Multi-omics profiling reveals homologous recombination repair hyperactivation as a therapeutic vulnerability in neuroendocrine cervical carcinoma

Abstracts

Neuroendocrine cervical carcinoma (NECC) is an aggressive and rare malignancy comprising 0.9–1.5% of cervical cancers. Its molecular basis remains poorly understood, limiting treatment options. A national multicenter cohort survival analysis showed that NECC patients had a 5.17-fold higher death risk than non-NECC patients and demonstrated broad chemotherapy resistance, which was further validated in both NECC cell lines and patient-derived organoids. To identify therapeutic targets, multi‑omics profiling of NECC tumors and matched normal adjacent tissues revealed pronounced DNA replication stress and hyperactivation of homologous recombination repair (HRR). These findings were validated in cells and patient-derived organoids using DR-GFP reporter and pRPA32 assays, suggesting that NECC may be susceptible to PARP inhibitor therapy. NECC cells exhibited moderate PARP inhibitor sensitivity, greater than that of resistant non‑NECC lines but less than that of highly sensitive A2780 cells. However, combining PARP inhibitors with cisplatin/carboplatin did not enhance efficacy, prompting the exploration of alternative combinatorial targets. Multi-omics analysis and in vitro validation identified Bloom syndrome helicase (BLM) as a key factor overexpressed in NECC that modulates PARP inhibitor sensitivity. Further in vitro and in vivo experiments confirmed that combining PARP and BLM inhibitors achieved potent anticancer effects. Mechanistically, cancer-associated fibroblasts (CAFs) infiltration in the tumor microenvironment upregulated the transcription factor E2F1 via IL6, with ChIP‒qPCR verifying that activated E2F1 directly binds the BLM promoter (hg38, chr15:90,715,346–90,717,346) to drive its expression, suppressing PARP inhibitor sensitivity. Our findings establish dual PARP and BLM inhibition as a promising therapeutic strategy for NECC by blocking HRR pathway compensation.

Introduction

Cervical neuroendocrine carcinoma (NECC) is a highly aggressive and exceptionally rare malignancy, accounting for only 0.9–1.5% of all cervical cancers.¹ Due to its propensity for early dissemination and resistance to conventional therapies, NECC is recognized as one of the most lethal gynecological malignancies.² Although recent efforts have improved the clinicopathological characterization of NECC,^3,4 the development of effective treatment strategies remains severely hampered by its rarity and the limited understanding of its molecular mechanisms.^1,2,5 To date, there is no standard therapeutic regimen specifically for NECC. Consequently, the 5-year survival rate of NECC patients has shown little improvement over the past decades, with more than 50% of patients experiencing disease progression after initial treatment.⁶ Therefore, deciphering the distinct molecular landscape of NECC is crucial for addressing these clinical challenges.

The advancement of high-throughput multi-omics technologies has provided critical technical support for systematically deciphering the molecular pathogenesis of malignant tumors and identifying potential therapeutic targets.⁷ However, due to the rarity of the disease and the challenges in obtaining specimens, multi-omics studies on NECC remain limited, with existing explorations largely confined to the analysis of mutational landscapes at the genomic level.^8–12 Recurrent somatic mutations have been identified in approximately 63% of NECC cases, with frequent alterations in genes such as PIK3CA, KRAS, and TP53.^8,10,13 However, the functional implications of these genomic aberrations remain poorly understood. The malignant phenotype of tumors is ultimately determined by protein expression and post-translational modification status.¹⁴ Consequently, systematic proteomic profiling is an essential approach to identify druggable targets in NECC. However, the absence of a comprehensive proteomic landscape for NECC, particularly studies integrating genomic alterations with functional protein expression and phosphorylation profiles, has significantly hindered the discovery of therapeutic targets. Thus, there is a pressing need for integrated proteogenomic analyses of well-annotated NECC specimens to elucidate gene expression patterns, regulatory networks, and functional biology, with the goal of uncovering disease-driving pathways and novel therapeutic vulnerabilities.

The immune microenvironment and its molecular features further exacerbate therapeutic challenges in NECC. Specifically, NECC frequently displays low or negative PD-L1 expression and a predominantly microsatellite-stable (MSS) phenotype,^15,16 which confers inherent primary resistance to mainstream immune checkpoint inhibitors and conventional targeted therapies. Interestingly, NECC tumors exhibit a high Poly ADP ribose Polymerase 1 (PARP1) positivity rate.¹⁶ PARP inhibitors are well established to confer significant clinical benefit in tumors with homologous recombination repair (HRR) deficiency;¹⁷ however, recent studies have challenged the traditional synthetic lethality paradigm, revealing that tumors with hyperactivated HRR pathways also benefit from PARP inhibitor therapy.^18,19 This benefit arises from a compensatory dependency on PARP-mediated DNA repair pathways driven by replication stress.^20–22 This insight offers a novel perspective for identifying therapeutic vulnerabilities in NECC, yet the HRR expression patterns, replication stress levels, and underlying molecular regulatory mechanisms in NECC remain poorly elucidated.

This study establishes a proteotranscriptomic atlas of NECC through integrated RNA sequencing, global proteomics, and phosphoproteomics profiling of primary tumor specimens. Systematic analysis identifies molecular drivers of NECC pathogenesis, defines their functional contributions to tumor progression, and reveals clinically exploitable vulnerabilities. Guided by these findings, a targeted therapeutic strategy was developed and validated in preclinical models. By characterizing the hyperactive DNA damage repair network—a dominant feature of NECC biology—this work provides a promising therapeutic target for the precise treatment of this aggressive malignant tumor.

Results

Clinical cohort and multi-omics analysis reveal chemoresistance and hyperactivated HRR in NECC

NECC is a rare yet highly aggressive histologic variant. Due to its rarity, current NCCN guidelines² recommend treatment strategies adapted from cervical squamous cell carcinoma and adenocarcinoma, with cisplatin-paclitaxel and cisplatin-etoposide being the most commonly used regimens; nevertheless, objective response rates remain modest and long-term survival is poor,⁸ underscoring the urgent need for biologically informed treatment strategies. To evaluate the sensitivity of NECC to first-line treatment strategies recommended by the NCCN guidelines, we enrolled eligible patients from the China Multicenter NECC Cohort (CMC-NECC, n = 715) and a non-neuroendocrine cervical cancer cohort (non-NECC, n = 1059) (Supplementary Fig. 1, Supplementary Fig. 2a and Supplementary Table 1). Following 1:2 nearest-neighbor propensity score matching, 112 NECC and 207 non-NECC patients were ultimately included for the assessment of chemotherapy regimen efficacy (Supplementary Table 2 and Supplementary Fig. 2b).

Kaplan‒Meier curves and Cox regression analysis revealed that NECC patients had a 5.17-fold higher risk of mortality after treatment than non-NECC patients (HR 5.17, 95% CI: 3.15–8.49, P < 0.001; Supplementary Tables 3 and 4 and Supplementary Fig. 2c–e). Compared to non-NECC patients treated with cisplatin/paclitaxel (TP) or carboplatin/paclitaxel (TC) regimens, NECC patients receiving TP/TC therapy showed significantly reduced survival rates, indicating treatment failure (P < 0.001, Fig. 1a). The etoposide/cisplatin (EP) regimen also did not improve survival outcomes for NECC patients (P = 0.02, Fig. 1a). In the NECC cohorts, compared to no chemotherapy, neither TP/TC-based chemotherapy (P = 0.45) nor EP-based chemotherapy (P = 0.69) was associated with an improvement in survival (Fig. 1b). These data underscore the limited sensitivity of NECC to current first-line chemotherapeutic regimens.

Fig. 1. Clinical Cohort and Multi-omics Analysis Reveal Chemoresistance and Hyperactivated homologous recombination repair in NECC.

a Survival analysis of NECC patients treated with TP/TC or EP versus non-NECC patients, censored cases are indicated in parentheses at each time point. b Survival analysis of NECC patients treated with or without platinum-based chemotherapy (TP/TC or EP regimen), censored cases are indicated in parentheses at each time point. c Drug sensitivity assays of TC-YIK (neuroendocrine cervical carcinoma), SiHa (cervical squamous cell carcinoma), HeLa (cervical adenocarcinoma), and small-cell lung cancer cell lines (NCI-H446 and DMS-53) to cisplatin, carboplatin, paclitaxel, and etoposide. d The cisplatin dose–response curve showed that NECC organoids were highly resistant to cisplatin (IC₅₀ > 500 μM). e KEGG enrichment analysis of upregulated genes in NECC tumor tissues (n = 10) reveals significant enrichment in DNA repair, cell cycle, and replication pathways. f Bar graph showing enrichment levels of different DNA repair pathways in NECC tissues (n = 10). g Compared with the squamous cell carcinoma cohort (GSE138080), NECC tumors exhibited significantly increased homologous recombination repair (HRR) activity. An external clinical NECC cohort (HRA002655) further confirmed that, relative to the squamous cell carcinoma cohort (GSE63514), NECC tumors display a molecular signature of elevated HRR activity. h An in vitro DR-GFP reporter assay showed that HRR activity was significantly higher in the TC-YIK NECC cell line than in non-NECC cells (SiHa and HeLa). i Volcano map depicting mRNA expression levels of key HRR genes (e.g., BLM, RAD51) in tumor vs. adjacent normal tissues (NATs). j, k RT‒qPCR and Western blot validation of BLM and RAD51 mRNA and protein expression in TC-YIK, SiHa, HeLa, and ECT1/E6E7 cells. l Pathway enrichment of concordantly upregulated genes/proteins (upper) and of hyperphosphorylated proteins (lower) in NECC tumors. m Multiplex immunofluorescence validation of multi-omics candidates in NECC. Scale bar, 100 µm. Statistical analyses were performed using the Mann–Whitney U test for two-group comparisons and the Kruskal–Wallis test for multi-group comparisons. n denotes independent biological replicates. Data are presented as mean ± SD. ns not significant, *P < 0.05, **P < 0.01, ****P < 0.0001

To further substantiate the chemorefractory phenotype of NECC, we performed parallel dose–response analyses in TC-YIK cells alongside cervical squamous carcinoma (SiHa), adenocarcinoma (HeLa), and small-cell lung cancer (NCI-H446, DMS-53) lines that share identical first-line regimens. TC-YIK consistently displayed the highest IC₅₀ values for every agent tested: cisplatin (>100 µM), carboplatin (>1 mM), paclitaxel (2.97 µM), and etoposide (>500 µM), corresponding to a minimum 485-, 8-, 623- and 23-fold increase, respectively, relative to the most sensitive comparator (Fig. 1c, Supplementary Fig. 3a and Supplementary Table 9). In the NECC organoid model (Supplementary Fig. 3b–d), similarly pronounced resistance to the antitumor effects of cisplatin was observed (IC₅₀ > 500 μM) (Fig. 1d). Moreover, TC-YIK cells showed minimal sensitivity to platinum-based doublet regimens, with no significant synergistic effect observed: the ZIP model revealed synergy scores below 2 for cisplatin/paclitaxel and cisplatin/etoposide combinations (Supplementary Fig. 3e–j), confirming an intrinsic, broad-spectrum chemoresistance that rationalizes the poor clinical outcomes observed in NECC patients treated with TP/EP.

To systematically characterize the molecular profile of NECC and explore potential therapeutic targets, we performed an integrated analysis of transcriptomics, proteomics, and phosphoproteomics on tumor tissues and paired adjacent normal tissues (NATs) from NECC patients. Quality control assessments confirmed high data quality across all omics platforms (Supplementary Fig. 4a–d and Supplementary Fig. 5a–e). Transcriptomic profiling identified 1671 significantly upregulated and 706 downregulated transcripts (Supplementary Fig. 6a and Supplementary Tables 10 and 11). Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses revealed that upregulated genes were prominently enriched in pathways related to the cell cycle, DNA repair, and DNA replication initiation, which were also associated with platinum resistance (Fig. 1e, Supplementary Fig. 6b and Supplementary Table 12), suggesting aberrant activation of cell proliferation and DNA repair processes in NECC. Given the central role of DNA damage response pathways in mediating resistance to platinum-based chemotherapies—a hallmark clinical challenge in NECC—we prioritized a detailed dissection of the DNA repair network. In contrast, downregulated genes were primarily involved in extracellular matrix organization, cell adhesion, and immune regulation (Supplementary Fig. 6c), consistent with the highly aggressive nature and immune-excluded phenotype of the tumor microenvironment of NECC.

Further analysis of DNA damage repair pathways revealed that HRR was the most prominently activated DNA repair mechanism in NECC (Fig. 1f, Supplementary Fig. 6e, f and Supplementary Tables 13–15). Compared with the non-NECC cervical cancer cohort (GSE138080), NECC tumors exhibited markedly elevated HRR activity. To validate this observation in an independent clinical cohort, we analyzed RNA-seq data from 18 NECC patient samples deposited in the Genome Sequence Archive (GSA) of the China National Center for Bioinformation (project ID: HRA002655). Consistent with the primary analysis, comparison with the control dataset GSE63514 further supported a high-HRR activity molecular signature in NECC tumors (Fig. 1g). Moreover, functional interrogation using an in vitro DR-GFP reporter assay confirmed that HRR activity in the TC-YIK NECC cell line was significantly higher than that in non-NECC cervical cancer cell lines, including SiHa and HeLa cells (Fig. 1h). Analysis of our NECC transcriptomic and proteomic data revealed that key HRR factors, including Bloom syndrome helicase (BLM) and radiation sensitive protein 51 (RAD51), were significantly upregulated at both the mRNA and protein levels (Fig. 1i, Supplementary Fig. 6d and Supplementary Table 16). Consistent with this, their expression in the NECC-derived TC-YIK cell line and organoids was substantially higher than that in non-NECC cervical cancer cell lines or organoids (Fig. 1j, k and Supplementary Fig. 7a, b). BLM, an ATP-dependent DNA helicase, not only facilitates RAD51-mediated DNA repair but may also contribute to tumor proliferation by maintaining genomic stability.^23–25 These findings suggest that hyperactivation of the HRR pathway may be a potential therapeutic target for this aggressive malignancy.

Multi-omics analyses revealed coordinated activation of the HRR pathway at the transcriptional, proteomic, and phosphoproteomic levels in NECC. Proteomic analysis demonstrated that proteins consistently upregulated in NECC tumor tissues were significantly enriched in cell cycle- and HRR-related pathways (Fig. 1l up, Supplementary Fig. 4e–g and Supplementary Tables 17–24). Consistent with the transcriptomic data, these proteins showed elevated expression in TC-YIK cells compared to other cervical cancer cell lines and normal cervical epithelial cells. Furthermore, pathway analysis revealed heightened activity of mitochondrial function-related proteins in NECC (Supplementary Fig. 4h). These mitochondrial proteins were associated with poor patient prognosis (HR > 1, P < 0.05; Supplementary Fig. 4i), suggesting a potential role in energizing accelerated cell cycle progression and HRR.

Phosphoproteomic profiling further uncovered kinase-mediated regulatory mechanisms underlying NECC pathogenesis. A total of 433 phosphoproteins (containing 736 sites) were specifically hyperphosphorylated in tumor tissues (Supplementary Fig. 5d and Supplementary Table 25). These differentially hyperphosphorylated proteins were predominantly enriched in cell cycle regulation and chromatin remodeling (Fig. 1l down and Supplementary Fig. 5f). Kinase-substrate enrichment analysis (KSEA) identified significant hyperactivation of kinases, including CLK1 and CSNK2A1/2 (Supplementary Fig. 5g–k, Supplementary Tables 26–30). CSNK2A1/2-mediated phosphorylation of key G2/M transition regulators, such as MCM2, was particularly prominent (Fig. 1m, Supplementary Fig. 5l, m and Supplementary Table 31). To integrate multi-omics data and delineate regulatory networks, we performed protein–protein interaction (PPI) analysis, which revealed significant functional connectivity between MCM family members (such as MCM2) and core HRR factors such as BLM and RAD51 (Supplementary Fig. 5n, o). Multispectral immunofluorescence confirmed the elevated in situ expression of BLM and RAD51 in NECC tumor tissues, along with elevated expression and distinct subcellular localization of both total and phosphorylated MCM2 (Fig. 1m and Supplementary Fig. 8a). We further intervened with a CK2 inhibitor (CX-4945, silmitasertib, GLPBio/GC13037), and observed a reduction in MCM2 phosphorylation, accompanied by a marked decrease in HRR activity in NECC cells (Supplementary Fig. 8b, c), indicating that MCM2 phosphorylation is critical for sustaining the heightened HRR activity in NECC cells. Collectively, our multi-omics data and in vitro experiments demonstrate coordinated, multi-layered activation of the HRR pathway in NECC patients—spanning transcriptional, proteomic, and phosphoproteomic levels—centered on a BLM-RAD51-MCM molecular network. This hyperactive DNA repair machinery may represent a therapeutic vulnerability in NECC.

Targeting hyperactivated homologous recombination repair reveals therapeutic potential of PARP inhibition in NECC, with synergistic efficacy upon BLM blockade

Elevated HRR activity is often an adaptive repair response activated by cancer cells to cope with lesions accumulated under DNA replication stress.^20,22,26,27 GO and KEGG enrichment analyses of transcriptomic data revealed that NECC samples with high HRR activity were significantly enriched for DNA replication–related pathways (Fig. 1e and Supplementary Fig. 6b), indicating a highly replication-active state. This finding further suggests that NECC may experience heightened replication stress, leading to persistent accumulation of DNA damage. Prior studies have reported that increased replication stress can render cancer cells more dependent on the HRR pathway for survival—even in the absence of canonical homologous recombination deficiency (e.g., BRCA1/2 mutations).²⁸ To test this premise, we assessed the replication stress marker pRPA32 by immunofluorescence and found that replication stress was markedly higher in the NECC cell line TC-YIK than in the non-NECC cervical cancer cell lines SiHa and HeLa (Fig. 2a). Consistently, the NECC organoids also exhibited higher levels of replication stress than the non-NECC organoids (two cervical squamous cell carcinomas and one cervical adenocarcinoma) (Fig. 2b). Collectively, these data indicate that NECC is under sustained replication stress, thereby driving a heightened reliance on HRR-mediated DNA repair.

Fig. 2. Elevated replication stress and consequent high HRR activity in NECC present a therapeutic vulnerability to PARP inhibition.

a Immunofluorescence analysis showed that, compared with non-NECC cells (SiHa and HeLa), NECC cells (TC-YIK) exhibited higher expression of the replication stress marker pRPA32, accompanied by increased DNA damage markers such as 53BP1 foci. Scale bar, 10 µm. b Compared with non-NECC organoids, NECC organoids showed higher expression of the replication-stress marker pRPA32, accompanied by increased DNA damage markers such as 53BP1 foci. Scale bar, 25 µm. c Olaparib sensitivity profiling across cervical, pulmonary, ovarian, and engineered cell line models. The assay included TC-YIK, SiHa, HeLa, CaSki (cervical squamous carcinoma), NCI-H446 and DMS-53 (pulmonary neuroendocrine), A2780 and SK-OV-3 (ovarian) cell lines, along with TC-YIK derivatives with BLM overexpression (TC-YIK_BLM-OE) or knockdown (TC-YIK_BLM-sh). d Dose-response curve of olaparib in NECC organoids. e The in vitro DR‑GFP reporter assay indicated that hydroxyurea (HU)‑treated TC‑YIK cells exhibited a marked increase in replication stress and a significant elevation in homologous recombination repair (HRR) activity, whereas olaparib‑treated cells showed pronounced suppression of these processes. f Hydroxyurea-induced replication stress significantly enhances the sensitivity of TC-YIK cells to PARP inhibitors, accompanied by upregulation of DNA damage and replication stress markers (53BP1 foci and pRPA32). Scale bar, 10 µm. g PARP1/2 expression levels positively correlate with olaparib response in neuroendocrine tumor models. This dependency was quantified using the DepMap database, indicating a potential predictive biomarker for PARP inhibitor sensitivity in this context. h PARP1/2 mRNA levels are elevated in neuroendocrine cell lines relative to non-neuroendocrine lines, as revealed by CCLE analysis. Specifically, PARP1 expression is higher in small cell lung carcinoma (SCLC) than in non-small cell lung carcinoma (NSCLC) and lung adenocarcinoma (LASC), while PARP2 shows no intergroup difference. In gastric cancers, both PARP1 and PARP2 levels are increased in small-cell neuroendocrine carcinoma (GSCNEC) compared to non-neuroendocrine gastric carcinoma (GC). Notably, PARP1 expression remains low in cervical adenocarcinoma (CAC), ovarian cancer (OC), and normal cell lines. i RT‒qPCR (lower panels) and western blot (upper panels) confirm elevated PARP1/2 mRNA and protein levels in NECC TC-YIK cells compared to HeLa, SiHa, and ECT1/E6E7 cells. j Hydroxyurea combined with PARP inhibitors can enhance the antitumor effect on TC-YIK cells, with increased expression of pCHK1. k Olaparib combined with cisplatin treatment of TC-YIK cells, did not have a good synergistic effect. Statistical analyses were performed using the Mann–Whitney U test for two-group comparisons and the Kruskal–Wallis test for multi-group comparisons. n denotes independent biological replicates. The data are presented as mean ± SD. ns not significant, *P < 0.05, **P < 0.01, ****P < 0.0001

PARP inhibitors, such as olaparib, constitute a class of anticancer agents that primarily target DNA repair pathways. We performed in vitro sensitivity profiling of PARP inhibition in NECC cell lines and organoid models. Non-NECC, as well as the PARP inhibitor-tolerant ovarian cancer cell line SK-OV-3, exhibited strong resistance to olaparib (all IC₅₀ > 50 µM). In contrast, the NECC cell line TC-YIK was sensitive to olaparib (IC₅₀ = 2.6 µM). Similar sensitivity was observed in small-cell lung cancer lines with neuroendocrine lineage features, including NCI-H446 (IC₅₀ = 4.3 µM) and DMS-53 (IC₅₀ = 42.2 µM). Quantitatively, TC-YIK, NCI-H446, and DMS-53 were approximately 19-fold, 12-fold, and 1.2-fold more sensitive to olaparib, respectively, than resistant cancer cell lines (Fig. 2c). Consistent with the cell-based results, olaparib also exerted a stronger antitumor effect in NECC organoids (IC₅₀ = 7.47 µM) than in non-NECC organoids (cervical adenosquamous carcinoma IC₅₀ > 50 µM) (Fig. 2d).²⁹ Collectively, these findings provide initial support for the therapeutic feasibility of PARP inhibition in NECC.

To elucidate the mechanism by which PARP inhibitors treat NECC, we performed whole-exome sequencing (WES) on 10 NECC tumors and NATs tissues and assessed homologous recombination deficiency (HRD) associated genomic scars. NECC showed a low HRD-positive rate (12.5%, Supplementary Table 37); therefore, PARP inhibitors may exert antitumor effects in NECC through alternative mechanisms. Given that multi-omics analyses and in vitro assays establish sustained replication stress together with elevated HRR activity in NECC, we hypothesized that PARP inhibitor efficacy may be linked to interference with HRR under replication stress. Consistent with this model, PARP inhibitor treatment of the NECC cell line TC-YIK significantly reduced HRR activity (Fig. 2e), decreased the expression of the core HRR factor RAD51 (Supplementary Fig. 9a), and increased DNA damage readouts, including 53BP1 foci (Fig. 2f). Together, these data indicate that PARP inhibitors can suppress HRR and thereby promote DNA damage accumulation, providing a mechanistic basis for their antitumor effects in NECC. We also analyzed PARP1/2 expression in NECC. High PARP1/2 expression correlated significantly with increased olaparib sensitivity in neuroendocrine models (Fig. 2g and Supplementary Fig. 9b). Moreover, we interrogated the DepMap database and found that neuroendocrine carcinoma cell lines originating from different organizations exhibited higher PARP1/2 expression than squamous/adenocarcinoma lines of the same organization (Fig. 2h). Subsequent molecular validation confirmed that both the mRNA and protein levels of PARP1/2 were significantly elevated in the NECC cell line TC-YIK relative to cervical squamous carcinoma (SiHa), adenocarcinoma (HeLa), and normal cervical epithelial (ECT1/E6E7) cells (Fig. 2i). The high expression of PARP1/2 in NECC cells can recruit a sufficient amount of PARP inhibitors, providing a material basis for their antitumor activity. To further interrogate how PARP inhibition functions under replication stress, we treated the NECC cell line TC-YIK with hydroxyurea, a DNA replication stress inducer. Hydroxyurea-induced replication stress potentiated the antitumor effect of PARP inhibitors, as evidenced by activation of the ATR pathway (increased pCHK1, Fig. 2f, j), indicating heightened replication stress, together with increased DNA damage readouts (marked elevation of 53BP1 foci). In this context, PARP inhibition further impeded RAD51 recruitment and suppressed HRR activity, ultimately exacerbating DNA double-strand breaks and triggering tumor cell death. In parallel, we induced DNA damage by ultraviolet irradiation and found that PARP inhibitor treatment prevented RAD51 recruitment, compromised HRR function, and impaired DNA damage repair, leading to tumor cell death (Supplementary Fig. 9a).

While PARP inhibitors demonstrate promising antitumor activity against NECC, it is noteworthy that the sensitivity of the NECC cell line TC-YIK remained 3.6-fold lower than that of the a BRCA-mutant, PARP inhibitor-hypersensitive ovarian cancer line A2780 (IC₅₀ = 0.72 µM), confirming a phenotype of moderate sensitivity (Fig. 2c and Supplementary Fig. 9c). However, the treatment of TC-YIK cells with platinum combined with olaparib still did not show a good synergistic effect (ZIP synergy score: 3.725, Fig. 2k), prompting the investigation of compensatory resistance mechanisms. These results indicate that PARP inhibitors have potential utility in NECC, although intrinsic limitations may constrain their efficacy.

Notably, multi-omics, experimental, and DepMap database validation consistently indicated significant overexpression of the BLM helicase in NECC. This overexpression was negatively correlated with sensitivity to PARP inhibitor treatment. (Figs. 1j, k and 3a, b, h and Supplementary Fig. 6f). Previous evidence suggests that BLM overexpression may confer PARP inhibitor resistance by stabilizing replication forks and compensating for HR function. BLM expression may represent a key mediator of intrinsic PARP inhibitor resistance in NECC. Supporting this, BLM overexpression in TC-YIK cells significantly reduced olaparib sensitivity (Fig. 2c and Supplementary Fig. 9c, d), indicating that BLM may be a key mediator affecting the therapeutic sensitivity of NECC to olaparib. Consistently, the high-BLM NECC line TC-YIK proved at least 1.85-fold more sensitive to the BLM inhibitor (IC₅₀ = 54 µM) than the low-BLM cervical squamous/adenocarcinoma lines (IC₅₀ > 100 µM). Moreover, enforced BLM overexpression in TC-YIK (TC-YIK-BLM-OE) further lowered the IC₅₀ to 4.6 µM (95% CI: 1.7–14 µM), yielding an 11.7-fold increase in sensitivity relative to parental TC-YIK and confirming that BLM abundance directly determines BLM inhibitor responsiveness (Fig. 3c and Supplementary Fig. 9e). In vitro pharmacodynamic studies confirmed that the BLM inhibitor synergized strongly with olaparib, markedly enhancing cytotoxicity in TC-YIK cells (ZIP synergy score: 19.349; Fig. 3d and Supplementary Fig. 9f). In addition, we generated a lentiviral shRNA construct targeting BLM (shRNA-BLM) and transduced it into TC-YIK cells (Supplementary Fig. 9g). BLM knockdown markedly increased NECC cell sensitivity to PARP inhibitor treatment (Fig. 2c), further indicating that BLM inhibition can substantially enhance the antitumor efficacy of PARP inhibition. We also treated TC-YIK cells with BLM-IN-1, an inhibitor targeting other functional sites of BLM. Similarly, we found that co-treatment with BLM-IN-1 significantly increased the antitumor sensitivity to PARP inhibitors (IC₅₀ = 5.81, Fig. 3d, e). These results underscore dual targeting of the PARP–BLM axis as a highly promising therapeutic strategy for NECC. The organoid experiment also confirmed that the combination of olaparib and BLM significantly inhibited the activity of the organoids (Fig. 3f, g and Supplementary Fig. 7c, d).

Fig. 3. Combined inhibition of PARP and BLM synergistically improves antitumor efficacy in NECC via overcoming BLM‑mediated resistance.

a BLM mRNA levels are elevated in neuroendocrine cell lines relative to non-neuroendocrine lines, as revealed by CCLE analysis. Specifically, BLM expression is higher in small cell lung carcinoma (SCLC) than in non-small cell lung carcinoma (NSCLC) and lung adenocarcinoma (LASC). In gastric cancers, BLM level is increased in small-cell neuroendocrine carcinoma (GSCNEC) compared to non-neuroendocrine gastric carcinoma (GC). b DepMap analysis reveals that BLM expression is negatively correlated with olaparib response in neuroendocrine tumor models. c ML216, a BLM inhibitor, exhibits differential sensitivity across cervical and pulmonary cell line models. The assay included cervical cell lines TC-YIK, SiHa, HeLa, CaSki, and normal cervical cell line ECT1/E6E7, along with pulmonary neuroendocrine lines NCI-H446 and DMS-53, and the BLM-overexpressing TC-YIK derivative (TC-YIK_BLM-OE). d Olaparib combined with either BLM inhibitor (ML216 or BLM-IN-1), exhibited synergistic inhibitory effects in TC-YIK. e Dose‒response curve of the BLM inhibitor BLM-IN-1 in TC-YIK cells. f Dose‒response curve of the BLM inhibitor ML216 in NECC organoids. g ATP activity assays demonstrate that ML216 or olaparib monotherapy significantly reduces cell viability compared to control. The combination of olaparib and ML216 yields a greater inhibitory effect than either agent alone. h Immunofluorescence shows that ML216 or olaparib alone suppresses RAD51 and raises γH2A.X; combined treatment maximizes DNA damage in TC-YIK. Scale bar, 10 µm. Statistical analyses were performed using the Mann–Whitney U test for two-group comparisons and the Kruskal–Wallis test for multi-group comparisons. n denotes independent biological replicates. The data are presented as mean ± SD. ns not significant, *P < 0.05, **P < 0.01, ****P < 0.0001

E2F1-driven BLM transcriptional activation confers PARP inhibitor resistance in NECC

Integrated multi-omics analyses and in vitro validation revealed significant overexpression of BLM in NECC, which was identified as a key factor limiting the efficacy of PARP inhibitors. To elucidate the transcriptional regulatory mechanism by which BLM affects the therapeutic efficacy of PARP inhibitors, we performed transcription factor prediction by integrating public databases (TRRUST, Cistrome DB, TCGA-CESC), which identified E2F1 as the most prominent candidate regulator of BLM (Fig. 4a). Our transcriptomic profiling demonstrated that E2F1 mRNA expression was significantly higher in NECC tumor tissues than in NATs (P < 0.001, Fig. 4b), and the expression of BLM was closely related to E2F1 in our transcriptome data and the TCGA database (Fig. 4c). Based on the RNA-seq data of NECC, we used single-sample gene set enrichment analysis (ssGSEA) to analyze the E2F1 transcriptional activity score in NECC tissues. The results showed that E2F1 transcriptional activity was significantly increased in tumor tissues (P < 0.01), and its activity was strongly positively correlated with gene expression levels (R = 0.66, P = 0.0014), indicating that E2F1 in NECC tissues is in a transcriptionally active state (Fig. 4d). We further analyzed RNA-seq data from NECC tissue samples deposited in the Genome Sequence Archive (GSA) of the China National Center for Bioinformation (project ID: HRA002655) for external validation. The analysis of the external validation dataset also indicated that E2F1 transcriptional activation is present in NECC tissues (Fig. 4e).

Fig. 4. E2F1-driven BLM transcriptional activation confers PARPi resistance in NECC.

a Integrative prediction identifies E2F1 as the top BLM regulator. b RNA‒seq analysis reveals significantly elevated E2F1 mRNA levels in tumors compared with matched NATs tissues) (P < 0.001). c E2F1 expression positively correlates with BLM levels both in our cohort and in the TCGA-CESC dataset. d E2F1 transcriptional activity, assessed by ssGSEA, is significantly elevated in NECC tumors versus normal tissues, and strongly correlates with E2F1 expression (R = 0.66, P < 0.01). e Analysis of the external validation dataset HRA002655 confirms that E2F1 transcriptional activity (ssGSEA) is significantly elevated in NECC tumors compared to normal tissues, and remains strongly correlated with E2F1 expression (R = 0.75, P < 0.0001). f Public ChIP‒seq peaks show E2F1 occupancy at the BLM promoter in multiple cancers. g JASPAR motif analysis identifies a high-confidence E2F1 binding site at chr15:90,715,346–90,717,346 (relative score 0.996). h–j ChIP–qPCR reveals stronger E2F1 binding to the BLM promoter in NECC TC-YIK cells than in non‑NECC cell lines (P < 0.001). k RT‒qPCR (upper) and Western blot (lower) confirming the highest E2F1 in TC-YIK cell lines. l shRNA-mediated E2F1 knockdown dose-dependently reduces BLM mRNA and protein levels. Statistical analyses were performed using the Mann–Whitney U test for two-group comparisons and the Kruskal–Wallis test for multi-group comparisons. n denotes independent biological replicates, and data are presented as mean ± SD. ns not significant, *P < 0.05, **P < 0.01, ****P < 0.0001

To validate the direct transcriptional regulation of BLM by E2F1, we interrogated ENCODE and Cistrome Chromatin ImmunoPrecipitation followed by sequencing (ChIP‒seq) datasets, which confirmed E2F1 binding at the BLM promoter region across multiple cancers, including cervical carcinoma (Fig. 4f). JASPAR motif analysis further identified a high-affinity E2F1 binding site within the BLM promoter (chr15:90,715,346–90,717,346; relative score = 0.996, Fig. 4g). Using ChIP‒qPCR, we validated E2F1 binding to the BLM promoter in TC-YIK, SiHa, HeLa, and ECT1/E6E7 cells, with the highest binding efficiency observed in the NECC cell line TC-YIK (Fig. 4h–j), and E2F1 and BLM expression in the NECC cell line TC-YIK was also higher than that in other cervical epithelial-derived cell lines (Fig. 4k, P < 0.001), indicating NECC tumor-specific enrichment.

Functional validation using lentivirus-mediated E2F1 knockdown resulted in a dose-dependent reduction in BLM expression (Fig. 4l), confirming E2F1 as a key upstream regulator of BLM. Collectively, these results demonstrate that E2F1 directly binds to the BLM promoter and drives its overexpression, thereby modulating sensitivity to PARP inhibition in NECC. These findings provide mechanistic insight into the upstream regulation of BLM and its role in therapy response.

CAFs-remodeled tumor microenvironment licenses PARP inhibitor resistance via E2F1-BLM activation in NECC

Integrated multi-omics analysis revealed pronounced extracellular matrix (ECM) remodeling in NATs, accompanied by substantial enrichment of multiple cancer-associated fibroblasts (CAFs)—including vCAF, rCAF, and mCAF (Fig. 5a–f and Supplementary Tables 32–36). These CAFs exhibited high metabolic and secretory activity, and CAFs subtype scores showed a strong positive correlation with ECM degradation pathway activity (Fig. 5g). Immunohistochemical staining further confirmed that CAFs were predominantly localized at the tumor–normal tissue interface (Fig. 5i).

Fig. 5. The CAFs-remodeled tumor microenvironment licenses PARPi resistance via E2F1-BLM activation in NECC.

a xCell deconvolution analysis reveals enrichment of cancer-associated fibroblasts (CAFs) subsets—including vascular (vCAF), resting (rCAF), and myofibroblastic (mCAF) types—in NATs relative to tumors. b Immune, stromal, and microenvironment scores higher in NATs (paired t-test). c Gene set enrichment analysis (GSEA) shows significant enrichment of the extracellular matrix disassembly pathway in NATs compared with tumors (NES = 1.85, P = 0.02). d CAFs subtype abundance between NATs and tumors (P < 0.05). e Heat map of CAFs marker expression in NATs and tumors. f ECM-degrading RNA molecules elevated in NATs. g Positive correlation between CAFs scores and ECM-disassembly GSVA (P < 0.05). h The schematic diagram shows that the accumulation of CAFs leads to high BLM expression in tumors, which is associated with tumor proliferation and drug resistance. i α-SMA immunohistochemistry in NECC samples reveals that CAFs are most abundant in NATs, moderately present in tumor tissues, and least frequent in normal distant tissues (NDTs). Scale bar, 100 μm. j BLM knockdown reverses CAFs-induced proliferation and significantly increases NECC cell sensitivity to PARP inhibitors. Scale bar, 100 µm. k Immunofluorescence staining shows that co‑culture with CAFs increases BLM and RAD51 foci formation in TC‑YIK cells, attenuates olaparib efficacy, and yet co‑treatment with olaparib and the BLM inhibitor ML216 maintains potent DNA damage (γH2A.X). Scale bars, 10 µm. Statistical analyses were performed using the Mann–Whitney U test for two-group comparisons and the Kruskal–Wallis test for multi-group comparisons. n denotes independent biological replicates, and data are presented as the mean ± SD. ns not significant, *P < 0.05, **P < 0.01, ****P < 0.0001

To investigate whether CAFs directly modulate the DNA damage response in tumor cells, we conducted CAF-tumor cell co-culture experiments. The results demonstrated that CAFs co-culture induced expression of BLM increases and hyperactivation of the HR repair pathway (Fig. 5k), significantly attenuating the efficacy of the PARP inhibitor olaparib in the NECC cell line. In contrast, combined treatment with the BLM inhibitor potently enhanced DNA damage. We cocultured BLM-depleted TC-YIK cells with CAFs to investigate subsequent phenotypic changes. BLM knockdown significantly reversed the CAFs-induced proliferation of TC-YIK cells (Fig. 5j). Consistently, CCK-8 assays showed that BLM depletion significantly increased the sensitivity of NECC cells to PARP inhibitor treatment (Fig. 2c), functionally establishing BLM as a critical determinant of CAFs-mediated PARP inhibitor tolerance in TC-YIK cells. These findings suggest that targeting the CAFs-regulated DNA repair axis may represent a novel therapeutic strategy. Collectively, these results indicate that CAFs in NATs contribute to NECC progression and therapy resistance by promoting ECM remodeling and modulating the DNA damage response (Fig. 5h).

To further elucidate the mechanism by which CAFs regulate DNA repair in the tumor microenvironment, we knocked down E2F1 in TC-YIK (NECC) and ECT1/E6E7 (normal cervical) cells and cocultured them with three examples of cervical cancer-derived CAFs isolates (CAF#1-CAF#3) or normal fibroblasts (NFs; NF#1-NF#3). RT‒qPCR and Western blot analyses revealed that, compared to NFs, CAFs markedly upregulated both E2F1 and BLM expression at the mRNA and protein levels in TC-YIK cells (Fig. 6a, b and Supplementary Fig. 10a, b, P < 0.0001), thereby establishing E2F1 as a key transcriptional regulatory factor for the upregulation of BLM induced by CAFs. To define the upstream signaling mechanism by which CAFs increase E2F1 activity in NECC cells, we established a CAFs–NECC co-culture system and profiled cytokines in conditioned media using a solid-phase antibody array (ARY022B) (Supplementary Fig. 10c). This analysis identified eight candidate CAFs-secreted factors, with IL-6 and CD40 showing the most prominent increases (Supplementary Fig. 10d). Integrating these results with NECC transcriptomic data, we found that IL-6 and CD40 were concordantly enriched in both the array readouts and transcriptome-based predictions; moreover, the transcriptome revealed coordinated upregulation of IL-6 together with its receptor IL6R and the signal-transducing subunit IL6ST in NECC tissues (Supplementary Fig. 10e). Prior evidence indicates that the IL-6/IL6R/IL6ST signaling axis can activate downstream pathways involved in the regulation of DNA replication.³⁰ Collectively, these findings suggest that CAFs may activate E2F1 signaling through IL-6 secretion.

Fig. 6. CAFs drive E2F1-dependent BLM upregulation to enhance proliferation and blunt olaparib response in TC-YIK NECC cells.

qRT-PCR (a) and Western blot (b) showing that co-culture with CAFs (#1–3) markedly increases E2F1 and BLM mRNA/protein levels in TC-YIK cells; this induction is abolished by E2F1 knockdown (shE2F1 #1/#3). c, d EdU incorporation assays demonstrate that CAFs enhance cell proliferation compared with normal fibroblasts (NFs), an effect that is reversed by the BLM inhibitor ML216 (n = 3). Scale bars, 100 µm. e, f CCK‑8 assays reveal that co‑culture with CAFs enhances the proliferative capacity of TC‑YIK NECC cells, an effect suppressed by BLM inhibitor (BLMi) treatment. g, h Quantification of γH2A.X foci following 4 Gy irradiation reveals that CAFs reduce baseline DNA damage, whereas treatment with the BLM inhibitor ML216 restores γH2A.X accumulation, indicating impaired homologous recombination repair capacity. Scale bars, 10 µm. Statistical significance was determined by two‑way ANOVA with Tukey’s post‑hoc test. For comparisons between two groups, the Mann–Whitney U test was used; for multi‑group comparisons, the Kruskal–Wallis test was applied. n denotes independent biological replicates, and data are presented as the mean ± SD. ns not significant, *P < 0.05, **P < 0.01, ****P < 0.0001

We next assessed the functional impact of CAFs on the E2F1/BLM axis using EdU incorporation and CCK-8 assays. Co-culture with CAFs significantly increased the proportion of EdU-positive cells compared to the control and NF co-culture groups. This effect was reversed upon treatment with the BLM inhibitor (Fig. 6c–f and Supplementary Fig. 10f–i), demonstrating that CAFs can promote the proliferation of NECC tumor cells, while inhibiting BLM can weaken the promoting effect of CAFs. Immunofluorescence detection of γH2A.X foci revealed that CAFs co-culture mitigated ultraviolet-induced DNA damage in TC-YIK cells. However, subsequent BLM inhibitor treatment substantially increased DNA damage (Fig. 6g, h and Supplementary Fig. 10j, k), suggesting that CAFs repair DNA damage in NECCs by regulating BLM expression and promoting tumor cell proliferation. Underscoring the role of BLM activation in CAFs-mediated repair protection and highlighting the potential of combined BLM and PARP inhibition to exacerbate DNA damage in NECC.

Collectively, these results demonstrate that CAFs, as key constituents of the tumor microenvironment, activate the E2F1-BLM signaling axis to enhance HRR capacity, thereby dampening NECC sensitivity to PARP inhibition. These findings provide new mechanistic insights into how stromal components contribute to PARP inhibitor therapy resistance and lay the groundwork for targeting the NECC CAFs-tumor crosstalk network through combined PARP and BLM inhibition.

Combined PARP and BLM inhibition synergistically suppresses tumor growth and overcomes CAFs-mediated resistance in vivo

To evaluate the in vivo antitumor efficacy of combined PARP and BLM targeting and its ability to overcome tumor microenvironment-mediated resistance, we established two mouse xenograft models. In the TC-YIK monoculture mouse model (Supplementary Fig. 11a, b), monotherapy with either olaparib or BLM inhibitor only partially suppressed tumor growth compared to the vehicle control, whereas the combination of olaparib and BLM inhibitor significantly reduced both tumor volume and weight (P < 0.001, Supplementary Fig. 11c–f), demonstrating strong synergistic antitumor efficacy in vivo.

To model therapy resistance mediated by CAFs in the tumor microenvironment, we developed a cotransplantation model using TC-YIK cells together with CAFs/NFs (Fig. 7a, b, Supplementary Fig. 11g, h). Compared to cografting with NFs, CAFs significantly accelerated tumor growth, underscoring the protumor role of CAFs in NECC progression. Importantly, under CAFs-rich conditions, the efficacy of olaparib monotherapy was markedly compromised, reaffirming CAFs-induced desensitization to PARP inhibition. In contrast, the olaparib and BLM inhibitor combination treatment maintained robust tumor suppression (Fig. 7c), indicating that dual targeting of PARP and BLM effectively counteracts microenvironment-mediated resistance by CAFs.

Fig. 7. Dual PARP/BLM inhibition overcomes CAFs-induced microenvironment-mediated drug resistance in NECC in vivo.

a BALB/c nude mice were subcutaneously coinjected with TC-YIK cells and CAFs or NFs (1:1; 2 × 10⁶ cells total). When tumors reached ~50 mm³, mice were randomized to receive vehicle, ML216 (25 mg/kg/day i.p.), olaparib (50 mg/kg/day i.g.), or their combination for 21 days (n = 7 per arm). b Comparison diagrams of tumor tissues in each group. c Body weight of mice, tumor growth curves, endpoint tumor volumes, and tumor weights. d IHC of endpoint tumors: combination therapy markedly reduces Ki67 and RAD51, while increasing γH2A.X foci, confirming aggravated DNA damage and impaired repair. Scale bars, 100 µm. e HE staining indicated that there was no organ injury in any group. Scale bars, 100 µm. Statistical analyses were performed using the Mann–Whitney U test for two-group comparisons and the Kruskal–Wallis test for multi-group comparisons. n denotes independent biological replicates, and data are presented as the mean ± SD. ns not significant, **P < 0.01, ****P < 0.0001

Immunohistochemical analysis of NECC mouse model tumor tissues revealed that the olaparib and BLM inhibitor combination group exhibited significantly reduced expression of the proliferation marker Ki67 and the homologous recombination protein RAD51, along with elevated levels of the DNA damage marker γH2A.X (Fig. 7d), suggesting potent inhibition of DNA repair and aggravation of DNA damage in vivo. Furthermore, the combination regimen did not cause significant loss of body weight or induce obvious pathological damage in major organs, including the heart, liver, spleen, lungs, or kidneys (Fig. 7e), supporting its favorable safety profile at the effective dosage.

In summary, these in vivo findings demonstrate that the combination of olaparib and a BLM inhibitor exerts significant synergistic antitumor effects and effectively reverses CAFs-mediated resistance in the tumor microenvironment. This combination represents a promising and translatable therapeutic strategy for NECC, with the potential to improve patient outcomes and survival.

Discussion

NECC remains a therapeutic challenge, characterized by early metastasis and a dismal prognosis.³¹ The lack of benefit from platinum-based chemotherapy, as observed in our and other cohorts,³² underscores the critical need to understand the molecular drivers and identify actionable therapeutic targets in NECC pathogenesis.³³ Our study, through an integrated proteotranscriptomic and phosphoproteomic analysis of treatment-naïve NECC specimens, unveils a previously unrecognized molecular hallmark: a hyperfunctional HRR state. This HRR hyperactivation, orchestrated around the BLM-RAD51-MCM2 axis and modulated by the tumor microenvironment, represents a pivotal therapeutic vulnerability in this aggressive disease. Our study reveals that HRR pathway hyperactivation in NECC is not merely a marker of DNA repair proficiency but rather a compensatory response to replication stress that creates a dependency on PARP-mediated repair. This explains the observed sensitivity to PARP inhibition, even in the absence of canonical HRR deficiency. However, the efficacy of PARP monotherapy is limited by compensatory upregulation of BLM, which stabilizes replication forks and serves as a resistance mechanism.

The dismal prognosis of NECC is starkly highlighted by our multicenter cohort analysis. Even after rigorous propensity-score matching, NECC patients faced a greater than fivefold increased risk of death compared to their cervical squamous or adenocarcinoma counterparts, with no survival benefit derived from standard adjuvant chemotherapy. This clinical observation of intrinsic chemoresistance was recapitulated in vitro, where the NECC cell line TC-YIK exhibited extraordinarily high IC₅₀ values across a panel of chemotherapeutic agents. Our multi-omics profiling provides a mechanistic basis for this phenotype, demonstrating significant co-upregulation at the transcriptomic, proteomic, and phosphoproteomic levels of key HRR components, including BLM, RAD51, and MCM2. We further confirmed, using functional assays, that NECC exhibits higher HRR activity. The convergence of these data strongly suggests that enhanced DNA repair capacity is a fundamental driver of NECC aggressiveness. This finding aligns with reported high genomic instability in NECC but moves beyond correlation by functionally validating HRR as a druggable axis.

Our study extends the current understanding of NECC biology in two key aspects. First, we delineate a precise signaling cascade wherein hyperactivated cell cycle kinases, such as CSNK2A1/2, lead to enhanced phosphorylation of MCM2, a critical step in the initiation of DNA replication and repair. This post-translational regulation synergizes with the transcriptional upregulation of BLM and RAD51, creating a robust HRR network. The coordinated hyperactivation of the HRR pathway at multiple molecular levels likely underlies the genomic instability and intrinsic chemoresistance observed in NECC. BLM helicase is known to resolve complex DNA structures to facilitate RAD51-mediated homologous recombination, a process critical for the repair of DNA interstrand cross-links induced by platinum agents.^25,34 Second, and more notably, we uncover a crucial role for the tumor microenvironment in sustaining this pro-repair state. We demonstrate that CAFs, a predominant stromal component in NECC, directly promote BLM expression in tumor cells by activating the transcription factor E2F1. While the tumor microenvironment is increasingly recognized as a modulator of therapeutic response, direct evidence of CAFs transcriptionally regulating key DNA repair enzymes in cancer cells remains sparse, positioning our work at the forefront of TME-DDR crosstalk research.³⁵ This CAF-E2F1-BLM axis represents a novel noncell-autonomous mechanism of HRR regulation, highlighting the therapeutic implications of targeting the tumor ecosystem alongside cancer cell-intrinsic pathways. We also validated that the upstream key signaling mechanism by which CAFs activate E2F1 in NECC tumor cells is IL6, which is a cytokine that can affect DNA replication.³⁰ There is existing evidence showing that IL6 can influence E2F1 activity.³⁶ Our experimental results also confirmed that CAFs can secrete IL6, and previous evidence has confirmed that IL-6, upon binding to its receptor IL-6R,^30,36 further recruits IL6ST, which then activates downstream signaling pathways and participates in the regulation of DNA replication and repair. It should be noted that our study only identified IL6 as a key cytokine for CAFs-mediated activation of E2F1 through transcriptomic analysis and an in vitro CAFs–NECC cell co-culture model, and more research is needed to fully elucidate the specific molecular mechanisms by which IL6 plays a role.

While HRR defects are classically associated with PARP inhibitor sensitivity in BRCA-mutant cancers, our findings illustrate that HRR hyperactivation can be exploited as a target in NECC. Although clinical evidence for PARP inhibitor monotherapy in NECC remains limited to case reports^37,38 and immunohistochemical studies showing high PARP1 expression (75-91% of cases),^39,40 our functional data demonstrate that NECC cells exhibit significant baseline sensitivity to olaparib, supporting its potential as a targeted therapy. This concept is further bolstered by recent studies in pulmonary neuroendocrine carcinomas showing PARP inhibitor efficacy⁴¹ and synergistic combinations in preclinical models.^42,43 Consistent with this, we observed moderate sensitivity to PARP inhibitor with olaparib in NECC models. However, monotherapy efficacy was limited, as BLM-mediated fork protection provided a rapid escape route, mirroring the modest objective response rates observed in early-phase PARP inhibitor trials. To overcome this adaptive resistance, we rationally designed a combination strategy simultaneously targeting PARP (with olaparib) and BLM (with the investigational inhibitor ML216). This dual blockade achieved potent synergistic effects in vitro and induced durable tumor regression in patient-derived xenograft models, including those incorporating CAFs. The efficacy of this combination in the presence of CAFs is particularly encouraging, suggesting its potential applicability across diverse tumor microenvironment contexts.

In vitro assays and NECC organoid models indicate that PARP inhibitor monotherapy elicits only a moderate antitumor response, whereas combining PARP inhibition in the context of high BLM expression yields substantially enhanced antitumor activity. Although prior work²⁵ and our in vitro and in vivo data support that BLM can stabilize replication forks and compensate for homologous recombination repair, thereby reducing cellular sensitivity to PARP inhibition, BLM inhibitors remain at the preclinical stage, and their clinical efficacy and safety have not yet been systematically evaluated in patients. Accordingly, future studies should incorporate patient-derived organoid drug response testing and clinical investigations to validate the antitumor activity of PARP inhibition in NECC and to rigorously assess the effectiveness and safety of combination regimens. Moreover, before considering PARP inhibitor–BLM inhibitor combination therapy, biomarker profiling should be performed, including BLM and PARP1/2 expression levels and HRD scores. Current evidence suggests that NECC patients with high expression of both BLM and PARP1/2 may be the subgroup most likely to benefit from this strategy.

The strategy of combining PARP inhibitors with BLM inhibitors for treating NECC, proposed in this study, is still in the preclinical research stage. Through preliminary validation of the efficacy and safety of the combined treatment in in vitro cell models, organoid models, and in vivo mouse models, a preliminary foundation has been laid for the translational application of this strategy. Future translational research should focus on advancing the following directions: (1) Systematic safety evaluation: Further clarify the safety profile and therapeutic window of the combined treatment in a broader range of cell lines, organoids, and animal models; (2) Development and validation of biomarkers: Screen and validate biomarkers that predict therapeutic efficacy, such as the expression levels of BLM, E2F1, and HRR pathway-related genes (RAD51, BRCA1/2, etc.), as well as dynamic pharmacodynamic markers such as γH2A.X and pRPA32, to precisely identify patient subgroups that could benefit from the combined therapy; (3) Efficacy verification of the combination regimen: Further confirm the synergistic antitumor effects of BLM inhibitors and PARP inhibitors in preclinical models, providing a basis for subsequent clinical trial design. Translating the “E2F1-BLM-HRR” regulatory axis into clinical practice faces multiple challenges, including the prospective validation and standardization of biomarkers (such as E2F1/BLM expression), early clinical development of BLM inhibitors, toxicity management of combination therapy, and resistance risks arising from tumor heterogeneity. At the same time, this axis also presents important opportunities: patient stratification could be achieved through E2F1/BLM expression levels, identifying subgroups that are sensitive to PARP inhibitor combination therapy, thus providing a basis for personalized treatment. Additionally, therapeutic strategies targeting this axis could explore more clinically tractable alternatives, such as using already approved Cyclin-dependent kinase 4/6 (CDK4/6) inhibitors (e.g., palbociclib) to indirectly downregulate BLM by inhibiting the RB-E2F1 pathway, or combining with ATR inhibitors to synergistically enhance replication stress. These approaches may provide more efficient paths for clinical translation.

Several limitations of our study warrant consideration. First, while our clinical cohort was substantial, all participants were of Han Chinese ancestry, and the rarity of NECC limited the sample size for robust subgroup analyses. Future validation in multi-ethnic, international cohorts is essential. Second, the BLM inhibitor remains a tool compound; the clinical translation of our strategy hinges on the development of clinical-grade BLM inhibitors and their thorough evaluation in patient-derived organoid models and toxicology studies. Third, while our multi-omics cohort included 10 patients, a substantial number, given the rarity of NECC, all key findings were rigorously validated functionally and in vivo, ensuring robustness. Our bulk omics approach and relatively short-term xenograft endpoints cannot fully capture tumor heterogeneity, long-term adaptive resistance, or dynamic immune-microenvironment interactions. Thus, the current rarity of NECC models restricts in vitro validation primarily to the TC-YIK cell line. To address this, we are actively establishing patient-derived organoids to enhance translational relevance. Finally, although PARP inhibitors are not yet standard in NECC, our work provides a strong preclinical rationale for biomarker-driven clinical trials.

In conclusion, our study through population cohort analysis and in vitro functional assays demonstrates that NECC exhibits pronounced resistance to first-line therapies and poor prognostic characteristics. Integrated multi-omics analysis reveals that the molecular mechanism underlying NECC progression stems from hyperactivation of the HRR pathway. Mechanistic studies indicate that the coordinated upregulation of core HRR components (including BLM and RAD51) at the transcriptomic, proteomic, and phosphoproteomic levels forms a BLM-RAD51-MCM regulatory network, which promotes tumor progression by sustaining genomic stability. Although HRR dependency confers synthetic lethality to PARP inhibitors in NECC, therapeutic efficacy is limited by compensatory BLM upregulation mediated through E2F1-dependent transcriptional activation. Notably, we found that CAFs in the tumor microenvironment promote excessive HRR activation via the E2F1-BLM signaling axis, serving as a critical stromal factor driving PARP inhibitor-acquired resistance. Importantly, in vitro and in vivo experiments confirm that dual targeting of PARP and BLM exhibits significant synergistic antitumor effects without observable toxicity, providing a promising combination therapy strategy for NECC clinical treatment.

Materials and methods

Study overview

This study employed a five-step framework to identify and validate novel therapeutic strategies for NECC. First, we established the CMC-NECC Cohort (n = 715) and non-NECC Cohort (n = 1059), tracking survival outcomes following first-line antitumor therapies, complemented by in vitro drug sensitivity assays to profile differential responses between NECC and conventional cervical cancer cell lines. Second, we performed multi-omics profiling (transcriptomics, proteomics, and phosphoproteomics) on treatment-naïve tumor specimens and paired NATs from 10 randomly selected CMC-NECC cases to define molecular signatures and identify druggable vulnerabilities. Third, candidate agents were rigorously screened in vitro for monotherapy efficacy and combinatorial potential. Fourth, integrated multi-omics and functional assays elucidated the mechanistic bases for optimized combination strategies. Finally, in vivo validation was conducted using cell line-derived xenograft (CDX) models to confirm therapeutic efficacy and dissect pharmacodynamic mechanisms. (Supplementary Fig. 1). This study was approved by the Institutional Review Board (IRB) of Fujian Maternity and Child Health Hospital (Approval No. 2024KY033) and conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participants prior to sample collection. The animal experiments were approved by the Animal Ethics Committee of Fujian Medical University (Approval No. IACUC FIMU 2023-0272).

Clinical validation cohort for chemotherapeutic response

To quantify the real-world effectiveness of empirical first-line chemotherapy in NECC, we analyzed patients treated with cisplatin, carboplatin, or etoposide from the CMC-NECC. The CMC-NECC cohort comprised histologically confirmed NECC cases diagnosed and treated between January 2008 and May 2024 across fifteen medical centers in China: Fujian Maternity and Child Health Hospital, Fujian Cancer Hospital, The First Affiliated Hospital of Fujian Medical University, Zhangzhou Affiliated Hospital of Fujian Medical University, Mindong Hospital Affiliated to Fujian Medical University, The First Hospital of Putian City, Jiangxi Maternal and Child Health Hospital, Xiangya Hospital of Central South University, Shaanxi Cancer Hospital, Gansu Maternal and Child Health Hospital, Hubei Maternal and Child Health Hospital, Shengjing Hospital of China Medical University, Shanghai First Maternity and Infant Hospital, Shenzhen Maternity and Child Healthcare Hospital, and Peking Union Medical College Hospital. Additionally, we recruited 1,059 patients diagnosed with squamous cell carcinoma or adenocarcinoma from the same 15 centers in China (2008-2020) to serve as the control group, comprising patients with non-NECC.

To minimize the influence of confounding factors and enhance the reliability of comparisons between the CMC-NECC and non-NECC groups, we employed propensity score matching (PSM), adjusting for variables including patient age, surgical approach, FIGO stage, lymphovascular space invasion (LVSI), lymph node involvement status, preoperative radiotherapy, and postoperative radiotherapy.

The first-line chemotherapeutic regimens evaluated in this cohort analysis were based on guideline-recommended combinations for cervical cancer, utilizing the agents cisplatin, carboplatin, and etoposide. The specific combination regimens and their abbreviations were as follows: TP/TC regimen, comprising a platinum agent (cisplatin or carboplatin) with paclitaxel; and EP regimen, comprising etoposide with cisplatin. For survival analysis, patients were categorized into distinct groups based on histology (NECC vs. non-NECC) and treatment received. Specifically, to compare chemotherapy efficacy between histologic types, we compared NECC patients receiving TP/TC versus non-NECC patients receiving TP/TC and NECC patients receiving EP. To assess survival benefit within the NECC population, we compared NECC patients receiving TP/TC-based or EP-based chemotherapy versus NECC patients not receiving any chemotherapy. Prognostic analyses were subsequently conducted between the matched NECC and non-NECC cohorts. Overall survival (OS) was defined as the time from diagnosis to death from any cause. Detailed baseline characteristics, treatment regimens, and follow-up data were extracted from medical records and are summarized in Supplementary Table 1, and the inclusion and exclusion criteria table is shown in Supplementary Fig. 2a.

Multi-omics discovery cohort

From the CMC-NECC cohort, treatment-naïve patients (no prior chemotherapy or radiotherapy) were selected. Ten participants were randomly chosen using a simple random sampling algorithm implemented in R software (v4.2.1). For each case, matched pairs of NECC tumor tissue and adjacent nontumour cervical tissue (NATs, ≤1 cm from tumor margin) were collected during surgical resection. Tumor and NATs tissues were sectioned consecutively (5–10 μm thick). The first and last sections (1st and 10th) were stained with hematoxylin and eosin (H&E) and evaluated by a pathologist to confirm the location and area of the tumor and NATs regions. Only samples with tumor and NATs cross-sectional diameters greater than 0.5 cm were considered valid for further experiments. Based on the H&E-stained sections, a “map” of NECC and NATs tissues was created to guide precise microdissection of the unstained “white sections.” Tumor regions were scraped into sterile, nuclease-free tubes. For NATs tissue, only regions within 1 cm of the tumor edge were scraped and stored in a separate tube. The minimum sample size for each omics analysis (e.g., transcriptomics, proteomics) was at least 5 white sections (10 μm thick) per sample. Clinicopathological data are presented in Supplementary Table 5. Fresh tumor tissue and NATs tissue were obtained intraoperatively, snap-frozen in liquid nitrogen within 20 min of devascularisation, and stored at −80 °C until simultaneous transcriptomic, global-proteomic, and phosphoproteomic analyses. Sample collection was approved by the institutional ethics committee, and all participants provided written informed consent.

Cell culture and reagents

The human cervical neuroendocrine carcinoma cell line TC-YIK, cervical squamous cell carcinoma line SiHa, adenocarcinoma line HeLa, immortalized cervical epithelial cell line ECT1/E6E7, and ovarian cancer lines A2780 and SKOV3 were purchased from iCELL or ATCC (all cell, consumables, reagents, antibodies, and their catalog numbers are provided in Supplementary Tables 6 and 7). CAFs and NFs were isolated from freshly resected cervical cancer and adjacent normal tissues, respectively, and authenticated by immunophenotyping (FAP⁺, α-SMA⁺).

Cells were cultured in complete media as recommended by the suppliers, at 37 °C in a humidified atmosphere with 5% CO₂. For co-culture experiments, TC-YIK or ECT1/E6E7 cells were seeded in the lower chamber of 0.4 μm Transwell inserts (Jet Biofil, China), while CAFs or NFs were cultured in the upper chamber. After 24 h of co-culture under serum-starved conditions, cells were harvested for downstream assays.

Vector construction and stable cell line transfection

Cell culture conditions were as described in Section cell culture and reagents. TC-YIK cells were transduced with shBLM lentiviruses (GV493/hU6–MCS–CBh–gcGFP–IRES–puromycin, Genechem, China) in the presence of polybrene (4–8 μg/mL) to stably knock down BLM expression; TC-YIK and ECT1/E6E7 cells were transduced with shE2F1 lentiviruses (GV493/hU6–MCS–CBh–gcGFP–IRES–puromycin, Genechem, China) to stably knock down E2F1 expression; and TC-YIK and ECT1/E6E7 cells were transduced with BLM overexpression lentiviruses (GV747/CMV enhancer–MCS–T2A–puromycin, Genechem, China) to stably overexpress BLM. Puromycin (1–2 µg/mL, Sigma) was used to select cells at 48 h after transduction, and cells were continuously selected for ~2 weeks to establish stable cell lines. The target sequences for shBLM were 5′-GCTACATATCTGACAGGTGAT-3′ (shBLM-1), 5′-CGAAGGAAGTTGTATGCACTA-3′ (shBLM-2), and 5′-GCCTTTATTCAATACCCATTT-3′ (shBLM-3); the target sequences for shE2F1 were 5′-ACCTCTTCGACTGTGACTTTG-3′ (shE2F1-1), 5′-TAAGAGCAAACAAGGCCCGAT-3′ (shE2F1-2), and 5′-CATCCAGCTCATTGCCAAGAA-3′ (shE2F1-3). The negative control sequence was 5′-TTCTCCGAACGTGTCACGT-3′ (CON313, Genechem, China). An empty-vector control lentivirus (CON563, Genechem, China) was used for BLM overexpression. Knockdown/overexpression efficiency was confirmed by qPCR and Western blot.

RNA sequencing (RNA-seq)

Total RNA was extracted using TRIzol reagent (Invitrogen) and quantified using a Qubit 4.0 fluorometer. RNA integrity was assessed using an Agilent 2100 Bioanalyzer (RIN ≥ 7.0). Libraries were prepared using the NEBNext Ultra RNA Library Prep Kit and sequenced on the Illumina NovaSeq 6000 platform (150 bp paired-end reads). Raw reads were processed using FastQC and Trimmomatic for quality control. Alignment to the human reference genome (GRCh38) was performed using HISAT2. Gene-level quantification was conducted using featureCounts, and differential expression analysis was performed using DESeq2 (|log₂FC| > 1, FDR < 0.05). Gene Ontology (GO) and KEGG pathway enrichment analyses were carried out using DAVID v6.8.

Genomics

WES was performed to interrogate coding regions enriched by probe hybridization and to profile coding-region variants relevant to tumor genomics. Raw sequencing data underwent routine quality assessment and filtering to remove low-quality reads, followed by evaluation of sequencing error-rate distribution and base-quality profiles. Clean reads were aligned to the human reference genome (B37/GRCh37), and alignment-based metrics, including sequencing depth and target-region coverage, were summarized to ensure data quality. Variants were then identified from the aligned data, and single-nucleotide polymorphisms (SNPs) were summarized as part of the standard variant-detection output.

HRD status in tumor tissues and cell-line samples was assessed using an amplicon-based targeted NGS assay for HRR-related gene variants and genomic scar scoring (GSS). The panel covered coding regions and exon–intron junctions of HRR-related genes and incorporated genome-wide SNP loci for GSS calculation. Reads were analyzed against the human reference genome (hg19), and variants were reported following HGVS nomenclature. HRD positivity was defined as GSS ≥ 45; samples with GSS < 45 were considered HRD-positive only if pathogenic/likely pathogenic BRCA1/2 variants were detected, otherwise HRD-negative. For Formalin-fixed paraffin-embedded (FFPE) tissue specimens, quality control metrics included Q30, BRCA/CDS coverage, BAF noise, and depth noise, and only samples meeting predefined QC criteria were interpreted.

Proteomics and phosphoproteomics

FFPE tissue sections were deparaffinized, rehydrated, and microdissected under H&E guidance. Proteins were extracted using the FFomic method as previously described.⁴⁴ After tryptic digestion, peptides were desalted and fractionated using high-pH reverse-phase liquid chromatography.

Phosphopeptides were enriched using TiO₂ beads. Samples were analyzed on an EASY-nLC 1200 system coupled to a Q Exactive HF-X mass spectrometer (Thermo Fisher Scientific). Raw MS data were processed using the iProX platform for peptide identification and quantification. Proteins with ≥2 unique peptides and FDR < 1% were retained. Phosphorylation sites were localized using ptmRS. Differential expression analysis was performed using Perseus (p < 0.05, |log₂FC| > 0.58). KSEA was conducted using the KSEAapp R package.

I-SceI–based HDR reporter assay

Cells (SiHa, HeLa, and TC-YIK) were seeded in 24-well plates and transiently cotransfected with the HDR reporter plasmid and the I-SceI endonuclease plasmid (GeneChem, China). Eight hours after transfection, the medium was replaced with complete medium, and the cells were cultured for an additional 48 h. The HR repair efficiency was quantified as the percentage of GFP-positive cells, indicative of successful HDR repair.

Flow cytometry analysis

Cells were treated with the following drugs 48 h after plasmid transfection: CK2 (1 μM, treated for 24 h), hydroxyurea (2 mM, treated for 2 h followed by a 2-h recovery), and PARP1 inhibitor (2.6 μM, treated for 24 h). After drug treatment, GFP-positive cells were quantified using flow cytometry on a BD LSRFortessa™ instrument (BD Biosciences) to assess HRR/HDR repair efficiency. Flow cytometry data were analyzed using FlowJo (v10.8.1).

Cytokine profiling

Cytokine profiling was performed using a membrane-based array (Proteome Profiler Human XL Cytokine Array, ARY022B, R&D Systems). Culture supernatants were collected from CAFs, NFs, TC-YIK cells conditioned with CAFs- or NFs–derived supernatants, TC-YIK cells alone, and CAFs‒TC-YIK co-cultures. Supernatants were harvested after 72 h of culture/conditioning or 72 h of co-culture, clarified by centrifugation, and analyzed with three biological replicates per condition (n = 3). Membranes were incubated with samples overnight, followed by biotinylated detection antibodies and streptavidin-HRP for chemiluminescent detection and imaging. Spot intensities were quantified by transmission scanning, duplicate spots were averaged, background was subtracted, and relative differences were compared across conditions.

Western blotting (WB)

Total protein was extracted using RIPA lysis buffer (Yaenzyme Biotech Co., Ltd. ShangHai, China) supplemented with protease inhibitor cocktail (Roche, Switzerland) and phosphatase inhibitors (Yaenzyme). Protein concentration was quantified using the BCA Protein Assay Kit (Yaenzyme). Equal amounts of protein (20 μg per lane) were separated by SDS-PAGE (Yaenzyme) and transferred onto PVDF membranes (Millipore, USA). Membranes were blocked with 5% nonfat milk in TBST for 1 h at room temperature, followed by overnight incubation at 4 °C with primary antibodies against BLM, RAD51, PARP1, PARP2, E2F1, γH2A.X, phospho-Chk1-S345, and GAPDH. After washing with TBST, the membranes were incubated with HRP-conjugated secondary antibodies (1:50,000, Tagene Biotech Co., Ltd., Xiamen, China) for 1 h at room temperature. Protein bands were visualized using ECL chemiluminescent substrate (Yaenzyme) and imaged using the ChemiDoc Imaging System (Bio-Rad). Band intensities were quantified using ImageJ software and normalized to GAPDH. All Western blot experiments were performed in biological triplicates to ensure reproducibility.

Quantitative real-time PCR (qRT‒PCR)

Total RNA was extracted using TRIzol reagent (Invitrogen) according to the manufacturer’s instructions. RNA concentration and purity were assessed using a NanoDrop 2000 (Thermo Fisher). cDNA was synthesized from 1 μg of total RNA using the PrimeScript RT Reagent Kit (Takara, Japan). qPCR was performed using SYBR Green PCR Master Mix (Applied Biosystems) on the ABI 7500 Real-Time PCR System. Each reaction was run in triplicate, and relative gene expression was calculated using the 2^(−ΔΔCt) method, with GAPDH as the internal control.

The sequences of primers used in this study are listed in Supplementary Table 8. All qPCR assays were performed in three independent biological replicates to ensure data reliability.

Cell immunofluorescence

Cells were seeded on glass coverslips in 24-well plates. Cells were treated with CK2 inhibitor (1 nM) for 24 h, hydroxyurea (HU, 2 mM) for 2 h followed by a 2 h recovery, and/or a PARP2 inhibitor (2.6 μM) for 24 h prior to immunofluorescence staining. After treatment, the cells were fixed with 4% paraformaldehyde for 15 min, permeabilized with 0.3% Triton X-100 for 10 min, and blocked with 5% BSA for 1 h. Coverslips were incubated overnight at 4 °C with primary antibodies against BLM, E2F1, PARP1, PARP2, 53BP1 or phospho-RPA32/RPA2 (Ser8). After three washes, coverslips were incubated with Alexa Fluor 488- or 647-conjugated secondary antibodies (1:500, Beyotime) for 1 h at room temperature. DAPI was used for nuclear staining. Coverslips were mounted with anti-fade mounting medium (Beyotime). Images were acquired using a Zeiss LSM 980 confocal microscope and analyzed using ImageJ with the Cellpose plugin for automated cell segmentation and fluorescence quantification, ensuring unbiased analysis. All imaging and quantification were performed in a blinded manner to reduce observer bias.

Organoid immunofluorescence

Organoid samples were processed as FFPE sections and stained using a TSA-based multiplex immunofluorescence workflow. Sections were baked at 65 °C for 2 h, deparaffinized in xylene, and rehydrated through graded ethanol (95%, 85%, and 75%; 5 min each), followed by PBS washes (3 × 5 min) and a brief PBST wash (30 s). Antigen retrieval was performed in EDTA buffer (pH 9.0) or citrate buffer (pH 6.0) by microwaving (buffer preheating for 10 min, followed by heating with slides for 25 min at medium-low power), and sections were then washed with PBS (3 × 5 min). Endogenous peroxidase activity was quenched with 3% H₂O₂ for 10 min at room temperature, followed by PBS washes (3 × 5 min). Sections were blocked with normal nonimmune goat serum (100 μL per section) for 20 min at room temperature. Primary antibodies were applied and incubated overnight at 4 °C, using the same antibody panel and dilutions as described in Section cell immunofluorescence (BLM, E2F1, PARP1, PARP2, 53BP1, and phospho-RPA32/RPA2 (Ser8); all 1:500). After PBST washes (3 × 5 min), sections were incubated with an HRP-polymer secondary reagent for 30 min at room temperature and washed with PBS (5 × 5 min). TSA fluorophore development was performed by incubation with tyramide fluorophore reagents (e.g., 520/570) for 5 min, followed by PBST washes (3 × 5 min). Nuclei were counterstained with DAPI for 10 min at room temperature in the dark, followed by PBS washes (3 × 5 min), and sections were mounted with anti-fade mounting medium prior to fluorescence imaging. For multiplex staining, the antibody–HRP-Polymer–TSA cycle was repeated sequentially for each target as needed.

Chromatin immunoprecipitation (ChIP)

ChIP assays were performed using the SimpleChIP Enzymatic Chromatin IP Kit (CST, USA) according to the manufacturer’s instructions. Briefly, cells were crosslinked with 1% formaldehyde for 10 min, quenched with glycine, and lysed. Chromatin was sheared to 200–500 bp fragments using a Covaris ultrasonicator. After preclearing with Protein G magnetic beads, chromatin was immunoprecipitated with anti-E2F1 antibody (CST, #3742) or control IgG overnight at 4 °C. Immunocomplexes were washed, eluted, and reverse-crosslinked. DNA was purified and analyzed by qPCR using primers targeting the BLM promoter region. Primer sequences are listed in Supplementary Table 8. ChIP‒qPCR results are presented as % of input and fold enrichment over IgG control, as recommended by ENCODE and CST guidelines.

Drug sensitivity assays

Cells were seeded in 96-well plates (5 × 10³ cells/well) and treated with increasing concentrations of cisplatin, carboplatin, etoposide, paclitaxel, olaparib, or BLM inhibitor for 24–48 h. Cell viability was assessed using the CCK-8 assay. IC₅₀ values were calculated using GraphPad Prism 9.5.1. Synergistic effects were evaluated using the ZIP model (synergyFinder R package).

Organoid drug sensitivity assay (ATP-based luminescence)

Cervical cancer organoids were plated in 96-well plates and treated with compounds as indicated. For screening, 29 compounds were tested across six concentrations with three technical replicates per concentration, together with medium-only controls (n = 3) and vehicle controls containing 0.1% DMSO (1:1,000; n = 3). Organoids were treated for 3 days with a medium change on day 2, and bright-field images were collected at days 0–3 (40× and 100×). Viability was quantified on day 3 using the CellCounting-Lite® 3D Luminescent ATP assay and normalized to the corresponding controls.

For dose–response profiling, organoids were treated for 48 h with olaparib (0.1–50 μM), BLM (1–100 μM), or cisplatin (0-100 μM) (three replicates per condition). Olaparib and BLM were prepared in DMSO and cisplatin in water; matched vehicle controls (0 μM) were included. Luminescence was measured at 48 h to fit dose–response curves and calculate IC₅₀ values. For efficacy validation, organoids were treated for 48 h with single agents or combinations (olaparib 500 μM; BLM 500 μM; cisplatin 200 μM), followed by bright-field imaging and ATP-based viability measurement (three replicates per condition).

DNA damage and repair assays

DNA double-strand breaks (DSBs) were assessed by immunofluorescence staining of γH2A.X and RAD51 foci. Cells were irradiated (4 Gy) or treated with drugs, fixed after 6–24 h, and stained with primary antibodies against γH2A.X (ABclonal/AP0687) and RAD51 (Abcam/ab88752). Foci were counted in ≥100 cells per group using ImageJ.

EdU cell proliferation assay

Cell proliferation was assessed using the EdU Cell Proliferation Kit (Beyotime, China) according to the manufacturer’s instructions. Briefly, cells were seeded in 24-well plates and incubated with 50 μM EdU for 2–6 h at 37 °C, a duration optimized in pilot experiments to ensure labeling during the log-phase of growth. After fixation with 4% paraformaldehyde and permeabilization with 0.3% Triton X-100, the cells were incubated with Click reaction cocktail for 30 min. Nuclei were counterstained with Hoechst 33342. Images were captured using a Zeiss LSM 980 confocal microscope. EdU-positive cells were quantified using ImageJ and expressed as a percentage of total nuclei.

Mouse xenograft model

To evaluate the antitumor efficacy of BLM and PARP inhibition in vivo, two independent xenograft experiments were performed using female BALB/c nude mice (4–6 weeks old, Beijing Vital River Laboratory Animal Technology Co., Ltd.). All animals were housed under specific pathogen-free (SPF) conditions with free access to food and water. Animal procedures were approved by the Ethics Committee of Fujian Maternity and Child Health Hospital and conducted in accordance with institutional guidelines.

TC-YIK monoculture xenograft

TC-YIK cells (1 × 10⁶ cells/mouse) were resuspended in a 1:1 mixture of PBS and Matrigel (Corning) and subcutaneously injected into the right flank of each mouse. When tumor volumes reached approximately 50 mm³, mice were randomly assigned into four treatment groups (n = 4 per group):

Control group: received daily intraperitoneal (i.p.) and intragastric gavage (i.g.) injections of vehicle; BLM inhibitor (B-i) group: received ML216 at 25 mg/kg/day via i.p. injection; PARP inhibitor (P-i) group: received olaparib at 50 mg/kg/day via i.g.; Combination group (B-i and P-i): received both ML216 (25 mg/kg/day) and olaparib (50 mg/kg/day).

TC-YIK and CAFs/NFs cotransplantation xenograft

To investigate the role of CAFs in modulating treatment response, TC-YIK cells were mixed with CAFs or NFs at a 1:1 ratio (total 2 × 10⁶ cells/mouse) and coinjected subcutaneously. When tumors reached 50 mm³, mice were similarly randomized into the four treatment groups described above (n = 7 per group).

Tumor monitoring and endpoint analysis

Tumor length (L) and width (W) were measured every 2 days using a vernier caliper, and tumor volume was calculated as V = 0.5 × L × W². After 14 days of treatment, all mice were euthanized, and tumors were excised and weighed. Part of the tumor tissue was fixed in 4% paraformaldehyde for H&E staining and immunohistochemical (IHC) analysis (e.g., TUNEL, Ki67, γH2A.X, RAD51, BLM, α-SMA); the remainder was snap-frozen for molecular analyses. Major organs (heart, liver, spleen, lungs, kidneys) were also collected for H&E staining to assess potential drug-related toxicity.

Cancer-associated fibroblast staining in tumor tissues

Tumor tissues were collected from the TC-YIK and CAFs/NFs cotransplantation xenograft model. After fixation in 4% paraformaldehyde for 24 h, the samples were paraffin-embedded and sectioned at 4 μm thickness. Sections were deparaffinized in xylene and rehydrated through a graded ethanol series. Antigen retrieval was performed using citrate buffer (pH 6.0) at 95 °C for 20 min. After blocking with 5% normal goat serum for 30 min, the sections were incubated overnight at 4 °C with primary antibodies against the CAFs markers α-SMA and FAP. After washing with PBS, the sections were incubated with Alexa Fluor 488- or 647-conjugated secondary antibodies (1:500, Beyotime) for 1 h at room temperature. Finally, the sections were counterstained with DAPI for 10 min and mounted with anti-fade mounting medium (Beyotime). Stained sections were analyzed by microscopy using a Zeiss LSM 980 confocal microscope.

Bioinformatics and data integration

Multi-omics data integration was performed using R 4.2.1. RNA-seq and proteomics data were matched by gene symbol, and correlation analyses were conducted using Pearson correlation. Pathway enrichment was performed using DAVID 6.8, and GSEA KSEA was conducted using the KSEAapp package. Transcription factor prediction was performed using JASPAR, hTFtarget, and CistromeDB. Tumor microenvironment deconvolution was performed using the xCell algorithm. Drug sensitivity data were extracted from the DepMap and CCLE databases. All bioinformatics workflows were performed under standard quality control and normalization protocols. All computational analyses were performed in at least two independent pipelines to verify reproducibility. All bioinformatics packages, versions, and parameters used in this study are listed in the Supplementary Software and Algorithms.

Bioinformatics analysis

Publicly available transcriptome datasets were obtained from the GEO database (GSE138080, CESC cohort) and the Genome Sequence Archive (GSA; HRA002655, NECC cohort). Expression matrices were quantile-normalized using normalizeQuantiles in limma (v3.60.3). Differential expression was performed with limma (v3.60.3) using thresholds of P < 0.05 and |log2FC| > 1 and visualized with ggplot2 (v3.5.2) and ComplexHeatmap (v2.20.0). Tumor microenvironment infiltration scores were inferred using xCell (v1.1.0); samples were stratified into fibroblast-high (CAF+) and fibroblast-low (CAF−) groups by the median fibroblast score, followed by differential expression analysis (P < 0.05, |log2FC| > 1). Pearson correlations between E2F1 expression and all genes were computed in Python using pandas; positively correlated genes were intersected with CAFs-associated DEGs to define an E2F1-associated candidate set. Pathway activity was quantified by ssGSEA using GSVA (v1.52.3) with MSigDB gene sets (Hallmark E2F Targets, DNA replication, and HRR); E2F Targets ssGSEA scores were used as a proxy for E2F1 transcriptional activity, and ssGSEA scores were also computed for HRR and DNA replication gene sets (including BRCA1/2, ATM, and MCM family genes). Statistical analyses were conducted in R (v4.4.1) using two-sided Wilcoxon rank-sum tests for two-group comparisons, with ggplot2 and ggpubr (v0.6.0) for visualization; P < 0.05 was considered significant. HRD analysis on paired tumor–normal sequencing data used Sequenza (v3.0.0) to infer purity, ploidy, and CNVs; depth and allele frequencies were extracted with sequenza-utils, processed with 50-bp binning and quality filtering, and optimal solutions were selected via the Bayesian model; scarHRD (v0.1.1) was then used to compute HRD-sum scores from LOH, LST, and TAI.

Statistical analysis

Survival curves were estimated using the Kaplan‒Meier method and compared with the log-rank test. For Kaplan‒Meier plots, the numbers at risk were provided at annual time points, with the numbers censored during each interval indicated in parentheses. The hazard ratios (HR) and their 95% confidence intervals (CI) for the effect of histology (NECC vs. non-NECC) and treatment regimens on OS were calculated using univariable and multivariable Cox proportional hazards regression models. All statistical analyses were performed using R software (version 4.2.1) with the “survival” and “survminer” packages. All experiments were independently repeated at least three times unless otherwise stated. Data are presented as the mean ± standard deviation (SD). Statistical comparisons between two groups were performed using two-tailed Student’s t-test. Multiple group comparisons were analyzed using one-way ANOVA followed by Tukey’s post hoc test. All statistical analyses were performed using GraphPad Prism version 9.0 and R version 4.2.1. A two-sided P value of <0.05 was considered statistically significant.

Author contributions

Conceptualization: B.H.D., J.T.Y., and P.M.S.; data curation: B.H.D., Y.W., Y.X.H., T.Y., X.H.L., X.H.C., Y.H.L., S.H.Z., X.Y.Y., and S.Y.C.; formal analysis: Y.W., Y.N.Q., Y.X.H., X.H.L., Y.L.C., and S.X.X.; funding acquisition: P.M.S. and B.H.D.; methodology: B.H.D., Y.W., Y.N.Q., L.H.C., Y.X.H., T.Y., X.H.L., X.H.C., Y.L.C., Y.H.L, S.H.Z., S.X.X, X.Y.Y, S.Y.C, Y.Z., L.L.,. X.W.P., and J.T.Y.; project administration: B.H.D. and P.M.S.; visualization: Y.W. and B.H.D.; writing–original draft: B.H.D. and Y.W.; writing–review and editing: B.H.D., Y.W., Y.N.Q., L.H.C., Y.X.H., Y.Z., L.L., X.W.P., J.T.Y., and P.M.S.; all the authors have read and approved the article.

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request. The raw sequence data reported in this paper have been deposited in the Genome Sequence Archive (Genomics, Proteomics & Bioinformatics 2025) in the National Genomics Data Center (Nucleic Acids Res 2025), China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences (accession number: HRA016702), which are publicly accessible at https://ngdc.cncb.ac.cn/gsa-human. The proteomic and phosphoproteomic data reported in this paper have been deposited in the OMIX, China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences (https://ngdc.cncb.ac.cn/omix), under accession no. OMIX014862. All other supporting data for the findings reported in this study are available within the article and its Supplementary Information files.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41392-026-02822-1.