platform and the definitiveness of its findings in future applications across diverse transcription factors and disease contexts.

In summary, our study establishes IP-SNPs-seq as a versatile and robust technology for functional SNP annotation and identifies AR-rs7600820-ODC1 as a critical regulatory axis in prostate cancer. By bridging genetic susceptibility, AR signaling, and MYC-driven oncogenesis, these findings not only advance mechanistic understanding of prostate cancer risk but also point to actionable biomarkers and therapeutic strategies. Together, the methodological and biological insights provided by this work hold broad significance for translating genetic discoveries into mechanistic and clinical advances across human disease.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Conceptualization, P.Z., G.-H.W., and L.W.; methodology, W.X., Q.-X.Z., and L.Q.; software, W.X., Q.Z.; validation, Q.-X.Z., L.Q., Z.W., T.W., D.D.; formal analysis, W.X. and P.Z.; investigation, Q.-X.Z., L.Q.; resources, L.W., G.-H.W., and P.Z.; data curation, G.-H.W. and P.Z.; writing—original draft preparation, P.Z. and G.-H.W.; writing—review and editing, W.X., Q.-X.Z., L.Q., Z.W., T.W., D.D., Q.Z., L.W., G.-H.W., and P.Z.; visualization, W.X. and P.Z.; supervision, L.W., G.-H.W. and P.Z.; project administration, G.-H.W. and P.Z.; funding acquisition, L.W., G.-H.W. and P.Z. All authors have read and agreed to the published version of the manuscript.

Funding

This work was supported by the National Natural Science Foundation of China (82372628 to P.Z.; 82311530050 and 82073082 to G.-H.W.), the Shanghai Interactional Collaborative Project (23410713300), the National Key Research and Development Program of China (2022YFC2703600), Sigrid Jusélius Foundation, Syöpäjärjestöt, and Fudan University Recruit Funding to G.-H.W., as well as the National Institute of Health (R01 CA250018-01) to L.W. The APC was funded by the National Natural Science Foundation of China (82372628).

Data availability

All data are available in the main text and the supplementary information, or from the corresponding author upon reasonable request.

Declarations

Competing interests

The authors declare no competing interests.