Mes/Stem-A ovarian cancer subtypes predict sensitivity to ATR inhibition

Abstract

High-grade serous ovarian cancer (HGSOC) remains one of the most common and lethal gynecological malignancies. Distinct gene expression molecular subtypes (GEMS) of HGSOC, with associated differences in outcome, have previously been described: Epithelial-A (Epi-A)/C3/Differentiated, Epithelial-B (Epi-B)/C2/Immunoreactive, Stem-like B (Stem-B)/C6, Mesenchymal (Mes)/C1, and Stem-like A (Stem-A)/C5/Proliferative. Notably, the Mes and Stem-A OC subtypes are associated with more advanced-stage tumors and poorer clinical outcomes, representing a molecular subtype of unmet clinical need. Hallmarks of HGSOC also include a high prevalence of genomic instability and increased replication stress. In this study, we demonstrate that HGSOC Mes/Stem-A (MSA) cell lines exhibit elevated levels of replication stress and increased sensitivity to ATR inhibitors. Additionally, through DNA fiber analysis, we observed significantly higher replication fork rates and ssDNA accumulation in MSA subtype cells. Consistent with the high levels of replication stress identified in these subtypes, MSA patient-derived xenograft models further validated the therapeutic vulnerability to ATR inhibition. The characteristic replication stress profile of the MSA subtype suggests its utility as a candidate biomarker for identifying patients most likely to benefit from ATR inhibition.

Introduction

High grade serous ovarian cancer (HGSOC) is one of the most lethal gynecological cancers [1, 2]. One of the salient features of HGSOC is the elevated levels of genomic instability, associated with a ubiquitous prevalence of TP53 mutations (>95%) and frequent homologous recombination repair defects (HRD; ~50%) [3], resulting in increased replication stress (RS) [4]. Briefly, RS is the cellular response to stalled replication forks or single-stranded DNA breaks, initially coated with RPA, leading to downstream activation of the DNA damage sensor ataxia telangiectasia and Rad3 related (ATR) kinase. Subsequently, checkpoint kinases such as CHK1, WEE1, are activated, eventually directing the cell to arrest cell cycle progression and initiate repair of stalled replication forks [5]. Due to rapid proliferative nature of cancer cells, targeting of these cell cycle checkpoint kinases render exhaustive replication fork breakage, culminating in replication catastrophe and apoptosis [6]. The concept of oncogene-induced RS as a driver of genomic instability was pioneered by seminal studies demonstrating that oncogene activation triggers DNA damage and activates the DDR early in tumorigenesis [7–9]. This intrinsic stress creates a specific vulnerability in cancer cells, providing the foundational rationale for using checkpoint kinase inhibitors to selectively eliminate transformed cells [10]. Recently, selective ATR inhibitors (such as berzosertib, ceralasertib, camonsertib, tuvusertib, elimusertib) have been investigated in phase I/II clinical trials as monotherapy or in combination with chemotherapy [11]. Furthermore, strategies utilizing ATR inhibitors in combination with poly(ADP-ribose) polymerase inhibitors to overcome platinum-resistance have also been investigated [12, 13]. Generally, studies using ATR inhibitors in ovarian carcinoma have shown varying levels of efficacy with significant challenges associated with drug toxicity. Hence, predictive biomarkers in this context would be crucial to help improve patient selection and the therapeutic index for this class of therapy.

Inter- and intra-tumor heterogeneity in OC also represents a significant hurdle in the context of precision therapy. Distinct gene expression molecular subtypes (GEMS) of OC, with associated differences in outcome, have previously been described: Epithelial-A (Epi-A)/C3/Differentiated, Epithelial-B (Epi-B)/C2/Immunoreactive, Stem-like B (StemB)/C6, Mesenchymal (Mes)/C1 and Stem-like A (Stem-A)/C5/Proliferative [14–16]. Notably, Mes and Stem-A OC subtypes are associated with more advanced-stage tumors and poorer clinical outcomes and represents a molecular subtype of unmet clinical need.

In this study, we demonstrated that OC cell lines and tumors of Mes/Stem-A (MSA) subtypes harbor an elevated RS signature and that MSA OC cell lines are selectively sensitive to ATR inhibitors elimusertib and camonsertib in vitro. Through immunoblot analyses and DNA fiber analyses, we observed the presence of endogenously elevated levels of RS in MSA cell lines and increased sensitivity to ATR inhibition. Furthermore, patient-derived xenograft tumors with the MSA signature also showed better responses to elimusertib.

Materials and methods

Cell lines and in vitro proliferation assay

Cell lines used in the study were grown as previously described [17]. The 2D anti-proliferative activity of elimusertib was evaluated by plating the cells in their appropriate growth medium in 96 well format, and by incubating them with increasing concentrations of elimusertib (1 pM–10 µM) for three days. We measured cell viability either using the Cell Counting Kit-8 (CCK-8, Dojindo Molecular Technologies, Inc.) and absorbance was recorded with a plate reader. We performed all measurements at least in biological triplicates and normalized with respect to the control group. The half-maximal inhibitory concentration (IC₅₀) was determined using a four-parameter logistic regression model. For the 3D antiproliferative assay, cell suspension with appropriate cell number was mixed with equal volume of 20% Matrigel (diluted in growth media) and seeded on ultra-low attachment for spheroid formation. After 7–10 days of spheroid growth, increasing doses of elimusertib were added to spheroids and antiproliferative activity was assessed using CellTiter-Glo 3D Cell Viability Assay (Promega, Cat. No. G9681).

For the colony formation assay, cells were plated in a 6-well plate and incubated at 37 °C for 24 h. Afterward, cells were exposed to elimusertib (2 µM) for 24 h. Subsequently, cells were allowed to recover in fresh medium from 5 to 7 days. Colonies were then stained by applying a 1X crystal violet solution (prepared from a 10X stock containing 8% glutaraldehyde for fixation and 0.5% crystal violet in 20% methanol for staining) for 2 h.

Western blotting

Whole cell lysates were prepared using RIPA buffer and resolved in standard reducing SDS-PAGE followed by blotting on PVDF membranes. Immunoblots were incubated overnight with suitable primary antibodies: AMBRA1 (Santa Cruz Biotechnology, #sc-398204), p21 (Santa Cruz Biotechnology, #sc-6246), pATR (Ser428, CST, #2853), pCHK1 (Ser345, CST, #2348), pRPA (Ser4/Ser8, Bethyl Laboratories, #A300-245A), γH2AX (Ser139, Abcam, #ab11174), and Tubulin (Abcam, ab7291) diluted in 2% BSA in Tris-buffered saline with 0.1% Tween (TBST). Infrared dye-conjugated secondary antibodies from Li-COR Biosciences, Lincoln, NE, USA at 1:10,000 diluted in TBST: IRDye 800-CW goat anti-mouse or anti-rabbit (#926-32210 and #926-32211) and IRDye 680LT goat anti-mouse or anti-rabbit (#926-68020 and #926-68021). Blots were scanned using the Odyssey infrared imaging system (Li-COR), and the resulting images were transferred to greyscale. Quantification of protein band densities were analyzed using the ImageJ software [18] and quantified as described using Davarinejad’s protocol [19].

Detection of ssDNA using immunofluorescence

The detection of ssDNA was conducted following a protocol adapted from ref. [20]. Briefly, cells were labeled with 60 µM BrdU for 20 h, followed by treatment with 2 µM elimusertib for 7 h. Cells were then rinsed once with ice-cold PBS and pre-extracted with ice-cold pre-extraction buffer for 5 min on ice. After aspiration of the buffer, cells were washed again with ice-cold PBS and fixed in freshly prepared ice-cold 4% paraformaldehyde (PFA) in PBS. The cells were washed twice more and blocked with 3% BSA in PBS for 1 h at room temperature.

Subsequently, cells were incubated with anti-BrdU antibody overnight, followed by washing and staining with a fluorescent secondary antibody for 1 h. Nuclear DNA was counterstained with diluted DAPI solution (1:2500) for 5 min. The cells were then mounted with ProLong® Antifade. Immunostained cells were visualized using a ZEISS Axio Imager.Z1 fluorescence microscope, and fluorescence signal quantification was conducted using ImageJ software.

DNA fiber assay and analysis

The method followed the protocol outlined in ref. [21]. Briefly, cells were pulse-labeled with 5-chloro-2’-deoxyuridine (CldU) at 60 μM and 5-iodo-2’-deoxyuridine (IdU) at 250 μM. Labeled cells were harvested, lysed, and spread on inclined glass slides, then fixed in methanol-acetic acid and denatured in HCl. DNA spreads were blocked and stained with primary anti-BrdU (BU1/75) (Abcam, ab6326) antibodies, followed by Alexa Fluor-labeled secondary antibodies. Single-stranded DNA was stained with an anti-ssDNA antibody and an Alexa Fluor-labeled secondary. Slides were mounted using ProLong® Gold Antifade, and images were acquired with a ZEISS Axio Imager.Z1 microscope.

To quantify DNA replication dynamics, at least 100 DNA fibers were analyzed in ImageJ. Fork speed was measured by determining the length of labeled tracks in micrometers (μm), dividing by the labeling time, and converting to kilobases (kb) by multiplying the length in μm by 2.59. For origin firing analysis, the percentage of new origin firing was quantified by calculating the frequency of origin firing events—specifically, the number of green-only tracts (indicating origins that fired during the second labeling period) relative to the total number of events.

Alkaline comet assay

We used the OxiSelect™ Comet Assay kit (Cell Biolabs, Inc., #STA-351-5) to perform alkaline comet assay following manufacturer’s protocol. Briefly, adherent cells were removed from dish by scraping with a cell scraper. A cell suspension of 1 ×10⁵ cells/mL were mixed with agarose at 1:10 ratio (V/V) and immediately transfer 75 μL/well onto the top of the agarose base layer on the glass slide. Maintaining the slides horizontally, the slides were subjected to electrophoresis with pre-chilled alkaline solution at 1 volt/cm, for 30 min at 4 °C in the dark. Following electrophoresis, the slides are dried, stained with a Vista Green DNA dye, and visualized by epifluorescence microscopy. Comets were analyzed with OpenComet (v1.3.1) [22]. Tail DNA% calculated using manufacturer’s protocol. At least thirty cells were analyzed per sample pooled from biological triplicates.

Generation of cell line-derived xenograft models

All in vivo xenograft studies adhered to the SingHealth Animal Care and Use Committee (IACUC:2018/SHS/1356) guidelines on animal use and handling. Usage of patient-derived materials has been approved by the National Health Group Domain Specific Review Board (DSRB), Reference: 2013/00705. Ten primary, treatment-naïve, ovarian cancer patients’ tumor sample were used for this study. Freshly isolated tumor tissue was individually minced into fine fragments that would pass through an 18-gauge needle. Fine tumor fragments were then mixed 1:1 (V/V) with Matrigel (Collaborative Research, Bedford, MA) to give a total volume of 0.2 ml (~10 million cells) per injection. The tissue mixture was subcutaneously injected in the flank of female immunodeficient, interleukin -2 receptor-γ-null (NOD-SCID-IL-2rγ) mice (8 weeks, 20–22 g, Envigo). No statistical method was used to predetermine sample size. When the tumors reached a predetermined size of 100 mm³, mice were randomized into treatment and control groups (n = 6–8 mice/group) and treated with elimusertib. Elimusertib was formulated in PEG400/Ethanol/water (60:10:30) and treatment was conducted in 3 days on/4 days off, twice daily, p.o. Tumor response was assessed by measuring tumor volume (length x width) using a caliper. Animal body weight was monitored as a measure of treatment-related toxicity. Tumor volume and body weight were determined 2–3 times per week. Changes in body weight were considered as a measure of treatment-related toxicity (>10% loss of body weight was considered critical and treatment dosing was held until recovery, >20% loss of body weight was considered toxic and treatment was terminated). Mice were euthanized when showing signs of toxicity (>20% body weight loss), or when tumors reached a maximum area of 225 mm². Tumor growth inhibition is presented as T/C ratio (treatment/control), calculated with tumor volume at study end. Relative tumor growth inhibition based on tumor volume (T/C ratio) was calculated using the following formula:

$$T/Cratio=\frac{{Tr\_volume}_d-{Tr\_volume}_0}{{Ctrl\_volume}_d-{Ctrl\_volume}_0}$$

${\mathit{Tr\_volume}}_d$ = tumor volume of treatment group at day d; ${\mathit{Tr\_volume}}_0$ = tumor volume of treatment group before treatment or day 0; ${\mathit{Ctrl\_volume}}_d$ = tumor area of control group at day d; and ${\mathit{Ctrl\_volume}}_0$ = tumor area of control group before treatment. T/C ratio <0 to <0.3 was considered a good response. T/C ratio >0.3 – <0.6 was considered a moderate response. T/C ratio >0.6–0.8 was considered a weak response (no biological relevance). T/C ratio 0.9–1.0 meant the treatment was inactive (no biological relevance).

Statistical analyses

All statistical analyses were performed using GraphPad Prism and data were presented as mean with standard error of mean. Two-tailed unpaired Student’s t test with Welch’s correction was performed to compare differences between two individual groups unless otherwise stated. A p-value of less than 0.05 was defined to be statistically significant and is indicated as follows—ns not significant, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Results

OC cell lines of specific ovarian cancer subtypes are sensitive to ATR inhibitors – Elimusertib & Camonsertib

By selectively knocking down cell cycle checkpoint proteins (ATM, ATR, CHK1 and CHK2) in the DDR/RSR pathway, McGrail et al. derived a replication stress response (RSR) gene signature that could predict the risk of breast cancer development [23]. Using this signature, we investigated the extent of replication stress among OC molecular subtypes [15] in a meta-analysis of expression datasets from 142 cultured epithelial OC cell lines and 3261 human ovarian cancers. Across the GEMS of OC cell lines, we observed that cell lines of Mes and Stem-A subtypes have enriched RS score in comparison to cell lines of Epi-A, Epi-B and Stem-B subtypes (Fig. 1A). Ovarian tumors with Mes and Stem-A subtypes are associated with worse clinical outcomes, disease recurrence and metastasis [15, 24]. A similar analysis of RS score in 3261 human ovarian cancers revealed a parallel trend of a significantly enriched RS score in the Mes and Stem A tumors group in comparison to the non-Mes/Stem A group (Fig. 1B). Hence, for this study, we regarded both Mes and Stem-A (MSA) as a high RS group, and other molecular subtypes such as Epi-A, Epi-B and Stem-B as a non-MSA (NMSA) lower RS group.

Fig. 1. MSA group have enriched replication stress and are sensitive to ATR inhibitors.

Box and whisker plots showing DNA replication stress response score of OC molecular subtypes in OC cell lines (A) and in OC tumor (B). On the right, same data shown in NMSA and MSA groups. Line/dot plots showing drug sensitivity profiles of NMSA cell lines (C, F) and MSA cell lines (D, G) for ATR inhibitors: elimusertib (C, D) and camonsertib (F, G). On the right, dot plots showing inhibitory concentrations (IC₅₀) of elimusertib (E) and camonsertib (H). Unpaired t-test with Welch’s correction is used for statistical analysis. For all statistical tests, *p < 0.05; **p < 0.01; ****p < 0.0001; ns not significant.

Augmented RS signatures observed in the MSA group led us hypothesize whether these tumors show relative sensitivity to RS pathway kinase inhibitors. Therefore, we assayed for anti-proliferative activity of two ATR inhibitors, elimusertib and camonsertib in our OC cell line panel of the NMSA group (JHOS2, PEO4, OVCAR8, OVCA420 and OVCA429) and MSA group (OVK18, A2780, OV17R, SKOV3, TYKNU, EFO21, DOV13 and HEYA8), and identified that these cell lines showed varying sensitivities to both drugs. Cell lines of the NMSA group showed lower sensitivity to elimusertib, where JHOS2 being the most resistant cell line with IC₅₀ of 571.6 nM (Fig. 1C, Supplementary Table S1). In the case of the MSA group, all the cell lines showed high sensitivity to elimusertib, with TYKNU being the most sensitive cell line having IC₅₀ of 13.2 nM (Fig. 1D, Supplementary Table S1). Similarly, in the case of camonsertib, cell lines of the NMSA group showed lower sensitivity with OVCAR8 being the least sensitive cell line with IC₅₀ of 346.8 nM (Fig. 1F, Supplementary Table S1) and cell lines of the MSA group showed high sensitivity with SKOV3 being the most sensitive cell line with IC₅₀ of 25.2 nM (Fig. 1G, Supplementary Table S1). Overall, cell lines of the MSA group showed a significantly increased sensitivity to elimusertib (Fig. 1E) and camonsertib (Fig. 1H) when compared to cell lines of the NMSA group. Interestingly, such selective chemosensitivity of the MSA group was not observed when treated with paclitaxel and carboplatin (Supplementary Fig. S1). Therefore, these results collectively suggest that the presence of high endogenous RS in Mes and Stem-A OC subtypes render selective sensitivity to ATR inhibitors targeting RSR pathway.

Conventional two-dimensional cell culture conditions have often shown translational limitations with variations in drug response, especially due to the lack of stromal components and dynamic tumor microenvironment. In particular, studies on drug screening and evaluating pharmacodynamics of anticancer drug candidates have demonstrated that 3D models tend to provide better comparability to clinical outcomes [25]. Therefore, we assessed the sensitivity of OC cell lines to elimusertib cultured on Matrigel. Our results show that OC cell lines of the MSA and NMSA groups formed 3D spheroids of comparable sizes when cultured on Matrigel (Fig. 2A). Subsequent treatment of 3D spheroids with elimusertib revealed that MSA cell line-derived spheroids showed significantly increased sensitivity in comparison to NMSA cell line-derived spheroids (Fig. 2B). Therefore, OC cell lines recapitulated the selective elimusertib sensitivity in 2D and 3D culture conditions highlighting that cell lines of the MSA group are more susceptible to ATR inhibition.

Fig. 2. MSA cell lines are sensitive to elimusertib in 3D.

A Images of cell-line-derived 3D spheroids of MSA (in blue) and NMSA (in red) cell lines. Spheroids were grown for 7–10 days followed by elimusertib treatment. Scale bar, 50 μm. B Line plot showing elimusertib sensitivity curves of MSA (in blue) and NMSA cell lines (in red) in 3D. On the right, dot plot showing MSA cell lines have significantly reduced IC₅₀ in comparison to the NMSA cell lines. Unpaired t-test with Welch’s correction is used for statistical analysis. **p < 0.01.

Activation of ATR-CHK1 axis in the MSA group

Since the MSA group of cell lines and patient tumors have shown high RS signature and increased sensitivity to ATR inhibitors (Fig. 1), we assessed basal levels of checkpoint kinases along the ATR–CHK1 axis [phosphorylated ATR (pATR), phosphorylated CHK1 (pCHK1), phosphorylated RPA (pRPA) and gamma-H2AX (γH2AX)]. Through immunoblot assays, we observed that the NMSA cell lines show reduced expression of pCHK1, pRPA and γH2AX levels in comparison to the MSA group (Fig. 3A). Quantification of protein expression showed that in spite of a non-significant enrichment of pATR and pCHK1 levels in the MSA cell lines, the expression of pRPA and γH2AX were significantly higher in the MSA group of cell lines (Fig. 3B). Next, we correlated the expression of pRPA and γH2AX to the elimusertib IC₅₀ of respective cell lines (Fig. 3C) and identified a significant correlation between high endogenous pRPA levels and elimusertib sensitivity (rho = −0.5175; p = 0.0444). Also, we observed a more significant correlation between endogenous γH2AX levels and elimusertib sensitivity (rho = −0.7413; p = 0.0039). These correlations further confirm that MSA cell lines with a lower IC₅₀ of elimusertib have higher endogenous pRPA/γH2AX expression. Since, γH2AX is fundamentally an established biomarker of DNA double-strand breaks (DSBs) [26], we performed comet assay to determine the underlying basal genomic instability. Our results from comet assay indicate that cell lines of the MSA panel had significantly higher levels of DNA in comet tail when compared to cell lines of the NMSA panel (Fig. 3D). These results suggest that in the MSA group, cells are either driven by high endogenous RSR potentially mediated through activated pRPA levels or fueled by DSBs as evidenced through elevated γH2AX levels and the percentage of DNA in the comet tail, preferentially rendering sensitivity to RSR targeting agents such as elimusertib.

Fig. 3. MSA cell lines have high levels of endogenous replication stress and DNA damage.

A Immunoblots showing basal expression of pATR, pCHK1, pRPA and γH2AX in NMSA and MSA OC cell lines. SKOV3 cells treated with 8 mM hydroxyurea for 4 h is used as a control sample. Tubulin is used as a loading control. B Dot plots showing normalized indicated protein levels (densitometric values, quantified through ImageJ) were compared between NMSA and MSA cell lines. Unpaired t-test with Welch’s correction is used for statistical analysis. ns, non-significant; *p < 0.05; ***p < 0.001. C Dot plots showing correlations between elimusertib IC50 (y-axes) and densitometric values [x-axes, arbitrary units (a.u.)] of pRPA (left) and γH2AX (right). Pearson correlation coefficients (r) and P values (p) are denoted. D Representative images of tail moment from Comet assays of the NMSA (in red) and MSA (in blue) cell lines. Scale bar, 50 μm. On the right, dot plot showing Comet tail DNA level (in %) for NMSA (red) and MSA (blue) cell lines. Two-tailed nested t-test is used for statistical analysis. **p < 0.01.

MSA group display altered replication fork dynamics and robust accumulation of single-stranded DNA

The observations of elevated endogenous RS factors in the MSA cell line panel prompted us to further investigate the replication fork dynamics at single-molecule resolution, using a DNA fiber assay as outlined in Fig. 4A. We selected cell lines from the MSA group (A2780, TYKNU, OVCAR10, OVK18) and the NMSA group (OVCA429, OVCA420, OVCAR8) for the DNA fiber labeling experiments. Results from a colony formation assay reinforced the conclusion that MSA cell lines are highly sensitive to ATR inhibition (Supplementary Fig. S2A). To study replication fork dynamics, we examined the extent of origin firing and relative fork speeds in these representative MSA and NMSA cell lines. It is known that fork rate and origin firing are interdependent variables that influence cellular outcomes on RS and genomic stability [27]. Excessive origin firing is known to be accompanied by a reduction in fork speed [28], whereas, molecular alterations that slows down replication forks activates the firing of “dormant” origins in a compensatory response to ensure the timely replication of the entire genome [29]. The crucial regulation of replication fork speeds is exemplified in cancers harboring different genetic perturbation driving high replication fork speeds and genomic instability [30–33]. To determine the percentage of replication forks stochastically fired, cells were pulse-labeled with CldU for 15 min, followed by IdU for another 15 min. New replication origins are determined by incorporation of the second label (IdU) (Fig. 4A). We demonstrated that newly fired replication origins are significantly elevated in the MSA cell lines compared to the NMSA cell lines (Fig. 4B, C, Supplementary Fig. S2B). Quantification of the percentages of new replication origins revealed a consistent elevation in MSA lines when compared to the NMSA cell lines (Fig. 4C, Supplementary Fig. S2B). Intriguingly, we observed that fork speeds were increased in the MSA cell lines relative to the NMSA cell lines (Fig. 4D, Supplementary Fig. S2C), despite a higher percentage of origin firing observed in the MSA cell lines (Fig. 4C), highlighting potential uncoupling between origin firing and the rate of replication fork progression, contributing to increased RS.

Fig. 4. MSA cell lines show increased replication fork dynamics.

A Schematic of DNA fiber labeling assay. Single-molecule DNA fiber analyses using nucleotide analogs (CldU and IdU) can evaluate an array of replication forks parameters such as the frequency of newly fired origins and replication fork speed. Created in BioRender. B Representative images of DNA fibers of NMSA (OVCA429, OVCA420, OVCAR8) and MSA (A2780, TYKNU, OVCAR10) cell lines. White arrows on DNA fibers in MSA cell lines denote origin firing. C Percentage of new origin firing (green fiber only) quantified after sequential labeling of DNA by CldU and IdU, respectively, for 15 min. Bar graph shows significant difference in origin firing in both NMSA and MSA cell lines (n > 100). D Replication tracts length in both NMSA and MSA cell lines were measured, and fork speed was calculated as mentioned in methods and plotted (n > 100). E Immunoblots showing basal expression of AMBRA1 and p21 in NMSA and MSA OC cell lines. SKOV3 cells treated with 8 mM hydroxyurea for 4 h is used as a control sample. Tubulin is used as a loading control. Single-stranded DNA (ssDNA) levels were detected in OVCA429 (NMSA) and A2780 (MSA) cell line pair (F) and OVCA420 (NMSA) and TYKNU (MSA) cell line pair (G) with control and elimusertib (2 µM) treatment. Representative images on the left show BrdU localization, while dot plots on the right display fluorescence intensity of ssDNA with (+) and without (−) treatment (n > 300). Statistical significance between two samples was determined using Student’s unpaired t test or Mann–Whitney U test. ***p < 0.001; ****p < 0.0001.

Intriguingly, this observed disparity in replication fork kinetics across OC cell lines prompted us to investigate into the molecular determinants involved. Recent studies elucidate that the cyclin-dependent kinase inhibitor p21 (also known as p21^CDKN1A) is essential for maintaining normal replication fork speed by regulating the recruitment of DNA polymerases; consequently, the loss of p21 expression leads to aberrant fork acceleration [33, 34]. Furthermore, AMBRA1 (Autophagy and Beclin 1 Regulator 1) serves as the linking substrate receptor within the CRL4 (Cullin-RING Ligase 4) E3 ubiquitin ligase complex, directing the ubiquitylation and subsequent turnover of D-type cyclins [35–37]. Consequently, the loss of AMBRA1 leads to chronic RS, rendering these cells highly dependent on the CHK1-mediated S-phase checkpoint for viability [36]. Our assessment of AMBRA1 and p21 expression in our panel of OC cell lines revealed that cell lines of the MSA subtype, which exhibits high fork velocities, uniformly lacked expression of AMBRA1 and p21, with the exception of trace levels of p21 in HEYC2 and HEYA8 (Fig. 4E). In contrast, NMSA subtype cell lines (JHOS2, CAOV3) showed basal AMBRA1 expression, while others (PEO4, OVCA420, UWB1.289) expressed p21. These data suggest that the presence of these proteins may optimally regulate fork progression and prevent the aberrant acceleration observed in NMSA cell lines.

Expanding on the extent of RS observed in MSA cells (Fig. 3), we next evaluated OC cell lines for the presence of single-stranded DNA (ssDNA). We utilized an immunofluorescence-based approach involving 5-bromo-2’-deoxyuridine (BrdU) pulse-labeling followed by detection under non-denaturing conditions. By omitting the DNA denaturation step, we ensured the anti-BrdU antibody specifically recognized exposed ssDNA regions, a hallmark of genomic instability, while remaining inaccessible to intact double-stranded DNA [38]. For comparison, we also included cells treated with elimusertib to induce extensive ssDNA foci formation. Fluorescence microscopic images show that elimusertib treatment in NMSA and MSA cell lines induced a significant increase in BrdU fluorescence intensity when compared to the respective untreated cells (Fig. 4F, G). Interestingly, we observed a significant enrichment of BrdU single-stranded DNA signals in MSA cell lines A2780 (Fig. 4F) and TYKNU (Fig. 4G) at untreated conditions in comparison to that of the NMSA cell lines, highlighting the presence increased ssDNA endogenously. Collectively, these results clearly demonstrate that the MSA group exhibits high levels of RS, as evidenced by altered replication fork dynamics and excessive accumulation of single-stranded DNA, which corroborates their selective sensitivity to ATR inhibitors.

MSA signature in elimusertib-responsive PDXs

To evaluate the selective sensitivity of elimusertib in vivo, we generated ten ovarian cancer patient-derived xenografts (PDXs) by subcutaneously injecting freshly isolated tumor fragments in the flank of interleukin-2 receptor-γ-null (NOD-SCID-IL-2rγ) mice (Fig. 5A). Following randomized grouping, mice were treated with elimusertib (40 mg/kg). All PDXs treated with elimusertib showed reduced mean tumor weight in comparison to vehicle control-treated group (Fig. 5B). Following tumor harvest, we determined gene expression profiles of the PDXs through microarray and utilized the ovarian subtype signatures derived [15] to assign the molecular subtype to each PDX. Subsequently, tumor response to elimusertib was determined based on Treatment/Control (T/C) ratios of tumor volume at cull. By comparing the elimusertib T/C percentage of PDXs with a poor prognosis-associated subtype (MSA) versus that of PDXs with a better prognosis-associated subtype (NMSA), a significant difference in elimusertib sensitivity (t-test, p = 0.0138) was observed between the two groups, with MSA PDXs showing higher sensitivity to elimusertib than the NMSA PDXs (Fig. 5C). Although all established PDX models demonstrated measurable tumor growth inhibition following elimusertib treatment, the magnitude of response varied considerably. Given that PDX engraftment may enrich for more proliferative, replication-stressed tumor-cell populations, this cohort may be biased toward tumors with intrinsic susceptibility to ATR pathway targeting. Therefore, our analysis focuses on relative differences in sensitivity rather than a strict sensitive versus resistant categorization.

Fig. 5. PDXs responding to elimusertib have MSA signature.

A Schematic of patient-derived xenografts model. Created in BioRender. B Data plot showing mean tumor weight derived from OC patient-derived xenografts (PDXs) treated with vehicle control (VC) and elimusertib. C Data plot showing treatment/control response (T/C ratio) to elimusertib (y-axis) in NMSA and MSA signature carrying PDXs.

Discussion

In this study, through two-, three-dimensional cell culture models, patient-derived xenografts and single molecular analyses, we have assessed DNA RSR on the molecular subtypes of HGSOC have unveiled three crucial findings. Firstly, we observed an enrichment of RSR pathway (ATR-CHK1 axis) in the MSA cell lines. Secondly, MSA cells show increased replication fork dynamics that render increased RS. Thirdly, patient-derived tumors and cell lines with the MSA signature are selectively sensitive to ATR inhibitors.

Within the expanding field of DDR-targeted treatments, ATR inhibitors represent a critical therapeutic frontier for ovarian cancer (OC), particularly for patients who have progressed on PARP inhibitors (PARPi). Notable findings from the CAPRI trial (NCT03462342) showed that ceralasertib plus olaparib achieved an objective response rate (ORR) of 48.5% in platinum-sensitive recurrent ovarian cancer, suggesting that ATR inhibition can restore sensitivity to PARPi independent of baseline HRD status [12, 39, 40]. This clinical utility is further bolstered by the phase II ATARI trial (NCT04065269), which exploited the synthetic lethal relationship between ARID1A loss and ceralasertib to achieve a clinical benefit rate of 47% in treatment-resistant gynecological cancers [41]. Furthermore, the MYTHIC trial (NCT04855656) is currently evaluating camonsertib plus the PKMYT1 inhibitor lunresertib, showing significant activity in biomarker-selected (e.g., CCNE1-amplified) platinum-resistant ovarian and endometrial cohorts [42]. Additionally, phase II data (NCT02595892) confirmed that adding berzosertib to gemcitabine significantly improved median progression-free survival compared to gemcitabine alone (22.9 vs. 14.7 weeks; HR 0.57) [43]. Collectively, these advancements underscore the potential of ATR inhibitors to redefine treatment protocols for ovarian cancer by addressing significant unmet needs in both PARP-inhibitor-resistant and platinum-refractory settings.

Mes subtype tumors have enriched EMT signatures and fibrinolysis pathway upregulation, whereas Stem-A subtype tumors are enriched with a chromatin modification signature [15]. Using non-transformed human mammary epithelial cells and patient-derived cancer stem cells, McGrail et al. have demonstrated that cancer stemness is associated with high RSR defects [23]. Additionally, such heightened RSR defects in other non-hypermutated cancers such as breast, brain, prostate and kidney show better response treatment with immune checkpoint blockade, emphasizing that the assessment of RS would be crucial for treatment modalities [44]. In breast cancer, localization of stem cell self-renewal factor BMI1 to stalled replication fork has been shown to be crucial for RAD51-dependent response to replication stress [45]. Sub-populations of tumor cells (from multiple cancer types) with high expression of EMT transcription factor ZEB1 induces DNA replication stress, elucidated by increased replication rate accompanied by higher γH2AX as well as pRPA and RAD51 foci in S phase [46]. Recently, we have shown a direct correlation between the enrichment of nuclear phosphorylated CHK1 levels and increased sensitivity to ATR inhibitor elimusertib [47]. Incidentally, in this study, we observed high expressions of pRPA and γH2AX levels in MSA group of cell lines. Additionally, a dynamic crosstalk between DNA damage response (elicited by both single and double-stranded DNA breaks) and epithelial-mesenchymal transition signaling network in circulating tumor cells of breast cancer has also been described [48]. Hence, while the underlying mechanism(s) for substantially high levels of RS in MSA tumors remains unclear, these recent data suggest that the EMT signature in MSA subtypes may directly lead to increased RS. While our data provides a comprehensive landscape of how baseline RS markers and MSA gene signature align with therapeutic response. Future functional studies such as CRISPR-mediated knockout of key RSR genes or the inducible overexpression of subtype-specific drivers will be essential to definitively establish a causal link.

One of the potential outcomes of this study is the use of molecular subtype stratification in screening HGSOC patients for targeted therapeutic intervention. In a previous evaluation of transcriptomic intra-tumor heterogeneity on 3,431 ovarian cancer samples, we showed that the majority of recurrent or relapsed tumor samples had enrichment or clonal conversion to MSA subtype irrespective of the primary tumor subtype [24]. Another study using patient-matched samples of primary, metastatic and recurrent high-grade serous carcinoma has demonstrated that upper abdominal/omental metastases harbour a Mes subtype, irrespective of the primary OC subtype [49]. Furthermore, an evaluation of data from 329 HGSOC patients revealed that the Mes subtype serves as an independent predictor of severe postoperative complications [50]. These studies further highlight the significance of the MSA signature in the context of metastatic and recurrent ovarian cancers associated with poorer outcomes. It is important to note that the transcriptional molecular subtypes used in this study, are derived from the bulk tumor biopsy and thus represent a composite signature of both tumor-intrinsic cell state and the extrinsic influence of the tumor microenvironment (TME). For example, Mes (MSA) and Epi-B (NMSA) show significant enrichment for stromal and immune signatures, respectively; MSA subtype show enrichment of angiogenesis. Therefore, our findings reflect biologically and clinically distinct disease entities that are defined not only by the tumor-cell intrinsic characteristics, but also its TME. Furthermore, our study suggests that the MSA signature could potentially be used to better select patients for treatment with ATR inhibitors, which, given the known toxicities associated with use of ATR inhibitors [51–53], could improve their risk-benefit ratio. However, the main caveat for such an effort would be greatly dependent on the source of the tumor (primary vs. metastasis) used for stratification. Recent studies have similarly demonstrated functional association between OC molecular subtypes and drug response. Preclinical studies on targeting the Gas6/AXL pathway, non-canonical Wnt/PCP pathway and microtubule inhibitor have shown selective sensitivity in MSA subtypes [15, 54–56]. In a randomized controlled phase III trial of primary ovarian cancer, bevacizumab selectively improves outcomes in patients of Stem-A (proliferative) subtype [57]. Collectively, these studies indicate that molecular subtype-assisted stratification could be an appropriate biomarker-based therapeutic strategy that warrants further investigation in the clinic.

Through single-molecule DNA fiber assay, we have shown that MSA subtype have increased endogenous replication fork dynamics, which was correlated with increased sensitivity to ATR inhibitors in our study. Excessive replication fork speed has been described as a causal link for inducing RS and DNA damage response [33]. Another study in neuroblastoma has reported that increased RS through high MYCN expression results in sensitivity to ATR inhibition [58]. Furthermore, testing of VE-822 (ATR inhibitor) and prexasertib (CHK1 inhibitor) on ovarian tumor-derived organoids revealed a direct association between replication fork instability and sensitivity to both aforementioned drugs [59]. While RS is traditionally associated with fork stalling, our data reveal that in specific ovarian cancer subsets, stress emerges from unregulated fork acceleration. This ‘fast-fork’ phenotype is mechanistically supported by our observation of reduced p21 and AMBRA1 levels especially in the MSA subtype. As established by Maya-Mendoza et al., p21 serves as a critical rheostat for replication rate, while AMBRA1 prevents RS by maintaining Cyclin D stability and preventing premature S-phase progression [33, 36]. The loss of these homeostatic controls in MSA subtype tumors forces a dependency on the ATR-mediated DNA damage response to manage the resulting genomic instability, thereby creating a predictable yet potent therapeutic window for ATR inhibitors.

In summary, we have shown that heightened RS in MSA group of HGSOC render increased sensitivity to ATR inhibitors, which suggests that a molecularly stratified approach to targeting ATR, could lead to improved outcomes in patients with HGSOC.

Author contributions

DSPT conceived, administered and supervised the project. VS, FZ, GAE, TZT, KTK, AT, and CL designed and performed experiments. AMW, HTH, MQ, and CWY conducted and supervised in vivo experiments. VS, GAE, FZ, TZT, CFC, and DSPT investigated, visualized and analyzed data. CFC and DSPT acquired the funding. VS wrote the manuscript. VS, TZT, GAE, CFC, and DSPT performed manuscript review and editing. All the authors have read and approved the final manuscript.

Funding

This research was supported by the National Research Foundation Singapore and the Singapore Ministry of Education under its Research Centres of Excellence initiative. This work was supported by funding from the National Medical Research Council (NMRC) Clinician Scientist Award Senior Investigator Grant (CSASI21jun-0003), the Pangestu Family Foundation Gynaecological Cancer Research Fund. This work was supported by the National University Cancer Institute of Singapore (NCIS) and the NUS Centre for Cancer Research (N2CR) seed grant (NR22NMR112OM., DSPT & CFC).

Competing interests

DSPT received research funding from AstraZeneca, Bayer, and Karyopharm and honoraria from AstraZeneca, MSD, Tessa Therapeutics, Novartis, Bayer, and Genmab. AMW was a permanent employee of BAYER AG at the time when the studies were performed.

Ethics approval and consent to participate

Statement for animal work: All in vivo xenograft studies adhered and performed in accordance to the SingHealth Animal Care and Use Committee (IACUC:2018/SHS/1356) guidelines on animal use and handling. The approval has been obtained from the committee and all methods were performed in accordance with the relevant guidelines and regulations. Statement for patient material: Usage of patient-derived materials has been approved by the National Health Group Domain Specific Review Board (DSRB), Reference: 2013/00705. Written informed consent was obtained from each patient. All methods were performed in accordance with the relevant guidelines and regulations.

Supplementary information

The online version contains supplementary material available at 10.1038/s41388-026-03830-9.