Specific Mechanisms of Chromosomal Instability Indicate Therapeutic Sensitivities in High-Grade Serous Ovarian Carcinoma

The vast majority of solid tumors exhibit chromosomal instability (CIN), the continual gain and loss of chromosomes or parts of chromosomes (1, 2). CIN can drive tumor heterogeneity and clonal evolution, and is thought to contribute to chemotherapy resistance in many cancer types including ovarian cancer (3, 4). Knowledge of the defective cellular pathways that underlie CIN would enable strategies to target cancer cells using synthetic lethal or CIN-limiting approaches (5), in addition to providing new diagnostic or prognostic tools. However, to date, the causes of CIN in cancer remain ill defined. Defective chromosome attachment to the mitotic spindle due to aberrant mitotic microtubule (MT) dynamics can contribute to CIN in cancer cell lines (6–9), potentially driven by alterations in spindle protein abundances (6) or genetic alterations in Aurora A, BRCA1, or Chk2 (9). Loss of retinoblastoma protein (pRB) leading to cohesion defects and CIN has been demonstrated in a sarcoma cell line (10). Studies classifying CIN mechanisms in representative, cancer-specific cell line panels are currently limited to colorectal cancer (8, 11), where both DNA replication stress (the slowing or stalling of DNA replication) and elevated MT assembly rates were shown to contribute to CIN (8, 11).

High-grade serous ovarian carcinoma (HGSC) represents an important clinical challenge; despite initial positive responses to first-line platinum therapy, most patients relapse, leading to a poor overall survival for this disease (12). The genomes from both HGSC patient tumors and ascites-derived HGSC cells bear the scars of CIN as evidenced by highly aberrant genomic landscapes (Supplementary Fig. S1A; refs. 3, 13–15). There has been extensive interest in inferring potential cancer mutational mechanisms from tumor and cancer cell line genomes, at both single-nucleotide variant (16) and chromosome-scale aberrations, particularly in ovarian cancer (17, 18). However, apart from a high prevalence of mutations in homologous recombination (HR) genes, and near ubiquitous TP53 mutations (14), genetic drivers of CIN remain to be elucidated in HGSC. Moreover, it has been shown that BRCA-mutated tumors can acquire HR-reactivating mutations (19), highlighting the need for functional analysis in defining ongoing CIN mechanisms. Genetic aberrations that may contribute to CIN in HGSC are Aurora A amplification and cyclin E (CCNE1) amplification (20). Overexpression of cyclin E is linked to worse patient outcome (21) with combined high CCNE1 expression and genomic amplification exhibiting higher genomic instability (22). RB1 mutations are also present in 17.5% HGSC tumors (15).

To date, functional characterization using appropriate cell line models for HGSC is lacking. Recent advances including genomic approaches have defined multiple distinct subtypes of ovarian cancer, and have allowed the classification of available tumor-derived cell lines into suitable models (23–25). We therefore undertook the first comprehensive functional characterization of a curated panel of HGSC cell lines to define mechanisms driving CIN. We demonstrate that all cell lines exhibit extensive, ongoing CIN in the form of high rates of numerical and structural chromosome defects, and chromosome segregation errors. We find gross defects in multiple pathways controlling chromosome stability, and that either suppressing MT dynamics using low doses of paclitaxel, or limiting replication stress using nucleoside supplementation, reduces chromosome segregation errors and CIN. Furthermore, we show that functional analysis of CIN rates and types can inform sensitivity to standard therapies relevant to HGSC. In-depth genomic and gene expression analyses revealed potential genome stability regulators correlated with specific CIN mechanisms, or responses to therapy, providing a platform to investigate these associations in patient datasets. These new insights provide a first step toward designing new approaches to treat HGSC, including determining whether limiting CIN could prevent CIN-driven chemotherapy resistance in HGSC (5), and guiding development of biomarkers for appropriate therapy choices.

HGSC and fallopian cell lines were sourced as detailed in Supplementary Table S1 and maintained at 37°C and 5% CO₂. Their identities were confirmed by short tandem repeat profiling (ATCC). HCT116, SW620 (kind gift from C. Swanton), and Cov318 were maintained in DMEM High Glucose (Sigma); Kuramochi, AOCS1, Ovkate, Ovsaho, and Snu119 were maintained in RPMI (Sigma). All medium were supplemented with 10% FBS and 100 U penicillin/streptomycin. G164 cells were grown in DMEM F12 (Sigma) supplemented with 5% human serum (H4522, Sigma) and 100 U penicillin/streptomycin. FNE1/FNE2 were maintained in FOMI media (University of Miami, Coral Gables, FL) supplemented with cholera toxin (C8052, Sigma). H2B-RFP (red fluorescent protein) stable cell lines were generated after transfection with lentiviral construct H2B-RFP (Addgene, 26001) and flow sorting RFP-positive cells. Cells were routinely tested for Mycoplasma using MycoAlert PLUS Mycoplasma Detection Kit (LT07-710, Lonza) and visual inspection using DAPI staining at the microscope. Cells were passaged for a maximum of 8–12 weeks (∼10–14 passages).

Cells were seeded into 96-well dishes. The next day, additional media was added, supplemented with either Embryomax nucleosides (final concentration of 10×) or low-dose paclitaxel (final concentration of 1 nmol/L) or chemotherapy agents (at indicated final concentrations). Plates were imaged over the course of 1 week using an IncuCyte live cell analysis system to calculate the percentage of confluency over time. The fold change in confluency as a growth ratio (final confluency/starting confluency for that cell line grown under that condition) was reported.

Cells were arrested in colcemid for 2 hours, collected then resuspended in hypotonic solution (0.2% KCl, 0.2% sodium citrate) for 7 minutes at 37°C. Cells were pelleted and resuspended in freshly prepared 3:1 methanol-glacial acetic acid, then dropped onto slides.

Cells were seeded onto slides at low density to ensure growth of colonies from single cells. Colonies were grown with/without nucleosides for 4 weeks then fixed for FISH. Cells in each colony were imaged and scored for centromere number, and percentage cells deviating from modal value for centromere of that colony was calculated.

The 100× Embryomax nucleosides (ES-008-D, Merck Millipore) were diluted in medium to 10× final concentration. Taxol (paclitaxel, P045, Cambridge Bioscience) was dissolved in DMSO and used at 1 nmol/L final concentration for low-dose rescue of CIN, or at higher doses for chemotherapy response. Monastrol (Sigma) was dissolved in DMSO and used at 100 μmol/L. Olaparib (AZD2281, Gmbh) was dissolved in DMSO.

Cells grown on coverslips were fixed with PTEMF [0.2% Triton X-100, 0.02 mol/L PIPES (pH 6.8), 0.01 M EGTA, 1 mmol/L MgCl₂, 4% formaldehyde]. After blocking with 3% BSA, cells were incubated with primary antibodies according to suppliers' instructions. Antibodies were obtained from Abcam [β-tubulin (ab6046), CenpA (ab13939), Centrin 3 (ab54531), cyclin A2 (ab16726), Hec1 (ab3613), RPA (ab79398)], Antibodies Incorporated [CREST (15-234-0001)], Bethyl Lab [Mad2 (A300-300A)], Millipore [H2AX (05-636)], Santa Cruz Biotechnology [53BP1 (sc-22760), and Rad51 (sc-398587)]. Secondary antibodies used were goat anti-mouse AlexaFluor 488 (A11017, Invitrogen), goat anti-rabbit AF594, AF488 (A11012, A11008, Invitrogen), and goat anti-human AF647 (109–606–088-JIR, Stratech or A21445, Invitrogen). DNA was stained with DAPI (Roche) and coverslips mounted in Vectashield (Vector H-1000, Vector Laboratories).

FISH was carried out according to manufacturer's instructions. In brief, cells on slides were fixed in 3:1 methanol:acetic acid, then put through an ethanol dehydration series (2 minutes in 70, 90, 100% ethanol) then air dried. Probe was added to slides that were heated to 72°C for 2 minutes, then left at 37°C overnight in a humid chamber. The next day, slides were washed in 0.25×SSC at 72°C for 2 minutes, then 2×SSC, 0.01% Tween at room temperature for 30 seconds. Slides were stained with DAPI then coverslips were mounted with Vectashield. Pan-centromere probe was purchased from Cambio (1695-F-02) and Centromere Enumeration Probes from Cytocell.

Metaphase spreads from each cell line were hybridized with the multiplex-FISH (M-FISH) probe kit 24XCyte (Zeiss MetaSystems) following the manufacturer's instructions. Briefly, the slides were incubated 30 minutes in 2×SSC buffer at 70°C, then allowed to cool at room temperature for 20 minutes. Following a 1-minute wash in 0.1×SSC, the cells were denatured in NaOH 0.07 mol/L for 1 minute, then washed in 0.1×SSC and 2×SSC. The cells were dehydrated in an ethanol series, and air dried. The probe mix was denatured at 75°C for 5 minutes, and preannealed at 37°C for 30 minutes. A total of 6 μL of probe mix were applied to each slide, under a 18 × 18 mm² coverslip. The slides were incubated for 3 days at 37°C, then washed for 2 minutes in 0.4×SSC, at 72°C, and 30 seconds in 2×SSC, 0.05% Tween20, at room temperature, and finally mounted in DAPI/Vectashield (VectorLabs). Images were acquired on an Olympus BX-51 microscope for epifluorescence equipped with a JAI CVM4+ progressive-scan CCD camera, and analyzed using the Leica Cytovision Genus v7.1 software (Leica). A minimum of 25 metaphases were karyotyped for each cell line.

Fibers were prepared as described previously (26). In brief, cells were pulse labeled with 25 μmol/L CldU and 250 μmol/L IdU (Sigma) for 20 minutes. Cells were harvested and then lysed using 0.5% SDS, 20 mmol/L Tris-HCl pH 7.4, 50 mmol/L EDTA. Fibers were spread on slides and DNA detected using rat anti-BrdU and CldU, with secondary antibodies as above.

Images were acquired using an Olympus DeltaVision RT microscope (Applied Precision, LLC) equipped with a Coolsnap HQ camera. Three-dimensional image stacks were acquired in 0.2 μm steps, using Olympus 100 × (1.4 numerical aperture), 60 × or 40 × UPlanSApo oil immersion objectives. Deconvolution of image stacks and quantitative measurements was performed with SoftWorx Explorer (Applied Precision, LLC). H2B-RFP-labeled cells were live imaged in 4-well imaging dish (Greiner Bio-one). A total of 20 μm z-stacks (10 images) were acquired using an Olympus 40 × 1.3 numerical aperture UPlanSApo oil immersion objective every 3 min for 8 h using a DeltaVision microscope in a temperature and CO₂-controlled chamber. Analysis was performed using Softworx Explorer. MT assembly assays (see below) were performed in part using an Eclipse Ti-E inverted microscope (Nikon) equipped with a CSU-X1 Zyla 4.2 camera (Ti-E, Zyla; Andor), including a Yokogawa Spinning Disk, a precision motorized stage, and Nikon Perfect Focus, all controlled by NIS-Elements Software (Nikon).

MT dynamics were analyzed as described previously (8). Briefly, assembly rates were calculated by tracking EB3-GFP protein foci in living cells. Cells were seeded onto glass-bottom dishes and transduced with virus containing pEGFP_EB3 (gift from S. Godinho) Cells were treated with Eg5 (Kif11) inhibitor monastrol (67 μmol/L, Sigma) for 2 hours. Cells were then imaged on the DeltaVision microscope or using an Eclipse Ti-E inverted microscope (Nikon). Cells were imaged every 2 seconds using four sections with a Z-optical spacing of 0.4 μm. Average assembly rates (μm/minute) were calculated using data for 20 individual MTs per cell for 10–20 cells.

Cell lysates were prepared using lysis buffer [20 mmol/L Tris-HCl (pH 7.4), 135 mmol/L NaCl, 1.5 mmol/L MgCl₂, Triton 1%, Glycerol 10%, 1× Protease inhibitor (Roche)]. Immunoblots were probed with antibodies against p53 (Santa Cruz Biotechnology, sc126), Aurora A (Cell Signaling Signaling, 12100), Vinculin (Cambridge Bioscience 66305), and cyclin E (Abcam ab3927) and developed using goat anti-mouse (Cell Signaling Technology, 7076S) or goat anti-rabbit (Santa Cruz Biotechnology, sc-2004) IgG HRP-conjugated antibodies, using a Chemidoc (GE Healthcare).

Cells were treated using an industrial cabinet X-ray device (RS-2000, Rad Source Technologies) for specific times calibrated to deliver 2 Gy ionizing radiation (IR) and then fixed after 2 or 24 hours recovery before immunofluorescence with antibodies against cyclin A and Rad51.

Cells were fixed in 4% formaldehyde for 7 minutes, permeabilized with 0.2% Triton X-100 for 2 minutes, stained with DAPI, and analyzed using BD FACS Diva 8.2. RPE1 cells were used to calibrate FACS analysis to generate a profile for DNA signal peaks, corresponding to a diploid cell line. RPE1-H2B-RFP and parental RPE1 cells were then mixed and analyzed together, for a direct comparison, to verify that H2B tagging did not alter the expected peaks. RPE1-H2B-RFP cells were then mixed with known near-diploid or aneuploid cell lines (HCT116 and SW1116) to verify that the FACS analysis could distinguish that RFP-positive cells gave the expected profile compared with the RFP-negative cells when analyzed together. RPE1-H2B-RFP cells were then mixed with individual HGSC cell lines, to determine whether each HGSC cell line overlapped the diploid DNA signature or differed, indicating aneuploidy.

Sample processing and whole-genome sequencing were carried out by Edinburgh Genomics. Samples were processed using Illumina TruSeq Nano libraries and sequenced with Illumina HiSeq X instruments to an average depth coverage of 30×. Because of the nature of our cell line samples, matched normal tissue/blood samples were unavailable for analysis. For further details including downsampling and absolute copy-number estimation, somatic mutation and copy-number variation, and cancer genome breakpoint analysis methods, see Supplementary Methods.

RNA was extracted from cell lines using RNEasy Kits (Qiagen) for three biological repeats. RNA sequencing (RNA-seq) was performed by Bart and the London Genome Centre on the Illumina NextSeq 500 platform, generating on average approximately 1.5 million single-end reads of 75 bp in length per sample. RNA-seq data have been deposited in Gene Expression Omnibus under the accession number GSE155310. For analysis methods, see Supplementary Methods.

Statistical tests were carried out where indicated in figure legends, as either an unpaired t test with Welch correction, or a one-way ANOVA with Dunnett multiple comparison (every cell line compared with FNE1). Asterisks denote the significance value between experimental conditions adhering to the following nomenclature: P < 0.05 (*); P < 0.01 (**); P < 0.001 (***); P < 0.0001 (****). For values close to P = 0.05, actual P values are given. All calculations were carried out using software (Graphpad Prism 8.0).

Cancer-derived cell lines have proven a useful resource to investigate ongoing mechanisms driving CIN (6, 8, 11, 27). We assembled a panel of seven HGSC lines, five from the top 10 suitable cell line models for HGSC (Cov318, Kuramochi, Ovkate, Ovsaho, Snu119; ref. 23) and two obtained from recent confirmed patients with HGSC [AOCS1 (28) and G164; Supplementary Table S1]. As tissue-type specific controls, we obtained two, h-TERT-immortalized fallopian tube serous epithelial cell lines [FNE1 and FNE2 (29)], representing the likely tissue of origin of HGSC (30). We performed whole-genome sequencing (Materials and Methods) to characterize the extent of genomic alteration in the HGSC lines. First, to visualize genomic gains and losses, we performed copy-number segmentation and computed the DNA copy-number profiles (Fig. 1A; see Materials and Methods for details). Similar to HGSC genomes available in the The Cancer Genome Atlas dataset (14, 23), the HGSC lines displayed complex copy-number profiles (Fig. 1A). Our analysis also computed the most likely models of ploidy (Supplementary Fig. S1B), and we further confirmed these using chromosome counting from metaphase chromosome spreads (Fig. 1B and C; Supplementary Fig. S1C) and FACS ploidy analyses (Supplementary Fig. S1D). There was a notable range in ploidy between cell lines, varying between near-diploid (2n) to near-tetraploid (4n; Fig. 1C). To examine chromosome alterations at a single-cell level, we performed M-FISH on metaphase chromosome spreads for FNE1 and Kuramochi cell lines. As expected, FNE1 were near diploid (Fig. 1D; Supplementary Table S1). In contrast, Kuramochi cells displayed a high prevalence of numerical and structural alterations that were highly heterogeneous between individual cells (Fig. 1D; Supplementary Fig. S1E). Metaphase chromosome spreads analyzed with centromeric FISH probes also demonstrated the presence of structural chromosome aberrations (dicentric and acentric chromosomes) in all HGSC cell lines (Supplementary Fig. S1F).

We next analyzed genetic mutations relevant to HGSC. We verified the known BRCA2 nonsense mutation in Kuramochi (23), but did not identify any BRCA1/2 mutations of known pathogenicity in other cell lines (Supplementary Table S2; Supplementary Fig. S5A). Some lines exhibited BRCA1 or BRCA2 copy-number alteration (Supplementary Fig. S1J) although gene expression was not significantly altered compared with the FNE controls except for BRCA2 in Ovsaho (Supplementary Fig. S1G and S5C). TP53 mutations occur in 96% of HGSC tumors (14). Accordingly, all seven HGSC cell lines exhibited TP53 mutations and aberrant p53 protein expression (including premature stop codons in AOCS1 and Ovsaho) compared with FNE1 cells (Supplementary Fig. S1H and J; Supplementary Table S2). Five cell lines demonstrated CCNE1 copy-number gain and corresponding changes in RNA expression, and all seven lines showed overexpression at protein level compared with FNE1 (Supplementary Fig. S1I and J). Our HGSC cell line panel thus recapitulates key genomic features of HGSC tumors, and encompasses a range of ploidy and genomic alterations.

The diversity of chromosome alterations between individual Kuramochi cells, and the prevalence of structural and numerical chromosome alterations in all HGSC lines, suggested that CIN was ongoing. Indeed, live cell imaging of cell lines stably expressing mRFP-tagged Histone H2B revealed the frequent occurrence of chromosome mis-segregation events in all HGSC cell lines (Fig. 1E and F). Chromosome segregation errors were further examined using high-resolution imaging of fixed cells to gain insights about the nature of missegregating chromatin (Fig. 1G–I). Anaphase lagging chromatin was sometimes negative for CREST-reactive serum (marks centromeric proteins) and Hec1 kinetochore proteins, suggesting that some missegregation events were precipitated by structural chromosome alterations (Fig. 1G, I, and J). To verify the acentric nature of lagging chromatin, we performed FISH using all-centromere-targeted probes in the AOCS1 cell line. This confirmed the presence of missegregating chromatin devoid of centromeric DNA sequence (Fig. 1K and L). We were struck by the high frequency of chromosome missegregation in some lines, for example, Cov318 exhibited errors in 50% of cells, with each cell typically displaying multiple errors, often of different types (Fig. 1J). HGSC cell lines thus exhibit continual missegregation of both intact and structurally abnormal chromosomes, contributing to their high rates of numerical and structural CIN.

We next tested whether control of mitosis was perturbed in HGSC, by analyzing mitotic progression kinetics using live cell imaging. Slight congression delays have been reported in colorectal cancer (8) but it was hitherto unknown whether this occurs in other cancer types. Four HGSC cell lines exhibited significant delays in mitosis, as measured by both the time from nuclear envelope breakdown to anaphase onset (Fig. 2A and B) and the time from chromosomal congression to the metaphase plate (Fig. 2C) There was no obvious correlation between higher cell line ploidy and slowed congression, suggesting this phenomenon is unconnected with the number of chromosomes present. The overall prolonged time in mitosis suggested that congression errors were capable of mounting a robust mitotic checkpoint response. Accordingly, live cell imaging revealed that all of the four HGSC cell lines tested were able to efficiently arrest mitosis following treatment with nocodazole to depolymerize all MTs (Fig. 2D). Moreover, immunofluorescence revealed the expected presence of two key mitotic checkpoint proteins, Mad2 and BubR1, on uncongressed chromosomes in all cell lines tested (Supplementary Fig. S2A and S2B). Prolonged delays in mitosis have been linked to defective sister chromatid cohesion and chromosome missegregation (31). To evaluate this, we measured intercentromere distance, which provides a measure of sister chromatid cohesion defects (32). This suggested chromosome cohesion was normal in all HGSC lines with the possible exception of Cov318 (Supplementary Fig. S2C and S2D). We note that it was not possible to directly assess cohesion fatigue in these cell lines as their slow chromosome congression rates confound these analyses.

The significant delays in chromosome congression, and the presence of some apparently whole (centric), lagging chromosomes in anaphase (Figs. 1I and 2C) suggested that the mitotic machinery was disrupted in HGSC. Centrosome abnormalities have been detected in ovarian cancer (27), and can promote the formation of multipolar spindles (33). Although multipolar spindles can resolve to a pseudobipolar spindle in a process known as centrosome clustering (34), this can elevate the frequency of incorrect chromosome attachments to the spindle and increase chromosome segregation errors (33, 35). We therefore quantified centrosome and spindle defects from mitotic cells using antibodies against centrioles (centrosome cores; two per centrosome). Most cell lines exhibited a significant percentage of cells with supernumerary (>4) centrioles (Supplementary Fig. S3A and S3B). Cell lines with extra centrosomes also displayed multipolar spindles in prometaphase cells (Supplementary Fig. S3C and S3D). Multipolar spindles usually resolved to pseudobipolar spindles by anaphase, because most cells underwent bipolar cell division (Supplementary Fig. S3E). To investigate whether centrosome amplification was associated with elevated chromosome segregation errors in HGSC, we compared error rates between anaphase cells with four, or more than four centrioles. In the two lines tested, AOCS1 and G164, the presence of extra centrosomes tended to correlate with a higher rate of chromosome segregation errors, although this was only statistically significant for G164 (Supplementary Fig. S3F and S3G), suggesting a potential small contribution of extra centrosomes to CIN in HGSC.

Elevated MT dynamics leading to delayed chromosome congression, chromosome segregation errors, and CIN were recently described in colorectal cancer as a result of either Aurora A overexpression (8) or the deregulation of the Chk2–BRCA1 axis (9, 36). Our panel of HGSC cell lines carries potential Aurora A defects due to a common amplification of the chromosomal region carrying the Aurora A gene [AURKA, 20q (23); Fig. 1A; Supplementary Fig. S3H], that is also common in HGSC tumors (23). Indeed, Aurora A levels were upregulated at both transcriptional and protein level in all our cell lines (Supplementary Fig. S3I). We therefore tested whether MT dynamics were altered in HGSC cell lines, by transiently expressing the MT tip-tracking protein EB3 tagged with GFP and filming MT growth. To allow accurate quantification of mitotic spindle MT dynamics, this assay was carried out in cells treated with the Eg5 inhibitor monastrol to generate monopolar spindles, as described previously (Fig. 2E; Supplementary Movie S1; ref. 8). We included the colorectal cancer cell lines HCT116 (CIN-negative) and SW620 (CIN-positive) as controls for normal, and elevated MT assembly rates, respectively (8). MT assembly rates were significantly elevated compared with FNE1 cells in all HGSC cell lines except Kuramochi (Fig. 2F) with most lines notably displaying mean MT assembly rates well above CIN-positive SW620 colorectal cancer cells. We sought to determine whether this phenomenon was causative for CIN in HGSC. We treated cells with a low dose of the MT stabilizing agent, paclitaxel (Taxol), previously demonstrated to suppress MT assembly rates in colorectal cancer cell lines (8). It has been previously established that monopolar mitotic spindles frequently orient asymmetrically when MT assembly is elevated (37), thus providing a proxy readout for elevated MT assembly rates. All seven HGSC lines demonstrated asymmetric monopolar spindles at a higher incidence than the FNE2 control, which was reduced when cells were treated with low-dose paclitaxel (Fig. 2G and H), and rescue of abnormal MT dynamics was directly confirmed for two cell lines using the EB3-GFP tip-tracking assay (Fig. 2I). Reducing abnormal MT dynamics also reduced chromosome segregation errors in all five HGSC cell lines tested (Fig. 2J and K) independent of any effect on proliferation (Supplementary Fig. S3J), similar to the effect in colorectal cancer cell lines (8). This demonstrates that, for this HGSC cell line panel, aberrant MT dynamics contribute to chromosome missegregation and that low-dose paclitaxel represents a viable strategy for experimentally reducing CIN in HGSC.

Replication stress, the slowing or stalling of DNA replication, is known to occur in multiple cancer types (38, 39) and was previously shown to contribute to CIN in colorectal cancer by generating acentric fragments and chromatin bridges (11). The presence of acentric lagging chromatin, structural chromosome defects, and the known roles of HR proteins frequently mutated in HGSC (such as BRCA1 and BRCA2) in protecting the DNA replication fork (40–42), suggested that HGSC cell lines may experience replication stress. To test this, we directly examined replication fork speed using single DNA fibre pulse labeling. All HGSC cell lines exhibited reduced replication fork rates compared with control FNE1 cells (Fig. 3A). We next examined whether this reduced fork speed would correlate with known hallmarks of replication stress, namely, elevated prometaphase DNA damage as indicated by γH2AX foci, 53BP1 bodies in G₁ cells and ultrafine anaphase bridges (11, 43). Most cell lines exhibited elevated levels of γH2AX, but increased levels of ultrafine bridges or formation of 53BP1 bodies were seen in only a few cell lines (Fig. 3B–D).

We next examined whether elevated replication stress contributes to chromosome missegregation and CIN in HGSC, by reducing replication stress using nucleoside supplementation as described previously (11, 44). This reproducibly reduced segregation errors compared with untreated cells in three cell lines (Ovsaho, Cov318, and G164). In contrast, Snu119, Kuramochi, and AOCS1 were overall nonresponsive (Fig. 3E and F; Supplementary Fig. S4). To test whether suppression of replication stress could translate into a reduction in karyotypic heterogeneity, we grew colonies from single cells and performed centromeric FISH to measure within-clone deviation in centromere number. Replication stress does not generally induce whole chromosome missegregation, and can cause preferential missegregation of specific chromosomes (45). Moreover, HGSC cell lines frequently demonstrated multiple error types per cell (Fig. 1J) and we therefore expected a weak, if any, reduction in centromere number deviation. Nonetheless, Ovsaho and Kuramochi displayed reduced karyotypic heterogeneity for one or both chromosomes scored following single-cell colony derivation in the presence of nucleosides (Fig. 3G and H). Cov318 did not show a reduction in colony mode deviation despite reduced chromosome segregation errors. Effects of nucleoside supplementation on chromosome segregation and karyotypic heterogeneity were independent of any effects on proliferation (Supplementary Fig. S3J). Taken together, these data demonstrate the presence of replication stress that contributes, in at least a subset of lines, to chromosome segregation errors and CIN in HGSC. However, the manifestation of the canonical hallmarks of replication stress, and the response to nucleoside supplementation, varied between cell lines.

We wondered whether this variation could be explained by differences in responses to replication stress. We tested this by treating cells with low doses of the DNA polymerase poison aphidicolin to provide exogenous replication stress (43). Cov318 cells had high basal levels of 53BP1 bodies that were further elevated in response to aphidicolin, whereas Kuramochi, Ovkate, and AOCS1 cells failed to recruit 53BP1 in response to both their endogenous and exogenous replication stress (Fig. 3C, I, and J). This suggests that some HGSC cell lines may either have an increased capacity to repair stalled forks before they result in damage, or a failure in labelling replication stress-induced DNA damage via the canonical pathway observed in model cellular systems (43) and other cancers (11).

PARP inhibitors (PARPi) recently emerged as a new therapy to treat cancers with HR repair deficiencies (HRD) caused by mutations in HR genes such as BRCA1/2 (46). Given that many of the HGSC cell lines demonstrated replication stress that might result in increased stalled forks, we wondered whether this might result in sensitivity to PARPis. We therefore treated all cell lines with a range of olaparib doses and monitored cell growth using IncuCyte cellular imaging (Fig. 4A). Most HGSC cell lines demonstrated similar sensitivity to FNE1 cells (proliferation reduced by 50% at 1–10 μmol/L olaparib). However, AOCS1 and Ovkate were more resistant, suggesting a differential HR capability between these and the other HGSC cell lines. We therefore directly tested cellular responses to ionizing radiation using a Rad51 focus formation time-course assay, a common measure of HR efficiency (Fig. 4B; refs. 47, 48). FNE1 cells showed an increase in Rad51 foci within 2 hours, and by 24 hours, foci began to decrease in number (Fig. 4C). This pattern was recapitulated in all HGSC cell lines, with the notable exception of AOCS1, that showed a marked delay in formation of Rad51 foci (Fig. 4C). Collectively, these data suggest that the relative resistance of AOCS1 and Ovkate to olaparib was not due to HRD-induced sensitivity in the other cell lines. A delayed Rad51 response in AOCS1 cells may, therefore, reflect a general failure of replication stress and/or DNA damage response rather than HR capability directly. We observed reduced replication stress responses (such as low γH2AX and 53BP1 foci despite reduced replication fork speeds, and failure to induce 53BP1 foci upon aphidicolin treatment) in AOCS1 and Ovkate (see Fig. 3B, C, and J). We suggest that an attenuated replication stress response might therefore serve as a novel indicator of resistance to PARPi therapy.

Paclitaxel is a mainstay chemotherapy agent used to treat patients with HGSC, but can be very neurotoxic (49). There are currently no biomarkers for which patients would benefit most from paclitaxel as a chemotherapy. Because paclitaxel stabilizes MTs, we reasoned that cell lines with elevated MT assembly rates might be inherently resistant to paclitaxel. Indeed, Kuramochi, the cell line that exhibits near-normal MT assembly rates (see Fig. 2F), was the most sensitive to paclitaxel, with a proliferation response similar to FNE1 and FNE2 (Fig. 4D). Other cell lines showed increasing resistance to paclitaxel that positively correlated with their MT assembly rates although this did not reach significance (Fig. 4E). This suggests that methods to assess elevated MT assembly rates in cancer patient samples could be further explored as an indicator for paclitaxel effectiveness.

Given the links between functional CIN readouts and chemotherapy sensitivities, we were motivated to thoroughly assess whether genomic, genetic, or transcriptomic analyses (more readily available from patient samples) might indicate these key CIN features, and also shed light on the precise deregulated pathways culminating in the observed functional phenotypes. For this purpose, we performed single-nucleotide variant (SNV) calling using the Genome Analysis Toolkit workflow, utilizing a Panel of Normals to stringently remove common SNPs, as matched normal samples were not available. We thus identified a list of likely somatic mutations (Supplementary Table S2; Supplementary Fig. S5A). Most cell lines carried two to seven potentially CIN-related mutations; however, none of these obviously cosegregated with specific phenotypes (Supplementary Fig. S5A). G33 bore 13 mutations in potential CIN genes, although the overall genome mutation rate was similar to the other HGSC cell lines (Supplementary Fig. S5B). Next, we analyzed the transcriptome of each cell line using RNA-seq. We collated all significantly altered CIN-related genes (Supplementary Table S3;Supplementary Fig. S5C). Of these, RB1, TopBP1, CCNE1 (cyclin E), and genes from MCM and GINS DNA replication complexes were frequently deregulated (Supplementary Fig. S5C). We noted that AOCS1 was the only cell line to show significant elevated expression of 53BP1. We also looked for gene expression changes that were linked to particular phenotypes. For PARPi resistance (Fig. 4A), we examined genes whose expression was altered (relative to both FNE1 and 2) in both Ovkate and AOCS1, but not altered in the five HGSC cell lines that showed olaparib sensitivity (Supplementary Fig. S5D). We found no statistically significant gene ontology pathway enrichment; however, we did note that there were five significantly altered genes with DNA replication and repair functions (Supplementary Fig. S5D). One of these was TLK1, a kinase responsible for assembly of nucleosomes on replication forks with a potential role in PARPi response (50). Similarly, an analysis of the cell lines with strongest paclitaxel resistance (Fig. 4D) revealed a cluster of eight genes involved in MT regulation, and five DNA replication-related genes were commonly altered in cell lines with an attenuated response to replication stress (Supplementary Fig. S5D). When comparing between cell lines with differential responses to nucleoside-mediated segregation error rescue (Fig. 3F), we found an enrichment in genes regulating nucleotide metabolism, which might explain why some cell lines had better CIN reduction than others. For the congression defects observed in four cell lines (Fig. 2C), the most relevant significantly enriched pathway was actin cytoskeleton reorganization (Supplementary Fig. S5D). As well as concentrating on known CIN-related pathways, we also identified the significantly altered genes and pathways in those HGSC cell lines sharing specific CIN phenotypes, analyzed independently from the FNE controls (Supplementary Fig. S6A and S6B; Supplementary Table S4). Overall genetic and transcriptomic analysis did not reveal any clear causative pathways to CIN, but these data provide a resource for future identification of genes related to those phenotypes, and to potentially identify novel roles in CIN for other genes.

Here we have performed the first systematic and comprehensive functional analysis of mechanisms driving CIN in a panel of representative HGSC cell lines. All seven lines demonstrate extensive ongoing CIN in the form of chromosome mis-segregation that is associated with multiple mechanisms, including elevated MT assembly rates, centrosome amplification, and replication stress (see Fig. 5A and B for a summary and correlations between phenotypes across all cell lines). A striking finding of our study is that compared to colorectal cancer cell lines, HGSC cell lines exhibit a greater number of cooperating CIN mechanisms, and each CIN mechanism often operates at more extreme levels (e.g., slower replication fork rates and higher MT assembly rates than colorectal cancer cell lines). Moreover, multiple CIN mechanisms likely exist within single HGSC cells as evidenced by the presence of chromosome segregation errors of multiple types per cell (Fig. 1G–J). This complexity explains the partial reductions in CIN obtained using CIN limiting experiments (low-dose paclitaxel and nucleoside supplementation), and will be important to consider when designing approaches to target CIN therapeutically in this disease. We also noted that HGSC lines were not entirely uniform in the extent to which each CIN mechanism was manifested. Interestingly we discovered that these differences in characteristics or severity of specific CIN mechanisms were related to sensitivity to paclitaxel and PARP inhibition, mainstays of HGSC treatment. Whole-genome sequencing and transcriptomic analysis revealed potential genetic drivers of CIN in HGSC that can be harnessed in future mechanistic studies. Given these findings, we suggest that assessing CIN mechanisms in patients with HGSC either functionally or by association with specific mutational or transcriptomic signatures may provide new biomarkers to predict sensitivity to paclitaxel and PARPis.

Centrosome amplification has been observed in many cancer types, although its origin is still unclear. Whole-genome doubling (WGD) events caused by cytokinesis failure could generate both centrosome amplification and increases in chromosome ploidy. Indeed, WGD is estimated to occur particularly frequently in HGSC (51). However, many cells exhibited centriole numbers exceeding eight (the expected consequence of cytokinesis failure; Supplementary Fig. S3G) and some centrosome amplified lines were close to diploid (e.g., Kuramochi), suggesting potential alternative routes to centrosome amplification. One such cause could be replication stress itself, since slow replication can cause extra centrosomes (52). Centrosome amplification correlated with increased segregation errors in the cell lines tested, but cells with normal centriole numbers also exhibited high error rates, suggesting this is not the main driver of CIN in HGSC (Supplementary Fig. S3F).

Most HGSC lines displayed defects in chromosome congression, and all cell lines except Kuramochi exhibited markedly elevated MT assembly rates. Notably, all of the cell lines we tested demonstrated significantly reduced segregation errors after treatment with low doses of paclitaxel to restore aberrant MT assembly (Fig. 2K), suggesting this is a major contributor to HGSC CIN. In colorectal cancer, elevated MT assembly rates are proposed to occur as a result of overactive Aurora kinase A (9, 36) or a deregulated BRCA1–Chk2 axis (8, 9). None of our eight cell lines exhibited BRCA1 point mutations. Some lines displayed evidence of partial BRCA1/2 copy-number loss but we saw no significant reduction in BRCA1/2 expression at the RNA level except in Ovsaho (Figs. S1G, S1J, and S5C). All lines except Snu119, however, displayed copy-number gain of the AURKA locus (Supplementary Fig. S3H and S3I), and all cell lines showed Aurora Kinase A overexpression at mRNA and protein level, suggesting that in this panel at least, AURKA gene amplification may be a key driver of MT overassembly, mitotic abnormalities, and CIN in HGSC. However, despite displaying Aurora Kinase A upregulation, Kuramochi did not exhibit significantly elevated MT assembly rates. Kuramochi was also notably the most sensitive to higher, clinically relevant doses of paclitaxel. Discovering the cause of these differential phenotypes and whether functional biomarkers could be derived (such as a clinical measure of MT assembly rates) may lead to the first clinical biomarker for paclitaxel sensitivity.

It is also possible that other genetic lesions could generate elevated MT assembly rates, for example, DNA-PKcs has recently been identified as upstream of the BRCA1–Chk2 axis (53). Cep72 has been implicated in MT assembly defects and is significantly overexpressed in Ovkate (41). Moreover, comparison of transcriptomic features of cell lines that were most affected by congression delays identified an enrichment of 11 genes in actin reorganization, that might indicate a systemic cytoskeletal defect. Supernumerary centrosomes could also contribute to elevated MT assembly rates as a result of increased MT nucleation (54), suggesting the intriguing possibility that extra centrosomes might mediate chromosome missegregation via mechanisms over and above their canonical role in promoting abnormal geometry during spindle formation. Finally, a connection between replication stress and elevated MT assembly rates has also been described in colorectal cancer (55) and it would be interesting to examine this further in HGSC.

Alongside mitotic defects, all seven HGSC cell lines also exhibited some degree of replication stress, as evidenced from single molecule DNA fibre analysis and other hallmarks of replication stress. Oncogene-induced replicative and mitotic stress can result from a myriad of known oncogenes and tumor suppressor genes (56, 57); however, HGSC harbors relatively few recurrent mutations. Similarly, our panel of cell lines shared few mutations in common (Supplementary Fig. S5A), with the exception of TP53. Cyclin E has been noted as a common defect in HGSC (20). Five cell lines showed CCNE1 amplification at gene level, six had significantly increased RNA expression and all showed increased protein expression (Supplementary Fig. S1I); however, this did not seem to correlate with any of the specific replication stress phenotypes we investigated herein and may simply play a role in promoting replication stress and CIN in general in HGSC (Fig. 5A; ref. 58). Replication stress has been successfully treated with exogenous nucleoside supplementation in a variety of models (11, 44, 59, 60). We found that nucleosides had a positive, or at worst only neutral, effect on segregation errors in HGSC cell lines. Crucially, we saw no negative effect in the FNE1 control. This suggests that nucleoside supplementation could be investigated further, perhaps as a prophylactic treatment for women with familial risk of developing HGSC, or given alongside therapy to reduce the likelihood of developing treatment-resistant disease.

Unexpectedly, many of the cell lines did not entirely follow the canonical response (11, 43) to replication stress. AOCS1 and Ovkate were among those cell lines particularly unusual in lacking multiple replication stress markers including γH2AX and 53BP1 bodies and when further challenged with aphidicolin, both remained incapable of mounting a DNA damage response (Fig. 3B, C, and J). Strikingly, these two cell lines were resistant to PARP inhibition compared with other cell lines and the FNE controls. Moreover, this was connected, in the case of AOCS1, to a defective Rad51 recruitment to IR-induced DNA damage, which would normally be taken to indicate HR deficiency. All other cell lines recruited Rad51 with normal kinetics. In line with this, there was no significant loss of expression at RNA level of known HR genes (except for RAD51D in Kuramochi and BRCA2 in Ovsaho; Supplementary Fig. S5C) despite the reduced copy number of BRCA1 and 2 in multiple cell lines and the known mis-sense mutation detected in Kuramochi (Supplementary Fig. S1G, S1J; Supplementary Fig. S5A and S5C; ref. 23). Originally, the success of PARPis was attributed to the synthetic lethality involved in disabling the break excision repair function of the PARP pathway, leaving HRD tumors unable to repair single-stranded DNA breaks. More recent work has pointed to an increased complexity of possible mechanisms, with PARP inhibition resulting in stalled replication forks that cannot be restarted in a BRCA-minus background, and thus become vulnerable to degradation due to excessive MRE11 nuclease activity (61). New “backup” pathways have also been described in cancer cells, such as reliance on Rad52 to promote Rad51 loading in BRCA-deficient cells (62), or the downregulation of PTIP to reduce MRE11 recruitment (61). Because replication stress is also likely to result in stalled forks that need repair or restart, these different pathways to stalled fork resolution may also underpin the variation in replication stress response observed within our HGSC cell line panel. It will be interesting to test whether inhibition of Rad52 would resensitize AOCS1 and Ovkate to PARP inhibition. These studies, alongside our findings that AOCS1's delayed Rad51 focus formation correlated with PARP inhibition resistance rather than sensitivity, suggest that consideration of replication stress levels and responses may also need to be taken into consideration when predicting clinical PARPi sensitivity, and when interpreting functional assessments of HR competency.

AOCS1 carried the highest number of mutations in known CIN-related genes (Supplementary Fig. S5A). We also noted that AOCS1 was the only cell line to show elevated gene expression of 53BP1 despite an inability to form 53BP1 bodies (Fig. 3J; Supplementary Fig. S5C), which might lead to an imbalance in the nonhomologous end joining/HR responses to DNA damage and contribute to PARPi resistance. Intriguingly, AOCS1 is also the only cell line with significant loss of p53 expression at both RNA (Supplementary Fig. S5C) and protein (Supplementary Fig. S1H) levels, whereas the other cell lines expressed mutant alleles of full length or truncated (Ovsaho) protein. Loss of p53 expression has previously been linked with aberrant Rad51 recruitment (63) and AOCS1′s nonsense allele (W146*) has been observed in a subset of patients with HGSC(64). It would be worth investigating the extent to which this mutation plays a role in AOCS1's unique phenotypes both experimentally and in clinical datasets.

The lack of matched normal patient DNA samples made some genetic analyses difficult. Herein we used various approaches to circumvent this issue. However, future studies will benefit from the acquisition of patient-specific matched normal DNA samples to permit more extensive analyses to link genotype to CIN phenotypes. Another potential caveat of this study is that there may be differences in the behavior of cell line models compared with cells within tumors in patients. To mitigate against this, we selected a panel of HGSC cell lines whose genetic and genomic features recapitulated HGSC tumors (23–25). We also enhanced the strength of our analysis by comparing our RNA expression profiles and CIN phenotypes not just between cancer cell lines but also against two appropriate, tissue-specific control cell lines. Moreover, the behavior of this cell line panel was notably different to previously characterized colorectal cancer cell line panels that did not display overt chromosome congression delays or supernumerary centrosomes (8, 11). Therefore, tumor-derived cell lines still represent the best currently available model to functionally connect genetic lesions, genomic alterations and CIN mechanisms. Moreover, a landmark study has recently shown extensive cell division abnormalities from patient-derived HGSC cancer cells, at similar rates to what we have observed in our cell lines (13). This provides an important complementary validation of our findings, and suggests the CIN mechanisms identified herein are likely to be bona fide HGSC CIN mechanisms, that can be validated in vivo in future studies.

This study provides new understanding of the nature of CIN in HGSC, is directly comparable with existing knowledge of CIN mechanisms in other cancer types, and moreover lays the groundwork for future studies to validate mechanisms driving HGSC CIN in vivo. Our findings also have the potential to facilitate future research into synergizing with patient-specific CIN mechanisms as a therapeutic strategy.

N. Tamura reports grants from Wellbeing of Women during the conduct of the study. N. Shaikh reports grants from Pancreatic Cancer Research Fund and Cancer Research UK during the conduct of the study and grants from AstraZeneca outside the submitted work. D. Muliaditan reports grants from Wellbeing of Women during the conduct of the study. C.M. Green reports grants from Wellcome Trust during the conduct of the study. K. Curtius reports grants from UKRI during the conduct of the study. S.E. McClelland reports grants from Astra Zeneca outside the submitted work. No potential conflicts of interest were disclosed by the other authors.

N. Tamura: Conceptualization, formal analysis, investigation, writing-review and editing. N. Shaikh: Conceptualization, formal analysis, investigation, writing-review and editing. D. Muliaditan: Formal analysis, investigation, methodology, writing-review and editing. T.N. Soliman: Investigation, writing-review and editing. J.R. McGuinness: Investigation. E. Maniati: Investigation. D. Moralli: Investigation. M.-A. Durin: Investigation. C.M. Green: Supervision. F.R. Balkwill: Resources. J. Wang: Resources, supervision. K. Curtius: Supervision, writing-review and editing. S.E. McClelland: Conceptualization, resources, supervision, funding acquisition, investigation, writing-original draft, project administration, writing-review and editing.

We would like to thank D. Bowtell for intellectual input and sharing of unpublished data, C. Swanton for the kind gift of SW620 and HCT116 cell lines, and S. Godinho for the kind gift of EB3-GFP constructs. We thank R. Basto and S. Taylor for sharing unpublished data and helpful discussions. We would also like to thank all members of the McClelland laboratory for useful discussions and reading of the manuscript. We thank the CRUK Flow Cytometry Core Service at Barts Cancer Institute (Core Award C16420/A18066). We acknowledge Tan A. Ince, Live Tumor Core (LTCC) at University of Miami for FNE1 and FNE2 cell lines.

N. Tamura was funded by Barts Charity (487/2133) and a Wellbeing of Women project grant (RG2040). D. Muliaditan and T.N. Soliman were funded by a Wellbeing of Women project grant (RG2040). N. Shaikh was funded by the Pancreatic Cancer Research Fund (PCRF) and a CRUK Pioneer Award (C35980/A27846). C.M. Green, D. Moralli, and M.-A. Durin were funded by Wellcome core award (090532/Z/09/Z). F.R. Balkwill was funded by European Research Council (ERC322566) and Cancer Research UK (A16354, A25714). E. Maniati and J. Wang acknowledge support from Cancer Research UK Centre of Excellence Award to Barts Cancer Centre (C16420/A18066). K. Curtius was funded by a UKRI Rutherford Research Fellowship.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.