Genotype-driven variations in lncRNA expression underlie predisposition to high-grade serous ovarian cancer

Graphical abstract

Highlights

• •lncRNAs contribute to a greater proportion to HGSOC heritability than PCGs.
• •We mapped lnc-eQTLs and PCG-eQTLs in HGSOC tumors.
• •We prioritized causal genes in HGSOC tumors and whole blood.
• •We characterized a single-cell full-length transcriptome atlas in HGSOC tumors.
• •We validated the role of CRHR1-IT1 in promoting apoptosis and inhibiting the growth of HGSOC cells.

Abstract

Introduction

New effective early diagnosis strategies and therapeutic targets are desperately needed for high-grade serous ovarian cancer (HGSOC). Previous functional interpretation of HGSOC genome-wide association studies (GWAS) have largely overlooked the role of long non-coding RNAs (lncRNAs).

Objectives

Our study aims to identify novel causal genes, particularly lncRNAs, that contribute to HGSOC susceptibility and to explore whether their may provide preliminary insights relevant to early detection or therapeutic development.

Methods

We mapped genotype effects on the expression of both protein-coding genes (PCGs) and lncRNAs in 348 HGSOC tumor samples. By performing Bayesian colocalization analysis, we prioritized causal genes in HGSOC tumors and whole blood, respectively. Next, we generated a single-nucleus full-length transcriptome map of HGSOC tumor using snRandom-seq. Furthermore, we conducted in vitro assays to validate the functional role of lncRNA CRHR1-IT1 in HGSOC.

Results

Our findings reveal that genetic variants significantly drive expression changes in both PCGs and lncRNAs. Notably, lncRNAs contribute a greater proportion of HGSOC heritability than PCGs. Using Bayesian colocalization analysis, we prioritized 22 high-fidelity causal genes in HGSOC tumors and whole blood, including ARL17A and CRHR1-IT1. Through snRandom-seq, we characterized the single-cell expression patterns of candidate susceptibility PCGs and lncRNAs in HGSOC tumors. Furthermore, we showed that elevated CRHR1-IT1 expression promotes apoptosis and inhibits the growth of HGSOC cells in vitro.

Conclusion

Our study highlights the significant regulatory role of genetic variants on both protein-coding genes and lncRNAs in HGSOC. While these findings enhance our understanding of HGSOC susceptibility and may inform future efforts toward biomarker discovery or immune-related therapeutic exploration.

Introduction

Invasive epithelial ovarian cancer (EOC) is among the leading causes of gynecologic cancer in women worldwide, with greater than 314,000 new diagnoses and approximately 207,000 deaths each year [1]. Histological subtypes of invasive EOC vary in epidemiological [2], [3], [4] and genetic risk factors [5], clinical responses to platinum-based therapy [6], [7], and cells of origin [8], [9]. High-grade serous ovarian cancer (HGSOC) is the most prevalent histotype, accounting for 60–70 % of EOC cases and the majority of EOC mortality [10], [11]. Current treatments mostly rely on traditional cytoreductive surgery combined with platinum-based chemotherapy, and the overall 5-year survival rate is still below 20 % for HGSOC due to the scarcity of effective early diagnosis strategies and the rapid development of chemoresistance [12], [13], [14], [15], [16], [17]. Therefore, new oncogenes and tumor suppressors that could serve as effective diagnostic and therapeutic targets for HGSOC, are desperately needed. To prioritize the best targets, it is of critical importance to understand the germline genetic architecture of HGSOC, as previous studies consistently confirmed that genetically supported drug targets could substantially contribute to the successful development of new therapeutics [18], [19].

HGSOC has a significant germline genetic component, with an estimated heritability around 20 to 40 % [15], [17], [20], [21]. Known germline loss-of-function mutations in susceptibility genes, such as BRCA1, BRCA2, TP53, BRIP1, MSH6, and RAD51C/D, only account for about 10 % of cases [22], [23], [24], [25], [26], [27]. The majority of genetic risk factors for HGSOC is due to common low-penetrance-susceptibility alleles [28]. Large-scale genome-wide association studies (GWASs) have so far identified 19 loci, where common single-nucleotide polymorphisms (SNPs) show statistically significant reproducible association with HGSOC predisposition [29], [30], [31], [32], [33], [34], [35], [36], [37], [38], suggesting that germline genetic variants influence tumorigenesis. However, interpreting the likely biological mechanism of GWAS is challenging, as more than 90 % of risk variants lie in the non-protein-coding regions of the genome. Expression quantitative trait locus (eQTL) analysis has been a powerful approach to map risk variants to disease-causing genes [39], [40], [41]. By integrating of eQTL and GWAS, several previous studies have successfully prioritized putative susceptibility genes for HGSOC risk loci, such as HOXD9 [42], TERT [42], CHMP4C [28], ANKLE1/ABHD8 [29], and GNL2 [43].

Although long non-coding RNAs (lncRNAs) dominate the transcriptome in human cells, most efforts to interpret the functional consequences of non-coding risk variants have primarily focused on their regulatory effects on protein-coding genes (PCGs), greatly neglecting lncRNAs, which have been increasingly shown to play critical roles in chemoresistance, tumorigenesis, tumor progression, and metastasis [29], [44], [45], [46], [47], [48], [49], [50], [51]. LncRNAs are likely to act as scaffolds, guides, or decoys to alter DNA-protein binding or protein–protein interactions [49], [52], resulting in dysregulation in cell proliferation, differentiation and apoptosis [53], [54]. Previous investigations of the correlation between disease-associated SNPs and lncRNA expression have successfully nominated several potential disease-causing lncRNAs, such as PCAT1 in promoting prostate cancer cell growth [55] and LINC01475 in influencing ulcerative colitis susceptibility [56]. Therefore, systematic interrogation of the regulation from germline variants on lncRNAs (lncRNA-eQTLs) could provide new insights into how risk variants contribute to HGSOC pathogenesis via non-coding genome and improve new therapeutic targets identification.

In this study, we generated a comprehensive atlas of eQTLs for both PCGs and lncRNAs, utilizing 348 HGSOC tumor samples with matched high-quality genotype and gene expression data from the Cancer Genome Atlas (TCGA) [57]. Our findings indicate that germline variants play crucial roles in modulating the expression of not only PCGs but also lncRNAs, which could further contribute to HGSOC predisposition. Using Bayesian colocalization analysis [58], we prioritized 10 susceptibility genes for HGSOC, comprising 6 PCGs and 4 lncRNAs, where genotype-driven gene expression changes are highly associated with HGSOC risk in high-grade ovarian tumors. Five of these genes, such as ARL17A and CRHR1-IT1, also showed an association with HGSOC predisposition in whole blood. We characterized the expression patterns of susceptibility PCGs and lncRNAs using single-nucleus resolution transcriptome maps of the human HGSOC tumor. In addition, we conducted in vitro assays and validated the functional role of CRHR1-IT1 in promoting apoptosis of ovarian cancer cells. Our study may help inform future efforts toward early detection and the development of new therapeutic strategies for HGSOC.

Materials and methods

Genotype and RNA-seq data processing for eQTL analysis

We downloaded matched genotype and quantified gene expression data for 350 primary HGSOC tumor samples from the TCGA data portal (https://tcga-data.nci.nih.gov/tcga/) [57]. The TCGA germline variants data were genotyped using the Affymetrix Genome-Wide Human SNP Array 6.0. Quality control of raw genotype data was performed using PLINK (v2.0) [59] (https://www.cog-genomics.org/plink2/). We excluded samples with misgendering, contamination, or low call rates. SNPs on sex chromosomes, monomorphic SNPs, and those with a minor allele frequency (MAF) 1 %, Hardy–Weinberg equilibrium P-value <1.0E-06, or call rate <95 % were removed as well. In total, we excluded 2 contaminated samples with high PIHAT values (>0.2). This left 348 individuals and 771,697 SNPs for further analysis (Table S1). To better understand the genetic background of the TCGA HGSOC samples, we performed Principal Components Analysis (PCA) on the genotype data of these 348 HGSOC subjects, along with 2,504 genotyped data from the 1,000 Genomes Project Phase 3 [60], using Genome-wide Complex Trait Analysis (GCTA) [61]. Most samples in our study clustered amongst the European populations, as shown in Fig. S1. The first 2 principal components (PCs) were used in the linear regression model for eQTL analysis.

To increase eQTL discovery power, autosomal variants were pre-phased by SHAPEIT2 [62] and subsequently imputed with IMPUTE2 [63], using 1,000 Genomes Project Phase 3 as the reference panel. Before imputation, indels and SNPs absent or incompatible with the 1,000 Genomes Project dataset were removed. Imputed SNPs with MAF <1 %, Hardy-Weinberg Equilibrium P-value <1E-06, and imputation confidence score INFO (a measure of r²) <0.4 (estimated by SNPTEST [64] were further excluded. Consequently, 8,579,445 high-quality autosomal SNPs were used for eQTL analysis.

The TCGA level 3 RNA-seq data were aligned to the human genome based on GENCODE v22 annotations [65]. Next, we normalized the downloaded raw read counts across samples using the TMM method and then estimated gene expression levels as transcripts per million (TPM). Only genes with at least 0.1 TPM in more than two samples were identified as expressed and used for downstream eQTL analysis.

eQTL mapping

We performed cis-eQTL (further referred to as eQTL) mapping using 348 primary HGSOC samples for both protein-coding genes (PCGs) and lncRNAs (Table S2). Expression measurements for each PCG or lncRNA were quantile-normalized and then transformed into a rank-based standard normal expression using inverse normal-transformed (INT). To remove the effects of latent confounders, we estimated 40 and 60 PEER [66] factors on INT expression by adjusting for age, batch effect, and top 2 genotype PCs for PCG-eQTL and lncRNA-eQTL identification, respectively. We selected the number of PEER factors that maximized the detection of ePCGs and elncRNAs (Fig. S2), that is, PCGs and lncRNAs significantly associated with at least one SNP. eQTL mapping was then performed using the INT gene expression as the outcome and SNP dosage as the independent variable, adjusting for age, batch effect, top 2 genotype PCs, and the estimated PEER factors. Nominal P-values were calculated for all SNPs located within ±1 Mb of PCGs/lncRNAs using MatrixEQTL [67] with an additive linear regression mode. A permutation procedure was then implemented using FastQTL [68] with the setting “-permute 10000”. We adopted Beta distribution-adjusted empirical P-values to calculate q using Storey’s q method [69], and identified eGenes with a q cutoff ≤0.05. Next, we defined an empirical P threshold, P_t, as the empirical P of the PCG/lncRNA closest to the 0.05 FDR threshold. P_t was used to estimate the nominal P cutoff for each gene based on the Beta distribution model of the minimum P distribution f(P_min), generated from the permutations for the gene. SNPs having a lower nominal P than the gene-level cutoff were identified as eSNPs, i.e., SNPs that significantly regulated at least one PCG or lncRNA [40].

Differentially expressed genes identification

To minimize computational batch effects causing by different data analysis pipelines in TCGA and GTEx (v7; https://gtexportal.org/home/) [70] cohorts, we downloaded the recomputed gene expression data in the ovaries of the TCGA HGSOC patients and GTEx healthy samples using a consistent analysis protocol from UCSC Xena (https://xena.ucsc.edu/) [71]. The estimated gene expression was quantified as transcripts per kilobase million (TPM). Genes with TPM <0.1 in <5 % of samples were excluded from downstream analysis. Differentially expressed genes (DEGs) between HGSOC tumor and healthy samples were identified using a linear model combined with an empirical Bayes method in the R package limma [72]. This linear regression model adopted gene expression as the outcome and disease status as independent variable, adjusting for age. Genes with a Benjamini-Hochberg FDR <0.05 were identified as DEGs.

Colocalization analysis

To evaluate the posterior probabilities of inherited HGSOC risk and expression of both PCGs and lncRNAs sharing common genetic causal variants in a given region, we performed colocalization analysis on summary data of HGSOC GWAS [33] and eQTL data from HGSOC tumors and whole blood (from GTEx v7) [70] using coloc v5.2.3 [58]. We adopted the Phelan et al. HGSOC GWAS study based on 13,037 cases and 40,941 controls [33]. HGSOC associated variants with a P-value <5E-08 were selected. Accounting for the inflation caused by GWAS variants representing the same signal, we performed LD pruning using swiss (https://github.com/statgen/swiss) by discarding variants in LD r² ≥0.8 with the lead SNP at each locus. Genomic regions within ±100 kb of each pruned GWAS variant, that overlap with ePCGs or elncRNAs, were used in the coloc analysis. Posterior probability was then estimated using summary data of HGSOC GWAS and PCG-eQTLs/lncRNA-eQTLs to identify the colocalization regions. In the coloc results, H3 tests the hypothesis that both traits (e.g., HGSOC risk and gene expression) are associated but with independent causal variants, whereas H4 tests the hypothesis that both traits share the same causal variants (colocalization). In this study, coloc examines whether HGSOC risk and gene expression are associated and share the same causal variants (H4). We used PP4 (the posterior probability of H4) >0.8 as the threshold of colocalization, following our previous work and multiple high-impact colocalization studies that adopt the same posterior probability thresholds to ensure high-confidence signals [58], [73], [39], [40], [41]. Notably, PP4 is a Bayesian posterior probability rather than a frequentist P-value; therefore, multiple testing correction is not required, as coloc directly quantifies the probability of shared causal variants within each tested region under predefined priors. This approach is consistent with the methodological guildelines of coloc and with standard practice in the colocalization literature [58].

Sample procurement for snRandom-seq

Six high-grade tumor specimens used for single-nucleus Random-seq were obtained from the Biobank of Zhejiang Cancer Hospital (Table S3). One specimen was collected from the intestinal surface during primary debulking surgery, and one was obtained from adipose tissue during surgery for platinum-resistant recurrence. The remaining four specimens were collected from ovarian tissues during primary debulking surgery. All specimens were confirmed as high-grade tumors by pathologically review.

snRandom-seq-based FFPE single-nucleus transcriptome library construction and sequencing

We utilized the VITApilote High-Throughput Single-Cell Transcriptome Kit for FFPE samples (M20 Genomics, Cat# R20123124) for pre-library processing, library construction, and purification. The VITAcruizer Single-Cell Preparation System DP400 (M20 Genomics, Cat# E20000131) was employed for single-cell partitioning, encapsulation, and nucleic acid capture. The experiments were carried out according to the manufacturer’s instructions for the snRandom-seq workflow. The main workflow was as follows: FFPE samples were cut from paraffin block and washed twice with 1 mL Xylene (Aladdin, Cat# X112054) for 5 min at room temperature to remove the paraffin. The samples were then gently rehydrated by immersing them in a graded series of ethanol solutions (100 %, 90 %, 70 %, 50 %, and 30 %; Aladdin, Cat# E130059), starting with 100 % ethanol and ending with 30 % ethanol and subsequently transferred to water. The rehydrated tissue sections were permaeablized under mild conditions to maintain nuclear integrity, followed by in situ blocking of single-stranded DNA. Random primers were then added to initiate in situ reverse transcription of total nuclear RNA within the tissue section, generating cDNA fragments representative of the nuclear transcriptome. These cDNA fragments were subsequently ligated to universal sequencing adapters within the nuclei prior to nuclear isolation. The tissue section was then lysed into individual nuclei by gentle mechanical disruption and enzymatic digestion, and the resulting suspension was filtered through a 20 μm cell strainer (pluriSelect, Cat#43-10020-20). The reverse-transcribed single-nucleus suspension was mixed with reagents and barcoding beads containing 30-nt cell barcodes and 8-nt unique molecular identifiers (UMIs). This mixture was encapsuled, captured and barcoded using the VITAcruizer Single-Cell Preparation System DP400, resulting in barcoded cDNA strands. The barcoded cDNA was then extended, purified, and PCR-amplified to generate sequencing libraries containing P5 and P7 adapters. The libraries were sequenced on a Novaseq 6000 platform (Illumina) with 150 bp paired-end reads (Table S3).

Compared with conventional droplet-based sc/snRNA-seq platforms (e.g., 10x Genomics), snRandom-seq employes a random-primed reverse transcription strategy that captures a broader spectrum of nuclear transcripts, including non-polyadenylated RNAs. This feature makes it particularly suitable for profiling lncRNA expression from preserved or FFPE tumor tissues.

SnRandom-seq data processing

Primer sequences and adaptors were trimmed from the raw data. The UMI (8nts) and cell-specific barcode (30nts) for each Read1 were then extracted. Sequenced barcodes were merged if they could be uniquely assigned to the same accepted barcode with a Hamming distance of 2 nts or less. Sequencing reads were aligned to the human genome based on GENCODE v22 annotations [65] using the STARsolo module of the STAR aligner (version 2.7.10a) with appropriate parameters [74]. The expression data were further analyzed with Seurat (v5.0.2) [75]. For quality control, we excluded features expressed in fewer than 5 cells. Low-quality cells with fewer than 200 detected features or more than 5,000 features were also excluded. In addition, cells with more than 40 % mitochondrial transcripts were filtered out. Samples were integrated and batch effects were removed using the FindIntegrationAnchors function and the RunHarmony function [75], [76], [77]. PCA was performed and the first 35 Harmony-adjusted PCs were selected to generate a k-nearest neighbor graph with k = 5. Uniform manifold approximation and projection (UMAP) embeddings with a resolution of 0.8 were generated to visualize the data [78].

Human gynecologic tumor scRNA-seq data

We used the RNA-seq count matrix of 75,532 cells profiled from human gynecologic tumors of 11 patients (GSE173682) [79]. Data processing and normalization were performed using R package Seurat (v5.0.2) [75]. Features expressed in fewer than 5 cells were removed, and gene expression values were normalized with the default Seurat workflow. The top 3,000 highly variable genes were selected for downstream scaling. Cluster assignment of the cells was based on the annotations in Fig. S4 of the published dataset [79].

Cell culture and transfection experiment

The human ovarian adenocarcinoma cell line OVCAR­3 (ATCC® HTB-161^TM) was purchased from ATCC (USA) and cultured in Roswell Park Memorial Institute 1640 Medium (RPMI-1640; Cat# 30­2001) supplemented with 20 % fetal bovine serum (Gibco, USA) and 0.01 mg/mL bovine insulin (Sigma, Cat# I6634). The human ovarian carcinoma cell line A2780 was a gift from Dr. Guangyi Jiang (Zhejiang Cancer Hospital) and cultured in the Dulbecco’s Modified Eagle Medium (DMEM, Cat# C11995500BT) supplemented with 10 % fetal bovine serum. Cells were incubated at 37 °C with 5 % CO₂, and the medium was changed every two days.

The CRHR1-IT1 sequence was synthesized according to the full-length sequence with an added poly-A tail based on the LINC02210 sequence, NR_026680.3 in NCBI (Table S4), and then subcloned into a pcDNA3.1 (+) vector (Genscript, Shanghai, China). OVCAR­3 and A2780 cells were seeded in a 24-well plate and incubated at 37 °C for 8 h. Next, the overexpression vector pcDNA3.1(+)-CRHR1-IT1 was transfected into OVCAR­3 and A2780 cells using Lipofectamine 3000 (Invitrogen, USA) and Opti-MEM (Thermo Fisher Scientific, USA), following the manufacturer’s protocol. Transfection conditions of OVCAR-3 included control (0 μg/well vector, null condition), mock (0 μg/well vector, baseline transfection reagents), vehicle (1 μg/well pcDNA3.1(+) vector), and CRHR1-IT1 overexpression at 0.25, 0.5, and 1 μg/well pcDNA3.1(+)-CRHR1-IT1 vector. Transfection conditions of A2780 included control (0 μg/well vector, null condition), mock (0 μg/well vector, baseline transfection reagents), vehicle (1 μg/well pcDNA3.1(+) vector), and CRHR1-IT1 overexpression at 0.03125, 0.0625, and 0.125 μg/well pcDNA3.1(+)-CRHR1-IT1 vector. After 6 h of transfection, cells were washed once with 1X PBS, fresh RPMI-1640 or DMEM medium was then added, respectively, and cells were cultured for 48 h post-transfection. RNA samples were collected from control, mock, vehicle, and CRHR1-IT1 overexpression conditions at the indicated vector concentrations at 24 and 48 h for sequencing. In total, 24 RNA-seq datasets from OVCAR-3 cells were obtained, with two biological replicates per transfection condition.

RNA isolation and quantitative real-time PCR (qRT-PCR)

Cells were lysed, and total RNA was isolated using the PureLink® RNA Mini Kit (Thermo Fisher Scientific, USA) according to the manufacturer's protocol. 1 μg of RNA was used to synthesize cDNA using the PrimeScript™ RT reagent Kit with gDNA Eraser (Takara, Japan). Subsequent qRT-PCR was performed using SuperReal PreMix Plus (SYBR Green) (Tiangen, China). The expression of mRNAs and lncRNAs was normalized to GAPDH. Relative gene expression was calculated by the 2^−ΔCT method. The primer sequences for GAPDH, CRHR1-IT1, CMYC, CRHR1, FAS, JUN, and PARP1 are presented in Table S5. All qRT-PCR reactions were performed using a CFX96 TOUCH/GelDoc XR + qRT-PCR machine (Bio-Rad, USA), with three technical replicates for each reaction.

Reads processing of collected RNA-seq data from OVCAR-3 cells

RNA was isolated and purified using TRIzol reagent (Invitrogen, Carlsbad, CA, USA). RNA quality was assessed with the Agilent Bioanalyzer 2100 (Agilent, CA, USA), with RIN scores above 7, and confirmed by electrophoresis with denaturing agarose gel. Poly (A) RNA was purified from 1 μg total RNA using Dynabeads Oligo (dT) 25-61005 (Thermo Fisher, CA, USA) through two rounds of purification. The poly (A) RNA was then fragmented into small pieces using the Magnesium RNA Fragmentation Module (NEB, cat.e6150, USA) at 94 °C for 5–7 min. The cleaved RNA fragments were reverse-transcribed to cDNA using SuperScript™ II Reverse Transcriptase (Invitrogen, cat. 1896649, USA). Strand-specific RNA-seq libraries were prepared using the dUTP method. 2 × 150 bp paried-end sequencing was then performed on the Illumina NovaSeq X Plus. Next, sequence quality was assessed using FastQC (v0.11.9). Adaptors and low-quality bases were trimmed using Trim-galore (v0.6.6). The remaining RNA-seq reads were then aligned to the human genome (hg19/GRCh37) using STAR (v2.7.10a) [80] based on GENCODE v24 annotations [81]. Gene expression levels were quantified using RSEM (v1.3.1) [82].

Weighted gene co-expression network analysis (WGCNA)

Weighted Gene Co-Expression Network Analysis (WGCNA) was performed on 24 RNA-seq data using the R package WGCNA [83]. The top 50 % variablely expressed genes were used. Modules with at least 30 genes were selected. The expression correlation between CRHR1-IT1 and genes in each module was then estimated. Genes with the absolute value of gene significance (GS) > 0.75 and module membership (MM) >0.75 were defined as key genes for that module.

Functional enrichment analysis and network construction

Functional enrichment analysis was performed using Metascape (https://metascape.org/) with default settings [84]. Network analysis was implemented in ConsensusPathDB (https://cpdb.molgen.mpg.de) [85], which integrates various interaction networks, including binary and complex protein–protein interactions, genetic interactions, metabolic pathways, signaling pathways, gene regulatory networks, drug-target interactions, and biochemical pathways.

Incucyte apoptosis assay for CRHR1-IT1

After 6 h of cell transfection, cells were washed once with 1X PBS and added fresh RPMI-1640 or DMEM medium, respectively. CellEvent™ Caspase-3/7 Green Detection Reagent (Thermo Fisher Scientific, USA) was then added to the cells, which were subsequently incubated in an Incucyte SX5 incubator for 42 h.

Statistical analysis of qRT-PCR and cell apoptosis assays

Statistical analysis of qRT-PCR and cell apoptosis assays was performed using GraphPad Prism 8.0. Data are presented as the mean ± SE, and experiments were repeated at least three times. Two-sided P-values were calculated using unpaired t-test analysis for the qRT-PCR assay. For the cell apoptosis assay, two-sided P-values were estimated using one-way ANOVA. Results were considered significant at *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001.

Results

Genotype-driven expression changes of PCGs and lncRNAs

To understand the contribution of germline variants to expression changes, we investigated the association between genotype and local expression variations of both PCGs and lncRNAs (Fig. 1A). After stringent quality control (Materials and methods), we selected 348 HGSOC tumor samples (Table S1) with matched high-quality genotype and gene expression data to identify PCG-eQTLs. And for lncRNA-eQTL analysis, we adopted quantified expression data of lncRNAs from the same 348 HGSOC tumor samples. We performed eQTL analysis using a linear regression model for PCGs and lncRNAs, respectively, and interrogated SNPs located within a ±1 Mb (cis) window of each PCG or lncRNA. The PCG-eQTL analysis identified 7,145 eGenes (that is, genes significantly associated with at least one SNP) at a Storey’s q-value ≤0.05. Our results successfully replicated the previously identified significant SNP-gene pairs in the ovary from the GTEx with π₁ = 0.99 (where π₁ represents the proportion of true positives) [86], aligning with our own and others’ previous identification of genotype-driven expression changes (eQTLs) from the same tissue [39], [41], [87]. The number of identified eGenes is also in line with previous eQTL study in 44 human tissues from the GTEx Consortium (Fig. S3) [86]. For example, we found that the genetic variant rs35478347 had a robust effect (two-sided P-value = 6.0E-06) on DHX58 expression (Fig. 1B). And the lncRNA-eQTL analysis identified 572 elncRNAs (that is, lncRNAs significantly associated with at least one SNP) (Table S6). For instance, we observed that the expression of the lncRNA LOC284581 (hg19 chr1:205,781,868–205,875,038) was significantly correlated with rs9438393 (two-sided P-value = 3.02E-74) in HGSOC tumor samples (Fig. 1C). Intriguingly, we found that the variance explained by the top associated SNPs decreased dramatically from that for lncRNA expression to protein-coding gene expression and to HGSOC risk (Fig. 1D), suggesting that expression changes of lncRNAs explain a higher portion of HGSOC heritability than PCGs. Moreover, the regulatory significance of the lead SNP on each ePCG or elncRNA increased for SNPs closer to the transcription start site (TSS), suggesting that germline variations in promoter regions have stronger regulatory effects on expression of both PCGs and lncRNAs (Fig. 1, Fig. 1). In addition, the regulatory effects of germline variants on lncRNAs were generally stronger than on PCGs (Fig. 1, Fig. 1), consistent with a previous study indicating that lncRNA eQTLs tend to have larger effect sizes than protein-coding eQTLs [56].

Fig. 1.

Genotype-driven expression changes of PCGs and lncRNAs contribute to HGSOC predisposition. (A) Schematic representation of PCG-eQTL and lncRNA-eQTL; genotype-expression association study. Expression changes regulated by a nearby SNP. (B) An example of PCG-eQTL. The genotype of SNP rs35478347 and the gene expression of DHX58 association in HGSOC tumors (N = 348 samples; two-sided P-value = 6.0E-06). (C) An example of lncRNA-eQTL. The genotype of SNP rs9438393 and the expression of the lncRNA LOC284581 association in HGSOC tumors (N = 348 samples; two-sided P-value = 3.02E-74). Center lines show the medians; box limits indicate the 25th and 75th percentiles; whiskers extend to the 5th and 95th percentiles. Two-sided P-value was calculated by linear regression model. (D) Distribution of the variance explained in expression of PCGs, lncRNAs, and HGSOC predisposition using the best eSNPs (the lead eQTL). Two-sided P-values were calculated using Student's t-test. Statistical significance is indicated by *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001. (E) The strength of association (y-axis) of the best eSNPs (the lead eQTL) decreases as the distance (x-axis) from the transcription start site of PCGs increases (N = 348 samples). (F) The strength of association (y-axis) of the best eSNPs (the lead eQTL) decreases as the distance (x-axis) from the transcription start site of lncRNAs increases (N = 348 samples). (G) Both genetic variants and environmental factors contribute to differentially expressed patterns between HGSOC tumor and healthy samples. Genotype-driven expression changes can be uncovered by eQTL analysis, while differentially expressed gene (DEG) identification detects HGSOC-associated genes. (H) Volcano plot. The x-axis is the negative base 2 log of the fold change in gene expression between HGSOC tumor samples (N = 419) and healthy samples (N = 88). The y-axis represents the negative base 10 log of the significance P-value estimated from DEG identification. (I) The number of eGenes influenced by at least one genetic variant, identified by eQTL analysis. DEGs were identified between HGSOC tumor samples (N = 419) and healthy samples (N = 88). Overlapping DEGs, including 404 PCGs and 94 lncRNAs, could be driven by underlying germline variants.

Given the substantial genotype-driven expression changes, we next investigated the potential role of germline variants in HGSOC predisposition (Fig. 1G). First, we identified differentially expressed genes (DEGs) in the ovaries of the HGSOC patients (N = 419 from TCGA) and healthy samples (N = 88 from GTEx). We identified 2,689 up-regulated and 4,616 down-regulated genes as DEGs (Fig. 1H), with 76.9 % being protein-coding genes and 19.9 % being lncRNAs (Fig. S4). To ascertain the contribution of germline variants in HGSOC predisposition, we overlapped our DEGs and eQTL results. We found that about 6.82 % of the identified DEGs, including 404 PCGs and 94 lncRNAs, could be driven by underlying germline variants (Fig. 1I and Table S7). For instance, the eGene DHX58, whose expression changes were regulated by germline variant rs35478347 (Fig. 1B), also showed significant differences (two-sided P-value = 1.31E-43) in expression between HGSOC tumors and healthy samples (Fig. S5).

In summary, our results suggested that underlying germline variants play important roles in driving expression changes of both PCGs and lncRNAs, which could further contribute to the HGSOC-associated DEGs.

Genotype-driven expression changes are enriched for HGSOC-associated GWAS loci

As the vast majority of GWAS-identified HGSOC risk loci reside in non-coding regions of the genome, these variants might function through regulatory effects on the expression of putative causal PCGs or lncRNAs (Fig. 2A). To understand the underlying biological mechanisms of the observed statistical association between SNPs and HGSOC predisposition, we first overlapped genome-wide significant loci in HGSOC GWAS [33] and in our eQTL signals for both PCGs and lncRNAs, respectively. Genotype-driven expression changes of 51 PCGs and 20 lncRNAs showed associations with HGSOC risk (Fig. 2B). Several putative HGSOC risk genes, such as HOXB4, LERK1, MYO19, MEIOB, MRRF, C3orf33, and CCDC23, have been identified as associated with epithelial ovarian cancer in previous independent studies [28], [88], [89], [90], [91], [92], [93]. Although the top significant GWAS signals were enriched on chromosomes 3, 4, 8, and 9, we observed a strong enrichment of candidate susceptibility genes for HGSOC on chromosome 17 after overlapping with eQTLs. These loci are adjacent to the known EOC and breast cancer marker gene BRCA1 on chromosome 17. Intriguingly, out of these 51 PCGs, 16 genes (about 31 %) have also been previously implicated as candidate breast cancer susceptibility genes (Table S8), including MAPT [28], [94],LGALS7 [95],TBC1D3 [96],IQCG [97] and CCL4 [98] etc. In addition, the elncRNA ERVK13-1 (Endogenous Retrovirus Group K13 Member 1) has previously been shown an association with breast cancer [99]. It appears that HGSOC and breast cancer share an extensive expression-based pleiotropy at these loci, consistent with previous observations [28], [34].

Fig. 2.

Integrative analysis of HGSOC GWAS with tumor and blood eQTLs reveals candidate susceptibility genes and regulatory networks. (A) Illustration of the integration analysis of HGSOC GWAS with eQTL data for PCGs and lncRNAs. (B) Miami plots showing overlapped loci that reached genome-wide significance in HGSOC GWAS and in our eQTL data for both PCGs (top) and lncRNAs (bottom), respectively. The orange dots define prioritized PCGs and lncRNAs. The x-axis represents chromosomal location. The y-axis represents the negative base 10 log of the significance P-value. Two-sided P-value was obtained from the HGSOC GWAS study [33]. (C) Illustration of the Bayesian colocalization analysis. This method integrates GWAS and eQTL data to identify loci where the causal variants associated with HGSOC and gene expression are shared (colocalization). (D) We identified 15 protein-coding genes and 7 lncRNAs, where HGSOC predisposition (GWAS) and gene expression (PCG-eQTL/lncRNA-eQTL) share causal genetic variants. Green circles denote colocalized genes identified in HGSOC tumors. Yellow circles denote colocalized genes identified in whole blood. (E) An example of DEG, ARL17A, between HGSOC tumor samples (N = 419) and healthy samples (N = 88). Two-sided P-value = 3.98E-48. (F) An example of differientially expressed lncRNA, CRHR1-IT1, between HGSOC tumor samples (N = 419) and healthy samples (N = 88). Two-sided P-value = 5.30E-45. Center lines show the medians; box limits indicate the 25th and 75th percentiles; whiskers extend to the 5th and 95th percentiles; outliers are represented by dots. (G) Expression correlation between the lncRNA KANSL1-AS1 (x-axis) and BRCA2 (y-axis) in HGSOC tumor samples (N = 348). Two-sided P-value = 0.04. (H) Pathway analysis performed by GIANT 2.0 [113], suggesting that colocalized genes KANSL1, LRRC37A2, and ARL17A, were strongly co-expressed in CD8⁺ T cells and enriched for chromatin organization and regulation of nitrogen compound metabolic process.

Next, we performed a Bayesian statistical framework established in colocalization (coloc) analysis [58] to prioritize disease-causing genes, whose genotype-driven expression changes are associated with GWAS signals (Fig. 2C), by integrating HGSOC GWAS with the ovarian tumor eQTL data. Our analysis nominated 10 susceptibility genes for HGSOC, including 6 PCGs and 4 lncRNAs (Fig. 2D and Table S9). Consistently, all 6 PCGs, CHMP4C, ARL17A, ARL17B, LRRC37A, LRRC37A2, and MAPT, have been previously identified as HGSOC susceptibility genes using a transcriptome-wide association study (TWAS) [28], [100], [101]. This cross-method concordance, based on distinct statistical assumptions, strengthens the robustness of our colocalization findings. Notably, around ±100 kb window of SNP rs142045106, we identified one protein-coding gene MAPT and 3 lncRNAs, including CRHR1-IT1, KANSL1-AS1, and LOC102724391, as putative disease-causing genes for HGSOC. This suggests a likely functional convergence of MAPT and its adjacent lncRNAs at this locus. Emerging studies have reported that MAPT expression showed an association with estrogen receptor expression in breast cancer, and an imperfect estrogen-response element has also been identified in the promoter region of MAPT [94], [102], [103]. In addition, the tau protein encoded by MAPT, was found almost exclusively bound to microtubules, which plays a pivotal role in mitosis in cancer development as well [104].

Germline variant-regulated transcriptional signals in whole blood and ovarian tumors provide insights into HGSOC biology

HGSOC is often diagnosed at a very late stage and is characterized by an immune-suppressive tumor microenvironment (TME) with limited tumor-infiltrating lymphocytes, rendering it largely refractory to checkpoint inhibitor immunotherapies [105], [106]. To explore whether transcriptional changes regulated by germline variants might offer biological insights relevant to HGSOC, we investigated putative disease-causing genes in whole blood by integrating HGSOC GWAS with whole blood eQTL data (from GTEx v7) [70] using Bayesian-test-based colocalization analysis [58]. As illustrated in Fig. 2D, our analysis nominated 12 PCGs and 5 lncRNAs in whole blood. Of these genes, HOXB2 [28], HOXB3 [107], [108], PLEKHM1 [109], RCCD1 [28], [34], and KANSL1 [28], [110], [111] have been previously implicated as susceptibility genes for epithelial ovarian cancer. Intriguingly, KANSL1 has been previously identified as an oncogene, whereas the colocalized lncRNA KANSL1-AS1 showed an opposite trend by assessing the 5-year survival time of 170 HGSOC patients [110]. In addition, the variant rs76640332 within KANSL1 loci was previously reported to be associated with lymphocyte percentage of leukocytes (two-sided P_GWAS = 5E-13) [112].

Next, we evaluated whether these susceptibility genes identified in whole blood could serve as circulating biomarkers for HGSOC. We overlapped the colocalized genes in the ovarian tumor and whole blood (Fig. 2D). ARL17A, LRRC37A2, and CRHR1-IT1 were found to show significant differences in expression between HGSOC tumors and healthy samples (Fig. 2, Fig. 2, Fig. S6). We also found that the expression of the overlapping elncRNA KANSL-AS1 correlated with the EOC marker gene BRCA2 expression in the TCGA HGSOC tumor samples (Fig. 2G). While these observations do not yet establish diagnostic utility, they suggest that colocalized genes identified in whole blood may represent accessible and promising candidates for future development as less-invasive biomarkers for HGSOC.

Using the genome-scale integrated analysis of gene networks in tissues (GIANT 2.0) [113], we found that the colocalized genes KANSL1, LRRC37A2, and ARL17A were strongly co-expressed in CD8⁺ T cells and enriched for chromatin organization and regulation of nitrogen compound metabolic process (Fig. 2H and Table S10). Nitrogen is one of the essential requirements for de novo synthesis of many important nitrogenous compounds, including amino acids and nucleotides, which support the growth of both cancer and immune cells [114]. The competition for available nitrogen between tumor cells and infiltrating CD8⁺ T cells might further alter the TME [114]. CD8⁺ T cells are pivotal to the efficacy of cancer immunotherapies [115]. Emerging studies have shown that variations in the fraction of infiltrating CD8⁺ T cells in the TME can lead to different outcomes in prognosis and treatment response in various malignancies, including breast cancer and HGSOC [116], [117], [118].

Single-nucleus resolution transcriptome maps for the human HGSOC tumor

Next, we characterized the expression atlas of PCGs and lncRNAs at single-cell resolution in human HGSOC tumors using snRandom-seq [119], [120], a droplet-based, highly sensitive, and full-length single-nuclei RNA-sequencing method optimized for FFPE samples. Because snRandom-seq captures full-length total RNAs with random primers (Fig. 3A), it enables more comprehensive detection of nuclear and long non-coding RNAs than poly(A)-based platforms such as 10x Genomics, thereby providing deeper insights into lncRNA regulation in tumor cells. We manually assigned the clusters to known cell types according to previously reported cell type marker genes [121], [122], [123], [124] (Fig. 3, Fig. 3). Our snRandom-seq analysis characterized the expression patterns of 10,991 PCGs and 1,481 lncRNAs at single-nucleus resolution in 83,188 human HGSOC tumor cells (Fig. 3, Fig. 3). Given the strong enrichment of candidate susceptibility genes on chromosome 17 (Fig. 2B) by overlapping HGSOC GWAS with tumor eQTLs, we investigated the single-cell expression patterns of these genes. More than 70 % of them showed high expression levels in the immune cells of HGSOC tumors (Fig. S7). Intriguingly, this dataset further confirmed the enriched co-expressed network (Fig. 2H) for the colocalized genes KANSL1, LRRC37A2, and ARL17A in T cells of HGSOC tumors (Fig. 3D), with KANSL1 showing strong T cell expression consistent with its reported association with the lymphocyte percentage of leukocytes [112]. In addition, we also observed that colocalized genes ARL17A, LRRC37A2, LRRC37A, and KANSL1-AS1, identified in both HGSOC tumors and whole blood (Fig. 2D), showed prominent expression in immune cells, cancer cells, and proliferative cancer cells. And MAPT, CHMP4C and ARL17B, identified in HGSOC tumors, were highly expressed in cancer cells and proliferative cancer cells (Fig. S8). These observations were largely confirmed in an independent single-cell dataset of human gynecologic tumors from 11 patients generated with 10x Genomics [79] (Figs. S9-S11), supporting the robustness of our single-nucleus full-length transcriptome map and its enhanced lncRNA coverage.

Fig. 3.

Single-nucleus resolution transcriptome maps for the human HGSOC tumor. (A) Experimental scheme for snRandom-seq. (B) UMAP plot of single-nucleus transcriptome data from human HGSOC tumors (N = 83,188 cells). Eleven distinct cell clusters were identified using previously reported cell type marker genes [121], [122], [123], [124]. (C) Violin plots showing the expression levels of representative marker genes across the 11 cell clusters. (D) Heatmap illustrating the expression patterns of genes involved in the enriched co-expression network for colocalized genes KANSL1, LRRC37A2, and ARL17A in T cells of HGSOC tumors. Mean expression values of the genes were calculated in each cluster. The color scheme represents the Z-score, derived from normalized gene expression levels, indicating relative expression compared to other cell types. (E) LocusZoom plots of HGSOC GWAS variants (top) [33] and HGSOC tumor lnc-eQTLs (bottom) on CRHR1-IT1. The x-axis shows the ±200 kb genomic region around rs146761208. The y-axis represents GWAS significance as −log₁₀(two-sided P-value). Each data point represents a variant, with color indicating r² (linkage disequilibrium). N = 53,978 individuals for HGSOC GWAS; N = 348 individuals for HGSOC tumor lncRNA-eQTLs.

The putative roles of CRHR1-IT1 and KANSL1-AS1 in ovarian cancer

Our analysis nominated two putative causal lncRNAs, CRHR1-IT1 and KANSL1-AS1, in chr17q21 for HGSOC (Fig. 2, Fig. 2). We identified 91 and 65 colocalized signals (with PP4 ranging from 0.80 to 1.00) (Table S11) for CRHR1-IT1 and KANSL1-AS1, respectively. 56 of these colocalized signals were observed from the same genomic regions. For instance, centered around the HGSOC GWAS SNP of rs146761208 (two-sided P_GWAS = 6.90E-10), colocalized signals were observed for both CRHR1-IT1 (Fig. 3E) and KANSL1-AS1 (Fig. S12). In this region, the HGSOC-associated GWAS variants were also the causal variants that influence the expression of CRHR1-IT1 and KANSL1-AS1. However, our scRNA-seq data of HGSOC tumors showed very low expression of the colocalized lncRNA CRHR1-IT1. Combining this with our previous observation shown in Fig. 2F, we hypothesized a potential protective role of CRHR1-IT1 in HGSOC development.

Functional insights into the lncRNA CRHR1-IT1

To validate the putative roles of CRHR1-IT1 in HGSOC, we performed quantitative reverse transcription polymerase chain reaction (qRT-PCR) analysis using gradient transfection in the HGSOC cell line OVCAR-3. We explored the biological pathways associated with CRHR1-IT1 expression changes using RNA-seq data collected from OVCAR-3 cells at 24 and 48 h after transfection. Transfection conditions included: control (0 μg/well vector, null condition), mock (0 μg/well verctor, baseline transfection reagents), vehicle (1 μg/well pcDNA3.1(+) vector), 0.25 (0.25 μg/well pcDNA3.1(+)-CRHR1-IT1 vector), 0.5 (0.5 μg/well pcDNA3.1(+)-CRHR1-IT1 vector) and 1 (1 μg/well pcDNA3.1(+)-CRHR1-IT1 vector) (Materials and methods). We found that transcript expression levels of CRHR1, the protein-coding gene corresponding to the lncRNA CRHR1-IT1 (CRHR1 intronic transcript 1), were markedly altered after transfection with pcDNA3.1(+)-CRHR1-IT1 in OVCAR-3 cells (Fig. S13). By performing a weighted gene co-expression network analysis (WGCNA) [83], we identified 9 modules in OVCAR-3 cells and assessed the correlation between the expression of genes within these modules and the lncRNA CRHR1-IT1 (Fig. 4, Fig. 4). The ‘green’ and ‘brown’ modules exhibited the most significant negative and positive correlations, respectively (Fig. 4B). Of the 357 genes in the ‘green’ module, 309 were identified as down-regulated in response to CRHR1-IT1, whereas 956 of the 1,251 genes in the ‘brown’ module were up-regulated by CRHR1-IT1 (Table S12). Functional analysis of these genes indicated enrichment in biological processes regulating cell growth and ovarian function, such as cell cycle, apoptosis, and oocyte meiosis (Fig. 4C). As shown in Fig. 4D, we observed that 309 down-regulated genes exibhited a dramatic decrease in expression at 24 h, while 956 up-regulated genes showed observable increase in expression at 48 h, following transfection with pcDNA3.1(+)-CRHR1-IT1 in OVCAR-3 cells, compared to the control, mock, and vehicle conditions. Interestingly, the up-regulated genes were likely involved in regulating cell death processes, such as the positive regulation of apoptotic process, whereas the down-regulated genes were enriched in cell cycle processes (Fig. 4E). These findings suggest that CRHR1-IT1 likely regulates HGSOC cell growth by initially inhibiting cell cycle progression and subsequently promoting apoptosis.

Fig. 4.

Co-expression network analysis indicates the potential role of CRHR1-IT1 in HGSOC development. (A) WGCNA cluster dendrogram and module assignment. Colors in the horizontal bar represent different modules, with genes clustered into 9 distinct modules. (B) Moduel-trait relationships. Pearson’s correlation between CRHR1-IT1 expression and each module is shown. Two-sided P-values represent the significance of these correlations. (C) Functional annotation (KEGG pathway) of genes in the ‘green’ and ‘brown’ modules, which exihibited the strongest correlations with CRHR1-IT1 expression. (D) Expression levels of 956 up-regulated genes and 295 down-regulated genes from the ‘green’ and ‘brown’ modules in OVCAR-3 cells. Expression values were calculated from 24 RNA-seq data collected at 24 and 48 h after transfection with vector concentrations of: control (0 μg/well vector, null condition), mock (0 μg/well vector, baseline transfection reagents), vehicle (1 μg/well pcDNA3.1 (+) vector), and increasing concentrations of pcDNA3.1(+)-CRHR1-IT1 vector: 0.25 μg/well, 0.5 μg/well, and 1 μg/well. (E) Functional annotation of up-regulated genes (top) and down-regulated genes (bottom) in the ‘green’ and ‘brown’ modules. (F) Expression correlation between JUN and CRHR1-IT1. The x-axis represents the log₂ transformed expression of CRHR1-IT1, and the y-axis is the log₂ transformed expression of JUN. N = 24 samples. (G) An example of DEG, JUN, between HGSOC tumor samples (N = 419) and healthy samples (N = 88). Two-sided P-value = 2.93E-23. Center lines show the medians; box limits indicate the 25th and 75th percentiles; whiskers extend to the 5th and 95th percentiles; outliers are represented by dots. (H) Network analysis performed by ConsensusPathDB [85], suggesting that JUN potentially regulates several apoptosis-related genes, including ATF3, DDIT3, TP53, BRCA1, and RARG. Seed nodes are shown in black, and intermediate nodes are shown in purple. ‘Andr_human_gene’ represents human androgen receptor.

Next, we identified potential transcription factors (TFs) that might regulate CRHR1-IT1 by combining results from multiple analytical approaches. First, using TRRUST v2 database [125], we found 37 enriched TFs, such as TP53, ATF6, and JUN (Fig. S14A), likely regulators of those 1,265 genes showing CRHR1-IT1-associated expression. Second, we prioritized CRHR1-IT1-associated genes by assessing gene significance (GS) and module membership (MM) using WGCNA [83]. Our analysis identified 48 key genes from the ‘green’ and ‘brown’ modules, which showed the strongest expression association with CRHR1-IT1 (Fig. S14B and Table S13). Third, we performed a sequence-based interaction prediction for CRHR1-IT1 using AnnoLnc2 [126], revealing 337 genes that may interact with CRHR1-IT1 (Table S14). We then evaluated the overlap of predicted TFs for CRHR1-IT1, and JUN was the only TF present in all three analyses (Fig. S14C). In addition, ChIP-seq data from various cell types confirmed that JUN binds to CRHR1-IT1, highlighting its potential regulatory role (Table S15) [126]. Our RNA-seq data and qRT-PCR results further suggested a strong positive correlation between CRHR1-IT1 and JUN expression (Pearson’s correlation = 0.92, two-sided P-value = 1.32E-10; Fig. 4F, Figs. S15 and S16). JUN, a crucial transcription factor, which has been previously reported to participate in diverse biological processes including apoptosis [127], [128], [129], [130], [131], [132], [133], was also found to be significantly down-regulated in HGSOC samples (Fig. 4G). Subsequent network analysis using the ConsensusPathDB [85] molecular interaction data from 31 public repositories (https://cpdb.molgen.mpg.de) revealed that JUN potentially regulate several apoptosis-related genes, such as ATF3, DDIT3, TP53, BRCA1, and RARG (Fig. 4H). Notably, ATF3 has been previously reported to promote apoptosis in cancer cells [134], [135], [136]. These findings suggest that CRHR1-IT1 may co-express with JUN to regulate apoptosis in HGSOC cells.

An increase in CRHR1-IT1 expression promotes apoptosis of HGSOC cells

To determine whether CRHR1-IT1 plays a role in the development of HGSOC cells, we performed a gradient transfection experiment in OVCAR-3 cells. CRHR1-IT1 expression was significantly increased after transfection with pcDNA3.1(+)-CRHR1-IT1 in HGSOC cells (Figs. S13A and S13B). Next, we used IncuCyte SX5 and Caspase-3/7 Green Detection Reagent to measure the fluorescence-activated cell intensity in transfected OVCAR­3 cells. We observed a dramatic increase in the proportion of apoptotic cells with higher transfection concentrations (Fig. 5, Fig. 5). And the rate of apoptosis accelerated as the concentration increased at 24 and 48 h after transfection (Fig. 5, Fig. 5). We also found that lncRNA CRHR1-IT1 slowed down the growth rate of non-apoptotic cells (Fig. S17). Moreover, our qRT-PCR results further confirmed that the expression of genes known to play important roles in apoptosis, including FAS, CMYC, and PARP1, were significantly correlated with lncRNA CRHR1-IT1 expression (Fig. 5, Fig. 5). These results suggested that elevated CRHR1-IT1 expression could promote apoptosis in OVCAR-3 cells. To validate the robustness of these findings, we repeated the gradient transfection and time-course apoptosis assays in another ovarian cancer cell line, A2780, and observed highly consistent results, including dose-dependent apoptosis induction and concordant regulation of apoptosis-related genes (Figs. S16A, S16B, S18, and S19). These results together support a conserved pro-apoptotic role of CRHR1-IT1 in two biologically distinct ovarian cancer cell models.

Fig. 5.

Increased expression of lncRNA CRHR1-IT1 promotes apoptosis in OVCAR-3 cells. (A) Representative photographs of cell apoptosis (caspase-3/7 detection reagent displayed as green fluorescence) after transfection with the pcDNA3.1 (+) or pcDNA3.1(+)-CRHR1-IT1 vectors for 24 or 48 h. Transfection conditions include: control (0 μg/well vector, null condition), mock (0 μg/well vector, baseline transfection reagents), vehicle (1 μg/well pcDNA3.1 (+) vector), and increasing concentrations of pcDNA3.1(+)-CRHR1-IT1 vector: 0.25 μg/well, 0.5 μg/well, and 1 μg/well. One image per condition is shown out of 48 analyzed images (16imagespertechnicaltriplicate) taken at 10× magnification. (B) Real-time growth monitoring of OVCAR-3 cells was performed using the Incucyte live imaging and analysis system, starting 6 h after transfection. Caspase-3/7 green fluorescence-positive signals represent early apoptotic cells, and negative phase signals represent non-apoptotic cells. Data point is the mean of 3 independent wells per time point ± SE. (C) The ratio of early apoptotic cells to non-apoptotic cells after transfection with the CRHR1-IT1_pcDNA3.1(+) vector for 24 h. (D) Ratio of early apoptotic cells to non-apoptotic cells after transfection with the CRHR1-IT1_pcDNA3.1(+) vector for 48 h. (E) Expression of FAS (24 h, top; 48 h, bottom) in OVCAR-3 cells at 24 and 48 h after transfection. (F) Expression of CMYC (24 h, top; 48 h, bottom) in OVCAR-3 cells at 24 and 48 h after transfection. (G) Expression of PARP1 (24 h, top; 48 h, bottom) in OVCAR-3 cells at 24 and 48 h after transfection. Transfection conditions include: control (0 μg/well vector, null condition), mock (0 μg/well vector, with baseline transfection reagents), vehicle (1 μg/well pcDNA3.1 (+) vector), and increasing concentrations of pcDNA3.1(+)-CRHR1-IT1 vector: 0.25 μg/well, 0.5 μg/well, and 1 μg/well. Results are shown as mean ± SE of three technical replicates per group. Significance levels are indicated as *p < 0.05, **p < 0.01, ***p < 0.001 and ****p < 0.0001, determined by unpaired t-test.

In summary, our findings highlight the significant role of the lncRNA CRHR1-IT1 in promoting apoptosis and inhibiting the growth of both OVCAR-3 and A2780 ovarian cancer cells, suggesting that CRHR1-IT1 could be a promising new therapeutic target for HGSOC.

Discussion

In this study, we defined genotype-driven expression changes for both PCGs (PCG-eQTLs) and lncRNAs (lncRNA-eQTLs). We demonstrated that lncRNAs account for a relatively higher portion of HGSOC heritability compared to PCGs. Our Bayesian-test-based colocalization analysis prioritized 10 HGSOC susceptibility genes, such as MAPT, CHMP4C, KANSL1-AS1, and CRHR1-IT1, in HGSOC tumors. We also identified 17 putative causal PCGs and lncRNAs in whole blood, which may provide preliminary clues relevant to early detection and immune-related therapeutic exploration for HGSOC, although these translational implications will require further experimental validation. Using the single-nucleus transcriptome map generated by snRandom-seq in human HGSOC tumors, we further characterized the cell-type-dependent expression patterns of the candidate susceptibility PCGs and lncRNAs. Finally, we focused on the lncRNA CRHR1-IT1 and demonstrated its crucial role in promoting apoptosis in HGSOC cells.

A central novelty of this work lies in estimating genotype-driven expression variation of HGSOC susceptibility lncRNAs. By defining lncRNA-eQTLs, we directly quantify the regulatory impact of germline variants on lncRNA expression. Our findings suggest that lncRNA expression changes could explain more variation in HGSOC heritability than PCGs. It is noteworthy that a considerable proportion of expression variation associated with HGSOC predisposition appears to be driven by underlying germline variants for both lncRNAs and PCGs. For example, the germline variant rs35478347 influences the expression of DHX58, a differentially expressed genes between HGSOC tumors and healthy samples. DHX58 encodes the RIG-I-like receptor LGP2, which has been previously reported to suppress RIG-I signaling [137] and protect tumor cells from ionizing radiation [138], [139]. These findings indicate that it will be critical to integrate genotype-driven expression changes of both lncRNAs and PCGs into future DEG identification to distinguish genetically driven from environmentally driven expression differences.

By integrating HGSOC-associated germline variants (from GWAS) with genotype-driven expression changes of PCGs and lncRNAs, we prioritized putative causal genes for HGSOC. Our initial overlapping results of HGSOC GWAS and eQTL data indicate that approximately 31 % of putative susceptibility genes for HGSOC are shared with breast cancer, including MAPT, LGALS7 and ERVK13-1, with strong enrichment on chromosome 17. Using Bayesian-test-based colocalization analysis, we identified 10 susceptibility genes for HGSOC with high fidelity. Notably, all 6 identified PCGs, CHMP4C, ARL17A, ARL17B, LRRC37A, LRRC37A2, and MAPT, showed genotype-driven expression changes associated with HGSOC predisposition in our study, which were independently replicated in a previous study using a different statistical method, TWAS [28], [100], [101]. TWAS employs distinct modeling assumptions, and this concordance provides additional, methodologically independent support for the robustness of our colocalization findings. We also demonstrate that further investigations of the identified colocalized genes in whole blood may yield new insights that are potentially relevant to early detection or immune-related therapeutic exploration for HGSOC. This is best illustrated by the shared colocalized genes, such as ARL17A, LRRC37A2, KANSL1-AS1, and CRHR1-IT1, whose expression and HGSOC predisposition are not only associated but also share common genetic causal variants in both tumor and whole blood. However, these implications remain preliminary and require extensive mechanistic studies and in vivo validation.

The snRandom-seq dataset further supports these findings by providing a more comprehensive view of lncRNA expression at single-nucleus resolution compared with conventional scRNA-seq platforms. Importantly, snRandom-seq also confers a distinct practical advantage: its random-primed reverse transcription strategy enables robust transcriptome profiling from FFPE and partially degraded tissues, allowing recovery of lncRNAs that are typically under-detected in poly(A)-based sc/snRNA-seq platforms. This FFPE compatibility is particularly valuable in HGSOC, where archival FFPE tumor specimens represent a major resource for translational and retrospective studies. These data expand our understanding of the cellular contexts in which susceptibility genes operate.

Our study has several limitations. First, the gene expression data used for eQTL mapping were not explicitly adjusted for tumor purity. However, we incorporated PEER factors [66] to account for hidden confounders, which may partially capture effects of tumor heterogeneity and non-tumor cell contributions. As such, the main conclusions are likely robust, as prioritized colocalized genes show consistent signals across independent datasets and methods, including TWAS. Second, while snRandom-seq provides enhanced sensitivity for non-polyadenylated transcripts and offers single-nucleus resolution, its focus on nuclear RNA limits the detection of cytoplasmic mRNAs relative to whole-cell approaches such as 10x Genomics. Third, although our findings suggest potential early-detection markers and therapeutic avenues, their clinical translation remains hypothetical. Future work may explore circulating lncRNAs as less-invasive biomarkers or RNA-targeted therapies as potential strategies. All these applications will require further mechanistic, in vivo, and preclinical studies to validate their feasibility.

Finally, our study highlights multiple lines of converging evidence suggesting a potential pro-apoptotic function of CRHR1-IT1 in HGSOC cells. Elevated CRHR1-IT1 expression promoted apoptosis and inhibited the growth of both OVCAR-3 cells and A2780 in vitro, demonstrating consistent effects across two biologically distinct ovarian cancer models. These in vitro results suggest a possible pro-apoptotic effect of CRHR1-IT1, but in vivo relevance and clinical significance remain to be established, and additional mechanistic studies will be necessary to fully elucidate the role of CRHR1-IT1 in HGSOC.

Conclusion

In conclusion, we identified genotype-driven expression changes of both PCGs and lncRNAs in HGSOC tumors. Our study highlights the importance of integrating lncRNA-eQTLs to understand the genetic underpinnings of HGSOC pathogenesis. We report a potential pro-apoptotic role of the lncRNA CRHR1-IT1 in vitro, though its in vivo and clinical significance remain to be established. These findings also suggest that lncRNAs could represent potential therapeutic avenues or circulating biomarkers, but their translational feasibility requires further mechanistic, in vivo, and preclinical investigation.

Consent for publication

Not applicable.

Data availability

Matched genotype and quantified gene expression data for 350 primary HGSOC tumor samples were downloaded from the GDC Data Portal as well (https://www.cancer.gov/ccg/research/genome-sequencing/tcga). The gene expression data in the ovaries of the TCGA HGSOC patients and GTEx healthy samples were downloaded from the UCSC Xena Toil recompute dataset [71] (“TCGA TARGET GTEx” unified RNA-seq cohort) available at https://xenabrowser.net/datapages/?dataset=TcgaTargetGtex_rsem_gene_tpm&host=https://toil.xenahubs.net. The eQTL data for lncRNAs and PCGs are available via a searchable public website https://zenodo.org/records/13760103. RNA sequencing data of OVCAR-3 cells and human HGSOC tumor snRandom-seq data generated in this study are publicly available in the GSA database (https://ngdc.cncb.ac.cn/gsa-human/) under accession numbers HRA009466 [140], [141] and HRA008122, respectively.

Ethics approval and consent to participate

Genotype, quantified gene expression data, and clinical records of the 350 HGSOC tumor samples used in this study comply with ethical regulations. Approval and informed consent for collection and sharing were already obtained by the relevant consortia from TCGA. HGSOC tumor specimens for snRandom-seq were obtained from the Biobank of the Zhejiang Cancer Hospital. The sample collection and overall study protocol were reviewed and approved by the Medical Ethics Committee of Zhejiang Cancer Hospital (Approval No.: IRB-2022-728), in accordance with established ethical guidelines for human subjects. All participants had provided written informed consent for biospecimen storage and future research use at the time of surgery; therefore, no additional study-specific consent was required.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article: