Proapoptotic activity of JNK-sensitive BH3-only proteins underpins ovarian cancer response to replication checkpoint inhibitors

Abstract

Recent studies indicate that replication checkpoint modulators (RCMs) such as inhibitors of CHK1, ATR, and WEE1 have promising monotherapy activity in solid tumors, including platinum-resistant high grade serous ovarian cancer (HGSOC). However, clinical response rates are generally below 30%. While RCM-induced DNA damage has been extensively examined in preclinical and clinical studies, the link between replication checkpoint interruption and tumor shrinkage remains incompletely understood. Here we utilized HGSOC cell lines and patient-derived xenografts (PDXs) to study events leading from RCM treatment to ovarian cancer cell death. These studies show that RCMs increase CDC25A levels and CDK2 signaling in vitro, leading to dysregulated cell cycle progression and increased replication stress in HGSOC cell lines independent of homologous recombination status. These events lead to sequential activation of JNK and multiple BH3-only proteins, including BCL2L11/BIM, BBC3/PUMA and the BMF, all of which are required to fully initiate RCM-induced apoptosis. Activation of the same signaling pathway occurs in HGSOC PDXs that are resistant to poly(ADP-ribose) polymerase inhibitors but respond to RCMs ex vivo with a decrease in cell number in 3-dimensional culture and in vivo with xenograft shrinkage or a significantly diminished growth rate. These findings identify key cell death-initiating events that link replication checkpoint inhibition to antitumor response in ovarian cancer.

Introduction

Ovarian, fallopian tube, and primary peritoneal cancers (termed “ovarian cancer” herein) constitute a heterogeneous group of neoplasms that annually affect 22,000 patients in the U.S. and have the highest mortality of any gynecological cancer [1]. Among these neoplasms, high-grade serous ovarian cancer (HGSOC) is the most lethal [2], accounting for over 200,000 cancer deaths per year worldwide [3]. Approximately 50% of HGSOCs are homologous recombination (HR) deficient (HRD) due to silencing or inactivation of HR repair genes [4]. Importantly, the HRD subset shows a better response to platinum-based chemotherapy. Moreover, addition of PARP inhibitor (PARPi) therapy has extended the duration of progression-free survival in HGSOC, particularly in cases with BRCA1 or BRCA2 mutations, reflecting the synthetic lethality of BRCA1/2 loss and PARP1 inhibition [5,6,7]. Nonetheless, even within this subset that derives the most benefit from PARPi therapy the majority of ovarian cancer patients relapse and die of their disease [8], highlighting the need for improved treatment.

Comprehensive characterization of ovarian cancers has shown that replication stress (RS), like HR deficiency, is widespread in HGSOC, creating another important vulnerability, albeit one that is not directly actionable [9, 10]. Moreover, a variety of studies have shown that treatment with inhibitors of replication checkpoint kinases, including ATR, CHK1, and WEE1, increase this RS, enhancing the effects of chemotherapies that induce replication stress or leading to cell death on their own [11, 12]. Consistent with these preclinical studies, promising clinical activity of combinations containing the current generation of replication checkpoint modulators (RCMs) has been observed in advanced solid tumors, including ovarian cancer [13,14,15,16]. Moreover, evidence of single agent activity has also been observed [17, 18]. Importantly, phase II trials of the CHK1i prexasertib in platinum-resistant or -refractory HGSOC demonstrated responses in cases that recurred after prior PARPi, suggesting that prexasertib might be beneficial in PARPi resistant disease [12, 18, 19]. Additionally, the WEE1i adavosertib has also been reported to have substantial activity in PARPi-resistant ovarian cancer, particularly in combination with olaparib (ClinicalTrials.gov identifier NCT03579316).

Because the efficacy of ATR, CHK1, and WEE1 inhibitor monotherapy in advanced ovarian cancer is modest, ranging from 6–30% in various studies [18, 20, 21], the identification of biomarkers that would allow selection of patients most likely to respond to RCMs would be beneficial. To this end, a variety of potential biomarkers of RCM sensitivity have been proposed, including HR deficiency, ATM inactivation, CCNE1 and/or MYC amplification, and biochemical evidence of RS as manifested by phosphorylation of CHK1, H2AX or RPA32 [12, 13, 22].

While previous efforts to identify predictive biomarkers for RCM response have focused on events proximate to replication fork disturbance, a large number of steps undoubtedly occur between enhancement of RS by RCMs and cancer cell death. Relatively little is known about these intervening steps. Although a recent study in acute monocytic leukemia cells indicated that RCMs kill through a pathway involving activation of the stress kinase JNK leading to JUN/FOS-mediated TNF gene activation and subsequent TNFα-induced triggering of the extrinsic apoptotic pathway [23], TNF-mediated processes were able to account for only part of the leukemic cell death, suggesting that a second pathway might contribute. In addition, it also has been unclear whether the same extrinsic apoptotic pathway contributes to RCM activity in solid tumors.

Here we examined the association between previously identified biomarkers of RS and response of a series of HGSOC patient-derived xenografts (PDXs) to paradigmatic inhibitors of CHK1 (prexasertib), ATR (ceralasertib), and WEE1 (adavosertib) ex vivo and in vivo. Because models lacking these RS biomarkers also responded to the RCMs, we additionally performed an in-depth analysis of steps leading to RCM-induced HGSOC cell death that implicated the intrinsic apoptotic pathway and identified an important role for BMF, a BH3-only protein previously implicated in detachment-induced apoptosis [24, 25], in RCM-induced killing of solid tumor cells. Collectively, this analysis identified key signaling steps contributing to RCM-induced cell death that have the potential to modulate RCM sensitivity or resistance.

Methods

Materials

All reagents and antibodies are listed in Supplemental Table S1.

Cell lines

Sources of cell lines are indicated in Supplemental Table S1. Cell lines were cultured in the following media: OVCAR8, OVCAR5, COV362, OV90, and IGROV1 cells in RPMI 1640 medium; CaOV3, PEO1, PEO1-derived PARP inhibitor resistant clones #2 and #3, and PEO4 cells in DMEM medium with high glucose supplemented with 10 μg/ml insulin, and 0.04 mM Dulbecco’s nonessential amino acids; and SKOV3 cells in McCoy’s 5A medium. Of these lines, SKOV3 has been classified as unlikely to be HGSOC based on genomic features and IGROV1 has been reported to be hypermutated [26]. All media contained 10% (vol/vol) heat-inactivated fetal bovine serum, 50 units/mL penicillin G, 50 μg/mL streptomycin, and 1 mM glutamine. All cell lines were authenticated by short tandem repeat profiling in the Mayo Clinic Cytogenetics Core Facility. In addition, PEO1 and PEO4 cells were confirmed to have their reported BRCA2 mutations by Sanger sequencing [27]. Additionally, all cell lines were tested with the MycoAlert Mycoplasma Detection Kit (Lonza) and found to be mycoplasma-free.

To generate FADD^−/− cells, oligonucleotides guiding to human FADD 228–237 (NM_003824.3) were synthesized, annealed, and cloned into the BsmBI site of lentiCRISPR-v2 plasmid (Addgene, Cambridge, MA). FADD targeting virus and empty vector (EV) were packaged by transfecting HEK293T cells with the packaging vector psPAX3, envelope vector pMD2.G, and lentiCRISPR-v2-FADD or EV using Lipofectamine 2000 (ThermoFisher, Waltham, MA). Two days after viral transduction, PE01 cells were selected with 3 μg/ml puromycin for a week, cloned by limiting dilution in 96-well plates, and assayed for gene interruption by immunoblotting. A similar process was used to generate BID^−/− cells and BAX^−/−/BAK^−/− (DKO) cells using oligos that target human BID 591–610 (NM_197966.2), human BAK 162–184 (AF520590.1), and human BAX 179–201 (NM_001291428.1).

For transient transfections, aliquots containing ~ 1 × 10⁷ cells were transfected with 40 µM siRNA and subjected to electroporation using a BTX830 square wave electroporator (BTX, San Diego, CA) delivering two 10-ms pulses at 280 V and harvested for immunoblotting or assayed for drug sensitivity as indicated in the individual figure legends.

Colony forming assays

For clonogenic assays, cells were plated for 12–16 h before drugs were added and monitored continuously for 10–14 days until colonies of ~ 50 cells formed in control wells. Plates were then stained with Coomassie blue so that visible colonies could be counted. Survival was calculated as the ratio of colonies treated with the drug compared with diluent.

RAD51 foci

Assays for irradiation-induced RAD51 foci were performed as previously described [28]. Cells plated on coverslips were irradiated (10 Gy) and incubated at 37 °C for 6 h before fixation in 4% (wt/vol) paraformaldehyde. Coverslips were incubated with primary antibody (1:500) in blocking buffer overnight at 4 °C. Coverslips were then washed 6 times with wash buffer (PBS containing 0.1% Triton X-100 and 0.1% bovine serum albumin), incubated with secondary antibody (Alexa fluor 488 goat anti-rabbit IgG and Alexa fluor 568 goat anti-mouse IgG, each at 1:1000) in blocking buffer for 1 h at 21 °C in the dark, washed, and counterstained with 1 µg/ml Hoechst 33258 in PBS. Samples were examined on a Zeiss Axiovert microscope with 63X/NA 1.4 lens and photographed using a Zeiss Axiocam MRm CCD camera using Zeiss Zen software. RAD51 foci were quantified manually in a blinded fashion in ≥ 100 cells per slide and considered positive if ≥ 10 Rad51 foci were visible.

Analysis of PDX generation and drug testing ex vivo

Patient-derived xenografts (PDXs) were generated in 6- to 7-week-old female SCID beige (C.B.17/IcrHsd-Prkdc^scid Lyst^bg−J; Envigo RMS, LLC, Indianapolis IN) mice as previously described in detail [28, 29]. All procedures were conducted in accordance with Animal Welfare Regulations and were approved by the Institutional Animal Care and Use Committee (IACUC) at the Mayo Clinic (A37615).

Sensitivity of these PDXs to various agents ex vivo was assessed as described [30]. In brief, PDXs were harvested from mice, minced into 2- to 4-mm pieces, and dissociated using a gentleMACS cell dissociator (Miltenyi Biotech) following the supplier’s protocol. Single-cell suspensions were then treated with the indicated drugs and plated in 24-well plates. Following 72 h or 5 days, cell viability was measured using CellTiter-Glo 3D (Promega). Cell viability was calculated for each concentration as an average of seven replicates and normalized to untreated vehicle controls.

Effects of WEE1i and ATRi in vivo

For all PDX studies, cryogenically preserved human ovarian cancer tumors were rapidly thawed and reestablished as previously described [29]. Four individual HGSOC models (PH061, PH063, PH0235 and PH354) were chosen based on previous PARP inhibitor response. Briefly, 0.1–0.2 cc of the minced tumor was prepared in 1:1 ratio with McCoy’s 5A Modified Medium (Cat # MT-10–050-CV, Corning Life Science) before intraperitoneal injection. PDX engraftment was followed by abdominal palpation and transabdominal ultrasound using a SonoSite S-Series ultrasound machine (Fujifilm SoniSite, Inc). Tumor diameter was assessed using the on-board SonoSite software real-time measurement tool and cross-sectional area was measured on exported images using ImageJ software after calibration as described.

When intraperitoneal tumor area reached 0.5–1 cm², mice were randomized to treatment arms. Olaparib was given by daily oral gavage in 15% methylcellulose (50 mg/kg as monotherapy). Ceralasertib 35 mg/kg or adavosertib 80 mg/kg was administered 3 days/week by oral gavage. The largest tumor cross-sectional area was measured twice weekly through day 28. Mice were euthanized individually when moribund or as a cohort at day 28. The primary endpoint was tumor area by ultrasound, normalized to the day 1 area of the same tumor and plotted as a ratio vs. time.

Statistics and reproducibility

Dose–response curves were performed in cell lines at least three times independently. Error bars in all plots represent mean ± sem of 3 independent experiments. *, ** and *** indicate p < 0.05, p < 0.01 and p < 0.001, respectively, using unpaired t tests. Relative IC₅₀ values were analyzed in Microsoft® Excel by linear regression. Where indicated, analysis of variance (ANOVA, one-way or two-way as appropriate) with post-hoc correction for multiple comparisons was performed using GraphPad Prism 9 software. Immunoblots were performed in indicated cell lines at least two times independently.

Additional methods

Methods for detection of apoptosis, cell fractionation, immunoblotting, qRT-PCR, RNAseq, and quantitative analysis of PDX growth ex vivo and in vivo have been previously published and are described in the Supplementary Methods.

Results

RCM monotherapy activity against PDXs ex vivo

To identify potential PDXs for in vivo testing, we first established short-term 3D cultures from 18 distinct PDXs with varied genomic backgrounds for testing ex vivo (Supplementary Table S2). These PDXs were all of the serous subtype, with 14 of the models established from primary tumor (Supplementary Table S3). Studies confirmed that PDX response to prexasertib was similar in snap frozen and fresh tissue (Supplementary Fig. S1A). Killing by prexasertib, ceralasertib, and adavosertib ex vivo varied across the models, with comparable median responses of 36% death with prexasertib, 34% with ceralasertib and 35% with adavosertib at 5 days (Fig. 1A-D), which is modest but would be considered a partial response in the clinical setting. Of the 18 PDXs, 10 had been previously assessed by clinical HRD tests. In these ovarian cancer PDXs treated ex vivo, response to RCM monotherapy did not correlate with HRD score (Fig. 1E).

Fig. 1

Analysis of RCM sensitivity in HGSOC PDXs ex vivo. A-D, Single-cell suspensions isolated from the indicated PDXs were plated and allowed to form 3D spheroids for 24 h. Cells were then treated with prexasertib, ceralasertib, or adavosertib for 72 h. Cell viability was assayed using RealTime-Glo™ MT Cell Viability Assay. Error bars, ± sem of three independent experiments. E, Association between % survival at 5 µM prexasertib (left, p = -0.02478) or ceralasertib and adavosertib (right, p = -0.4556 and 0.5437, respectively) with HRD score using Spearman correlation. F, Indicated protein levels across ovarian carcinoma PDX [ordered by % survival at 5 µM to prexasertib (Table S2), with most sensitive toward the left] with serial dilutions of Ovcar5 cells. G, PDXs that were tested for sensitivity to ceralasertib and adavosertib but not prexasertib (ordered by % survival at 5 µM to ceralasertib, with most sensitive on the right)

In subsequent experiments, several molecular features of stressed replication forks were examined (Fig. 1F and G). RS at baseline as manifested by high γH2AX and phospho-Ser^4/8-RPA32 was observed in all PDXs and did not correlate with response ex vivo. There was, however, a trend toward lower CHK1 levels in models that were more sensitive to the CHK1i prexasertib, although this did reach statistical significance (Fig. S1B, 𝜌 = 0.566, p = 0.059).

WEE1i-induced tumor regressions independent of CCNE or MYC amplification

Focal amplifications involving the CCNE1 and MYC loci, which are common in HGSOC, are harbingers of primary treatment failure and poor overall survival after standard-of-care platinum-based therapy [31, 32] but have been suggested to be associated with RCM sensitivity in previous studies [33,34,35]. To test this in vivo, four ovarian cancer PDX models that displayed moderate responses to CHK1, ATR, and WEE1 inhibitors ex vivo were treated with RCMs. All had TP53 mutations and were known to be PARPi-resistant (PARPi^res), with three models wildtype for BRCA1 and BRCA2 and one having a BRCA2 variant. Two of these PDXs (PH354 and PH235) were characterized as CCNE and MYC amplified (Table S2, highlighted in blue), with higher CCNE1 and MYC expression also detected by immunoblotting (Fig. S2A).

When mice bearing orthotopic xenografts were treated with diluent, the PARPi olaparib, the ATRi ceralasertib, or the WEE1i adavosertib over a 28-day period, treatment was well-tolerated, with no dose reductions required (Fig. S2B). Weekly transabdominal ultrasound indicated that none of the xenografts shrank below baseline in the olaparib arm (Fig. 2A-D), confirming that they are PARPi resistant by previously described response criteria [28].

Fig. 2

Response of HGSOC PDXs in vivo. Ceralasertib, adavosertib or olaparib were administered to mice bearing orthotopic HGSOC PDX models PH354 (A), PH235 (B), PH061 (C), or PH063 (D) with tumor growth measured by transabominal ultrasound. Following 4 weeks of treatment, tumor growth was observed for a minimum of 60 days unless the tumor met euthanasia criteria. The number of mice under observation is indicated below the x-axis, where text color indicates treatment arm. Shaded areas, 95% confidence intervals; pairwise comparisons for treatment arms are *, *** and ****: p < 0.05, p < 0.001, and p < 0.0001, respectively, for indicated global p value comparisons

Importantly, some of these PARPi-resistant PDXs responded to RCMs. In particular, adavosertib induced tumor shrinkage and prolonged survival in PH354 (CCNE^amp MYC^amp) and PH061 (diploid for CCNE1 and MYC), with more extensive shrinkage in PH354 than in PH061 (Fig. 2A, B and S2C). In contrast, neither the CCNE^amp MYC^amp PDX PH235 nor the nonamplified PDX PH063 showed shrinkage below baseline during adavosertib treatment, although statistically significant growth inhibition was observed (Fig. 2C, D and S2C; Supplementary Table S4).

Overall, ceralasertib was less efficacious across all models at the dose administered. The best response, which was prolonged disease stabilization, was observed in the CCNE^amp MYC^amp PDX PH354, with some blunting of growth in PH063 (Fig. 2A and D) and little effect in CCNE^amp MYC^amp PH235 or nonamplified PH061 (Fig. 2B and C and Supplementary Table S4).

Because the lack of apparent association between RCM response and CCNE or MYC amplification differed from reported results in other preclinical studies, we also analyzed the association of HGSOC PDX sensitivity to prexasertib ex vivo (Fig. 1A-E) with expression of CCNE1 and MYC (Fig. S2D and E). As indicated in Fig. S2F, MYC levels were lower in sensitive PDXs. These results suggested that factors in addition to CCNE1 and/or MYC overexpression might be needed to predict HGSOC response to RCMs, prompting further investigation RCM-induced signaling.

RCM sensitivity of ovarian cancer cell lines

As a starting point for these further studies, we assessed the clonogenic response of a panel of 11 ovarian cancer cell lines or clones, including nine of HGSOC origin, to prexasertib, ceralasertib, and adavosertib (Fig. 3A) [26, 36, 37]. Mean prexasertib IC₅₀ values (Fig. 3B) ranged from 1.9 nM (CaOV3) to 12 nM (OVCAR5). Likewise, adavosertib and ceralasertib sensitivities varied across these cell lines (Fig. 3B).

Fig. 3

Effect of RCMs on HR deficient and HR proficient ovarian cancer cell lines. A, Cells were treated continuously with CHK1i (prexasertib), ATRi (ceralasertib) or WEE1i (adavosertib) in a clonogenic assay. Results are expressed as a percentage of colonies formed relative to cells treated with diluent alone. Error bars, ± sem from three or more independent experiments. B, Relative IC₅₀s were estimated by linear regression from data in panel A. * indicates cell lines with a deleterious BRCA1 or BRCA2 mutation. C, To define HR status, cells were subjected to 10 Gy ionizing radiation, incubated for 6 h, fixed, co-stained for nuclear RAD51 and γH2AX, and imaged (100 cells/treatment). Scale bar = 5 µm D, Comparison of prexasertib IC₅₀ in cells with HRD (n = 3) versus HR proficiency (n = 8) using Wilcoxon rank sum test, p = 0.918. E, CHK1 expression by immunoblotting with cell lines on the left ordered by IC₅₀ to prexasertib and PEO1 cell line derivatives grouped to the right. F, Association between prexasertib IC₅₀ and CHK1 levels (r = 0.686). G, Whole cell lysates from indicated cell lines (ordered by IC₅₀ to prexasertib, with most sensitive on the left) were harvested for immunoblotting with β-actin (ACTB) as the loading control

Because of the established association between HR deficiency and sensitivity to platinum drugs as well as PARPis [4,5,6, 38,39,40,41], we assessed the relationship between RCM response and HR pathway integrity, as assessed by ionizing radiation (IR)-induced RAD51 foci, a standard assay to assess HR [42]. In contrast to the BRCA2 mutant PEO1, which fail to form IR-induced RAD51 foci, the most sensitive (CaOV3) and least sensitive (COV362 and OVCAR5) lines readily formed IR-induced RAD51 foci as summarized in Fig. 3C, Fig. S3A and S3B. Moreover, the HR deficient HGSOC PEO1 line and HR proficient PEO4 line derived from the same patient later in the course of disease had similar IC₅₀s for each of the RCMs tested (Fig. 3A and B). Consistent with these observations, we failed to observe a significant relationship between HRD and RCM sensitivity across these ovarian cancer cell lines (Fig. 3D and S3C).

Across the cell lines, there was a weak correlation between prexasertib and adavosertib IC_50s in the HR proficient lines (Fig. S3D) as has been reported in primary patient models of ovarian cancer [43]. In addition, there was a correlation between CHK1 levels and prexasertib IC₅₀ (𝜌 = 0.69, p = 0.047, Fig. 3E and F) as observed in the PDXs (Fig. S2G). In contrast, no significant association was observed between CHK1 levels and sensitivity to adavosertib or ceralasertib (Fig. S3E). Moreover, even though baseline levels of ATR and WEE1 also varied (Fig. 3G), they did not correlate with sensitivity to any of the three drugs.

Contribution of CDC25A to RCM-induced killing

Because HGSOC cells, which are usually TP53 mutant or null, rely extensively on the intra-S and G2-M checkpoint regulators ATR, CHK1, and WEE1 to coordinate DNA repair and cell cycle progression [44], we assessed RCM-induced cell cycle changes in four HGSOC lines. After a 24-h prexasertib exposure, three of the lines arrested in S phase at concentrations below their IC₅₀ values from the clonogenics; and the fourth arrested in G2 (Fig. 4A and B). WEE1i treatment induced similar cell cycle effects, whereas ATRi treatment caused only a minor increase in the S phase population in the lines (Fig. S4A and B).

Fig. 4

CDC25A contributes to CHK1i-induced apoptosis. A, Representative histogram showing cell cycle analysis in PEO1 cells treated with prexasertib for 24 h. B, Cell cycle distributions of PEO1, COV362, CaOV3, and Ovcar5 cells treated with increasing prexasertib for 24 h, stained with propidium iodide, and analyzed by flow cytometry. C, D, PEO1 (C) and COV362 (D) cells were treated with the caspase inhibitor Q-VD-OPh (5 µM) along with diluent, prexasertib (10 and 20 nM), ceralasertib (2 and 4 µM), or adavosertib (750 and 1000 nM) for 72 h and immunoblotted for indicated proteins. These concentrations were used in subsequent experiments unless otherwise specified. E, F, PEO1 and COV362 cells were treated with prexasertib (E) or LY2880070 (F) for 5 days and analyzed for Annexin V + cells by flow cytometry. G, Annexin V + cells in PEO1 cells treated with indicated siRNAs, followed by prexasertib treatment. siCDC25A pool vs siCTRL, ***p < 0.001; siCDC25A#1 vs siCTRL, ****p < 0.0001; siCDC25A#2 vs siCTRL, **p < 0.001 using two-way ANOVA with Dunnett’s multiple comparisons test. H, I, PEO1 cells (H) and COV362 cells (I) were treated for 72 h with diluent, prexasertib, ceralasertib, or adavosertib in the absence of caspase inhibitors and analyzed for cleavage of caspase substrates. Error bars in B and E–G, mean ± sem from 3 independent experiments. Arrows indicate cleavage products of caspase 9. * Indicates non-specific band

In further experiments, we examined key biological changes involved in S and G2 phase progression, focusing on the HGSOC cell lines PEO1 and COV362, which differ in genomic features and RCM sensitivity (Fig. 3B and S3B). In both lines, prexasertib induced a number of changes, including i) diminished CHK1 activity as manifested by reduced autophosphorylation at Ser²⁹⁶; ii) decreased CHK1-mediated CDC25A phosphorylation, which led to increased CDC25A levels; and iii) and diminished CDK2 inhibitory phosphorylation, which resulted in increased CDK2 activity that was manifested by increased phosphorylation of the CDK2 substrate NPM1 on Thr¹⁹⁹ [23] as shown in Fig. 4C and D (lanes 2 and 3 vs. lane 1). Similar results were observed with the ATRi ceralasertib (Fig. 4C and D, lanes 6 and 7 vs. lane 1). The WEE1i adavosertib abolished CDK2 phosphorylation at Tyr¹⁵, increased CDK2 activity, and increased RS, as manifested by increased γH2AX and phospho-RPA32, even though CHK1 activity (CHK1 Ser²⁹⁶ autophosphorylation) was unaffected (Fig. 4C and D, lanes 4 and 5 vs. lane 1). Not surprisingly, the common feature of the three RCMs was enhancement of RS.

Examination of later effects after drug treatment indicated that a 5-day exposure to prexasertib (Fig. 4E and S4C), the oral CHK1i LY2880070 [45] (Fig. 4F), ceralasertib (Fig. S4D), or adavosertib (Fig. S4E) induced annexin V staining, a hallmark of apoptosis, at the same concentrations that induced cell cycle arrest at 24 h. Indeed, PEO1 and CaOV3 cells started to show signs of apoptosis even at concentrations below their IC₅₀s in colony forming assays. In contrast, the more resistant COV362 and OVCAR5 cells did not reach 50% cell death even at the highest concentrations, again demonstrating decreased drug response. In further experiments, CDC25 knockdown blunted this apoptosis, confirming the contribution of CDC25A upregulation to RCM-induced killing (Fig. 4G and S4F). Additional studies showed that procaspase 3 and PARP1, two well-established caspase substrates [46], were also cleaved, further supporting the view that cells were undergoing apoptosis after RCM treatment (Fig. 4H and I).

Contribution of JNK activation to cell death

To begin to assess the mechanism of this RCM-induced apoptosis, we performed RNAseq in CHK1i-treated PEO1 and COV362 cells. Despite a difference in the number of significantly altered (|log₂(FC)|> 1, p < 0.05) transcripts, there was extensive overlap in the two lines (Fig. S5A). Multiple transcripts involved in cell death processes were upregulated (Fig. 5A), reflecting a stress gene signature (Fig. S5B) [47]. The upregulation of transcripts encoding components of the TNFα pathway (Fig. 5A purple labels and Fig. S5C), which has been implicated in RCM-induced apoptosis in monocytic leukemia cells [23], prompted us to examine the roles of JNK; the JNK substrate activator protein 1 (AP-1), a stress- and cytokine-activated transcription factor [48]; and the death receptor pathway in the HGSOC cells. CHK1i-induced changes observed in the HGSOC lines included upregulation of the AP-1 components JUN and FOS, death ligands TNFα and FASLG (Fig. 5B-D), as well as upregulation of phospho-JUN and JUN, reflecting activation of the JNK pathway (Fig. 5C, D, and S5C). These changes were also observed in HGSOC cells treated with ceralasertib or adavosertib, suggesting a shared mechanism of cell death. Importantly, siRNA-mediated knockdown of JNK1 and JNK2 protected cells from CHK1i-induced apoptosis (Fig. 5E, F, and S5D-S5F). In contrast to AML, however, RCM-induced apoptosis in HGSOC cell lines was barely affected by CRISPR/Cas9-mediated knockout of FADD or BID (Fig. 5G, H, S5G and S5H), two proteins that play essential roles in death receptor-induced apoptosis (Fig. S5I, S5J and ref. 50). Collectively these observations suggest that JNK signaling contributes to RCM-induced HGSOC cell death, but the upregulated death ligands do not.

Fig. 5

JNK and AP-1 stress responses promote CHK1i, ATRi and WEE1i apoptosis in ovarian carcinoma. A, Cells were treated with diluent or prexasertib (20 nM) for 48 h and analyzed by RNAseq. Biological replicates were prepared on three separate days. Heatmap comparing prexasertib treatment to vehicle in PEO1 (left) and COV362 (right). Differentially expressed apoptotic genes (|log2FC|> 2 and p < 0.05 by two-sided t-test, n = 3 per group) are noted in purple and orange. B, Aliquots of PEO1 and COV362 cells independent of those in panel A were treated for 48 h with prexasertib (20 nM) and assayed for the indicated mRNAs by qRT-PCR. C, D, PEO1 (C) and COV362 cells (D) treated with the caspase inhibitor Q-VD-OPh (5 µM) and diluent, prexasertib, ceralasertib, or adavosertib at the concentrations indicated for Fig. 4C were blotted for the indicated proteins. E, PEO1 cells were transfected with the indicated pools of siRNAs, treated with prexasertib for 5 days, and assayed for annexin V binding. F, Immunoblots to assess JNK levels in PEO1 cells transfected with the indicated siRNA pools. G, H, Pooled PEO1 cells transduced with empty vector (EV), FADD sgRNA (G) or BID sgRNA (H) were treated for 5 days with prexasertib and assayed for annexin binding. Error bars in B, E, G, and H, mean ± sem of 3 independent experiments. *, *** and ****: p < 0.05, p < 0.001 and p < 0.0001, respectively, for indicated comparison by unpaired t-test corrected for multiple comparisons. FADD^−/− vs PEO1 EV p = 0.1585; PEO1p vs PEO1 EV p = 0.0786; BID^−/− #2 vs PEO1 EV p = 0.1850; BID^–/– #4 vs PEO1 EV p = 0.0286 using two-way ANOVA with Dunnett’s test for multiple comparisons

Contribution of BH3-only proteins BIM, BMF and PUMA

Because death receptor pathway interruption had limited impact on RCM-induced death of HGSOC cells, we next assessed the contribution of the mitochondrial apoptotic pathway. According to current understanding, activation of this pathway involves trafficking of BH3-only proteins to the mitochondrial surface, where they inhibit anti-apoptotic BCL2 family members and, under certain circumstances, directly activate the mitochondrial permeabilizers BAX and BAK [49, 50]. Loss of BAX and BAK (Fig. S6A and S6B) protected cells from cell death induced by prexasertib (Fig. 6A, B, and S6C) as well as ceralasertib and adavosertib (Fig. 6C-F). In addition, prexasertib, ceralasertib and adavosertib induced loss of mitochondrial outer membrane potential (Fig. 6G, H) as determined by CMXRos staining [51], confirming the occurrence of mitochondrial outer membrane permeabilization, a feature of the mitochondrial apoptotic pathway, during RCM-induced death.

Fig. 6

CHK1i, ATRi, and WEE1i-induced apoptosis is BAX- and BAK-dependent. A-F, Pooled PEO1 (A, C, E) or COV362 cells (B, D, F) transduced with empty vector (EV) or two separate BAX/BAK double knockout clones (DKO) were treated for 5 days with increasing prexasertib (A, B), ceralasertib (C, D), or adavosertib (E, F) and assayed for annexin binding. Error bars, mean ± sem for 3 or more independent experiments. DKO#1, DKO#6 vs PEO1 EV ****p < 0.0001, DKO#2, DKO#13 vs COV362 EV ****p < 0.0001 using two-way ANOVA with Dunnett’s test for multiple comparisons. G, Representative histogram showing CMXRos + cells treated with prexasertib for 5 days. Arrows indicate decrease in the fluorescence peak in prexasertib treated cells. H, Quantitation in PEO1 cells treated with increasing prexasertib, ceralasertib, or adavosertib concentrations for 5 days, stained with MitoTracker CMXRos, and analyzed by flow cytometry

Further inspection of the RNAseq data (Fig. 5A, orange labels) and subsequent qRT-PCR (Fig. 7A) revealed prominently upregulated transcripts encoding proteins in the mitochondrial apoptotic pathway, including the BH3-only proteins NOXA, PUMA, and BIM. Consistent with these results, immunoblotting demonstrated increased NOXA as well as BIM_L and BIM_S, lower molecular weight BIM splice variants that are more potent killers [52], in RCM-treated HGSOC cell lines, although PUMA was not consistently upregulated at the protein level (Fig. 7B and C, lanes 2–7 vs lane 1). Moreover, mitochondrial fractions from COV362 and PEO1 cells were found to contain more BIM_L and BMF after prexasertib treatment (Fig. 7D, E), suggesting their release from sites of sequestration and translocation to the mitochondria [49, 50], events previously reported to occur downstream of JNK activation in cells undergoing anoikis [24, 25]. In addition, NOXA was increased in the COV362 mitochondria.

Fig. 7

CHK1i, ATRi, and WEE1i-induced apoptosis in ovarian cancer cells is driven by combined action of PUMA, BIM and BMF. A, PEO1 and COV362 cells were treated for 48 h with prexasertib (20 nM) and assayed for the indicated mRNA by qRT-PCR. Error bars, ± sem of 3 independent experiments. *, ** and ***, p < 0.05, p < 0.01, and p < 0.001 respectively, for indicated comparison by unpaired t-test corrected for multiple comparisons. B, C, PEO1 (B) and COV362 cells (C) were treated with the caspase inhibitor Q-VD-OPh (5 µM) and diluent, prexasertib, ceralasertib, or adavosertib as indicated for Fig. 4C for 72 h and immunoblotted for the indicated BH3 proteins. D, E, After PEO1 (D) and COV362 (E) cells treated with prexasertib (20 nM) + the caspase inhibitor Q-VD-OPh (5 µM) for 72 h, whole cell lysates, cytosol, and mitochondria were isolated and probed for the indicated proteins. GSTπ and MCL1 served as markers of the cytosolic and mitochondrial fractions respectively. F, 24 h after PEO1 cells were transfected with the indicated siRNAs, prexasertib (20 nM) was added for 5 days and cells were assayed for AnnexinV binding. Error bars, mean ± sem from 3 or more independent experiments. siPUMA-BIM and siPUMA-BMF vs siCTRL, p < 0.01; siPUMA-BIM-BMF-BID-NOXA (siPBBBN) vs siCTRL, p < 0.001; siPUMA-BIM-BMF (siPBB) vs siCTRL, p < 0.0001 using two-way ANOVA with Dunnett’s test for multiple comparisons. G, Immunoblots to assess BH3-only protein levels in PEO1 cells transfected with the indicated siRNA combinations. H, Schematic of BH3-only protein-mediated apoptosis induced by RCMs. I, Immunoblot analysis of potential response predictors in the two CCNE^ampMYC^amp PDXs PH354 and PH235 (top, blue) in comparison to WT PDXs PH061 and PH063 (bottom, black). Each lane contains tumor from a single PDX-bearing mouse after 8 days of the indicated treatments. Dashed line indicates removal of lanes from mice on a separate study

To assess the contribution of these changes to RCM-induced killing, BH3-only proteins were knocked down individually and in combinations. Depletion of the individual BH3-only proteins BIM, BMF, PUMA, or NOXA offered limited protection (Fig. S7A and S7B). In contrast, combined knockdown of PUMA + BMF or PUMA + BIM decreased prexasertib-induced apoptosis by > 40%, and knockdown of PUMA + BMF + BIM (siPBB) almost completely protected cells (Fig. 7F and G). Knockdown of additional BH3-only proteins, including BID and NOXA (siPBBBN), diminished apoptosis to the same extent as siPBB, suggesting that BMF, BIM, and PUMA are the critical BH3-only proteins in this process.

Collectively, these results establish a molecular framework linking RCM treatment to mitochondrial apoptosis in which CHK, ATR, or WEE1 inhibition leads to stabilization of CDC25A, S phase dysfunction, sustained JNK signaling, and activation of the BH3-only proteins PUMA, BIM, and BMF, leading to BAX/BAK-mediated apoptosis (Fig. 7H). With this framework in hand, we examined PDX samples harvested on day 8 of therapy during the HGSOC PDX study shown in Fig. 2. While it is not possible to assess changes in BH3-only protein expression in vivo because cells with the greatest BH3-only protein increase undergo apoptosis and are lost from the analysis in the absence of caspase inhibitors [53], examination of upstream mediators of this pathway (Fig. 7I) indicated that adavosertib induced complete CDK2 dephosphorylation, increased CDK2 activity and increased JNK phosphorylation in the models that showed objective tumor shrinkage (PH354 and PH061, lanes 4–6 vs lanes 1–3) but not in models that failed to shrink (PH235 and PH063). In accordance with the significant tumor slowing induced by ceralasertib in PH354 and PH061, the tumors harvested on Day 8 after ATRi also displayed reduced CDK2 activity and increased JNK phosphorylation (Fig. 7I, lanes 7–9 vs lanes 1–3, models PH354 and PH061), suggesting that levels of activated CDK2 and JNK in HGSOC tumors might be useful pharmacodynamic biomarkers that predict subsequent response to RCM therapy.

Discussion

The RCMs studied here, inhibitors of CHK1, ATR, and WEE1, have shown tantalizing but inconsistent clinical activity against HGSOC, suggesting the need for better understanding of their action to identify neoplasms most likely to respond. To facilitate the clinical development of these classes of agents, the present study has systematically examined the pathway of cytotoxic signaling that starts with the inhibition of replication checkpoint kinases and culminates in the activation of the mitochondrial apoptotic pathway in HGSOC cells (Fig. 7H).

One of the goals in studying RCM activity in model systems is to identify biomarkers that predict response. Even though an increasing number of clinical trials are interrogating expression of CCNE1 and MYC as determinants of RCM response [22, 54], our examination of multiple PDXs treated with RCMs in vivo or ex vivo indicated that HGSOC responses to RCMs appear to be independent of CCNE1 or MYC expression (Figs. 1, 2, and S2). Instead, the present study found that prexasertib responses are associated with low CHK1 protein levels (Fig. 3 and  S2), in agreement with results in lung cancers and acute leukemias [23, 55]. Thus, CHK1 levels might need to be evaluated in further studies right along with the recently proposed predictive biomarkers BLM, a helicase involved in ATR-mediated fork stabilization after stress, and the WEE1 regulator FAM112A [56, 57].

In addition to predictive markers, there is great interest in improved pharmacodynamic biomarkers of RCM activity. The present study implicates CDC25A as a potentially important contributor to CHK1i anti-cancer effects that is upregulated during CHK1i and WEE1i therapy (Fig. 4). Moreover, examination of the transcriptomic and signaling changes during RCM treatment identified a series of transcripts, constituting part of a stress gene signature [47], that were upregulated in both of the HGSOC cell lines characterized by RNAseq after drug treatment (Fig. 5 and S5). Consistent with these results, we observed that RCMs activated the JNK pathway, which plays a critical role in RCM-induced cytotoxicity as indicated by the effects of JNK knockdown (op cit.). Further analysis indicated that mRNAs encoding a number of death ligands, including TNFα and FASL, were upregulated by RCM treatment. Notably, however, interruption of the extrinsic apoptotic pathway afforded HGSOC cells little protection from RCM-induced cell death (Fig. 5). This lack of involvement of death ligands in RCM-induced killing of HGSOC cells is in contrast to the role of JUN-mediated TNF transactivation in RCM-induced killing of acute monocytic leukemia cells [23], illustrating variation in cytotoxic pathways activated by RCMs in different cell types. Instead, RCM-induced cell death in HGSOC cells depends on the mitochondrial apoptotic pathway, as indicated by the loss of mitochondrial membrane potential (Fig. 6H), which reflects mitochondrial permeabilization, as well as the protective effects of knocking out BAX and BAK (Fig. 6A-F) or knocking down the BH3-only proteins BIM, BMF, and PUMA (Fig. 7F and S7), three established initiators of the mitochondrial pathway [49, 58] that traffic to mitochondria during RCM treatment (Fig. 7D and E). Importantly, involvement of BMF, a BH3-only protein previously implicated mainly in anoikis, also illustrates a way in which the downstream portion of the RCM cytotoxicity pathway is also potentially unique compared to other antineoplastic agents. Indeed, the present observations identify critical steps distal to effects at replication forks that must occur for RCMs to be cytotoxic, suggest additional signaling events that can be interrogated to determine whether components of the replication checkpoint have been successfully inhibited, and imply that factors capable of inhibiting the mitochondrial pathway, e.g., amplifications of the loci encoding BCLXL or MCL1 that are commonly observed in ovarian cancer, might need to be considered in identifying patients for treatment with RCMs.

Analysis of the HGSOC PDXs treated in vivo indicated that H2AX phosphorylation was detectable at baseline and increased with RCM treatment in three of four PDXs, including two PDXs that did not show regressions (Fig. 7). In contrast, CDK2 phosphorylation was completely inhibited only in the two PDXs that shrank during adavosertib treatment and not in the two that failed to shrink (Fig. 7I), establishing inhibition of CDK2 phosphorylation as a potentially important pharmacodynamic indicator of adavosertib action as well. More broadly, the present analysis identified upregulation of CDC25A, dephosphorylation and activation of CDK2, and activation of JNK1/2 as evaluable steps that might be productively interrogated in future RCM clinical trials.

While the present studies utilized prexasertib, adavosertib and ceralasertib, it is important to recognize that these are well-established inhibitors that can be viewed as paradigm drugs. Prexasertib, which has exhibited promising activity in HR proficient ovarian cancer [23], remains in clinical trials as ACR-368 (Clinical Trials.gov identifier NCT05548296). Although adavosertib and ceralasertib are being supplanted by orally bioavailable agents for further testing as monotherapies and in combinations, it is anticipated that the newer agents will induce the same signaling described here.

In summary, the present results provide new mechanistic insight into therapies targeting the DNA damage response and provide preclinical data to support a series of potential biomarkers for further testing as RCMs undergo additional clinical development.

Conclusions

While numerous studies over the past decade have contributed to current understanding of how replication checkpoint modulators impact cell cycle regulation, there is limited understanding of how the resulting changes ultimately kill susceptible cells. Recent trials indicate the clinical benefit of prexasertib monotherapy in the setting of platinum-resistant CCNE^amp cancer, the response is still heterogeneous, and outcomes are dismal. Our findings suggest a model in which replication stress results in the activation of JNK and AP-1 to effect two sets of distinct changes, i) upregulation of PUMA and BIM mRNAs and ii) displacement of BIM and BMF from sites of sequestration so they can directly and/or indirectly activate BAX/BAK at the mitochondrial outer membrane. Measurement of all of these steps and perhaps more will be required to understand the which ovarian cancer patients will benefit from these therapies.