AZD4635 Targets cAMP/CREB Axis to Salvage PARPi-Induced Immune Evasion and Enhance Antitumor Efficacy in Ovarian Cancer

Abstract

Background/Objectives: Poly(ADP-ribose) polymerase inhibitors (PARPis) have significantly transformed the treatment landscape for ovarian cancer; however, their clinical efficacy is often limited by poor response rates and the emergence of resistance. Recent studies have revealed that in ovarian cancer cells resistant to PARPi, the expression levels of adenosine receptors are upregulated. Accumulation of adenosine activates adenosine A2A receptor (A2AR) on immune cells, leading to immune suppression and immune escape. We hypothesize that this is a key factor limiting the efficacy of PARPi and driving the development of resistance. Therefore, the rational combination of PARPi with A2AR antagonists (A2ARas) may represent a highly promising anticancer strategy. Methods: To assess the effects of the PARPi AG14361 and the A2ARa AZD4635 on ovarian cancer growth and the immune microenvironment, we conducted in vitro and in vivo experiments and utilized single-cell RNA sequencing (scRNA-seq) to construct a high-resolution immune landscape. Results: AG14361 significantly inhibited ovarian cancer growth both in vitro and in vivo, accompanied by the accumulation of cyclic adenosine monophosphate (cAMP) and activation of the cAMP/cAMP response element-binding protein (CREB) pathway in mouse cells and tumor tissues. However, compared to monotherapy, the combination of AG14361 and AZD4635 significantly enhanced antitumor activity by inhibiting cAMP accumulation and the cAMP/CREB pathway. More importantly, the combination therapy of PARPi and A2ARa reduced the infiltration of immunosuppressive cells (such as regulatory T cells and M2 macrophages) while increasing the infiltration of cytotoxic T cells and granzyme B-positive cells, thereby creating a more favorable immune microenvironment for tumor clearance. Single-cell analysis revealed distinct functional subpopulations of macrophages and T cells, highlighting the complexity of immune heterogeneity and the potential for targeting specific immune cell subpopulations to enhance therapeutic efficacy. Conclusions: These findings suggest that the combination therapy of PARPi and A2ARa is a highly promising strategy that overcomes PARPi-induced immune escape by targeting the cAMP/CREB axis, thereby synergistically enhancing antitumor effects and holding promise as an effective treatment for solid tumors.

1. Introduction

Ovarian cancer ranks as the third most prevalent malignancy within the female reproductive system and concurrently holds the dubious distinction of being the deadliest gynecologic cancer [1]. In 2020 alone, the global landscape witnessed an estimated 313,959 new diagnoses of ovarian cancer, with the grim toll of over 207,252 fatalities attributed to this disease [2]. Alarmingly, its incidence is on an upward trajectory in Eastern Europe and select regions of Asia [3]. Despite remarkable strides in ovarian cancer treatment, leveraging tumor debulking and platinum-based chemotherapy, the specter of high recurrence rates continues to haunt patients. A staggering 70% of patients experience relapse within a mere 3 years, with the intervals between subsequent recurrences progressively shortening until the emergence of platinum resistance becomes inevitable [4]. In this context, the quest for effective maintenance therapy drugs has emerged as a pivotal breakthrough in the ongoing battle to enhance the survival rates of ovarian cancer patients.

The advent of poly(adenosine diphosphate) ribose polymerase (PARP) inhibitors (PARPis) has undeniably revolutionized the therapeutic landscape of ovarian cancer. A wealth of clinical studies has consistently underscored the significant extension of progression-free survival (PFS) that PARPi bestows upon ovarian cancer patients [5,6]. The U.S. Food and Drug Administration (FDA) has now granted approval to three distinct PARPi—olaparib, niraparib, and rucaparib—for deployment as maintenance therapy in ovarian cancer patients [7]. Although PARPi has demonstrated encouraging potential in clinical applications, its limited sustained efficacy and the propensity for resistance development—both intrinsic and acquired—remain significant hurdles. Over 40% of patients with BRCA1/2 mutations fail to respond to PARPi, and many patients develop resistance after prolonged oral administration of these agents [8,9]. To date, multiple mechanisms have been identified that contribute to PARPi resistance [10]. Notably, a study by Chi et al. [11] revealed a marked upregulation of the inflammatory pathway and adenosine receptor in olaparib-resistant cells, implicating adenosine and adenosine signaling as key mediators of PARPi resistance.

Adenosine (ADO) is a metabolite that is highly enriched within the tumor microenvironment (TME) and has emerged as a pivotal immune inhibitory factor in tumors [12]. Under normal physiological conditions, adenosine and ATP are present at exceedingly low levels in the extracellular fluid [13]. However, inflammation, ischemia, or cancer can trigger the high-level release of ATP through various mechanisms [14]. Extracellular ATP is subsequently degraded by extracellular nucleotidases, most notably CD39 and CD73, ultimately yielding adenosine [15]. Notably, CD39 and CD73 are highly expressed on cells within the TME and are further upregulated under hypoxic conditions through HIF-1α-mediated mechanisms [16]. In addition to the canonical ATP degradation pathway via the CD39-CD73 axis, extracellular adenosine can also be generated from nicotinamide adenine dinucleotide (NAD⁺) through the CD38-ENPP1-CD73 pathway [15]. Wang et al. [17] demonstrated that PARP inhibitors induce an increase in intracellular NAD⁺ levels, a metabolite that critically modulates the cytotoxic potency of PARP inhibitors [18]. Extracellular adenosine then activates downstream pathways by binding to adenosine receptors (primarily A2A receptors) expressed on immune cells, including triggering increased adenosine cyclase activity, which in turn elevates intracellular cyclic adenosine monophosphate (cAMP) levels with potent immunosuppressive effects [19]. By activating protein kinase A (PKA), cAMP exerts profound effects on a wide range of immune cells and processes, inhibiting effector cell function and stabilizing immunosuppressive regulatory cells, ultimately promoting tumor cell escape from immune surveillance [20]. An increasing number of preclinical studies have confirmed that the CD39/CD73/A2AR pathway represents a novel target for immunotherapy [21,22]. Within this pathway, three distinct targets can be exploited to inhibit tumor immune escape. First, directly inhibiting the binding of adenosine to A2AR on immune cells can block adenosine-mediated immunosuppressive signaling. Second, inhibiting the function of CD39 or CD73 can reduce adenosine production, thereby diminishing its immunosuppressive effects. These strategies are aimed at disrupting the adenosine-mediated immunosuppressive network within the TME.

Currently, numerous clinical trials targeting the adenosine pathway are underway [12,20,22,23]. These trials include antibodies targeting CD39 and CD73, as well as small-molecule inhibitors of the A2A receptor. Early-stage clinical trials across various tumor types have demonstrated promising antitumor effects of these inhibitors. As more clinical trial results are published, the combination of blocking the adenosine pathway with immune checkpoint inhibitors, targeted therapies, traditional chemotherapy agents, radiation therapy, and endocrine therapy is expected to yield better clinical outcomes for cancer patients. We posit that the accumulation of adenosine within the TME may be a crucial factor underlying resistance to PARPi and the insensitivity of tumor cells to PARP inhibitors. Given this hypothesis, the concurrent application of PARPi and inhibitors targeting the adenosine pathway may provide innovative therapeutic avenues for ovarian cancer and other solid tumors. The CD39/CD73/A2AR pathway is activated by the binding of adenosine to the A2A receptor, which subsequently modulates the proliferation and function of immune cells. In this context, the novel A2AR antagonist AZD4635, which competitively binds to the A2A receptor, can effectively block A2AR-mediated signaling in tumor-infiltrating immune cells, thereby mitigating the immunosuppressive characteristics of the TME [24,25]. Integrating PARPi with an A2AR antagonist in a rational combination therapy regimen may thus represent a promising anticancer strategy, offering hope to ovarian cancer patients.

Here, we show that the A2AR antagonist AZD4635 markedly amplifies the antitumor efficacy of PARP inhibitors. Mechanistically, AZD4635 blocks adenosine–A2AR engagement, lowers intracellular cAMP, and suppresses CREB phosphorylation. This signaling switch re-invigorates the tumor immune microenvironment, counteracts PARPi-elicited immune escape, and unleashes a potent, synergistic anticancer response. These data provide a rationale for integrating A2AR antagonism into next-generation PARP-based combination regimens.

2. Materials and Methods

2.1. Cell Lines and Cell Culture

The mouse ovarian cancer cell line ID8 (Cat. No. FH1030) was obtained from Shanghai FuHeng Biotechnology Co., Ltd. (Shanghai, China). ID8 cells were cultured in DMEM (Gibco, NY, USA) supplemented with 10% fetal bovine serum (FBS; Gibco, NY, USA) and 1% penicillin/streptomycin (Gibco, NY, USA) at 37 °C in a humidified atmosphere containing 5% CO₂. Peripheral blood mononuclear cells (PBMCs) were isolated from C57BL/6 mice. Blood was collected via orbital puncture into anticoagulant tubes. The whole blood was mixed with an equal volume of PBS and carefully layered over an equal volume of Ficoll separation solution (Cytiva, Shanghai, China). The samples were centrifuged at 2000 rpm for 20 min. The intermediate layer containing the PBMCs was transferred to a new centrifuge tube and centrifuged again at 1500 rpm for 10 min. The cells were washed three times with PBS, resuspended in complete DMEM medium, and counted. The prepared PBMCs were reserved for follow-up experiments.

2.2. Antibodies and Chemicals

AG14361 (Cat. No. HY-12032) and AZD4635 (Cat. No. HY-101980) were obtained from MedChemExpress (Shanghai, China). Antibodies specific to Gapdh (Cat. No. 5174S), phospho-Creb (Cat. No. 9198S), Creb (Cat. No. 9197) were purchased from Cell Signaling Technology (CST; Danvers, MA, USA). Antibodies specific to Cd8 (Cat. No. bs-0648R), Cd4 (Cat. No. bs-0647R), and Foxp3 (Cat. No. bs-0269R) were purchased from Beijing Bioss Biotechnology Co., Ltd. (Beijing, China). Antibodies specific to Cd11b (Cat. No. R380675) and Cd206 (Cat. No. R360017) were purchased from Zenbio (Chengdu, China). Antibodies specific to Granzyme B (Cat. No. ab255598) was purchased from Abcam (Cambridge, UK).

2.3. Cell Viability Assay

ID8 cells were plated in 96-well plates at a density of 2 × 10³ cells/mL (100 μL per well) in complete medium. After overnight adherence, PBMCs were added at various ratios and co-incubated for 3 h. The co-cultures were then treated with AG14361 or (and) AZD4635 for 48 h. The medium was subsequently replaced with complete medium containing 10% CCK-8 (GlpBio Technology, Montclair, CA, USA), and the optical density at 490 nm was measured using a Multiscan MK3 microplate reader (ThermoFisher, Waltham, MA, USA) after an additional 2 h.

2.4. Colony Formation Assay

ID8 cells (500 cells per well, 1 mL per well) were seeded in 12-well plates with complete medium. After overnight adherence, PBMCs were added at various ratios and co-incubated with ID8 cells for 3 h. The co-cultures were then treated with AG14361 (10 μM) or (and) AZD4635 (32 μM) for durations determined by clone size, with the drug-containing complete medium refreshed every three days. On the last day, the medium was discarded, and the cells were washed once with PBS, fixed with 4% paraformaldehyde for 30 min at room temperature, washed once with PBS, and stained with crystal violet. The plates were photographed, and the number of colonies was counted to calculate the clone formation ratio using the formula: clone formation (%) = number of colonies/total number of seeded cells × 100%.

2.5. Enzyme-Linked Immunosorbent Assay (ELISA)

The levels of cAMP in ID8 cells and tumor tissues were determined using an ELISA kit (Cat. No. MM-0544M1) from Jiangsu Enzyme Immunoassay Industrial Co., Ltd. (Changzhou, Jiangsu, China), according to the manufacturer’s instructions.

2.6. Western Blot Analysis

Cell samples were washed twice with cold phosphate-buffered saline (PBS) and lysed with RIPA buffer (Beyotime; Shanghai, China) containing a protease inhibitor cocktail (Invitrogen; Grand Island, NY, USA) at 0 °C to extract proteins. Protein concentrations were determined using a BCA assay kit (Beyotime; Shanghai, China). The protein samples were separated by SDS-PAGE and transferred to PVDF membranes (Millipore, Bedford, MA, USA). Membranes were blocked with 5% skim milk at room temperature for 2 h, incubated with primary antibodies overnight at 4 °C, washed three times with TBST, and then incubated with secondary antibodies for 1 h at room temperature. Protein bands were detected using an enhanced chemiluminescence assay (ThermoFisher, Waltham, MA, USA) and imaged with a Champchemiluminmeter (Sagacreation; Beijing, China). Band intensities were quantified using ImageJ software (version 1.46).

2.7. RNA Extraction and Quantitative Real-Time PCR

Total RNA was isolated from cell samples using a reagent (Cat. RE-03113, FOREGENE, Chengdu, China). cDNA was synthesized from the RNA using PrimeScript RT Master Mix and SYBR green (TaKaRa, Japan). Relative expression levels were determined by the 2-ΔΔCt method with Gapdh as an internal control. Primers for real-time PCR were as follows: Cd39 forward: 5′-CTGTGGGGTTGACCCAGAAC-3′; reverse: 5′-TTGTGTGAGAAGAACCCGCA-3′; Cd73 forward: 5′-GTATCCGGTCGCCCATTGAT-3′; reverse: 5′-AAAGGCCTTCTTCAGGGTGG-3′.

2.8. Transfection

The siRNA for Cd39 and Cd73 were designed and obtained from Guangzhou Aiki Biotechnology Co., Ltd. Cd39, Cd73 siRNAs, and siRNA negative control (siRNA NC; Aiki Biotechnology, Guangzhou, China) were transiently transfected into ID8 cells using the Lipofectamine 3000 reagent (Invitrogen, Grand Island, NY, USA) according to the manufacturer’s instructions. Twelve hours post-transfection, the transfection medium was replaced with fresh complete medium, and cells were collected 48 h later for subsequent experiments. The siRNA sequences are as follows:

Cd39 (mouse):

• siRNA#1 (5′-GCACCAAGAGACACCCGUUUATT-3′),

• siRNA#2 (5′-UUGGCUUCUCCUCUAUCAUAGTT-3′),

• siRNA#3 (5′-CCUUCUGCAAGGCUAUCAUUUTT-3′);

Cd73 (mouse):

• siRNA#1 (5′-CCCAUUGAUGAACGCAACAAUTT-3′),

• siRNA#2 (5′-GCACUGGGAAAUCAUGAAUUUTT-3′),

• siRNA#3 (5′-AGUGUCGAGUGCCCAGUUAUGTT-3′).

2.9. Immunohistochemistry (IHC)

Fresh tissue samples were fixed in 4% paraformaldehyde and underwent dehydration using an automatic immunohistochemical stainer prior to paraffin embedding and sectioning. The immunohistochemical staining and statistical analysis of mouse tissues were conducted in accordance with standardized protocols [26]. The sections were incubated overnight at 4 °C with primary antibodies, followed by incubation with biotinylated anti-rabbit secondary antibody IgG, and subsequently stained with DAB solution. After restaining, dehydration, and clearing, the sections were mounted using neutral gum. The sections were examined under an inverted microscope, with five fields evaluated per section. Scoring was performed based on the intensity of the dye color and the number of positive cells.

2.10. Animal Experiments

This animal experimental protocol was approved by the Laboratory Animal Ethics Committee of Guangdong Provincial Medical Laboratory Animal Center (Approval No.: C202302-12; Approval Date: 28 February 2023). A total of 24 female C57BL/6 mice (6–8 weeks old) were purchased from Beijing Vital River Laboratory Animal Technology Co., Ltd. (Guangzhou, China; Production License No.: SCXK (Yue) 2022-0063) and housed under standard conditions. ID8 cells (1 × 10⁷ cells in 200 μL of medium) were injected subcutaneously into the right forelimb axilla of immunocompetent C57BL/6 mice. Tumor size was measured every 2 days with calipers, and tumor volume was calculated using the formula: 1/2 (length × width²). When tumors reached approximately 100 mm³, mice were randomized into control or experimental groups. The control group (6 mice) received saline by gavage, while the experimental group received AG14361 intraperitoneally (6 mice), AZD4635 by gavage (6 mice), or a combination of AG14361 and AZD4635 (6 mice). Mice were treated every 2 days, and body weight and tumor volume were monitored every other day. On day 3 after the final dose, mice were euthanized, tumors were weighed and photographed, and tumors were divided into two parts for storage in liquid nitrogen and 4% paraformaldehyde.

2.11. Transcriptomics Analysis

Total RNA was extracted using Trizol Reagent (Invitrogen, Carlsbad, CA, USA), and its concentration, quality, and integrity were assessed with a NanoDrop spectrophotometer (Thermo Scientific, Waltham, MA, USA). Three micrograms of RNA served as the starting material for RNA sample preparation. Sequencing libraries were constructed as follows: mRNA was isolated from total RNA using poly-T oligo-attached magnetic beads. Fragmentation was performed with divalent cations in an Illumina proprietary buffer at elevated temperatures. First-strand cDNA was synthesized using random primers and Super Script II, followed by second-strand synthesis with DNA Polymerase I and RNase H. The resulting cDNA ends were blunted via exonuclease/polymerase activities, and the enzymes were removed. After adenylation of the 3′ ends, Illumina PE adapter oligonucleotides were ligated. The library fragments were then purified using the AMPure XP system (Beckman Coulter, Beverly, CA, USA) to select for cDNA fragments of 400–500 bp. These fragments were enriched by PCR using Illumina PCR Primer Cocktail for 15 cycles. The final products were purified and quantified using the Agilent high sensitivity DNA assay on a Bioanalyzer 2100 system (Agilent, Santa Clara, CA, USA). The sequencing library was sequenced on the NovaSeq 6000 platform (Illumina, San Diego, CA, USA) by Shanghai Personal Biotechnology Cp. Ltd. (Shanghai, China).

2.12. Single-Cell RNA Sequencing (scRNA-Seq) Analysis

Fresh mouse tumor tissues were minced into approximately 1 mm³ pieces and suspended in DMEM medium (Invitrogen) supplemented with 10% fetal bovine serum (FBS; Gibco). Single-cell suspensions were prepared and Cd45⁺ cells were isolated using Cd45 MicroBeads, as previously described [27,28]. Approximately 10,000 Cd45⁺ cells per group were loaded onto the 10X Genomics GemCode Single-cell instrument to generate single-cell Gel Bead-In-Emulsion (GEMs). The GEMs were collected in a reservoir, where the gel beads dissolved to release barcode sequences. cDNA fragments were reverse transcribed and labeled. The gel beads were disrupted, and the emulsion was broken to release the cDNA. PCR amplification was performed using the cDNA as a template. The products from all GEMs were pooled to construct a standard sequencing library. Genedenovo Biotechnology Co., Ltd. (Guangzhou, China) performed GEMs generation, cDNA amplification, library preparation, and sequencing. The bioinformatics analysis pipeline followed the protocol described by Ren et al. [27]. Bioinformatic analysis was performed using Omicsmart, a real-time interactive online platform for data analysis (http://www.omicsmart.com, accessed on 1 March 2025).

2.13. Analysis of Survival Prognosis, Clinical Characteristics, and Tumor Immune Microenvironment

RNA sequencing (RNA-seq) data from the TCGA-OV (ovarian serous cystadenocarcinoma) project were obtained from the Cancer Genome Atlas (TCGA, https://portal.gdc.cancer.gov, accessed on 7 April 2025) database, based on the STAR workflow, and TPM-formatted data along with corresponding clinical information were extracted. Prior to analysis, the dataset underwent standardized quality control: samples with missing gene expression, incomplete key clinical information, or absent survival data were sequentially excluded. Subsequently, using 1.2-fold and 0.8-fold mean values of CD39/CD73 mRNA expression as cutoffs, samples were stratified into high- and low-expression groups, resulting in the inclusion of 66 TCGA-OV patients.

Kaplan–Meier analysis was employed to evaluate the correlation between CD39 and CD73 mRNA expression and survival outcomes. Proportional hazards hypothesis testing and survival regression modeling were performed using the survival package in R. Visualization of the results was achieved with the survminer and ggplot2 packages, with statistical significance assessed by the log-rank test (p < 0.05). Moreover, we examined the relationship between various clinical phenotypes and CD39 and/or CD73 mRNA expression, with baseline data visualized using ggplot2. Immune cell composition data for the 66 TCGA-OV patients were retrieved from the Cancer Immunome Atlas (TCIA, https://tcia.at/home, accessed on 7 April 2025). This database offers comprehensive immunogenomic analyses of next-generation sequencing (NGS) data from 20 solid tumors sourced from TCGA and other repositories.

2.14. Statistical Analysis

Data are expressed as the mean ± standard deviation (SD). Statistical analyses were conducted using Prism software (version 7.0; GraphPad, San Diego, CA, USA). Normality was assessed prior to analysis. Datasets following a normal distribution were analyzed using one-way ANOVA or Student’s t-test. p values for survival curves are assessed by log-rank test. For non-normally distributed datasets, comparisons were made using the Mann–Whitney U test or Kruskal–Wallis test. Significance was set at p < 0.05.

3. Results

3.1. Synergistic Tumor Suppression by PARPi AG14361 and A2AR Antagonist AZD4635 in Murine Ovarian Cancer Cells, Enhanced in PBMC Coculture

To elucidate the cytotoxicity of AG14361 and AZD4635 in murine ovarian cancer cells, we evaluated the viability of ID8 cells using the CCK-8 assay. Both AG14361 and AZD4635 dose-dependently reduced ID8 cell survival, with IC₅₀ values of 28 ± 2.19 μM and 122.3 ± 40.3 μM, respectively (Figure 1a–d). To further explore their effects on tumor immune activity, we cocultured ID8 cells with PBMCs at varying ratios (1:1, 2:1, and 4:1) and treated them with AG14361 or AZD4635. Compared to controls lacking PBMCs, AG14361- and AZD4635-treated ID8 cells exhibited enhanced viability inhibition when cocultured with PBMCs, with greater anticancer effects observed as PBMC numbers increased. However, under identical treatment conditions, AZD4635 demonstrated a weaker inhibitory effect on ID8 cell survival compared to AG14361. To explore the potential for synergistic anti-tumor effects between PARPi and A2AR antagonists, we utilized a clonogenic assay to assess the efficacy of AG14361 and AZD4635, both individually and in combination, in vitro. Our results revealed that the combination of AG14361 and AZD4635 significantly diminished the clonogenic potential of ID8 cells compared to either agent alone (Figure 1e–f). Furthermore, in a co-culture system comprising PBMCs and ID8 cells at a 1:1 ratio, both monotherapy and combination therapy exhibited enhanced tumor growth inhibition relative to controls lacking PBMCs. Based on these observations, we elected to employ a 1:1 co-culture system of PBMCs and ID8 cells in our subsequent experiments. Notably, while AZD4635 exhibited a relatively modest impact on clonogenic capacity compared to AG14361, its combination with AG14361 markedly potentiated the anti-proliferative efficacy of AG14361 against ID8 cells. This enhancement was particularly pronounced in the co-culture system incorporating PBMCs, suggesting that the addition of AZD4635 synergistically augments the in vitro anti-tumor effects of AG14361 in this context.

Figure 1.

A2ARa AZD4635 enhanced the efficacy of PARPi AG14361 in vitro and in vivo. (a,b) Chemical structure of AG14361 (a) and AZD4635 (b). (c,d) CCK-8 assay showing cytotoxicity of AG14361 (c) and AZD4635 (d) on ID8 cells co-cultured with PBMCs at indicated ratios. (e,f) Representative images (e) and quantification (f) of colony formation in ID8-PBMC co-culture systems following indicated treatments. (g) Schematic of ID8 syngeneic tumor model and treatment regimen. (h–j) Tumor images (h), growth curves (i), and endpoint tumor weights (j) for mice treated with vehicle, AG14361 (10 mg/kg), AZD4635 (25 mg/kg), or combination (n = 6 per group). (k) Mice weight growth curves after indicated treatments at different time points. Error bars are shown as mean ± SD from three independent repeats. * p < 0.05, ** p < 0.01; ns, no significance.

3.2. In Vivo Synergistic Antitumor Efficacy of Parpi AG14361 Combined with A2AR Antagonist AZD4635 in Murine Ovarian Cancer Models

To elucidate whether AZD4635 potentiates the antitumor efficacy of AG14361 in ovarian cancer in vivo, we established a syngeneic mouse model utilizing C57BL/6 mice. ID8 cells were subcutaneously inoculated into these mice, which were subsequently treated with either saline or the respective drugs every 2 days, as depicted in Figure 1g. Our findings revealed that both AG14361 and AZD4635 monotherapies significantly curtailed the growth of the ID8 tumor model (Figure 1h–i), echoing the outcomes of our in vitro studies. However, the tumor growth inhibition achieved by these single-agent therapies was modest. Strikingly, the combination of the PARP inhibitor AG14361 and the A2AR antagonist AZD4635 induced a pronounced reduction in tumor volume compared to the monotherapy cohorts, thereby highlighting the superior antitumor activity of the combination regimen over either AG14361 or AZD4635 administered alone. This observation was further corroborated by the analysis of tumor weights across the different groups (Figure 1j). Throughout the experimental duration, we meticulously tracked the body weights of the mice and observed no significant disparities among the groups (Figure 1k), suggesting that the overall health of the mice remained unaffected. Moreover, no overt adverse effects, such as vomiting or diarrhea, were noted in the mice during the entire study period. This absence of noticeable side effects implies that the dosages and administration methods employed for AG14361 and AZD4635 were well-tolerated by the mice. In conclusion, our study demonstrates that in the ID8 ovarian cancer model of C57BL/6 mice, AZD4635 significantly augments the antitumor growth effect of AG14361. The combination treatment regimen exhibits robust antitumor activity and is well-tolerated in vivo, without eliciting any discernible adverse reactions.

3.3. RNA-Seq-Based Transcriptomics Unveil Mechanisms of AG14361 and AZD4635 Combination Therapy

To elucidate the molecular underpinnings of the superior antitumor activity observed with the combination of AG14361 and AZD4635 compared to monotherapy, we conducted a comprehensive transcriptional analysis using RNA-seq on ID8 cells treated with the drug combination or left untreated (Figure 2a). Prior to investigating the molecular mechanisms, we assessed the reliability of our experimental design and the reproducibility among samples through correlation and principal component analysis (PCA) of the expression levels of the loaded samples. Correlation analysis reveals that the repeated samples in the control and treatment groups exhibit high similarity and strong correlation in their expression patterns (Figure 2b). In line with this, PCA results demonstrate that samples within each group are highly similar to one another, thereby corroborating the robustness of our experimental design (Figure 2c).

Figure 2.

Transcriptomic analysis elucidates the molecular mechanisms of AG14361 and AZD4635 combination therapy in ovarian cancer treatment. (a) A flowchart of RNA-seq technology. (b) Correlation heatmap representing the correlation of gene expression levels among samples using Pearson correlation coefficients. (c) A 3D principal component analysis plot, with the x-axis representing the first principal component, the y-axis representing the second principal component, and the z-axis representing the third principal component. Different colors indicate distinct groups. (d) A volcano plot was generated based on the results of differential expression analysis. (e) Heatmap clustering analysis. The x-axis represents genes, with each column corresponding to one sample. Red indicates high expression, while blue indicates low expression. Samples with similar expression patterns are grouped into clusters, distinguished by different colors. The right side of the heatmap displays the expression patterns of genes in each cluster across samples, with the red line indicating the average expression level of genes within each cluster. (f,g) KEGG enrichment analysis factor plot. The x-axis shows the enrichment factor (ratio of DEGs in a pathway to the total genes in that pathway), and the y-axis lists the KEGG pathways. Point size indicates the number of DEGs (upregulated or downregulated) enriched in each pathway, with color intensity reflecting significance levels. (f) displays the factor plot for upregulated DEGs, and (g) for downregulated DEGs. (h,i) GO-GSEA (h) and KEGG-GSEA (i) analyses were performed to explore the pathways potentially modulated by the combination of AG14361 and AZD4635.

RNA-seq analysis identified 898 differentially expressed genes (DEGs) exhibiting p-values below 0.05 and absolute log₂ fold changes of ≥1. Specifically, compared with the control group, the combination-treated ID8 cells exhibited 388 upregulated DEGs and 510 downregulated DEGs (Figure 2d and Supplementary file S1). Hierarchical clustering analysis was then performed to assess the expression patterns of these DEGs across different treatment groups, clustering genes and samples based on the correlation of their expression patterns. The clustering results were further categorized into nine distinct clusters according to the expression trends of the DEGs across various samples, with each cluster representing a unique expression trend (Figure 2e). Notably, DEGs in clusters 1–5 were upregulated in the combination-treated group relative to the control group, while those in clusters 6–9 were downregulated.

To further elucidate the biological significance of these DEGs, we performed KEGG pathway enrichment analysis on the RNA-seq data of the combination-treated ID8 cells. The analysis revealed that the 388 upregulated genes were significantly involved in several key signaling pathways, including the ErbB signaling pathway, HIF-1 signaling pathway, NF-kappa B signaling pathway, JAK-STAT signaling pathway, MAPK signaling pathway, IL-17 signaling pathway, and p53 signaling pathway (Figure 2f and Supplementary file S2). In contrast, the 510 downregulated genes were prominently associated with the Hippo signaling pathway, MAPK signaling pathway, and Wnt signaling pathway (Figure 2g). Additionally, we conducted Gene Set Enrichment Analysis (GSEA) to gain deeper insights into the functional impact of these DEGs. The Gene Ontology (GO)-GSEA results indicated that in ID8 cells treated with the combination of AG14361 and AZD4635, the target gene set related to “regulation of ATP-dependent activity (GO:0043462)” was significantly downregulated, while several immune-related pathways, such as “positive regulation of immune response (GO:0050778),” “positive regulation of immune system process (GO:0002684),” “regulation of interleukin-17 production (GO:0032660),” and “regulation of lymphocyte mediated immunity (GO:0002706),” were significantly upregulated (Figure 2h). Similarly, the Kyoto Encyclopedia of Genes and Genomes (KEGG)-GSEA showed that the target gene sets of the “B cell receptor signaling pathway (mmu04662)” and “cAMP signaling pathway (mmu04024)” were significantly downregulated, while immune-related pathways, such as the “HIF-1 signaling pathway (mmu04066),” “JAK-STAT signaling pathway (mmu04630),” and “Th17 cell differentiation (mmu04659),” were significantly upregulated (Figure 2i).

Collectively, these findings highlight the potentially critical role of these signaling and immune-related pathways in the enhanced antitumor efficacy of AG14361 when combined with AZD4635 in ovarian cancer. To a certain extent, these results elucidate the mechanism by which AZD4635 enhances the therapeutic effects of AG14361 in ovarian cancer by inhibiting adenosine generation and modulating key signaling and immune-related pathways.

3.4. AZD4635-Mediated Adenosine Antagonism Enhances AG14361 Efficacy in Ovarian Cancer via cAMP/Creb Pathway Regulation

Given that AZD4635 functions as an A2AR antagonist, we posited that its mechanism of action involves the inhibition of adenosine-mediated signaling pathways. To explore this hypothesis, we subjected the DEGs previously identified to further scrutiny, ultimately pinpointing 23 DEGs intricately linked to adenosine regulation, with their expression profiles depicted in Figure 3a. Upon conducting a combined GO-KEGG analysis on these genes, we unveiled that they are pivotal in the metabolic processes of ATP, ADP, and AMP, as well as in cAMP metabolism and binding, and predominantly participate in the cAMP signaling pathway (Figure 3b). This finding resonates strongly with our prior GO-GSEA and KEGG-GSEA analyses, exemplified by the mechanisms of “regulation of ATP-dependent activity (GO:0043462)” and the “cAMP signaling pathway (mmu04024)”. Moreover, through protein–protein interaction (PPI) analysis of these genes, we singled out Pde4a, Adcy7, and Rapgef3 as the core players in the cAMP signaling pathway (Figure 3c). Existing literature reports that the binding of adenosine to A2AR can increase cAMP concentration and activate a series of downstream signaling pathways, thereby exerting immune suppression and promoting tumor growth [29]. As an important intracellular second messenger, cAMP activates cAMP-dependent protein kinase A, which subsequently phosphorylates cAMP response element-binding protein (Creb) to regulate the expression of target genes, thus triggering a series of cellular events.

Figure 3.

AZD4635 enhances the anti-ovarian cancer efficacy of AG14361 by inhibiting the cAMP/Creb pathway. (a) Heatmap of adenosine-related DEGs in the transcriptome. (b) Factor plot from integrated GO and KEGG analysis of adenosine-related DEGs in the transcriptome. (c) Interaction network analysis of adenosine-related DEGs in the transcriptome. (d) The levels of cAMP in samples from in vitro and in vivo experiments, following various treatments, were measured using ELISA. (e,f) Western blot analysis and quantification of p-Creb and Creb levels in ID8 cells (e) and mouse tumor tissues (f) following exposure to vehicle, AG14361, AZD4635, or their combination. (g) The efficiency of siRNA knockdown of Cd39 or Cd73 genes was evaluated by qRT-PCR. (h,i) The impact of Cd39 and Cd73 genes was assessed by measuring cAMP levels in differently treated cell samples using ELISA. (j,k) Western blot analysis and quantification of p-Creb and Creb expression levels in ID8 cells following Cd39 (j) and Cd73 (k) knockdown, and/or AG14361 treatment. Error bars are shown as mean ± SD from three or six independent repeats. * p < 0.05, ** p < 0.01; ## p < 0.01; ns, no significance.

Building on these insights, we formulated the hypothesis that the enhancement of AG14361’s therapeutic efficacy in ovarian cancer by AZD4635 may be attributable to the regulation of the cAMP/Creb pathway. To validate this hypothesis, we embarked on measuring the cAMP levels across various treatment groups, both in vivo and in vitro, employing ELISA for quantification. The results revealed that AG14361 treatment elicited a significant upsurge in cAMP levels in ovarian cancer compared to the control group; however, this increase was effectively counteracted by the co-administration of the A2AR antagonist AZD4635, which curbed cAMP accumulation in ID8 cells and mouse tumor tissues (Figure 3d). Subsequently, we harnessed Western blotting to gauge the expression levels of Creb and its phosphorylated counterpart, p-Creb. The data indicated that AG14361 treatment engendered a marked escalation in p-Creb/Creb expression levels in ovarian cancer relative to the control group; yet, this elevation was thwarted by the co-administration of AZD4635, which repressed the p-Creb/Creb expression levels in ID8 cells and mouse tumor tissues (Figure 3e,f).

Given the essential roles of CD39 and CD73 in this pathway, we propose that modulating the conversion of ATP to adenosine is crucial for elucidating how AZD4635 enhances the therapeutic efficacy of AG14361 in ovarian cancer by inhibiting the cAMP/Creb pathway. To delineate the roles of Cd39 and Cd73 in adenosine generation and AG14361-mediated ovarian cancer therapy, we engineered three siRNAs targeting Cd39 and Cd73, respectively, to transiently suppress their expression in ID8 cells. After validating the silencing efficacy of the siRNAs through qRT-PCR (Figure 3g), we employed ELISA to quantify cAMP levels in ID8 cells, thereby evaluating the impact of Cd39 and Cd73 knockdown on adenosine generation. The results revealed that, compared to controls, cAMP levels were significantly diminished in cells subjected to Cd39 or Cd73 knockdown, whereas AG14361 treatment elicited a substantial increase in cAMP (Figure 3h,i). Notably, in Cd39 or Cd73 knockdown cells treated with AG14361, cAMP levels remained significantly elevated relative to their respective knockdown groups. Concurrently, under identical treatment conditions, we assessed the expression of the A2AR downstream effectors Creb and p-Creb using Western blotting. Relative to controls, Cd39 or Cd73 knockdown markedly reduced p-Creb/Creb expression, while AG14361 treatment robustly enhanced p-Creb/Creb levels. Consistent with the cAMP findings, the AG14361-induced upregulation of p-Creb/Creb was still pronounced in Cd39 or Cd73 knockdown cells compared to their knockdown counterparts (Figure 3j,k). These observations underscore that, within the CD39/CD73/A2AR axis, A2AR likely serves as the pivotal regulatory node of the cAMP signaling pathway. This insight indirectly substantiates the rationale and superiority of employing A2AR antagonists to augment PARPi efficacy in our study. Our data demonstrate that inhibiting A2AR function can mitigate the AG14361-induced surge in cAMP levels mediated by adenosine. Collectively, these results suggest that the accumulation of adenosine and the engagement of the CD39/CD73/A2AR axis may curtail the therapeutic potential of AG14361 in ovarian cancer. Importantly, this limitation can be surmounted through the co-administration of AZD4635.

3.5. CD39 and CD73 Expression Levels Correlate with Ovarian Cancer Prognosis and Immune Microenvironment

In the preceding results, we uncovered a significant correlation between the expression levels of CD39 and CD73 and the limited efficacy of AG14361 in treating ovarian cancer. This cascade ultimately impacts the regulation of the immune microenvironment and tumor growth. Therefore, elucidating the expression patterns and functions of CD39 and CD73 in clinical ovarian cancer patients is critical to understanding whether the A2AR antagonist AZD4635 can effectively address the limitations of AG14361 treatment. To this end, we harnessed the gene expression dataset of TCGA-OV from the TCGA database to perform co-expression heatmap analyses of CD39 (gene name ENTPD1) and CD73 (gene name NT5E) with the adenosine-related DEGs previously identified. Our analyses revealed that CD39 and CD73 exhibit similar expression patterns with several genes, including RAPGEF3, ADCY7, PDE4A, AK5, AK4, AMPD1, and CREB3L1 (Figure 4a). Notably, ovarian cancer patients with high expression of CD39 and CD73 also displayed elevated expression levels of these adenosine-related DEGs. These findings suggest that CD39 and CD73 may influence the immune microenvironment and tumor growth, and thus the efficacy of AG14361 treatment, by affecting adenosine-related genes or pathways.

Figure 4.

In TCGA-OV patients, survival duration, clinicopathological characteristics, and immune cell composition are associated with the expression of CD39 and/or CD73. (a) In the TCGA-OV cohort, a heatmap analysis was conducted to examine the co-expression of CD39 and CD73 with adenosine-related genes. The upper section presents primary variable values in bar format, while the lower section uses a heatmap to show other variable values and their trends relative to the primary variables. The analysis employed Spearman correlation. (b) Kaplan–Meier analysis was performed to examine the association between CD39 and/or CD73 expression and overall survival. (c) Correlation analysis of CD39 and CD73 mRNA levels with clinical stage, tumor subtype, age, residual tumor status, and primary therapy outcome in TCGA-OV patients. (d) Analysis of immune cell composition in TCGA-OV patients with high or low expression of CD39 and/or CD73 mRNA.

To further explore the relationship between CD39 and CD73 expression levels and ovarian cancer prognosis, we conducted Kaplan–Meier survival analyses on a cohort of 66 ovarian cancer patients from the TCGA-OV dataset (Supplementary file S3). The results demonstrated significant differences in overall survival between the high CD39 expression group (median survival time = 951 days) and the low CD39 expression group (median survival time = 1511 days), as well as between the high CD73 expression group (median survival time = 992 days) and the low CD73 expression group (median survival time = 1458 days) (Log-rank p values were <0.001 and 0.02, respectively) (Figure 4b). These data indicate that high expression of CD39 or CD73 is closely associated with poor prognosis in ovarian cancer patients. We further stratified the 66 ovarian cancer patients into four groups based on CD39 and CD73 expression levels: the double-low expression group, the double-high expression group, the group with low CD39 and high CD73 expression, and the group with high CD39 and low CD73 expression. Kaplan–Meier survival analyses revealed distinct survival curves among these groups. Specifically, the double-high expression group exhibited the shortest median survival time (822 days) and the worst prognosis, whereas the double-low expression group had the longest median survival time (1511 days) and the best prognosis. Moreover, patients with at least one low expression of CD39 or CD73 had significantly better survival than those with double-high expression (p < 0.05). These findings further underscore the critical roles of CD39 and CD73 in ovarian cancer progression and highlight the strong association between double-high expression of CD39/CD73 and poor prognosis in ovarian cancer patients.

Additionally, we analyzed the clinical data of ovarian cancer patients with high and low expression of CD39 or CD73 (Figure 4c). The results showed that patients with low expression of CD39 or CD73 were more likely to be in the early or intermediate stages of disease, whereas those with high expression were predominantly in advanced and more malignant stages. Regarding tumor location, bilateral involvement was common among ovarian cancer patients and was not significantly correlated with CD39/CD73 expression. However, among patients with unilateral involvement (left or right), high CD39 expression or low CD73 expression was more frequently observed on the left side. In terms of age, elderly ovarian cancer patients typically exhibited high expression of CD39 or CD73. Regarding treatment response, most patients with low expression of CD39 or CD73 achieved complete remission after treatment, whereas those with high expression often experienced partial remission or disease progression.

To further investigate the relationship between CD39 and CD73 expression levels and immune cell composition in ovarian cancer patients, we collected immune cell composition data for the 66 ovarian cancer patients from The Cancer Immunome Atlas database and performed group analyses based on CD39 and CD73 expression levels. The results showed that CD39 and CD73 expression significantly impacted the immune cell composition in ovarian cancer patients (Figure 4d). Specifically, compared with patients with low CD39 expression, those with high CD39 expression exhibited a significant increase in the proportion of macrophages (including both M1 and M2 types) and regulatory T cells, while the proportions of monocytes and CD4-positive T cells were significantly reduced. Similarly, compared with patients with low CD73 expression, those with high CD73 expression had a significant increase in the proportion of macrophages (particularly M2 type), monocytes, and B cells, whereas the proportions of macrophages (M1 type), CD4-positive T cells, and regulatory T cells were significantly decreased.

We further analyzed the immune cell composition of the four patient groups: the double-low expression group, the double-high expression group, the group with low CD39 and high CD73 expression, and the group with high CD39 and low CD73 expression. The results showed that, compared with the double-low expression group, the double-high expression group had a significant increase in the proportion of macrophages (M1 and M2 types), dendritic cells, and regulatory T cells, while the proportions of B cells, CD4-positive T cells, and monocytes were significantly reduced. These findings demonstrate that the immune cell composition in ovarian cancer is closely related to CD39 and CD73 expression levels. The expression of CD39 and CD73 plays a critical role in ovarian cancer progression. Their high expression may promote adenosine production and downstream pathway activation, thereby influencing the tumor immune microenvironment, driving tumor malignancy, and ultimately affecting treatment efficacy.

3.6. AZD4635 Augments AG14361 Antitumor Efficacy by Modulating Immune Cell Infiltration in the ID8 Syngeneic Mouse Model

Adenosine, a ubiquitous metabolite endowed with potent immunosuppressive properties, exerts a spectrum of effects within the TME: it curtails the proliferation of antigen-presenting cells, dampens the activation of effector T cells (Teff), catalyzes the activation of regulatory T cells (Tregs), redirects macrophage polarization from a pro-inflammatory to an anti-inflammatory and pro-angiogenic phenotype, and stifles the activity of natural killer (NK) cells. Tumor-infiltrating lymphocytes (TILs), as integral constituents of the TME, serve as proxies for gauging tumor aggressiveness and responsiveness to anticancer therapies. To decipher whether AZD4635 potentiates the antitumor effects of AG14361 by sculpting the immune cell landscape within ovarian cancer, we deployed a panel of specific antibodies against cardinal immune cell markers—Cd4 (marker for helper T cells), Cd8 (marker for cytotoxic T cells), Foxp3 (specific marker for regulatory T cells—Tregs), Cd206 (marker associated with M2 macrophages), and Cd11b (a general marker for myeloid-derived suppressor cells)—and assessed their expression in tumor tissues from an ID8 mouse model of ovarian cancer. We also monitored Granzyme B (Gzmb), a functional hallmark of cytotoxic T lymphocytes (CTLs) and NK cells (Figure 5a).

Figure 5.

The combination of AG14361 and AZD4635 enhances the infiltration and activation of immune cells within the tumor microenvironment. (a) Tumor tissues isolated from C57BL/6 mice were subjected to immunohistochemical analysis for the expression of tumor immune-infiltrating cell markers CD4, granzyme B, CD8, CD206, Foxp3, and CD11b. Scale bar = 10 μm. (b–g) Bar chart of statistical analysis of IHC results for tumor immune-infiltrating cell markers (CD4 (b), granzyme B (c), CD8 (d), Foxp3 (e), CD206 (f), and CD11b (g)) based on a scoring system. Error bars are shown as mean ± SD from six independent repeats. * p < 0.05, ** p < 0.01; ns, no significance.

In the ID8 syngeneic mouse model, AG14361 monotherapy left the numbers of CD4⁺ T cells and Granzyme B⁺ cells in tumor tissues largely unaltered compared to controls (Figure 5b,c). However, the addition of AZD4635 to AG14361 elicited a marked increase in both CD4⁺ T cells and Granzyme B⁺ cells. For CD8⁺ T cells, AG14361 treatment alone engendered a significant rise in their numbers within tumor tissues, a trend further amplified by the combination with AZD4635 (Figure 5d). Conversely, AG14361 monotherapy precipitated a significant influx of Tregs and M2 macrophages, an effect that was abrogated by AZD4635 co-treatment, which precipitated a substantial decline in the infiltration of these immunosuppressive cells (Figure 5e,f). Myeloid-derived suppressor cells (MDSCs) remained unperturbed by AG14361 alone but were significantly attenuated in the presence of AZD4635 (Figure 5g). Collectively, these results underscore that AZD4635 augments the antitumor efficacy of AG14361 by recalibrating the balance of immune cell infiltration within the ovarian cancer microenvironment.

3.7. Single-Cell Transcriptomics Unveiled the AG14361–AZD4635-Driven Rewiring of the Ovarian Cancer Immune Microenvironment

To elucidate the intricate immune cell composition in ovarian cancer tissues following treatment with AG14361 and AZD4635, we employed single-cell RNA sequencing on tumor samples from the ID8 mouse ovarian cancer model. Our analysis encompassed six samples—three controls and three treated—with the experimental design depicted in Figure 6a. By isolating CD45⁺ immune cells, we captured a total of 19,153 cells, which were subsequently subjected to non-linear dimensionality reduction using Uniform Manifold Approximation and Projection (UMAP). Through meticulous examination of clustering results and canonical marker gene expression, we identified seven major immune cell types: T cells (Cd3d, Cd3e, Cd3g), B cells (Cd19, Cd79a, Cd79b, Ms4a1), granulocytes (S100a8), dendritic cells (Bst2, Irf8, Siglech), macrophages (Cd163, Cd68, C1qa, Mrc1), NK cells (Gzma, Nkg7, Klrd1), and plasma cells (Jchain, Iglv1, Ighg1) (Figure 6b,c). Bubble plots visualized the distribution of these markers across cell subpopulations (Figure 6d), while heatmap analysis highlighted the top 5 upregulated genes in each cell type, such as Themis, Bcl11b, Icos, Trac, and Cd3e in T cells; Ms4a1, Fcmr, Bank1, Pax5, and Fcer2a in B cells; C3ar1, Pid1, Clec4a1, Fcgr1, and Arg1 in macrophages; and Ncr1, Gzma, Xcl1, Spry2, and Klre1 in NK cells (Figure 6e). Subsequently, we conducted a combined GO and KEGG analysis on the differentially upregulated genes among these subclusters, which provided compelling evidence for further elucidating the biological functions and mechanisms of these distinct cell subpopulations (Figure 6f). Comparative analysis of immune cell populations between control and treated groups revealed significant shifts in cell cluster proportions. Notably, combined treatment altered the abundance of B cells, T cells, macrophages, and NK cells (Figure 6g). Quantitative assessment of cell subpopulation abundance, via the ratio of observed to expected cell numbers (Ro/e), indicated increased enrichment of T cells, macrophages, NK cells, dendritic cells, and granulocytes following drug treatment, while B cell enrichment decreased (Figure 6h).

Figure 6.

Single-cell transcriptomics analysis identified diverse immune cell types in ovarian cancer tumor tissues from both control and drug combination-treated mice. (a) Schematic diagram of single-cell suspension acquisition and sequencing data analysis. (b) A UMAP plot illustrating the 8 clusters identified via integrated analysis, each comprising tumor samples from 3 control mice and 3 mice treated with the drug combination. Each point represents a single cell, colored according to cluster classification. (c) UMAP plot showing expression levels of selected genes. (d) Bubble plot showing marker genes across 8 cell clusters. Dot size indicates fraction of expressing cells, colored according to z-score scaled to expression levels. (e) Heatmap showing the expression signature of top 5 marker genes for 8 cell clusters. In the figure, each column corresponds to a cell, and each row to a gene. Gene expression levels across different cells are color-coded, with yellow representing higher expression and purple indicating lower expression. (f) The bubble plot illustrates the results of GO and KEGG enrichment analysis for differentially upregulated genes in the 8 cell clusters. (g) The bar chart shows the compositional proportions of the combined immune cell populations in each sample. Each stacked bar represents a sample, with its total cell count normalized to 1. (h) Ro/e heatmap: The abundance of cell subclusters in each group is quantified and assessed by calculating the ratio of observed to expected cell counts. (i) The violin plot illustrates the gene signature scores for the CD39/CD73/A2AR pathway and the cAMP signaling pathway in selected immune cell clusters. The y-axis indicates gene score enrichment values, and the x-axis denotes different subclusters. (j) The box plot displays the differential gene signature scores for the CD39/CD73/A2AR pathway and the cAMP signaling pathway across immune cell clusters in various groups. (k) The box plot illustrates the differences in gene signature scores associated with inflammation, stemness, metastasis, apoptosis, angiogenesis, and invasion among immune cells in various groups. ** p < 0.01, *** p < 0.001; ns, no significance.

Given the crucial roles of adenosine metabolism and the CD39/CD73/A2AR and cAMP signaling pathways in the therapeutic efficacy of AG14361 and AZD4635, we constructed gene sets targeting these pathways. The CD39/CD73/A2AR gene set not only encompasses the “adenosine gene signature” (AdenoSig), including Cxcl1, Cxcl2, Cxcl3, Cxcl5, Cxcl6, Cxcl8, Ptgs2, and Il-1β, but also includes “adenosine signaling score” markers, such as Pparg, Cybb, Col3a1, Foxp3, Lag3, App, Cd81, Gpi, Ptgs2, Casp1, Fos, Mapk1, Mapk3, and Creb1 [21]. Additionally, we further incorporated genes closely related to adenosine generation, such as Lyve1, Pdpn, Vegfc, AdoA2AR, Nt5e, and Entpd1. The cAMP signaling pathway gene set was derived from KEGG database annotations. The analysis of specific gene set enrichment unveils distinct patterns of enrichment across various immune cell types (Figure 6i). Notably, granulocytes and macrophages exhibit significantly higher scores within the CD39/CD73/A2AR pathway gene set compared to other cell types, suggesting a potential role in immune regulation mediated by this pathway. Similarly, NK cells and macrophages display relatively higher expression in the cAMP signaling pathway gene set, indicative of heightened activity in cAMP signal transduction. In contrast, T cells and B cells maintain relatively lower expression levels in these gene set scores. We further compared gene set enrichment scores between control and drug combination-treated groups (Figure 6j). Within the CD39/CD73/A2AR pathway gene set, B cells, T cells, and macrophages from the drug combination group exhibited reduced scores compared to the control group. In contrast, no significant differences were observed among groups for NK cells, dendritic cells, and granulocytes. For the cAMP signaling pathway gene set, most cell types, except macrophages and granulocytes, showed lower scores in the drug combination group.

Based on these findings, we propose that the combination of AG14361 and AZD4635 in treating ovarian cancer may inhibit adenosine production, thereby blocking the downstream CD39/CD73/A2AR and cAMP signaling pathways. This could effectively reverse the immune suppressive state within the tumor microenvironment, exemplified by increased T cell infiltration, thereby enhancing therapeutic efficacy. Conversely, AG14361 treatment alone may activate the cAMP signaling pathway in macrophages, inducing immune suppressive effects that persist even with AZD4635 co-treatment, thus limiting AG14361’s anti-tumor efficacy. Thus, macrophages may be a key factor in addressing AG14361’s efficacy challenges. Furthermore, we conducted gene set scoring analysis for key biological processes, including inflammation, stemness, metastasis, apoptosis, anti-angiogenesis, and invasion. These gene sets were consistently downregulated in the treated group (Figure 6k). This comprehensive analysis not only enriches our understanding of the mechanisms underlying drug combination therapy but also suggests that the combination of AG14361 and AZD4635 may broadly inhibit tumor progression by targeting multiple biological processes, providing valuable insights into the therapeutic potential of this drug regimen.

3.8. Heterogeneity of Macrophage Subpopulations in Ovarian Cancer and Their Response to AG14361 and AZD4635 Combination Therapy

As previously suggested, macrophages may be a crucial factor in addressing the efficacy challenges associated with AG14361. Considering the significant influence of macrophages on the efficacy of combination therapies, it is imperative to delve deeper into their roles. In this study, we employed scRNA-seq to re-cluster 2347 macrophages, thereby identifying four distinct subpopulations: M0, M1, M2, and M3. Each subpopulation exhibited a unique gene expression profile, with the M0 cluster characterized by high expression of Plpp3, Lhfpl2, and Ctsk; the M1 cluster by Clec10a, C1qa, Fcgrt, and Selenop; the M2 cluster by Isg15, Ccl5, and Ly6c1; and the M3 cluster by Ifitm1, Vcan, and S100a4 (Figure 7a and Figure A1a). Based on the highly upregulated marker genes depicted in the heatmap, we designated these clusters as Lhfpl2-, C1qa-, Ccl5-, and Vcan-macrophages, respectively. Violin plots further revealed that the expression specificity of each marker gene was largely confined to a unique subpopulation (Figure 7b).

Figure 7.

Characterization and identification of macrophage features in control and drug combination-treated groups. (a) The UMAP plot illustrates the integrated macrophage subclusters derived from six mouse tumor samples, comprising three from the control group and three from the drug combination-treated group. (b) The violin plot illustrates the expression levels of Lhfpl2, C1qa, Ccl5, and Vcan genes within macrophage subclusters. (c) The bubble heatmap illustrates the expression levels of signature genes in four macrophage subclusters, indicated by color, and the proportion of expressing cells, indicated by bubble size. (d) The bubble plot illustrates the results of GO and KEGG enrichment analysis for differentially upregulated genes in the four macrophage subclusters. (e) The numerical heatmap illustrates the distribution of signature scores for each type across different macrophage subclusters. These feature scores are calculated based on gene sets from published studies. (f,g) The box plot highlights the differential feature scores of various gene sets among different groups of macrophage subclusters. (h) The scatter plot displays the differentially expressed genes in macrophage subpopulations in the drug combination treatment group compared to the control group. (i) The bubble plot highlights the expression specificity of selected differentially expressed genes within each macrophage subcluster across various groups. (j,k) Pseudotime analysis of macrophages inferred using Monocle2 (j), with each point representing a single cell and clustering information displayed (k). (l) Pseudotime values (top panel) and differentiation states (bottom panel) from Monocle2 analysis were mapped onto the original UMAP plot. Dots indicate cells, with darker colors representing higher differentiation levels and distinct colors corresponding to different states. (m) Heatmap showing regulation heterogeneity of TF genes among macrophages in the control group and drug-treated group by SCENIC. (n) The heatmap shows regulon activity and distribution across different groups, with differences indicated by color. (o) The heatmap shows regulon activity and distribution across subclusters, with differences indicated by color. * p < 0.05, ** p < 0.01, *** p < 0.001; ns, no significance.

To further characterize these four macrophage clusters, we conducted differential expression analysis of upregulated genes in these cell populations (Figure 7c and Supplementary file S4). The M0_Lhfpl2 cluster was found to specifically overexpress the Lhfpl2 gene, a transmembrane protein associated with reproduction whose relationship with tumor-associated macrophages has been scarcely explored. This cluster also exhibited high expression of Arg2, a key regulator of macrophage anti-inflammatory responses. Additionally, the M0_Lhfpl2 cluster also exhibited high expression of genes associated with angiogenesis (Vegfa and Cd44) and phagosome proteolytic cathepsins (Ctsa, Ctsz, Ctsd, Ctsl, and Ctsk), suggesting that this subpopulation may play a key role in promoting tumor angiogenesis and immune suppression. Macrophages in the M1_C1qa cluster highly expressed complement genes C1qa/C1qb/C1qc, which are initiators of the classical complement pathway and are implicated in various immune functions, including pathogen recognition, immune complex clearance, and phagocytosis of apoptotic cell debris. Concurrently, this cell subpopulation highly expressed genes related to phagocytic function (Mrc1), lipid metabolism (Apoe), and local antioxidant function (Selenop), underscoring the importance of the M1_C1qa cell subpopulation in phagocytosis and metabolic regulation. Macrophages in the M2_Ccl5 cluster exhibited high expression of chemokines (Ccl2/3/4/5) and inflammatory-related factors such as tumor necrosis factor, which are instrumental in the recruitment, activation, and polarization of macrophages. Additionally, this subpopulation highly expressed the gene Isg15, which is associated with the induction of M2-like macrophages. The M3_Vcan cluster was characterized by high expression of Vcan and S100a transcripts (S100a4/6/10), previously associated with monocytes. Furthermore, this cell subpopulation highly expressed M1-like markers, antigen presentation-related MHC II molecules (H2-Aa, H2-Ab1, H2-Ea, H2-Eb1, and H2-DMa), and Cd74, which are crucial for antigen processing and presentation to CD4⁺ T cells. Subsequently, we performed an integrated GO and KEGG analysis on the differentially upregulated genes among these subclusters, which furnished robust and persuasive evidence for delving deeper into the biological functions and underlying mechanisms of these distinct cell subpopulations (Figure 7d).

To further dissect the functional roles of each macrophage subpopulation in ovarian cancer development and drug combination, we analyzed the scores of each cell based on markers for M1, M2, angiogenesis, and phagocytosis to explore the functional phenotypes of each cell subpopulation [30] (Supplementary file S5). The results revealed that the M0_Lhfpl2 cluster exhibited higher M2 markers, and the M1_C1qa and M2_Ccl5 clusters also showed higher M2 enrichment. However, intriguingly, the M2_Ccl5 cluster, in addition to high M2 enrichment, also exhibited higher expression of classical M1 markers, indicating that the M2_Ccl5 cluster co-expressed M1 and M2 gene markers (Figure 7e). To enrich this analysis, we also referred to the methods of Azizi et al. [31] and conducted a polarization scoring assessment of M1 and M2 for these macrophage clusters. The results showed that the M3_Vcan cluster exhibited a more pronounced trend towards M1 polarization, while the M1_C1qa and M2_Ccl5 clusters also demonstrated varying degrees of M1 polarization. In terms of M2 polarization trends, the results were consistent with the previous analysis: the M0_Lhfpl2 cluster exhibited higher expression of M2 marker genes. Macrophage functional phenotypes have been reported to exist in an in vitro M1/M2 dual-polarization state [32]. The practice of dividing populations using only M1/M2 macrophage marker genes is outdated and overly simplistic, which limits our in-depth study of the functions of macrophage clusters. Overall, these results further demonstrate the limitations of this in vitro polarization model. Additionally, the scoring results for angiogenesis and phagocytosis marker gene sets showed that the M1_C1qa exhibited significantly higher “phagocytosis scores,” closely related to phagocytosis, which is crucial for immune responses; in contrast, the M0_Lhfpl2 relatively showed enrichment of angiogenesis-related genes. We applied the aforementioned scoring model to the four macrophage clusters across different treatment groups (Figure 7f). The results revealed that the combination drug treatment significantly diminished the enrichment of M1 macrophage markers within the M1_C1qa cluster while concurrently augmenting the enrichment of M2 macrophage markers in the M0_Lhfpl2 cluster. Within the polarization model, only the M0_Lhfpl2 cluster exhibited a significant elevation in M2 macrophage polarization enrichment relative to the control group. Regarding angiogenesis scores, the treatment groups of the M0_Lhfpl2, M2_Ccl5, and M3_Vcan clusters all demonstrated significant enrichment to varying extents. In contrast, with respect to phagocytosis scores, only the treatment group of the M1_C1qa cluster manifested significant enrichment compared to the control group.

Given the more complex phenotypes of tumor-associated macrophages (TAMs) in the TME [33], to further explore the diverse functions and heterogeneous nature of these macrophage subpopulations, we utilized a consensus model of TAM diversity to select six TAM subpopulations, including interferon-preprocessed TAMs (IFN-TAMs), immune regulatory TAMs (Reg-TAMs), inflammatory cytokine-rich TAMs (Inflam-TAMs), lipid-associated TAMs (LA-TAMs), angiogenic TAMs (Angio-TAMs), and proliferative TAMs (Prolif-TAMs), and performed gene set scoring on our samples using their marker genes (Supplementary file S5). The results unveiled functional heterogeneity among these four macrophage clusters. IFN-TAMs were highly enriched in the M1_C1qa and M2_Ccl5 clusters, characterized by high expression of IFN-regulated genes and M1-like markers; Inflam-TAMs, marked by their expression characteristics of inflammatory cytokines, were highly enriched in the M2_Ccl5 cluster, followed by M0_Lhfpl2 and M3_Vcan clusters; LA-TAMs were not highly enriched in any of the four macrophage clusters, and Prolif-TAMs were also essentially not enriched; Angio-TAMs were highly enriched in the M0_Lhfpl2 cluster, consistent with the previous results, and the M2_Ccl5 cluster also highly enriched this subpopulation; most interestingly, the Reg-TAMs scoring showed that the M1_C1qa cluster was highly enriched in Reg-TAMs among the four macrophage clusters, and this subpopulation, similar to alternatively activated macrophages, may possess immune suppressive functions. Subsequently, we applied the consensus model of TAM diversity to the four macrophage subpopulations in different groups (Figure 7g). The results demonstrated that AG14361 combined with AZD4635 treatment reshaped the functional heterogeneity of macrophage populations in tumor tissues. For instance, the drug treatment group significantly reduced the enrichment of IFN-TAMs in the M0, M2, and M3 macrophage populations and the enrichment of Inflam-TAMs in the M0 and M1 macrophage populations, while significantly increasing the enrichment of Angio-TAMs in the M0 and M2 macrophage populations and the enrichment of Reg-TAMs in the M0, M1, and M2 macrophage populations. These alterations in macrophage subpopulation enrichment may underlie the limited efficacy of AG14361.

Utilizing the Ro/e index as our analytical framework, we uncovered marked alterations in the macrophage immune infiltration subtypes within the ovarian cancer tissues of mice subsequent to AG14361 and AZD4635 combination therapy. Notably, the M1_C1qa cluster was significantly enriched in the drug combination treatment cohort (p < 0.05), whereas the M0_Lhfpl2, M2_Ccl5, and M3_Vcan macrophage subtypes predominantly resided in the control group (Figure A1b,c). While the M1_C1qa macrophages exhibited elevated phagocytosis-related gene scores post-treatment, suggestive of enhanced tumor phagocytic capacity and a potential boon to the antitumor immune response, a concomitant high score for Reg-TAMs complicates this interpretation. This dual profile indicates that M1_C1qa macrophages may simultaneously secrete immunosuppressive cytokines, thereby dampening immune cell activation and function and aiding tumor cells in evading immune detection. Based on these findings, we propose that the M1_C1qa cluster serves as a critical regulatory node in enhancing the efficacy of the AZD4635 and AG14361 combination against ovarian cancer. Thus, a comprehensive characterization of this subgroup is imperative for optimizing the therapeutic outcomes of this drug regimen.

We also conducted intergroup differential gene analysis of these macrophage subpopulations. Identifying subgroup-upregulated genes is conducive to the development of molecular markers for cell subtypes and provides new insights into the core functional genes of special cell subtypes. The results revealed that there were varying numbers of differentially expressed genes in each cell subpopulation. To elucidate the transcriptional profiles of these subclusters, we identified the top 10 significantly upregulated genes within each subpopulation and visualized their expression patterns using heatmaps (Figure A1d and Supplementary file S6). Concurrently, we identified the top 5 most significantly upregulated and downregulated genes, which were visualized in a volcano plot to further accentuate the expression alterations of these key genes (Figure 7h). Compared with the control group samples, we found that macrophage scavenger receptors (Stab1, Msr1, Cd163), chemokine receptors (Ccr2, Ccr5), arginase isoenzymes (Arg1, Arg2), phagosome proteases (Ctsa, Ctsb, Ctsl), and Selenop were upregulated to varying degrees in each cell subpopulation in the drug combination-treated samples. However, chemokines Ccl2, Ccl5, and cytokine Il1b were downregulated in the drug combination-treated samples (Figure 7i). Meanwhile, we also utilized GSEA to study the differentially enriched pathways in macrophage clusters (Figure A1e). The results showed that cluster M0 was enriched in the adipogenesis (NES = 1.39, p < 0.01), coagulation (NES = 1.38, p < 0.05), IL6-JAK-STAT3 signaling pathway (NES = 1.34, p < 0.05), p53 pathway (NES = 1.28, p < 0.05), and mTORC1 signaling pathway (NES = 1.28, p < 0.05); cluster M1 was enriched in the TNFα/NF-κB signaling pathway (NES = −1.26, p < 0.05), and protein secretion (NES = 1.38, p < 0.05); cluster M2 was enriched in the epithelial–mesenchymal transition (NES = 1.41, p < 0.01), apical junction (NES = 1.32, p < 0.05), hypoxia (NES = 1.30, p < 0.05), myogenesis (NES = 1.29, p < 0.05), glycolysis (NES = 1.28, p < 0.05), and mTORC1 signaling pathway (NES = 1.27, p < 0.05); and cluster M3 was enriched in the mitotic spindle (NES = 1.29, p < 0.05).

To further explore the changes in macrophage subtype-specific gene expression in the tumor immune heterogeneity formed after drug combination treatment, we employed Monocle2 technology to reconstruct the pseudotime trajectory inference of all obtained macrophages (Figure 7j,k). Through this technology, the developmental trajectory of macrophages was divided into 13 developmental layers (states 1–13), with the M1_C1qa located at the starting point of the cell evolution in this atlas (Figure A1f), presenting a complex multi-branch structure: M1_C1qa as the root node, and the remaining macrophages clusters distributed at the terminal states of different branches. Notably, the M1_C1qa was largely concentrated at the starting point, with a small portion located at the terminal end of one branch. We also mapped the calculated pseudotime values and differentiation states back to the original UMAP dimensionality reduction map, and the results showed that the developmental trajectory of macrophages exhibited significant differences at the starting state. Specifically, cells with lower pseudotime values or in state 1 were essentially the M1_C1qa cluster, and as the pseudotime value increased, the degree of cell differentiation increased, revealing that M1_C1qa plays a crucial role in the development and progression of ovarian cancer after drug treatment (Figure 7l).

Additionally, using Monocle technology, based on the gene expression signals in all cells and the pseudotime values of each cell, we screened for genes differentially expressed over time to identify key genes related to the developmental differentiation process. The results showed that the gene expression trends varied among each cell cluster (Supplementary file S7). For example, in cluster 1, the expression abundance of genes such as Mrc1, C1qa, C1qb, Ccl8, and Selenop gradually decreased along the pseudotime axis, while in cluster 5, the expression abundance of genes such as Il1b, Cd52, and Ifitm3 gradually increased along the pseudotime axis (Figure A1g). Given the importance of branches 1 and 2 in the pseudotime trajectory, we analyzed these two branches using Monocle to identify differentially expressed genes in their pseudotime differentiation fates (Supplementary file S7). We also selected some genes of interest for heatmap analysis, including chemokines (Ccl2/4/5, Cxcl1/2/11), MHC-II molecules (H2-Aa, H2-Ab1, H2-Ea, H2-Eb1, and H2-DMa), phagosome proteolytic cathepsins (Ctsa, Ctsb, Ctsl), and complement genes (C1qa/C1qb/C1qc) (Figure A1h,i).

To assess the differences in transcription factor (TF) expression levels during macrophage differentiation, we performed single-cell regulatory network inference and clustering (SCENIC) analysis. After completing the SCENIC analysis, we used bar charts to display the number of predicted regulons, TFs, and target genes (Figure A1j). The regulon activity heatmap revealed differences in the activity of the same regulon among different cells, facilitating the identification of cell subpopulations specifically regulated by regulons. The results showed higher regulon enrichment in the M0_Lhfpl2 cluster (Figure A1k). Further analysis revealed that although some TFs were shared between the control and drug combination groups, they also exhibited unique TFs (Figure 7m). For example, Jund, Erg1, Maf, Klf4, Mitf, Zmiz1, and Runx1 were identified in both the control and treatment groups, but the openness in the treatment group was higher (Figure 7n). Meanwhile, the openness of these TFs also varied among different macrophage clusters, with higher openness in the M0_Lhfpl2 and M1_C1qa clusters (Figure 7o). Specifically, Jund exhibited specific functionality in M1_C1qa, while Zmiz1, Mitf, and Runx1 exhibited specific functionality in M0_Lhfpl2, and they may play important roles in the treatment of ovarian cancer with AG14361 combined with AZD4635.

In summary, our study underscores the pivotal role of macrophages in the combined treatment of ovarian cancer with AG14361 and AZD4635. Our observations reveal that macrophage infiltration in drug-treated samples is closely associated with genes and pathways related to scavenger receptors, phagocytosis, angiogenesis, and chemokine signaling. However, the activation of these pathways may have counteracted the enhancement of AG14361’s antitumor immune effect by AZD4635, ultimately influencing the overall outcome of the combination therapy.

3.9. Heterogeneity of T Cell Subpopulations in Ovarian Cancer and Their Response to AG14361 and AZD4635 Combination Therapy

In our study, we delved into the characteristics and distribution of lymphocytes within the samples, in addition to examining the features of macrophages. Through batch-corrected UMAP and clustering analysis of all T cells (n = 6417) from both the control and drug combination treatment groups, we were able to delineate seven distinct T cell subclusters (Figure 8a). Utilizing T cell markers Cd8A, Cd8b1, Cd4, and Cd40lg, we further identified and separated two CD8⁺ T cell subclusters (TC_C1 and TC_C3) and four CD4⁺ T cell subclusters (TC_C0, TC_C2, TC_C4, and TC_C5) (Figure 8b). Moreover, within the T cell population, we uncovered a double-negative T (DNT) cell cluster (TC_C6), which was defined by the absence of CD4 and CD8 expression but notable for the expression of T cell receptor genes Trdv4, Trgv2, and Trgv6 (Figure 8c).

Figure 8.

Characterization and identification of T cell features in control and drug combination-treated groups. (a) The UMAP plot illustrates the integrated T cell subclusters derived from six mouse tumor samples, comprising three from the control group and three from the drug combination-treated group. (b) UMAPs of all T cells, colored by the expression of CD4⁺ and CD8⁺ T cell markers. (c) Dot plot showing the expression of classic marker genes in each subcluster. (d) The dot plot illustrates the normalized average expression levels of selected genes associated with T cell function across different cell subpopulations (left panel: CD4⁺ T cells; right panel: CD8⁺ T cells). (e) Heatmap showing the normalized average expression of selected myeloid cells function-associated genes in each cell subpopulation. (f) The numerical heatmap illustrates the distribution of diverse feature scores across various T cell subclusters, calculated based on gene sets from published studies. (g) The box plot highlights the differential feature scores of various gene sets among different groups of T cell subclusters. (h) UMAP plots illustrate the integrated T cell subpopulations within the control (top panel) and treatment (bottom panel) groups, respectively. (i) The violin plot displays the enrichment levels of exhaustion-related gene sets across different subclusters. (j) The bubble heatmap depicts the expression patterns of immune checkpoint genes among the seven T cell subpopulations across different groups. (k) The scatter plot displays the differentially expressed genes in T cell subpopulations in the drug combination treatment group compared to the control group. (l) Pseudotime analysis of T cells using Monocle2 (left panel), where each point represents a single cell and indicates its state information (right panel). (m) Pseudotime values (left panel) and differentiation states (right panel) from Monocle2 analysis were mapped onto the original UMAP plot. Dots indicate cells, with darker colors representing higher differentiation levels and distinct colors corresponding to different states. (n) Heatmap showing regulation heterogeneity of TF genes among T cells in the control group and drug-treated group by SCENIC. (o) The heatmap shows regulon activity and distribution across subclusters, with differences indicated by color. (p) The heatmap shows regulon activity and distribution across different groups, with differences indicated by color. * p < 0.05, ** p < 0.01, *** p < 0.001; ns, no significance.

To further interrogate the properties of these subclusters, we conducted marker gene heatmap analyses. The results unveiled unique gene expression profiles for each subcluster. For instance, the C0 subcluster was characterized by high expression of Igfbp4, Lef1, S1pr1, and Dusp10; the C1 subcluster by Klrc1, Ly6c2, Ly6c1, and Ccl5; the C2 subcluster by Tnfsf8, Tbc1d4, Smco4, and Eea1; the C3 subcluster by Fgf13, Sidt1, Cd8b1, and Cd8a; the C4 subcluster by Myb, Ift80, Ikzf2, and lzumo1r; the C5 subcluster by Foxp3, Itgae, Tnfrsf9, and Tnfrsf4; and the C6 subcluster by Trdv4, Blk, Abi3bp, and Scart1 (Figure A2a). We delved deeper into the specificities of CD8⁺ and CD4⁺ T cells. As the marker gene heatmap unfolded, it revealed that the marker genes within the CD4⁺ T cell cluster largely echoed the findings from our earlier analyses. Yet, the results pertaining to CD8⁺ T cells turned out to be particularly intriguing: the C1 subcluster was marked by high expression of Gzmb, S100a4, Ccr2, S100a6, and Ifng, whereas the C3 subcluster was distinguished by high expression of Ccr7, Sell, Aff3, Cmah, and Lef1 (Figure A2b,c). By examining the expression of functional markers and marker genes, we successfully identified and characterized T lymphocyte subclusters (Figure 8d). Among CD4⁺ T cells, we distinguished three functional and phenotypic states: TC_C0 and TC_C4 were naive CD4⁺ T cells (Tn; highly expressing Ccr7, Sell, Lef1, Tcf7), but TC_C4 also highly expressed T cell exhaustion markers Ctla4 and Tnfrsf9; TC_C2 were T helper (Th) cells producing interferon-gamma-like (Th1-like; highly expressing Ucp2, Arpc1b, Tbx21, and Ifng), but these cells also highly expressed Ctla4; TC_C5 were regulatory CD4⁺ T cells (Treg; highly expressing Foxp3, Il2ra, Ctla4, Tnfrsf9, and Tnfrsf18). Among CD8⁺ T cells, we defined TC_C1 as CD8⁺ cytotoxic T lymphocytes (CTLs; highly expressing Nkg7, Gzmk, Gzma, Gzmb, and Ifng), while TC_C3 was defined as naive CD8⁺ T cells (Tn; highly expressing Ccr7, Sell, Lef1, Il7r, and Tcf7). We also followed the method of Jie Xiong et al. [34] to conduct a heatmap analysis of expression functional markers and marker genes for the seven cell clusters, further validating the phenotypic state of each T cell subcluster (Figure 8e). Subsequently, we performed a combined GO and KEGG analysis on the DEGs among these subclusters, providing robust evidence for elucidating the biological functions and mechanisms of these distinct T cell subpopulations (Figure A2d and Supplementary file S8).

Given the complexity of T cell functions and phenotypic states, which cannot be fully captured by the aforementioned methods, we adopted a strategy inspired by the work of Azizi et al. [31]. We meticulously selected a panel of transcriptionally distinct genes to perform gene set scoring across the seven T cell clusters, visualizing the results through a heatmap (Figure 8f and Supplementary file S9). Our analysis revealed that TC_C0 is predominantly involved in the type II interferon response; TC_C1 exhibits high enrichment for genes associated with pro-inflammatory and cytolytic effector pathways, as well as the type II interferon response; TC_C2 is highly enriched for genes regulated by anergy and hypoxia/HIF; TC_C5 shows high enrichment for genes related to T cell terminal differentiation and anti-inflammatory responses; and TC_C6 is highly enriched for genes regulated by hypoxia/HIF. These findings suggest that these T cells may be differentially exposed to inflammatory, hypoxic, anergic, and cytotoxic effector pathways. To further explore these differences, we compared several characteristic gene sets between groups. The results indicated that in the drug combination treatment group, the anergy score of TC_C0 was significantly higher than that of the control group, the cytolytic effector pathway score of TC_C1 was lower, and the anti-inflammatory score of TC_C5 was higher (Figure 8g). These differential scores in the drug-treated T cells provide valuable insights into the mechanisms underlying the therapeutic effects of the drug combination.

Utilizing the Ro/e index, we observed significant shifts in the proportions of T cell immune infiltration subtypes in the ovarian cancer tissues of mice following treatment with AG14361 and AZD4635. Naive CD4⁺ T cells (TC_C0 and TC_C4), Th1-like cells (TC_C2), Tregs (TC_C5), and naive CD8⁺ T cells (TC_C3) were predominantly found in the control group, whereas CTLs (TC_C1) and γδ T cells (TC_C6) were more prevalent in the drug combination treatment group (Figure 8h and Figure A2e). These results highlight that combination therapy effectively inhibits the infiltration of immunosuppressive Tregs while enhancing the infiltration of CTLs, key immune cells for antitumor responses, thereby strengthening antitumor immunity, preventing immune evasion, and curbing ovarian cancer progression. Given that T cell effects are regulated by a balance between activation and exhaustion, and that T cell exhaustion is a critical mechanism for tumor immune evasion, we conducted exhaustion scoring for these T cell clusters using markers such as Pdcd1, Tox, Cxcl13, Tigit, Ctla4, Tnfrsf9, Havcr2, and Lag3. Data showed that T cells in ovarian cancer tissues after drug combination treatment exhibited weaker exhaustion characteristics compared to the control group (Figure A2f). Moreover, exhaustion scores varied among T cell clusters, with TC_C5 showing the most significant exhaustion characteristics, followed by TC_C4, TC_C2, and TC_C6, while TC_C0, TC_C1, and TC_C3 exhibited less pronounced exhaustion (Figure 8i). We also assessed the expression of immune checkpoint genes that determine T cell cytotoxic function. These genes, including Cd274, Ctla4, Icos, Lag3, Cd27, Tnfrsf18, and Vsir, were predominantly enriched in Tregs (TC_C5), with a small subset (Lgals1 and Lgals3) enriched in γδ T cells (TC_C6) (Figure 8j). Comparative analysis between groups revealed that Cd274, Ctla4, Icos, Lag3, Cd27, and Vsir were highly expressed in Tregs of the drug combination treatment group, whereas Lgals1 and Lgals3 were highly expressed in γδ T cells of the control group. Collectively, these results suggest that AZD4635 enhances the antitumor effect of AG14361 by inhibiting Treg infiltration and promoting their exhaustion, thereby weakening their role in tumor immune evasion. However, the high expression of immune checkpoint genes in Tregs may also indicate a potential mechanism for tumor cell insensitivity to PARP inhibitors, which could offset or limit the maximum therapeutic efficacy of the drug combination.

To further elucidate the impact of the drug combination on T cell subclusters, we performed differential expression analysis between the two groups. The results revealed that the drug combination treatment significantly reshaped the gene expression profiles of T cell subclusters (Supplementary file S10). We employed volcano plots (top 5) and heatmaps (top 3) to visualize the expression patterns of upregulated and downregulated genes within each subcluster. Notably, genes such as Pf4, Saa3, Apoe, Cish, S100a6, Lyz2, Ndst1, and Appbp2 exhibited significant upregulation, while Mrpl43, Nme2, Prdx2, Igkc, Pold4, Rfxap, Pop5, Pts, and Dtymk were significantly downregulated (Figure 8k and Figure A2g). GSEA pathway analysis highlighted significant enrichment of pathways such as EMT, complement, inflammatory response, Kras signaling, IL2/STAT5 signaling, TGF beta signaling, and hypoxia in the T cell subclusters following drug combination treatment, with a concomitant reduction in the oxidative phosphorylation pathway (Figure A2h).

To decipher the developmental trajectories of T cells, we utilized the Monocle 2 algorithm to perform pseudotime ordering on CD8⁺ and CD4⁺ T cells (Figure 8l,m and Figure A2i). This pseudotime analysis, based on transcriptional similarity, revealed distinct developmental processes among the T cell subclusters. The analysis traced the progression from state 1, encompassing naive CD4⁺ T cells (TC_C0) and naive CD8⁺ T cells (TC_C3), through branching into state 2 (TC_C1, TC_C2, TC_C5, TC_C6) and state 3 (TC_C0, TC_C2, TC_C4, TC_C5). Notably, exhausted T cells were highly enriched in the later stages of pseudotime, indicative of a transition from activation to exhaustion. By leveraging Monocle technology to identify genes differentially expressed over time, we pinpointed key genes associated with developmental differentiation. The gene expression trends varied markedly among the clusters: in cluster 1, naive T cell markers (Tcf7, Sell, Ccr7, Lef1) and ribosomal proteins showed decreasing expression along the pseudotime axis; in cluster 2, cytotoxic effector molecules and pro-inflammatory cytokines exhibited increasing expression; and in cluster 3, exhaustion-related markers initially increased before declining (Figure A2j and Supplementary file S11).

To explore the regulatory networks underlying these changes, we employed SCENIC analysis. The bar charts and heatmaps revealed variable enrichment of regulatory factors across the T cell subclusters (Figure A2k and Figure 8n). Fil1, Elf1, and Ets1 were consistently identified across all subclusters, while Nfkb1, Tcf7, Foxo1, and Stat3 exhibited differential openness. Notably, TCF7 showed higher openness in TC_C0 and TC_C3, whereas Stat3 was most open in TC_C2 (Figure 8o). Post-treatment, the openness of Tcf7 decreased, while that of Stat3 increased (Figure 8p). These findings suggest that Stat3 and Tcf7 may play pivotal roles in the antitumor effects of AG14361 and AZD4635 combination therapy in ovarian cancer.

3.10. Heterogeneity of B Cell Subpopulations in Ovarian Cancer and Their Response to AG14361 and AZD4635 Combination Therapy

In subsequent analyses, we conducted an in-depth examination of B cells (n = 3779) and identified five distinct clusters (Figure 9a). During the initial phase of cluster annotation, we leveraged classic markers Cd19 and Ms4a1 (Cd20) for identification, as these markers are expressed throughout various stages of B cell development (Figure 9b). In the second phase, we adopted the approach of Weisel et al. [35] to assign phenotypic characteristics of naive B cells and memory B cells to these clusters. Specifically, when M exceeds N, it indicates higher expression of the relevant genes in memory B cells (MBCs); conversely, when M is less than N, it suggests more significant expression in naive B cells (NBCs). Our analysis revealed that clusters B_C0, B_C1, and B_C3 exhibited higher naive B cell markers, while cluster B_C4 displayed higher memory B cell markers (Figure 9c).

Figure 9.

Characterization and identification of B cell features in control and drug combination-treated groups. (a) The UMAP plot illustrates the integrated B cell subclusters derived from six mouse tumor samples, comprising three from the control group and three from the drug combination-treated group. (b) UMAPs of all B cells, colored by the expression of B cell markers. (c) Dot plot showing the expression of marker genes in each subcluster. (d) Heatmap showing the expression signature of top 5 marker genes for 5 cell clusters. In the figure, each column corresponds to a cell, and each row to a gene. Gene expression levels across different cells are color-coded, with yellow representing higher expression and purple indicating lower expression. (e) The bubble plot illustrates the results of GO and KEGG enrichment analysis for differentially upregulated genes in the five B cell subclusters. (f) Box plots illustrate the significance of differences in cell subpopulation counts between groups. (g) Monocle2 was used to perform pseudotime analysis of T cells (left panel), with each point representing a single cell and displaying its state (middle panel) and clustering information (right panel). (h) The scatter distribution plots showing the expression variation in some specific genes in each state during the pseudotime. (i) The scatter plot displays the differentially expressed genes in B cell subpopulations in the drug combination treatment group compared to the control group. ** p < 0.01, *** p < 0.001; ns, no significance.

Next, we utilized differentially expressed markers to further assign functional phenotypes to these clusters, including activation, proliferation, and regulatory B cells (Bregs). Notably, cluster B_C2 was characterized by high expression of proliferation markers such as Stmn1, Mki67, and Hmgb2. We also classified these clusters using activation markers Cd69 and Cd83, finding that clusters B_C1 and B_C0 highly expressed these markers. Given the critical immunomodulatory role of Bregs in tumor progression, it is essential to explore the related phenotypes of these clusters in depth. Although no single marker can specifically identify all Bregs, the combined use of multiple markers (such as Havcr1, Cd5, and Cd1d) can aid in their identification and study. We observed that cluster B_C4 relatively highly expressed Cd5 and Cd1d, which are markers of regulatory B10 cells. Additionally, clusters B_C1 and B_C0 upregulated MHC-II genes, indicating their potential for antigen presentation.

To further interrogate the distinctiveness of these five B cell subclusters, we conducted a marker gene heatmap analysis, revealing unique gene expression profiles for each subcluster (Figure 9d). To elucidate the biological functions and molecular characteristics of these subclusters, we performed GO and KEGG analyses on their DEGs (Supplementary file S12). Notably, subclusters B_C1, B_C2, and B_C3 exhibited enhanced activity in B cell receptor (BCR) signaling, B cell activation, immune response activation, and T cell activation regulation. Specifically, B_C2 showed increased activity in lymphocyte proliferation, phagocytosis, and cell cycle progression (Figure 9e). Additionally, we analyzed the proportion of B cell immune infiltration subtypes after treatment with AG14361 in combination with AZD4635. Compared to the control group, we observed a significant decrease in the immune infiltration of B_C0 and B_C3, while B_C1 showed a significant increase (Figure 9f).

To validate our cluster annotations, we performed trajectory analysis to delineate their developmental trajectories. Pseudotime analysis revealed that proliferating B cells (B_C2) gradually differentiated into memory B cells (B_C4) or naive B cells (B_C0 and B_C3), with naive B cells eventually forming activated B_C1 (Figure 9g). The expression patterns of key genes further corroborated these findings: proliferative genes such as Mki67 and Stmn1 decreased over time, while naive B cell markers (Cxcr4) and memory B cell markers (Pecam1) increased (Supplementary file S13). Similarly, the expression of the activation marker Cd83 also showed an upward trend (Figure 9h).

Finally, we performed differential expression analysis on B cell subclusters from both the control and drug combination treatment groups. Volcano plots (top 5) demonstrated that the drug combination treatment significantly altered the gene expression patterns of B cell subclusters (Figure 9i and Supplementary file S14). For instance, genes such as Pf4, Saa3, Lgmn, Lyz2, Ctsb, and Ckap4 were significantly upregulated, while Gle1, Gng12, Prdx4, Nrip1, Ift172, and Akna (p < 0.05) were significantly downregulated.

4. Discussion

The introduction of PARP inhibitors has revolutionized the treatment paradigm for cancers harboring homologous recombination deficiencies, with ovarian cancer emerging as a quintessential example. However, the clinical utility of PARP inhibitors is frequently undermined by suboptimal long-term efficacy and the emergence of resistance, presenting formidable challenges in the management of ovarian cancer. Chi et al. [11] recently uncovered that in ovarian cancer cells resistant to olaparib, the expression of adenosine receptor genes is markedly elevated, implicating the adenosine signaling pathway in the development of PARP inhibitor resistance. Given that PARP inhibition precipitates an increase in intracellular nicotinamide adenine dinucleotide levels—which can be metabolized into adenosine—and that the resultant adenosine accumulation can bind to the A2A receptor on immune cells to activate the cAMP pathway [22,23,36], thereby fostering tumor immune suppression and immune evasion. These direct and indirect lines of evidence suggest that the accumulation of adenosine and the activation of the adenosine signaling pathway may be pivotal factors impeding the long-term efficacy of PARP inhibitors and driving drug resistance. However, the precise strategies to effectively surmount this issue through combination therapies remain to be fully elucidated.

A substantial body of experimental evidence has established that adenosine concentrations within the tumor microenvironment are significantly elevated compared to normal tissues [37,38]. Yet, the pronounced increase in adenosine levels elicited by PARP inhibitor treatment is equally noteworthy. Given adenosine’s rapid metabolism, characterized by a plasma half-life of approximately 10 s, direct quantification of adenosine levels in biological samples poses considerable challenges [21]. Consequently, the cAMP product, mediated by the binding of adenosine to its receptor A2A, has emerged as a canonical indicator for evaluation [39,40]. Clinical PARP inhibitors are constrained by drug resistance and limited durability. AG14361 drives ATP accumulation [41], providing an ideal tool to dissect the synergistic mechanisms of combined A2AR antagonism and inform clinical optimization strategies. Our investigations reveal that the PARP inhibitor AG14361 exerts a profound inhibitory effect on the in vitro and in vivo growth of ovarian cancer. This inhibition is concomitant with the accumulation of cAMP in mouse cells and tumor tissues, which subsequently activates the cAMP/CREB pathway. Although AG14361 continues to suppress ovarian cancer growth despite augmenting adenosine levels, this phenomenon is not isolated. Research by Sureechatchaiyan et al. [42] has demonstrated that adenosine can augment the sensitivity of human ovarian cancer cells to cisplatin, mediated by increased pAMPK and diminished pS6K levels. We speculate that the initial accumulation of adenosine by PARPi does not overshadow its anticancer effects; however, as time progresses and adenosine continues to accumulate, its pro-tumor effects ultimately eclipse the anticancer effects of PARPi. To amplify the therapeutic activity and sensitivity of AG14361, we propose a combinatorial approach incorporating PARPi with A2AR antagonists. Our results unveil that the cAMP surge mediated by AG14361-induced adenosine can be effectively mitigated by the A2AR antagonist AZD4635. This inhibition yields a striking suppression of ovarian cancer growth. This indicates that AZD4635 enhances the antitumor effects of AG14361 by inhibiting the binding of adenosine to the A2AR and thereby blocking the cAMP-CREB pathway. Transcriptomic analyses further corroborate these molecular mechanisms, demonstrating that AZD4635 enhances AG14361’s antitumor effects through negative regulation of the cAMP signaling pathway (ES = −0.31, p < 0.05). Numerous studies have highlighted that the cAMP-dependent pathway is one of the most critical signaling cascades in malignant ovarian cells, with ovarian cancer growth and metabolism being heavily contingent upon alterations in cAMP-PKA-CREB axis signaling [43]. Dimitrova et al. [44] through a novel systems biology approach, identified that high activation of the cAMP-CREB axis is highly correlated with platinum resistance in ovarian cancer. Inhibition of CREB phosphorylation significantly enhances the sensitivity of resistant cells to platinum therapy while selectively targeting ovarian cancer stem cells responsible for platinum resistance and tumor recurrence. Collectively, these findings underscore that modulation of the adenosine-A2AR-cAMP-CREB axis is key to reversing PARP inhibitor treatment sensitivity and resistance.

Additionally, our transcriptomic analyses have uncovered that AZD4635 significantly potentiates the anti-tumor efficacy of AG14361, potentially through modulating ATP metabolism. To facilitate tumor growth, cancer cells often induce a hypoxic microenvironment, which results in the accumulation of extracellular ATP [45]. Subsequently, this ATP is metabolized into adenosine within the tumor microenvironment via the sequential actions of the ectoenzymes CD39 and CD73. This cascade represents the predominant pathway for adenosine generation in tumors. Thus, the sources of adenosine can be targeted at three key nodes: CD39, CD73, and A2AR. Our results demonstrate that knockout of CD39 or CD73 significantly inhibits cAMP accumulation in ovarian cancer cells and suppresses the cAMP-CREB signaling pathway. However, functional defects in CD39 or CD73 did not abrogate the cAMP accumulation and activation of the cAMP-CREB signaling pathway induced by PARPi AG14361. This suggests that the pathways by which AG14361 induces adenosine production are multifaceted. In addition to the well-characterized CD39/CD73 catalysis of ATP to generate adenosine, alternative pathways for extracellular adenosine generation exist: one involves the conversion of NAD⁺ to ADPR by CD38, followed by the conversion of ADPR to AMP by CD203a, with subsequent conversion of AMP to adenosine by CD73; another pathway involves the conversion of S-adenosyl-L-homocysteine (SAH) to adenosine by S-adenosyl-L-homocysteine hydrolase (SAHH) [23]. Collectively, these findings suggest that compared to merely targeting the sources of adenosine, blocking the binding of adenosine to A2AR and the activation of downstream signaling pathways may be crucial for enhancing the durability and tolerance of AG14361 treatment. Thus, A2AR emerges as the most critical regulatory node within the adenosine-A2AR-cAMP-CREB axis, thereby justifying the rational incorporation of an A2AR antagonist into our combinatorial therapy regimen.

Adenosine, a key immunosuppressive metabolite, exerts its effects by binding to the A2A receptor on immune cells, thereby activating the downstream cAMP signaling pathway. This cascade ultimately leads to the suppression of immune effector cell activity, facilitating immune surveillance escape by tumor cells [45]. Our transcriptomic analyses revealed that the enhanced anticancer efficacy and refined immune response regulation achieved with the combination of AZD4635 and AG14361 are intricately connected. Further support for this notion comes from in vivo experiments and immunohistochemical data, which collectively demonstrate that this combination therapy effectively overcomes the adenosine-mediated immunosuppressive microenvironment. Specifically, the dual treatment restores the effector functions of multiple immune cell populations, yielding significant antitumor effects in mouse models. Strikingly, the combination therapy reduces the infiltration of immunosuppressive cells, such as regulatory T cells and M2 macrophages, while simultaneously increasing the number of cytotoxic T cells and granzyme B-positive cells. This shift in immune cell composition creates a more favorable immune microenvironment, tipping the balance toward effective tumor clearance. Ultimately, these changes enhance the antitumor immune response elicited by AG14361 and effectively counteract the immune escape mechanisms employed by tumor cells, which are often driven by adenosine signaling.

Utilizing scRNA-seq, we have meticulously constructed a high-resolution, large-scale atlas of the single-cell immune landscape, revealing marked disparities in the immune microenvironment between the drug combination therapy cohort and the control group. Our findings underscore the pivotal roles of macrophages and T cells in modulating the enhanced efficacy of AZD4635 when combined with AG14361. These insights lay the groundwork for deeper biological inquiries and the development of novel immune checkpoint inhibitors or refined combination therapies. Through re-clustering of macrophages, we delineated four distinct functional subpopulations, elucidating their unique biological functions and phenotypic profiles. Notably, the M0_Lhfpl2 and M2_Ccl5 clusters exhibited elevated expression of angiogenesis-related genes and correspondingly higher Angio-TAMs scores. Angiogenesis, a cornerstone of cancer development and progression, is significantly influenced by VEGFA, as previously documented [46]. TAMs are recognized not only as substantial producers of VEGFA but also as key sources of additional pro-angiogenic factors. Wang et al. [47] demonstrated that TAMs can bolster the cancer stem cell properties of tumor cells via VEGFA secretion, further implicating VEGFA in immune suppression within the TME. The M0_Lhfpl2 cluster is characterized by specific expression of Lhfpl2, a gene with limited research precedence. Gong et al. [48] reported that LHFPL2 is selectively expressed in macrophages, with high-expression subgroups displaying enhanced M2 polarization, hypoxia, immune escape, and angiogenesis scores, thereby driving tumor progression. In contrast, the M2_Ccl5 cluster is distinguished by its secretion of multiple C-C chemokine ligands and elevated Inflam-TAMs scores. Liu et al. [49] highlighted that macrophage-derived CCL5 facilitates tumor cell immune evasion through the p65/STAT3-CSN5-PD-L1 axis. Additionally, CCL4 has been shown to amplify tumor progression by recruiting regulatory T cells and pro-tumor macrophages, thereby enhancing their tumorigenic potential [50].

In our investigation, the M0_Lhfpl2 macrophage cluster displayed elevated Angio-TAMs scores following combination therapy, despite no significant changes in its infiltration levels between the control and treatment groups. Conversely, the M2_Ccl5 cluster exhibited a marked reduction in infiltration after drug combination treatment, accompanied by high Angio-TAMs scores. Although the Inflam-TAMs scores did not show significant differences, Ccl5 secretion from the M2_Ccl5 cluster was relatively diminished compared to the control group following combination therapy. These observations highlight the profound impact of combination therapy on the function and phenotype of distinct macrophage subpopulations. The M3_Vcan cluster, characterized by high expression of MHC-II molecules (e.g., H2-Eb1, H2-Aa) and Cd74, exhibited traits associated with mature macrophages involved in antigen presentation. This suggests that macrophages may retain their classical antigen-presenting functions under different treatment conditions. These findings underscore the complexity of macrophage phenotypes and highlight the need for further exploration of their roles as therapeutic targets and in combination therapies.

Notably, we identified an M1_C1qa macrophage subpopulation with elevated Reg-TAMs scores and infiltration levels in the combination therapy group. Previous studies have implicated C1q⁺ macrophages as pro-tumor cell clusters [51,52]. Li et al. [53] demonstrated that C1q deficiency in macrophages rescues CD8⁺ T cell exhaustion and enhances the antitumor immune activity of CD8⁺ T cells and NK cells. Conversely, Mehta et al. [54] showed that olaparib-treated macrophages limit T cell proliferation and antitumor function, and that removal of M2 macrophages using anti-CSF-1 receptor antibodies improves the efficacy of olaparib. Our immunohistochemical analyses revealed that AG14361-induced M2 macrophage infiltration can be reversed by AZD4635, although infiltration levels remain higher than those in the blank control group. Collectively, these findings suggest that the M1_C1qa macrophage subpopulation mediates immune suppression, which may represent a vulnerability in the enhanced anticancer efficacy of the AZD4635 and AG14361 combination. A deeper characterization of this subpopulation is thus essential for optimizing the therapeutic synergy of this drug combination.

In our analysis of T cells, we uncovered seven distinct functional subpopulations, encompassing conventional CD4⁺ and CD8⁺ T cells as well as a rare CD3⁺CD4⁻CD8⁻ DNT cell subpopulation (TC_C6). The role of DNT cells in the tumor microenvironment, particularly in ovarian cancer, remains poorly understood, with few published reports. In our study, DNT cells exhibited elevated exhaustion scores. Differential gene expression analysis revealed that genes such as Abr, Ccl3, Appbp2, Ext1, and Homer1 were significantly upregulated in the drug combination treatment group, whereas genes including Cdc25b, Tm2d1, Cnpy2, Ndufs8, and Mrps18a were downregulated. We anticipate that our findings will offer valuable insights for future investigations into DNT subpopulations in ovarian cancer.

Notably, our results demonstrated that combination therapy reduced the infiltration of regulatory T cells (Tregs, TC_C5). Previous studies have shown that Treg infiltration in ovarian solid tumors or ascites is detrimental, as these cells promote tumor growth through the secretion of immunosuppressive cytokines [55]. Tregs have also been implicated in cancer progression and drug resistance across various stages of ovarian cancer [56]. Thus, strategies aimed at depleting or modulating Tregs hold promise for enhancing clinical efficacy and overcoming tumor resistance in ovarian cancer. However, despite reduced infiltration levels under combination therapy, some immune checkpoint molecules, particularly Ctla4, remained highly expressed. The inhibitory function of Tregs is regulated by CTLA-4, and CTLA-4 deficiency impairs Treg-mediated suppression both in vivo and in vitro [57]. Tekguc et al. [58] reported that Tregs can downregulate CD80/CD86 expression on antigen-presenting cells (APCs) via CTLA-4-dependent phagocytosis, thereby inhibiting APC stimulatory activity on T cells. Concurrently, CTLs have emerged as key players in cancer immunotherapy, with ongoing clinical trials focused on generating CTLs with potent antitumor activity [59]. Depletion of cytotoxic T cells markedly attenuates the antitumor efficacy of olaparib [17]. Our study revealed that combination therapy enhances CTL infiltration (TC_C1). Collectively, these findings suggest that our treatment regimen can restore T cell functionality and reinvigorate antitumor immune responses. Given the complex interplay between CTLs and Tregs, future studies should focus on the Ctla4⁺ Tregs subpopulation, with the goal of enhancing CTL numbers and functionality by inhibiting Treg activity, thereby sensitizing cancer cells to immune attack.

In addition to the aforementioned effects on T cells and macrophages, this therapeutic regimen also influenced B cell subsets, hinting at a potential role for B cells in mediating treatment responses. However, due to the relatively low abundance of other immune cell populations, such as NK cells, DCs, and granulocytes, we refrained from further analysis or interpretation to avoid drawing spurious conclusions. In summary, this study reveals the critical role of the ADO–A2AR signaling axis in modulating the efficacy of PARPi, offering a novel strategy for combination therapy in ovarian cancer. Given that CD39/CD73-mediated ADO accumulation represents one of the primary mechanisms of tumor immune evasion, ADO constitutes a shared bottleneck limiting both PARPi efficacy and contributing to immune checkpoint inhibitor (ICI) resistance. Therefore, blockade of the ADO–A2AR axis emerges as a pivotal approach to overcome the limited efficacy of the “PARPi + ICI” combination regimen. Future efforts could explore a triple-combination strategy—“PARPi + A2ARa + ICI”—to synergistically achieve a tripartite effect: targeted tumor cell killing, immune activation, and remodeling of the tumor microenvironment, thereby paving a new path toward precision therapy for BRCA1/2-mutated ovarian cancer.

This study has several limitations that need to be addressed in future research: First, the PBMCs used in the experiments were not subjected to systematic immune cell phenotyping, making it impossible to determine the baseline composition ratios of various immune cell subsets, which limits the in-depth interpretation of the in vitro experimental results. Second, the in vivo experiments did not employ an orthotopic tumor model, which differs from the actual clinical scenario of ovarian cancer development and makes it difficult to authentically recapitulate the interactions between tumors and local immune cells as well as stromal cells. Finally, the correlation analyses integrating survival outcomes, clinical characteristics, and tumor immune microenvironment were limited by a relatively modest sample size of 66 TCGA ovarian cancer cases, which may be insufficient to comprehensively capture the heterogeneity of immune signatures across diverse clinical subtypes.

5. Conclusions

Collectively, our findings support the emerging paradigm of targeted drug combination therapies for solid tumors that rely on adenosine as a primary immune suppression mechanism and exhibit homologous recombination deficiencies. This study provides compelling evidence for the synergistic antitumor effects of combining PARPi with A2AR antagonists in the treatment of ovarian cancer. Specifically, AZD4635 enhances the anticancer efficacy of PARPi by targeting the A2AR receptor and modulating the cAMP/CREB signaling pathway. This dual mechanism of action not only inhibits tumor cell proliferation but also remodels the tumor immune microenvironment to bolster antitumor immunity (Figure 10). Our results offer valuable insights for the development of innovative combination therapies targeting ovarian cancer.

Figure 10.

Schematic diagram illustrating the proposed mechanism by which the A2AR antagonist AZD4635 enhances the anti-ovarian cancer efficacy of the PARP inhibitor AG14361.

Abbreviations

The following abbreviations are used in this manuscript:

PARPi	Poly(ADP-ribose) polymerase inhibitors
NAD+	Nicotinamide adenine dinucleotide
A2AR	Adenosine A2A receptor
A2ARa	A2AR antagonists
scRNA-seq	Single-cell RNA sequencing
cAMP	Cyclic adenosine monophosphate
CREB	cAMP/cAMP response element-binding protein
ADO	Adenosine
TME	Tumor microenvironment
PKA	Protein kinase A
PBMCs	Peripheral blood mononuclear cells
ELISA	Enzyme-linked immunosorbent assay
PBS	Phosphate-buffered saline
RNA-seq	RNA sequencing
TCGA	the Cancer Genome Atlas
TCGA-OV	the Cancer Genome Atlas-ovarian serous cystadenocarcinoma
FBS	Fetal bovine serum
TCIA	the Cancer Immunome Atlas
NGS	Next-generation sequencing
SD	Standard deviation
PCA	Principal component analysis
DEGs	Differentially expressed genes
GSEA	Gene Set Enrichment Analysis
GO	Gene Ontology
KEGG	Kyoto Encyclopedia of Genes and Genomes
PPI	Protein–protein interaction
Tregs	Regulatory T cells
Teff	Effector T cells
NK cells	Natural killer cells
TILs	Tumor-infiltrating lymphocytes
GZMB	Granzyme B
MDSCs	Myeloid-derived suppressor cells
UMAP	Uniform Manifold Approximation and Projection
AdenoSig	Adenosine gene signature
TAMs	Tumor-associated macrophages
IFN-TAMs	Interferon-preprocessed TAMs
Reg-TAMs	Immune regulatory TAMs
Inflam-TAMs	Inflammatory cytokine-rich TAMs
LA-TAMs	Lipid-associated TAMs
Angio-TAMs	Angiogenic TAMs
Prolif-TAMs	Proliferative TAMs
Tn	Naive T cells
Th	T helper cells
Th1-like	Th cells producing interferon-gamma-like
CTLs	CD8+ cytotoxic T lymphocytes
MBCs	Memory B cells
NBCs	Naive B cells
Bregs	Regulatory B cells
BCR	B cell receptor
SAH	S-adenosyl-L-homocysteine
SAHH	S-adenosyl-L-homocysteine hydrolase
APCs	Antigen-presenting cells
ICI	Immune checkpoint inhibitor
TFs	Transcription factors

Supplementary Materials

The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/pharmaceutics18020257/s1, Supplementary file S1. The list of differentially expressed genes in ID8 cells subjected to combination drug therapy elucidated through transcriptomic analysis; Supplementary file S2. The list of KEGG-enriched pathways derived from the analysis of differentially expressed genes; Supplementary file S3. Clinical information and expression data of CD39 and CD73 for 66 ovarian cancer patients screened and organized from the TCGA-OV dataset; Supplementary file S4. A list of upregulated differentially expressed genes in macrophage subpopulations; Supplementary file S5. List of macrophage-related gene signatures; Supplementary file S6. List of differentially expressed genes in macrophage subpopulations between groups; Supplementary file S7. List of differentially expressed genes in macrophages during pseudo-time differentiation. Supplementary file S8. A list of upregulated differentially expressed genes in T cell subpopulations; Supplementary file S9. A list of T cell-related gene signatures; Supplementary file S10. List of differentially expressed genes in T cell subpopulations between groups; Supplementary file S11. A list of differentially expressed genes in T cell subpopulations that vary along the pseudotime axis; Supplementary file S12. A list of upregulated differentially expressed genes in B cell subpopulations; Supplementary file S13. A list of differentially expressed genes in B cell subpopulations that vary along the pseudotime axis; Supplementary file S14. List of differentially expressed genes in B cell subpopulations between groups.

Appendix A

Figure A1.

Re-clustering analysis of macrophages in the control and drug combination treatment groups reveals their distinct characteristics. (a) Heatmap showing the expression signature of top 5 marker genes for 5 cell clusters. In the figure, each column corresponds to a cell, and each row to a gene. Gene expression levels across different cells are color-coded, with yellow representing higher expression and purple indicating lower expression. (b) Ro/e heatmap: The abundance of cell subclusters in each group is quantified and assessed by calculating the ratio of observed to expected cell counts. (c) Box plots illustrate the significance of differences in cell subpopulation counts between groups. (d) The heatmap displays the top 5 differentially expressed genes in each macrophage subpopulation in the drug combination treatment group compared to the control group. (e) Compared with the control group, GSEA reveals the significant pathways associated with each macrophage subpopulation in the drug combination treatment groups. (f) Pseudotime analyses of macrophages, colored by cell states. (g) The heatmap displays differentially expressed genes along the pseudotime axis (x-axis). The right y-axis indicates the genes, while the left y-axis represents the clustering results, which show gene clusters with similar expression trends. (h,i) The pseudo-heatmap displays a diverse array of genes implicated in macrophage differentiation, specifically within branch 1 (h) and branch 2 (i). These genes are further organized into four distinct clusters based on their expression patterns. (j) The bar chart illustrates the quantities of regulons, transcription factors (TFs), and target genes that were identified following the completion of the SCENIC analysis. (k) The heatmap displays the activity values of the cell regulons and uses color bars for grouping, samples, and subgroups to show the heterogeneity and distribution characteristics in different dimensions. ** p < 0.01, *** p < 0.001; ns, no significance.

Figure A2.

Re-clustering analysis of T cells in the control and drug combination treatment groups reveals their distinct characteristics. (a) Heatmap showing the expression signature of top 5 marker genes for 7 cell clusters. In the figure, each column corresponds to a cell, and each row to a gene. Gene expression levels across different cells are color-coded, with yellow representing higher expression and purple indicating lower expression. (b) Heatmap showing the expression signature of top 5 marker genes for CD4⁺ T cell clusters. (c) Heatmap showing the expression signature of top 5 marker genes for CD8⁺ T cell clusters. (d) The bubble plot illustrates the results of GO and KEGG enrichment analysis for differentially upregulated genes in the T cell subclusters. (e) Ro/e heatmap: The abundance of cell subclusters in each group is quantified and assessed by calculating the ratio of observed to expected cell counts. (f) The violin plot displays the enrichment levels of exhaustion-related gene sets across different groups. (g) The heatmap displays the top 5 differentially expressed genes in each T cell subpopulation in the drug combination treatment group compared to the control group. (h) Compared with the control group, GSEA reveals the significant pathways associated with each T cell subpopulation in the drug combination treatment groups. (i) Pseudotme analyses of T cells, colored by cell clusters. (j) The heatmap displays differentially expressed genes along the pseudotime axis (x-axis). The right y-axis indicates the genes, while the left y-axis represents the clustering results, which show gene clusters with similar expression trends. (k) The bar chart illustrates the quantities of regulons, TFs, and target genes that were identified following the completion of the SCENIC analysis. (l) The heatmap displays the activity values of the cell regulons and uses color bars for grouping, samples, and subgroups to show the heterogeneity and distribution characteristics in different dimensions.

Author Contributions

The authors confirm contribution to the paper as follows: Conceptualization, H.S. and C.X.; methodology, B.P. and X.Y.; software, B.P. and X.Y.; validation, B.P. and X.Y.; formal analysis, B.P., X.Y., X.W., H.S., C.X. and N.Z.; investigation, B.P., X.Y., X.W., J.F. and Q.L.; data curation, B.P., X.Y., X.W., J.F. and Q.L.; writing—original draft preparation, B.P.; writing—review and editing, H.S., C.X. and N.Z.; visualization, B.P., X.Y., C.X. and H.S.; supervision, H.S., C.X. and N.Z.; project administration, H.S., C.X. and N.Z.; funding acquisition, H.S., C.X. and N.Z. All authors have read and agreed to the published version of the manuscript.

Institutional Review Board Statement

The animal study was approved by the Experimental Animal Ethics Committee of the Guangdong Provincial Medical Laboratory Animal Center (Approval Number: C202302-12, approved on 28 February 2023) and conducted in accordance with the protocol approved by the Guangdong Provincial Animal Care and Use Committee.

Informed Consent Statement

Not applicable.

Data Availability Statement

All data supporting the findings of this study are available within the paper and its Supplementary Information. Publicly available data were obtained from published TCGA (https://portal.gdc.cancer.gov, accessed on 7 April 2025) and TICA (https://tcia.at/home, accessed on 7 April 2025).

Conflicts of Interest

The authors declare no conflicts of interest.

Funding Statement

This work was supported by grants from the Special Project for Clinical and Basic Sci&Tech Innovation of Guangdong Medical University (GDMULCJC2024131), the Science and Technology Bureau of Foshan (FS0AA-KJ819-4901-0082), the Research Project Approval for National key clinical specialty discipline construction program of Breast cancer center in Hubei cancer hospital in 2024 (2024 HBCHBC C-C03), the Scientific Research Projects of Hubei Cancer Hospital (2024 HBCHYN12), and the Natural Science Foundation of Hubei Province (2025AFC117).