A living biobank of ovarian cancer ex vivo models reveals profound mitotic heterogeneity

Abstract

High-grade serous ovarian carcinoma is characterised by TP53 mutation and extensive chromosome instability (CIN). Because our understanding of CIN mechanisms is based largely on analysing established cell lines, we developed a workflow for generating ex vivo cultures from patient biopsies to provide models that support interrogation of CIN mechanisms in cells not extensively cultured in vitro. Here, we describe a “living biobank” of ovarian cancer models with extensive replicative capacity, derived from both ascites and solid biopsies. Fifteen models are characterised by p53 profiling, exome sequencing and transcriptomics, and karyotyped using single-cell whole-genome sequencing. Time-lapse microscopy reveals catastrophic and highly heterogeneous mitoses, suggesting that analysis of established cell lines probably underestimates mitotic dysfunction in advanced human cancers. Drug profiling reveals cisplatin sensitivities consistent with patient responses, demonstrating that this workflow has potential to generate personalized avatars with advantages over current pre-clinical models and the potential to guide clinical decision making. High-grade serous ovarian carcinoma is often associated with TP53 mutation and chromosomal instability (CIN). Here, the authors generate ex vivo cultures from biopsies and ascites of patients and perform characterization to evaluate CIN mechanisms and compare drug sensitivity with patient responses.

Full text

ARTICLE
A living biobank of ovarian cancer ex vivo models
reveals profound mitotic heterogeneity
Louisa Nelson1,11, Anthony Tighe1,11, Anya Golder 1, Samantha Littler1, Bjorn Bakker2, Daniela Moralli 3,
Syed Murtuza Baker4, Ian J. Donaldson 4, Diana C.J. Spierings 2, René Wardenaar2, Bethanie Neale5,
George J. Burghel6, Brett Winter-Roach7, Richard Edmondson1,8, Andrew R. Clamp9, Gordon C. Jayson1,9,
Sudha Desai 10, Catherine M. Green3, Andy Hayes 4, Floris Foijer 2, Robert D. Morgan1,9 &
Stephen S. Taylor 1*
High-grade serous ovarian carcinoma is characterised by TP53 mutation and extensive
chromosome instability (CIN). Because our understanding of CIN mechanisms is based
largely on analysing established cell lines, we developed a workflow for generating ex vivo
cultures from patient biopsies to provide models that support interrogation of CIN
mechanisms in cells not extensively cultured in vitro. Here, we describe a “living biobank”of
ovarian cancer models with extensive replicative capacity, derived from both ascites and solid
biopsies. Fifteen models are characterised by p53 profiling, exome sequencing and tran-
scriptomics, and karyotyped using single-cell whole-genome sequencing. Time-lapse
microscopy reveals catastrophic and highly heterogeneous mitoses, suggesting that analy-
sis of established cell lines probably underestimates mitotic dysfunction in advanced human
cancers. Drug profiling reveals cisplatin sensitivities consistent with patient responses,
demonstrating that this workflow has potential to generate personalized avatars with
advantages over current pre-clinical models and the potential to guide clinical decision
making.
https://doi.org/10.1038/s41467-020-14551-2 OPEN
1Division of Cancer Sciences, Faculty of Biology, Medicine and Health, University of Manchester, Manchester Cancer Research Centre, Wilmslow Road,
Manchester M20 4GJ, UK. 2European Research Institute for the Biology of Ageing (ERIBA), University of Groningen, University Medical Center Groningen,
9713 AV Groningen, The Netherlands. 3Wellcome Centre Human Genetics, University of Oxford, Roosevelt Drive, Oxford OX3 7BN, UK. 4Genomic
Technologies Core Facility, Faculty of Biology, Medicine and Health, University of Manchester, Michael Smith Building, Dover Street, Manchester M13 9PT,
UK. 5NIHR Manchester Biomedical Research Centre, Manchester University NHS Foundation Trust, Manchester Academic Health Science Centre,
Manchester, UK. 6Genomic Diagnostic Laboratory, St Mary’s Hospital, Central Manchester NHS Foundation Trust, Oxford Road, Manchester M13 9WL, UK.
7Department of Gynaecological Surgery, The Christie NHS Foundation Trust, Wilmslow Rd, Manchester M20 4BX, UK. 8Department of Gynaecological
Surgery, St Mary’s Hospital, Central Manchester NHS Foundation Trust, Oxford Road, Manchester M13 9WL, UK. 9Department of Medical Oncology, The
Christie NHS Foundation Trust, Wilmslow Rd, Manchester M20 4BX, UK. 10 Department of Histopathology, The Christie NHS Foundation Trust, Wilmslow
Rd, Manchester M20 4BX, UK.
11
These authors contributed equally: Louisa Nelson, Anthony Tighe. *email: stephen.taylor@manchester.ac.uk
NATURE COMMUNICATIONS | (2020) 11:822 |https://doi .org/10.1038/s41467-020-14551-2 | www.nature.com/naturecommunications 1
1234567890():,;
Content courtesy of Springer Nature, terms of use apply. Rights reserved

Ovarian cancer is the leading cause of gynaecological-
related mortality, accounting for ~152,000 deaths world-
wide annually1. The most prevalent subtype, high-grade
serous ovarian carcinoma (HGSOC), which is believed to originate
from the fallopian tube2–5, is particularly lethal because it develops
rapidly and often presents with advanced stage disease. Treatment
options are limited, typically cytoreductive surgery and plati-
num/paclitaxel-based chemotherapy6. While many patients initi-
ally respond well, most develop recurrent disease, yielding
relatively poor survival rates that have not changed substantially
for 20 years7.
HGSOC is characterised by ubiquitous TP53 mutation and
extensive copy number variation8,9. Recurrent amplifications of
MYC,PTK2 and CCNE1 are common, whereas PTEN is fre-
quently lost, and chromosome breakage events often inactivate
NF1 and RB110–12.BRCA1/BRCA2 are inactivated in ~20% of
cases, leading to homologous recombination (HR) defects10, but
DNA damage repair defects are more widespread12,13. Extensive
copy number variation implies chromosomal instability (CIN),
i.e. the gain/loss of chromosomes and/or acquisition of structural
rearrangements14. While p53 loss permits CIN, the underlying
primary causes remain poorly understood and are likely com-
plex15–17. Indeed, whole-genome sequencing of HGSOCs iden-
tified multiple CIN signatures, including foldback inversions, HR
deficiency and whole-genome duplication18,19.
CIN presents both challenges and opportunities when treating
HGSOC. By driving phenotypic adaptation, CIN accelerates drug
resistance; ABCB1 rearrangements have been identified in 18.5%
of recurrent tumours, enhancing drug-pump-mediated efflux of
chemotherapy agents12,20. However, CIN can be exploited to
develop synthetic-lethality-based strategies, pioneered by the use
of poly (ADP-ribose) polymerase (PARP) inhibitors to target
BRCA-mutant tumours21–27. Because of the paucity of actionable
driver mutations in HGSOC, synthetic lethality is an attractive
option and a better understanding of CIN may open up new
therapeutic strategies.
Delineating disease-specific CIN mechanisms and developing
novel therapeutic strategies requires models that reflect various
human cancers. While judiciously selected cell lines provide
tractable models to study cancer cell biology28, they under-
represent the genetic heterogeneity exhibited by tumours29 and
lack the clinical annotations necessary to correlate in vitro drug
sensitivities with in vivo chemotherapy responses. While patient-
derived xenografts are excellent translational resources30,31, high-
throughput drug profiling is difficult and the timescales involved
are challenging in terms of directing personalised treatment. By
contrast, living biobanks have the potential to more rapidly
generate well-characterised and tractable models suitable for
discovery research, drug screening and guiding clinical deci-
sions32–35. To develop clinically annotated models that recapi-
tulate HGSOC, we built a living biobank of ex vivo cultures. Here
we describe a workflow and exemplar panel of ovarian cancer
models (OCMs), and demonstrate their potential to study CIN
and drug sensitivity.
Results
Establishing a living biobank of ovarian cancer ex vivo models.
To build a living biobank, we established a biopsy pipeline, col-
lecting samples from patients diagnosed with epithelial ovarian
cancer treated at the Christie Hospital, and a workflow to gen-
erate ex vivo OCMs with extensive proliferative potential.
Between May 2016 and June 2019, we collected 312 samples from
patients with chemo-naïve and relapsed disease, either as solid
biopsies or as ascites (Fig. 1a). Using our standard workflow, thus
far we have generated 76 ex vivo cultures. Here, as proof of
principle, we describe 15 OCMs derived from 12 patients.
Average patient age at diagnosis was 59 years (range 25–81 years)
with a mean survival from diagnosis of 27 months (range
2–125 months; Supplementary Table 1). For 12 samples, ascites
were collected following treatment while two ascites and one solid
biopsy were chemo-naïve. Ten patients had HGSOC while two
had mucinous ovarian carcinoma. Longitudinal biopsies were
collected from three patients (Fig. 1a).
To establish cultures, red blood cells were lysed, the remaining
cellular fraction harvested by centrifugation, disaggregated if
necessary then plated in OCMI media (Fig. 1b). Serial passaging
and selective detachment eliminated white blood cells and yielded
separate tumour and stromal fractions, which were characterised
using phenotypic assays prior to next-generation sequencing and
functional profiling. The models are referred to using the OCM
prefix followed by the patient number and, if one of a series, the
biopsy number (Supplementary Fig. 1a). Models generated
independently from the same biopsy are distinguished by an
alphabetical suffix. Pilot experiments showed that standard media
formulations only supported proliferation of the stromal cells.
However, during the course of our pilot studies, Ince et al.
described OCMI media which enabled them to establish 25 new
patient-derived ovarian cancer cell lines36. In our hands, OCMI
also supported tumour cell proliferation, allowing us to routinely
generate primary cultures with extensive proliferative potential
(Supplementary Fig. 1a). Thus our observations confirm the
ability of OCMI media to routinely generate ex vivo ovarian
cancer models.
Characterisation of ex vivo models. To determine whether the
OCMs possess the expected hallmarks of ovarian cancer, we
characterised the cultures using an array of molecular cell bio-
logical approaches (Fig. 1b). Tumour and stromal fractions were
morphologically differentiated, with the epithelial appearance of
the tumour cells contrasting the fibroblastic stromal cells
(Fig. 2a). Time-lapse microscopy and Ki67 expression confirmed
both fractions were proliferative (Fig. 2b, c), and the veracity of
the separation workflow was confirmed with immunological
markers and p53 profiling (Fig. 2d, e and Supplementary Figs. 1a
and 2a). Tumour cells were typically positive for PAX8, EpCAM
and CA125, and failed to elicit a functional p53 response upon
Mdm2 inhibition (Supplementary Fig. 1a). Consistently, tumour
cells expressed p53 mutants and frequently overexpressed MYC
(Supplementary Figs. 1a and 2a). Some tumour cells however
were negative for one or more tumour markers despite har-
bouring TP53 mutations (Supplementary Fig. 1a), possibly
reflecting tumour heterogeneity and/or epithelial–mesenchymal
transition37. In light of these exceptions, tumour cultures were
defined as such if they had an epithelial morphology, expressed
PAX8, EpCAM and/or CA125, and/or had a TP53 mutation,
while stromal cells were defined as having a fibroblastic mor-
phology, strong vimentin staining and wild-type TP53.
Interestingly, OCM.64–3, generated from the third biopsy from
patient 64, exhibited phenotypic heterogeneity; some cells had
large, atypical nuclei and were negative for PAX8 and EpCAM,
while others were positive for both and had smaller nuclei
(Supplementary Fig. 2b). EpCAM/PAX8-positive cells were not
detected in OCM.64–1, established from the first biopsy, possibly
reflecting tumour evolution during treatment. By exploiting
EpCAM status, we separated the two sub-populations (Supple-
mentary Fig. 2c), revealing that only the EpCAM-negative
population (OCM.64–3Ep−) expressed high levels of MYC
(Supplementary Fig. 2a).
Two tumour cultures, OCM.69 and OCM.87, had wild-type
TP53 and a functional p53 response (Supplementary Figs. 1a and
ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14551-2
2NATURE COMMUNICATIONS | (2020) 11:822 | https://doi.org/10.1038/s41467-020-14551-2 | www.nature.com/naturecommunications
Content courtesy of Springer Nature, terms of use apply. Rights reserved

2a). Re-evaluation of OCM.69, which was also CA125 and EpCAM
negative, demonstrated stromal overgrowth so this culture was used
as a negative internal control for subsequent studies. By contrast,
OCM.87 was positive for PAX8, EpCAM and CA125 and thus
confirmed as a tumour model. To determine whether OCMs
reflected the primary tumours, we analysed archival tissue, either
from the original diagnostic biopsy or from primary cytoreductive
surgery (Fig. 1a). Formalin-fixed and paraffin-embedded archival
tumour blocks were available for eight patients and immunohis-
tochemistry (IHC) analysis correlated well with immunofluores-
cence analysis of the ex vivo cultures (Supplementary Fig. 1a, b). For
example, OCMs 61 and 72, the two mucinous tumours, were PAX8
negative in both contexts. By contrast, OCMs 46, 66 and the other
the HGSOC tumours were PAX8 positive, consistent with a
fallopian tube origin. Interestingly, 74, which yielded a PAX8-
negative OCM 9 years later, displayed focal PAX8 staining
indicating that heterogeneity already existed in the primary tumour.
Nevertheless, these observations demonstrate that the OCM models
possess the hallmarks of cancer cells and reflect their respective
primary tumours.
Exome and gene expression analysis. To determine if the models
displayed the genomic features typical of HGSOC, they were
interrogated by exome sequencing and RNAseq. Analysis of
exome variants showed that sequential cultures from the same
patient had similar mutational burdens (Fig. 3a). p53-proficient
OCM.87 displayed a highly elevated mutational load, possibly
indicating a tumour driven by a mismatch repair defect. By
contrast, the well-differentiated mucinous ovarian carcinoma
38
*
33
*
61
59
*
66
**
64
**
87
*
79
*
46
*
69
*
*
72
*
Paclitaxel
Carbo/cisplatin
VEGFi
Other
Ascites collected
Solid biopsy
Debulking surgery
Ex vivo culture
*
123
Time (yrs)
Age at
diagnosis
81
68
71
53
58
54
25
73
57
45
64
RIP
71
RIP
83
RIP
72 RIP
46
RIP
65
RIP
61
RIP
57
RIP
30
RIP
57
RIP
74
**
74
64
RIP
74
a
b
OCMI
Selective
detachment
Fluid
RBC WBC
Tumour
Stromal
–80 °C Lenti-GFP-H2B
Validation:
Microscopy
Flow cytometry
p53 status
NGS:
Exome
RNAseq
scWGS
Cell biology:
Time-lapse
Drug sensitivity
Cell fate
Patient
ascites
LN2
LN2
Primary tumour
Fig. 1 Establishing a biobank of ovarian cancer ex vivo models. a Patient timelines showing age at diagnosis and death, treatments and biopsy collections.
bWorkflow for processing and storage of stromal and tumour fractions. (RBC red blood cells, WBC white blood cells, LN
2
and −80 °C specifies long-term
storage, OCMI ovarian carcinoma modified Ince medium, NGS next-generation sequencing, VEGFi vascular endothelial growth factor inhibitor). See also
Supplementary Fig. 1 and Supplementary Tables 1 and 2.
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14551-2 ARTICLE
NATURE COMMUNICATIONS | (2020) 11:822 | https://doi.org/10.1038/s41467-020-14551-2 | www.nature.com/naturecommunications 3
Content courtesy of Springer Nature, terms of use apply. Rights reserved

model, OCM.61, had a relatively low mutation rate. Interrogating
genes known to be mutated in HGSOC confirmed the TP53
lesions and identified additional mutations in BRCA1,NF1 and
RB1 (Fig. 3b). Importantly, targeted amplicon sequencing of the
primary tumours revealed TP53 mutations identical to those
identified by the exome sequencing (Supplementary Table 2),
again demonstrating that the OCMs reflect the primary tumours.
Gene expression profiling showed that the tumour and stromal
cultures clustered into two distinct clades (Fig. 3c). Principal
component analysis (PCA) showed that the stromal cultures
clustered very closely, despite originating from 12 different patients
(Fig. 3d). While the PCA scores for the tumour cultures associated
less tightly, those derived from the same patient, e.g. OCM.66–1
and OCM.66–5, clustered very tightly. The two mucinous cultures
were also closely associated while p53-proficient OCM.87 was an
outlier. The phenotypic heterogeneity displayed by OCM.64–3
also manifested in the PCA; OCM.64–3Ep−associated more
closely with EpCAM-negative OCM.64–1butwasdetachedfrom
OCM.64–3Ep+. Taken together, these observations further confirm
the separation of distinct tumour and stromal populations,
and also highlight the phenotypic inter and intratumour
heterogeneity.
Single-cell transcriptomics. To further explore the phenotypic
heterogeneity, we turned to single-cell approaches, initially ana-
lysing chemo-naïve OCM.38a using a Fluidigm platform. Hier-
archical clustering identified two dominant clusters, Tumour A
Epcam-PE/Cy7
CD44-BV421 CD105-APC
Epcam-PE/Cy7 Epcam-PE/Cy7
CA125-AF488 CA125-AF488 CA125-AF488
106
106107
105
105
104
104
103
103
102
102
101
100
Tumour
Stromal
Stromal
Mixed
Mixed
87.392.0
70.886.4
37.6
45.8
47.3
43.4
ac
de
TumourStromal
p53
DNA
Merge
04896
Time (H)
0
20
40
60
Confluency (%)
Stromal
048 96
Time (H)
Tumour
0
20
40
60
b
Ctrl Taxol Cisplatin
Tumour
Stromal
Pax8
Ki67
DNA Merge
Pax8
Ki67
DNA Merge
Stromal
Tumour
TumourStromal
OCM.79
OCM.79 OCM.79
OCM.38aOCM.38a
OCM.66-5
OCM.38a
OCM.79
Fig. 2 Characterisation of ex vivo models. a Phase contrast images showing distinct morphologies of stromal and tumour cells. Scale bar, 200 µm. bTime-
lapse imaging measuring confluency showing suppression of proliferation by 1 µM cisplatin and 100 nM paclitaxel. cImmunofluorescence images showing
expression of PAX8 and Ki67. Scale bar, 50 µm. dFlow cytometry profiles quantitating the tumour markers EpCAM and CA125, and the stromal markers
CD44 and CD105. Numbers represent percentage of cells in the quadrant. eImmunofluorescence images of Nutlin-3-treated cells (OCM.79) showing
stabilisation of p53 in stromal cells but not tumour, and DNA sequence showing TP53 mutation in tumour cells (OCM.38a). Scale bar, 20 µm. Data in
panels aand care derived from analysis of OCM.79, while data in panels band dare derived from analysis of OCMs 38a, and 66-5 respectively. Panels
a,cand eare representative images from single experiments. Source data for panels b,cand dare provided as a Source Data file, including the
gating/sorting strategy for panel d. See also Supplementary Figs. 1 and 2.
ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14551-2
4NATURE COMMUNICATIONS | (2020) 11:822 | https://doi.org/10.1038/s41467-020-14551-2 | www.nature.com/naturecommunications
Content courtesy of Springer Nature, terms of use apply. Rights reserved

and Stromal A (TA and SA, Fig. 4a). Interspersed within SA were
8 cells from the tumour fraction (TB), presumably contaminating
stromal cells. Adjacent to TA was a small cluster from the stromal
fraction (SB), possibly reflecting tumour contaminants in the
stromal fraction. A PCA and pathway analysis resolved SA into
two clusters, SAa and SAb, and SB formed a third, distinct cluster
(Fig. 4b). By contrast, TA comprised two overlapping clusters,
TAa and TAb. This classification was supported by interrogating
specific genes, with the tumour cells expressing EPCAM,TP53
and MYC but negative for XIST and TSIX, consistent with loss of
the inactive X chromosome (Fig. 4c). Interrogating cell cycle
signatures showed that TAa and TAb had low and high G2/M
scores respectively (Fig. 4d). Moreover, genes involved in mitosis
and chromosome segregation were overdispersed in the tumour
cells (Fig. 4e, f), and the cells expressing high levels of mitotic
genes had high G2/M scores (Fig. 4g). Thus, the heterogeneity
exhibited by the tumour cells most likely reflects cell cycle stage.
To extend this analysis, we analysed OCMs 38b, 59, 74–1 and
79 using a 10x Genomics platform. Tumour and stromal cells
from the four pairs were mixed 3:1 and analysed in parallel. t-
SNE plots showed that the majority of cells from each sample
formed distinct clusters, whereas smaller fractions formed an
overlapping cluster (Fig. 5a). Based on the 3:1 mix, we reasoned
that the large distinct clusters represented the tumour cells while
the overlapping cluster corresponded to the stromal cells.
Consistently, the distinct clusters accounted for ~75% of the
cells while ~25% made up the overlapping cluster (Fig. 5b).
Moreover, cells in two of the distinct clusters did not express
XIST (Fig. 5c), consistent with loss of the inactive X chromosome.
Pathway analysis identified 10 different sub-clusters (Fig. 5d).
Seven were private to the tumour cells, with OCMs 38b and 79
dominated by single sub-clusters (1 (87%) and 2 (96%)),
OCM.74–1 composed of two sub-clusters (3 (69%) and 7
(30%)), and OCM.59 composed of three (6 (46%), 8 (17%) and
Stromal Tumour
69*
66–5
61
79
87
38
64–1
33
64–3
74–3
59
69
74–1
72
66–1
46-p5
46-p15
74–1
74–3
64–3–
64–1
33
64–3+
64–3
87
38
79
59
72
61
66–5
66–1
46-p14
46-p4
0 1×1042×1043×104
102
103
104
105
LOH
Somatic
p14
p4
5172
33
31
79
87-p53+
38b
59
3–
3+
69*
3
1
64
74
66
46
61
c
ab
–50 0 50 100
–60
–40
–20
0
20
40
PC1: 42% variance
PC2: 10% variance
69*
Stromal
61
72
Mucinous
p4
p14
46
15
66
13
74
13
3+
3–
64
38b 59
79
61
33
Tumour
d
564 genes
–15 0 15
Deletion
Insertion
Substitution
Missense
Nonsense
33
38b
46
59
64–1
64–3
64–3–
66–1
66–5
69
72
74–1
74–3
79
87
61
64
–
3+
TP53
BRCA1
BRCA2
BRAF
E2F7
EGFR
ERBB3
ITGAV
JAK2
KIT
KRAS
MAP3K5
MECOM
MLH1
MTOR
NCOR2
NF1
NOTCH2
NOTCH4
NUMBL
PDGFRB
PRKCI
PTEN
RB1
RBL2
RPTOR
STAT1
TSC2
87-p53+
Fig. 3 Exome and gene expression analysis. a Whole-exome sequencing showing somatic and loss of heterozygosity variants identified by referencing
tumour cells to their matched stromal counterparts. bSummary of mutations in genes associated with HGSOC. cHierarchical clustering and dprincipal
component analysis of global gene expression profiles, distinguishing stromal and tumour clades, and showing the close relationship of tumour samples
from the same patient. 69* is a stromal culture. Symbol colours in aand dserve to distinguish different OCM tumour samples. Source data for panels aand
dare provided as a Source Data file.
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14551-2 ARTICLE
NATURE COMMUNICATIONS | (2020) 11:822 | https://doi.org/10.1038/s41467-020-14551-2 | www.nature.com/naturecommunications 5
Content courtesy of Springer Nature, terms of use apply. Rights reserved

9 (37%)). By contrast, three sub-clusters were shared between the
stromal cells from all four patients; for example, 24%, 42% and
35% of the OCM.38b stromal cells fell into sub-clusters 4, 5 and
10 respectively. Thus, single-cell transcriptomics confirms that
despite originating from different patients, the stromal cells are
phenotypically similar while the tumour cells display marked
inter-patient heterogeneity. Further analysis will however be
required to evaluate the nature of this heterogeneity, including
whether or not it reflects differences in cell cycle stage.
Nevertheless, these data highlight an advantage of deriving
ex vivo models, namely the ability to analyse highly purified
tumour fractions unfettered by contaminating stromal cells and
the microenvironment.
Single-cell shallow whole-genome sequencing.Tokaryotypethe
ex vivo models, cultures were subjected to single-cell whole-genome
sequencing (scWGS). Analysis of stromal cultures showed that
they were largely diploid (Fig. 6a and Supplementary Fig. 3b). By
contrast, the tumour cells displayed profound deviations. Moreover,
the inter-cellular heterogeneity within any given culture was
g
TA cells
Mitotic genes G2/M
0
0.5
1.0
0510
b
–200 –100 0 100 200
–100
–50
0
50
Integrated analysis
PC 1: 42% variance
PC 2: 3% variance
TAb
TB
SB
TAa
SAa
SAb
a
f
Tumour B
Stromal B
Stromal A
Tumour A
15
10
5
0
UbcH10
BubR1
Bard1
Topo2 α
Eg5
Brca1
Nek2
CEP55
RRM2
Mklp2
Cenp-E
Mad2
PRC1
Bub1
KIF23
Survivin
CDK1
Hurp
Ki-67
Cap-G
CycB1
Cdkn2
KIF20B
HMGA2
CycB2
Ndc80
Nuf2
Knl1
Mps1
Cenp-K
Asp
Sgo2
HMMR
d
TAa TAb SAa SAb SB
EPCAM
ATR
PRKCI
PTK2
EIF5A2
KRAS
PIK3CA
KDM5A
HES1
TP53
GSK3B
CDH2
MYC
BCL2L1
CDH1
DLL3
NCSTN
NOTCH3
CCNE1
MDC1
IL33
TFPI
TSIX
XIST
POSTN
COL1A1
IGFBP5
HAS2
HSD11B1
COL3A1
ANXA10
DAB2
0
5
10
TumourStromal
c
0 0.5 1.0
0
0.5
1.0
G1
G2/M
TAb
TAa
–8
–10
–12
–14
412
log10 pValue
Fold enrichment
–6
–4
–2
816
80
Count
Log10
pvalue
–10
Mitosis and
chromosome
segregation
Mitotic
cell cycle
2420
Overdispersed tumour genes
Spindle
checkpoint
Chromosome
segregation
–35
–45
–55
24
log10 pValue
Fold enrichment
–25
–15
–5
35
700
Count
Log10
pvalue
–40
Catabolism
Viral
processes
6
Overdispersed stromal genes
Intracellular transport
Ribosome
biogenesis
–20
Metabolism
Biosynthesis
e
Fig. 4 Fluidigm single-cell transcriptomics. a Hierarchical clustering of gene expression profiles distinguishing stromal and tumour cells from chemo-naïve
OCM.38a. bPrincipal component analysis integrated with pathway analysis showing subpopulations of tumour and stromal cells. cHeat map showing
mean expression levels of selected genes in OCM.38a tumour and stromal sub-populations. dScatter plots of G1 score versus G2/M score for individual
cells within the TAa/TAb sub-populations. eGene ontology analysis of overdispersed genes in stromal and tumour cells. fNetwork analysis of
overdispersed genes in tumour cells. gHeat map of overdispersed genes showing that TAb cells expressing higher levels of mitotic genes and have high
G2/M scores. Source data for panels b–eand gare provided as a Source Data file.
ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14551-2
6NATURE COMMUNICATIONS | (2020) 11:822 | https://doi.org/10.1038/s41467-020-14551-2 | www.nature.com/naturecommunications
Content courtesy of Springer Nature, terms of use apply. Rights reserved

conspicuous, consistent with extensive CIN. Interestingly, four
features stood out whereby genomes were marked by whole-
chromosome aneuploidies, rearranged chromosomes, monosomies
or tetrasomies (Fig. 6a, b and Supplementary Fig. 3b). OCMs 38a,
46 and 79 were characterised by whole-chromosome and chro-
mosome arm aneuploidies (Fig. 6a and Supplementary Fig. 3a). By
contrast, OCMs 33, 59 and 66–1 also displayed rearrangements and
focal amplifications. OCMs 64–1, 87, 38b and, to some extent,
64–3Ep−displayed numerous tetrasomies, while OCMs 64–3Ep+
and 74–1/3 harboured several monosomies (Fig. 6a and Supple-
mentary Fig. 3b). Note that OCM.38a and OCM.38b, independent
models developed from the same biopsy sample, had very different
karyotypes; whether this reflects intratumour heterogeneity or
evolution ex vivo remains to be determined. The two mucinous
samples were very different; chemo-naïve OCM.61 was largely
disomic but OCM.72 displayed numerous aneuploidies and focal
amplifications (Supplementary Fig. 3b). Note that while OCM.61
was derived from a low-grade mucinous adenocarcinoma, OCM.72
was derived from a poorly differentiated tumour, indicating more
aggressive disease (Supplementary Table 1). The karyotypes of the
OCM.64–3 sub-clones were strikingly different; while 64–3Ep−
displayed trisomies and tetrasomies, 64–3Ep+harboured mono-
somies and disomies (Fig. 6a). Moreover, there was an interesting
symmetry; the monosomic and disomic chromosomes in 64–3Ep+
were typically disomic and tri/tetrasomic respectively in 64–3Ep−
.
While the relationship between these sub-clones remains to be
determined, the scWGS vividly highlights the profound CIN
exhibited between and within different ovarian cancer models.
M-FISH reveals highly rearranged chromosomes. To verify the
CIN highlighted by the scWGS karyotyping, we used two
orthogonal approaches, namely multiplex fluorescence in situ
hybridization (M-FISH) and quantitation of mitotic spindle poles.
Compared with HCT116, a near-diploid colon cancer cell line,
OCMs 38b, 66–1 and 79 were dominated by features consistent
with the scWGS, namely tetraploidies, rearranged chromosomes
and whole chromosome aneuploidies respectively (Fig. 7a).
OCM.59 was also dominated by rearranged chromosomes,
including recurrent and unique derivative chromosomes, chro-
mosome fragments, micro-chromosomes, dicentrics and ring
chromosomes (Fig. 7b). Interestingly, the primary tumour from
patient 59 was notable in that the IHC analysis revealed profound
nuclear atypia and multi-nucleated giant cells (Supplementary
Fig. 1c), indicating that the extensive CIN observed ex vivo was
present in vivo.
Immunofluorescence analysis of the stromal cultures and nine
established ovarian cancer cell lines showed that mitotic cells
were typically bipolar (Fig. 7c, d). By contrast, multipolar spindles
were prevalent in OCM tumour cells. We extended this analysis
to include eight additional OCMs generated during the latter part
of this study, including three recently described by us38, thereby
including an additional four chemo-naïve models. All eight
satisfied the working definition above, i.e. they had epithelial
morphologies, were positive for PAX8, and/or had a TP53
mutation. Interestingly, in four out of six chemo-naïve OCMs,
multipolar spindles were rare (OCMs 38, 118, 124 and 195),
consistent with CIN becoming more pervasive as the disease
evolves in response to cytotoxic chemotherapy12,14. Nevertheless,
the M-FISH and spindle pole quantitation supports the extensive
CIN observed by the scWGS.
Quantitating spindle poles also gave us an opportunity to
analyse CIN in tumour cells at much earlier passage. Because the
selective detachment workflow requires several passages, the
ex vivo cultures were typically analysed by passage 10. To analyse
earlier stages, frozen unseparated populations were recovered
(Fig. 1b) and exposed to the Mdm2 inhibitor Nutlin-3, thereby
79
74–1
38b
59
XIST
Tumour
Stromal
a
cd
b
1
2
3
4
5
6
7
8
9
10
ST
0
20
40
60
80
100
Cell count (%)
38b
59
74–1
79
1
2
3
6
7
8
9
4
5
10
TumourStromal
100 50 0
e
74–1
38b
79
59
Stromal
Fig. 5 10x Genomics single-cell transcriptomics. a t-Stochastic neighbour embedding (t-SNE) plot showing clustering of single cells from four OCM pairs,
with tumour and stromal cells mixed 3:1. bDot plots quantitating the percentage of cells in the stromal and tumour clusters. Line represents the mean (N=
4 biological independent samples i.e. n=1 for each of the four OCMs). c,dt-SNE plot from aoverlaid with XIST expression (c) and 10 sub-populations
identified by hierarchical clustering (d). eHeat maps quantitating the percentage of cells from each patient sample in the 10 sub-populations. Source data
for panels band eare provided as a Source Data file.
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14551-2 ARTICLE
NATURE COMMUNICATIONS | (2020) 11:822 | https://doi.org/10.1038/s41467-020-14551-2 | www.nature.com/naturecommunications 7
Content courtesy of Springer Nature, terms of use apply. Rights reserved

38a tumour46 tumour
123456789101112131415
16
17
18
19
20
21
22
X
38 stromal59 tumour64–3 Ep–64–3 Ep+66–1 tumour79 tumour 33 tumour
–1 0 1 2 3
0.00
0.01
0.02
0.03
Aneuploidy score
Structural score
Stromal
61 38a
46
87
79
64–3+
38b
64–3–
72
33
66–5
74–3
59
66–1
64–1
74–1
Diploid
WCA
Tetraploidy
Mono-
somies
Rearranged
0 1 2
3 4 5
6 7 8
9+10
0.4
0.2
0.1
0.01
0.05
Heterogeneity score
b
a
Fig. 6 scWGS karyotyping. a Genome-wide chromosome copy number profiles determined by single-cell whole-genome sequencing showing aneuploidies
and rearranged chromosomes in tumour cells. Each row represents a single cell, with chromosomes plotted as columns and colours depicting copy number
state. bBubble plot showing structural, aneuploidy and heterogeneity scores. See also Supplementary Fig. 3. Source data for panel bare provided as a
Source Data file.
ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14551-2
8NATURE COMMUNICATIONS | (2020) 11:822 | https://doi.org/10.1038/s41467-020-14551-2 | www.nature.com/naturecommunications
Content courtesy of Springer Nature, terms of use apply. Rights reserved

rapidly eliminating the p53-proficient stromal cells. OCMs 33, 46,
66–1, 74–1 and 79 were then analysed between passage zero and
two, showing an abundance of multipolar spindles (Fig. 7d).
Interestingly, for OCMs 33, 46 and 74–1, the frequency of bipolar
spindles increased at later passage, suggesting that continued
propagation ex vivo leads to the emergence of relatively stable
sub-clones more reminiscent of established cell lines. Never-
theless, our analysis of OCM cells very shortly following biopsy
isolation confirms a profound level of CIN, consistent with the
scWGS karyotyping.
Time-lapse microscopy reveals highly abnormal mitoses. The
karyotype heterogeneity and abnormal spindle poles numbers
suggests mitotic dysfunction. Indeed, the extensive copy number
variations exhibited by HGSOC predicts a high level of CIN. To
determine the extent of mitotic dysfunction, we introduced a
GFP-tagged histone then characterised the ex vivo models using
fluorescence time-lapse microscopy (Fig. 1b). Often, mitosis was
successful with chromosomes separating equally (Fig. 8a). Fre-
quently however, chromosome alignment was protracted and
segregation abnormal. While stromal cells completed mitosis
0
a
HCT116 38b
66–1 79
60
80
100
40
048 Der
1X
Chromosomes
Cells
T
4 8 12 16 20
ABCDEFGHIJKLM
Unique
Micro
Dic/Rg
0246
Derivative
chromosomes
46
92
138
*
T
b
015304560
25
50
75
100
0
Cells with ≥4 spindle poles (%)
Cells with two spindle poles (%)
Tumour
Stromal
Cell lines
2
13
1
16
011
115
33
74–1
46
79
66–1
Chemo-naïve
DNAArApH3
cd
T
TT
12
118
38
124
110
195
87
Fig. 7 M-FISH karyotyping. a Heat maps quantitating total chromosome count (T, range 40–>100) and individual chromosome counts (matrix, range
0–8), for OCMs 38b, 66–1 and 79, enriched for tetraploidy, rearranged chromosomes and whole-chromosome aneuploidy features respectively. Derivative
chromosomes indicated by white. HCT116, a near-diploid, stable cell line with two derivative chromosomes, is shown for comparison. bM-FISH analysis of
OCM.59. Heat map, exemplar chromosome spread and two exemplar M-FISH images. Heat map shows total chromosome count (T) and individual
chromosome counts (matrix, range 0–6), quantitating recurring derivatives (A to M), unique derivatives (U), DNA fragments and micro-chromosomes
(Micro), and other abnormal structures including dicentrics and ring chromosomes (Dic/Rg). The chromosome spread shows a micro (arrow) and
dicentric (arrowhead) chromosome while the M-FISH images show whole chromosome aneuploidies, rearranged chromosomes and different derivatives.
cImmunofluorescence images of cells stained to detect phospho-histone H3 (serine 10), Aurora A and the DNA (representative images from single
experiment). Scale bar, 10 µm. dQuantitation of mitotic spindle poles in stromal cells, OCM tumour cells and nine established cell lines. Numbers outside
the symbols indicate OCM culture while numbers inside the symbols indicate passage number. Orange arrows connect tumour samples from the same
OCM culture analysed at different passages. Source data for panels a,band dare provided as a Source Data file.
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14551-2 ARTICLE
NATURE COMMUNICATIONS | (2020) 11:822 | https://doi.org/10.1038/s41467-020-14551-2 | www.nature.com/naturecommunications 9
Content courtesy of Springer Nature, terms of use apply. Rights reserved

swiftly, mitosis in the tumour cells was protracted and exhibited a
profound range, often with skewed distributions (Fig. 8b), con-
sistent with spindle assembly checkpoint (SAC) delaying mito-
sis39. While cultures from the same patient had similar
characteristics (e.g. OCM.66–1/5), the OCM.64–3 sub-clones
were dissimilar; OCM.64–3Ep+cells, which have smaller nuclei
and monosomies, underwent mitosis faster than their EpCAM
negative counterparts (Fig. 8b).
Mitosis in the stromal cells was largely error-free (Fig. 8c). By
contrast, lagging chromosomes and anaphase bridges dominated
Stromal
46
87
3+
72
79
38
3–
1
33
1
3
5
59
1
101
102
103
104
Time in mitosis (min)
b
64 74 66
Bridge
Multipolar
Unaligned
Cytokinesis
Other
74–3
55.5
25.5
30.9
–
32.7
Bridge
Multipolar
Unaligned
Cytokinesis
Other
59
59.6
13.2
6.1
12.3
17.5
Bridge
Multipolar
Unaligned
Cytokinesis
Other
72
31.1
5.2
16.5
6.1
53.0
Bridge
Multipolar
Unaligned
Cytokinesis
Other
46
21.1
4.4
2.6
0.9
2.6
Bridge
Multipolar
Unaligned
Cytokinesis
Other
33
54.2
16.0
14.5
11.5
13.7
Normal
Bridge
only
All other
defects
0 100 200 300 400
0.0
0.2
0.4
0.6
0.8
Skew
46
87
64
72
79
38
74
33
59
66
Mean
101102103104
0
10
20
30
40
Time (min)
Relative frequency (%)
46
33
66–5
c
1
5
1
3
3+
3–
1
Successful mitosis Multipolar mitosis Cytokinesis/abscission
failure
Chromosome bridge/
lagging chromosome
Anaphase with unaligned
chromosomes
Other
e.g. cohesion fatigue
a
Bridge
Multipolar
Unaligned
Cytokinesis
Other
Stromal
9.9
0.7
–
–
–
6
12
18
24 h
ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14551-2
10 NATURE COMMUNICATIONS | (2020) 11:822 | https://doi.org/10.1038/s41467-020-14551-2 | www.nature.com/naturecommunications
Content courtesy of Springer Nature, terms of use apply. Rights reserved

the tumour cultures (Fig. 8c and Supplementary Fig. 4a, b), but
these events were more dramatic compared with those observed
in established CIN cell lines40,41. Cytokinesis/abscission failures
and multipolar mitoses occurred frequently, with OCMs 72 and
74–3 standing out. Daughter nuclei often reconvened long after
anaphase, consistent with DNA blocking abscission. OCM.33 had
a high degree of cohesion fatigue42, possibly accounting for the
high skew score (Fig. 8b); note that premature sister chromatid
separation prevents SAC satisfaction, enforcing a mitotic arrest43.
A corollary of this observation is that despite extensive mitotic
dysfunction, the SAC is intact. Indeed, cells exhibiting anomalies
took longer to complete mitosis and arrested when challenged
with paclitaxel (see below). Conversely, OCM.46 completed
mitosis relatively quickly and displayed the least number of
anomalies (Fig. 8b, c and Supplementary Fig. 4a). However,
despite SAC functionality, anaphase with unaligned chromo-
somes, a phenomenon very rarely seen in established cell lines,
was a recurrent feature. OCM.59 stood out with 12% premature
anaphases (Fig. 8c). Thus, the time-lapse data demonstrates that
mitosis in the ex vivo models is profoundly defective and
considerably heterogeneous, indicating that the analysis of
established cell lines underestimates the mitotic dysfunction in
advanced human cancers.
Disrupting tissue architecture can influence chromosome
segregation fidelity44. Therefore, we asked whether the OCMs
also displayed mitotic dysfunction when cultured as 3D
organoids45. Analysis of OCM.66–1 in 3D revealed aberrant
mitoses including anaphases with unaligned chromosomes
(Fig. 9a). We also observed a phenotype not seen in 2D, namely
chromosome ejection at anaphase, possibly reflecting the ability
of a 3D environment to better anchor ectopic spindle poles.
Importantly, the frequency of aberrant mitoses in 3D was similar
to 2D (Fig. 9b). Interestingly, the 3D mitoses were not as
protracted as those in 2D (Fig. 9c), suggesting that the 3D
environment might constrain the spindle leading to more rapid
SAC satisfaction.
Cell fate profiling. To understand how aberrant mitoses impact
cell fate and culture dynamics, we set out to determine pro-
liferation rates and post-mitotic cell fate. Doubling times ranged
from under 30 h for OCMs 46 and 87, to over 100 h for OCMs 59
and 74–1/3 (Fig. 10a). Fate profiles of the faster growing models
showed that most cells completed multiple cell divisions
(Fig. 10b). By contrast, in slow growing OCM.74–1, only 32% of
cells divided; 20% remained in interphase and 24% died without
entering mitosis. This anti-proliferative phenomenon was
observed to some extent in most of the cultures (Supplementary
Fig. 5). Taken together with the high frequency of abnormal
mitoses described above, a likely explanation is that prior divi-
sions generated daughters harbouring genomes incompatible with
continued cell cycle progression. Interestingly, 12% of cells in
OCM.74–1 fused with neighbouring cells. Although less frequent,
this occurred in several other cultures (Supplementary Fig. 5).
Fusion events typically involved daughter cells, suggesting that
abscission was not fully executed at the end of the previous cell
cycle46,47. Nevertheless, despite the high frequency of abnormal
mitoses, sufficient cells survived to yield proliferative cultures.
Drug sensitivity profiling. To determine drug sensitivity, we
measured culture dynamics in the presence of cisplatin and
paclitaxel (Fig. 10c). IC
50
values for cisplatin ranged ~7-fold
across the cohort, with OCMs 33 and 64–3Ep+the most sensitive
and resistant respectively (Fig. 10d). These values did not corre-
late with paclitaxel IC
50
values, which were less variable. While
the two cultures from patient 66 responded similarly to both
cisplatin and paclitaxel, the two OCM.64-3 sub-cultures diverged
considerably, with OCMs 64–3Ep−and 64–3Ep+having cisplatin
IC
50
values of ~0.6 µM and ~2.1 µM respectively. Despite
appearing karyotypically similar, the sequential cultures from
patient 74 also had distinct sensitivities, with OCM.74–1 more
resistant to both cisplatin and paclitaxel. The patients’tumour
responses to chemotherapy broadly correlated with ex vivo drug
sensitivities (see Supplementary Table 1). OCMs 33, 38b, and
74–3 had the lowest IC
50
values for cisplatin and were derived
from patients who achieved a radiological response and a sig-
nificant reduction in serum CA125 following platinum-based
chemotherapy. In contrast, OCMs 46, 59, 64–1, 66–1/5 and 79
originated from patients with progressive disease. Moreover, none
of these patients achieved an improvement in serum CA125 levels
during treatment. A notable exception was OCM.74–1, which
exhibited a cisplatin IC
50
suggestive of platinum-resistant disease
yet the patient had a partial radiological response and a sig-
nificant reduction in serum CA125. In this case, the in vivo
response could have resulted from the gemcitabine component of
her chemotherapy. Nevertheless, the congruence between the
patient tumour responses and the drug sensitivity of the ex vivo
cultures suggests that models generated by this workflow do
indeed reflect the patient’s tumours.
Heterogeneous responses to paclitaxel. Paclitaxel is routinely
used in the treatment of ovarian cancer. Previously, we showed
that paclitaxel-induced cytotoxicity in established cancer cells
lines is highly heterogeneous40. The OCMs also exhibited inter
and intra-culture variation (Fig. 10b and Supplementary Fig. 5).
For example, in 10 nM paclitaxel, 60% of cells in OCM.46
underwent an abnormal mitosis while at 100 nM, 32% underwent
slippage and 22% died in mitosis (Fig. 10b). OCM.87 exhibited a
similar behaviour; abnormal mitoses dominated in 10 nM, with
26% slippage and 22% death in mitosis at 100 nM paclitaxel. By
contrast, the fate profiles of OCM.66–1 were similar at both
concentrations despite an extended mitosis at 100 nM. Consistent
with its high IC
50
, 10 nM paclitaxel had a marginal impact on
OCM.74–1, only reducing the number of successful divisions
from 32 to 28%. Strikingly in most models, the number of cells
that died in interphase following slippage or an abnormal mitosis
was low, with an average of only 12% across the cohort. Never-
theless, these observations show that the ex vivo ovarian cancer
models represent a valuable resource for drug sensitivity profiling
and detailed mode of action studies.
Fig. 8 Time-lapse microscopy. a Examples of abnormal mitoses in tumour cells expressing GFP-H2B, showing images before and after anaphase onset
(representative images at multiple positions from single experiment). Scale bar, 10 µm. bAnalysis of time spent in mitosis, at least 100 cells measured from
nuclear envelope breakdown to anaphase onset. Rank ordered box-and-whisker plot with boxes, whiskers and “+”showing the interquartile range, 10–90%
range, and mean respectively. Line graph showing linear regression of the frequency distributions for OCMs 33, 46 and 66–5. Bubble plot of Hougaard’s
skew against the mean, with bubble size proportional to the variance. cQuantitation of mitotic anomalies with each column representing one cell and the
vertical grey bars representing the time each cell spent in mitosis. Pie charts show the number of normal mitoses, those with anaphase bridges only and all
other defects combined. Note that the stromal data is compiled from three cultures, namely OCMs 33, 66 and 79. See also Supplementary Fig. 4. Source
data for panels band care provided as a Source Data file, including number of biological independent samples for each OCM in panel B.
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14551-2 ARTICLE
NATURE COMMUNICATIONS | (2020) 11:822 | https://doi.org/10.1038/s41467-020-14551-2 | www.nature.com/naturecommunications 11
Content courtesy of Springer Nature, terms of use apply. Rights reserved

Discussion
Living biobanks are powerful resources, with the transformative
aspect coming from the ability to perform detailed phenotypic
studies on well-characterised models that accurately reflect a
patient’s tumour, and in turn, the ability to correlate ex vivo
observations with clinical chemotherapy responses32–34,48.As
such, living biobanks can potentially address limitations associated
with established cancer cell lines, and indeed, our analysis shows
that thus far, we have grossly underestimated the mitotic dys-
function in advanced human tumours. The biopsy pipeline and
workflow we describe here generates ex vivo ovarian cancer cul-
tures with extensive proliferative potential, rendering models
amenable to detailed cell cycle studies, including characterisation
of mitotic chromosome segregation and drug sensitivity profiling.
Efficient generation of proliferative cultures was facilitated by
adopting OCMI media36, extending the potential of this for-
mulation beyond generating cell lines to also creating tumour cell
cultures that can be analysed shortly following biopsy isolation; the
vast majority of analyses here were performed within 10 passages.
Importantly, by using conditions that allow immediate tumour cell
proliferation, bottlenecks that might otherwise select for distinct
sub-populations are minimised; indeed, OCMI media maintains
the genomic and transcriptomic landscapes of the original
tumours36. Consistently, the congruence of the gene expression
profiles and karyotypes of cultures generated from sequential
biopsies indicates that the workflow generates consistent and
reflective tumour models. At the same time, the ability of different
sub-cultures to emerge indicates that the models also potentially
reflect intra-tumour heterogeneity. Important next steps will be to
track genomic and phenotypic evolution during culture estab-
lishment and propagation. During the course of this work, addi-
tional methodologies were described to establish panels of ovarian
cancer models, either as 2D cultures and organoids35,45,49,50.
Another next step will be to compare genome evolution and CIN
in these different culture conditions. Moreover, it will be important
to characterise the genomes as the primary cultures evolve ex vivo
in to established cell lines. The reduction in spindle pole numbers
at later passages suggests that more stable subclones might be
selected for rapidly once the tumour cells are liberated from the
in vivo microenvironment.
The workflow characterising the models involved a com-
plementary array of orthogonal approaches including expression
of tumour markers, p53 profiling, exome sequencing, global
transcriptomics and scWGS-based karyotyping. Our analysis
highlights the risk of relying only on the expression of a small
number of tumour markers51, which is perhaps not surprising in
light of the extensive heterogeneity exhibited by HGSOC. And
importantly, while the case of OCM.69 highlights the technical
challenges during the early phase of culture establishment, it also
illustrates the veracity of the workflow. We recognised that this
culture was outgrown by stromal cells upon p53 profiling and
closer inspection of cell biological parameters. This assessment
was confirmed by the exome and RNAseq analysis. Thus far, of
the 312 samples from 135 patients, we have attempted to generate
cultures from 290, yielding 76 OCMs, i.e., a success rate of 26.2%.
These OCMs are derived from 44 patients, yielding a per patient
success rate of 32.6%. In some cases, however, when the first
attempt failed, we were able to generate a tumour culture from a
0303545
115 125 130 140 175
Bridge
Multipolar
Unaligned
Cytokinesis
Other
66–1
2D
66–1
3D
Bridge
Multipolar
Unaligned
Cytokinesis
Other 101
102
103
Time in mitosis (min)
S2D3D
35.3
2.6
2.6
9.5
5.2
33.6
10.6
2.7
6.2
2.7
12
24 h
6
12 h
bc
a
Fig. 9 Mitosis in 3D. a Z-stack projections showing examples of abnormal mitoses in OCM.66–1 from three biological replicates when cultured in 3D.
Numbers show minutes after imaging initiated. Scale bar, 20 µm. Arrowhead shows unaligned chromosomes at anaphase, arrow shows an ejected
chromosome. bQuantitation of mitotic anomalies with each column representing one cell and the vertical grey bars representing the time each cell spent in
mitosis. cViolin plot showing the time spent in mitosis for OCM.66–1 when cultured in 3D. Lines show the median and interquartile ranges. The 2D data
from Fig. 8b is for comparison only. Source data for panels band care provided as a Source Data file.
ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14551-2
12 NATURE COMMUNICATIONS | (2020) 11:822 | https://doi.org/10.1038/s41467-020-14551-2 | www.nature.com/naturecommunications
Content courtesy of Springer Nature, terms of use apply. Rights reserved

subsequent attempt, facilitated by the availability of frozen,
unseparated cells (Fig. 1b). Important next steps will be to define
workflow modifications that increase the first-attempt success
rate. Preliminary observations suggest that serum source and
plating surface can be important factors. All the OCMs described
here were generated in low-oxygen conditions, but we note that
several of the cell lines generated by Ince et al.36 are cultured in
atmospheric oxygen, suggesting that oxygen concentration may
also be a factor.
The scWGS-based karyotyping was particularly informative,
in terms of validating and comparing the different models.
In particular, we identified four karyotype features whereby
33
38b
74–3
87
64–3–
46
66–5
72
66–1
79
64–1
74–1
59
64–3+
0
6
12
18
30
60
33
38b
74–3
87
64–3–
46
66–5
72
66–1
79
64–1
74–1
59
64–3+
0
1
2
3
4
–2 –1 0 1
0
100
200
300
Log (μM)
Area under curve
–1 0 1 2 3
0
100
200
300
Log (nM)
IC
50
46 - Cisplatin
IC50= 0.85 μM
46 - Taxol
IC50= 6.3 nM
024487296
0
2
4
6
8
Time (h)
Normalised GOC
0 24487296
0
2
4
6
8
Time (h)
46 - Cisplatin 46 - Taxol
Cisplatin (μM) Taxol (nM)
46
87
72
79
64–1
64–3–
38b
66–1
66–5
33
64–3+
59
74–1
0
50
100
150
Doubling time (h)
a
0 0.05 0.10 0.20 0.39 0.78 1.56 3.13 6.25 12.5 25 50
0 0.49 0.98 1.95 3.91 7.81 15.63 31.25 62.50 125 250 500
Cisplatin (μM)
Taxol (nM)
c
d
74–3
200
Death in mitosis
Slippage
Interphase
Abnormal mitosis
Mitosis No mitotic entry
Death in interphase
Fusion
Fission
4666–1
DMSO 10 nM Taxol 100 nM Taxol
74–187
0 24487296
Time (h)
024487296
Time (h)
024487296
Time (h)
024487296
Time (h)
b
28
18
18
30
76
10
12
60
10
10
6
22
32
22
14
88
42
24
24
6
38
22
22
8
8
32
8
24
20
12
16
22
16
34
8
6
6
88
6
26
6
46
10
6
10
26
26
22
8
*
%
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14551-2 ARTICLE
NATURE COMMUNICATIONS | (2020) 11:822 | https://doi.org/10.1038/s41467-020-14551-2 | www.nature.com/naturecommunications 13
Content courtesy of Springer Nature, terms of use apply. Rights reserved

genomes were enriched for either whole-chromosome aneu-
ploidies, rearranged chromosomes, monosomies or tetrasomies.
Integrating these classes with recently described CIN signatures
is an important future step18,19. By comparing the genomes of
single cells, the scWGS-based karyotyping also illustrates the
profound heterogeneity within the cultures, indicating perva-
sive CIN. The proliferative nature of the cultures also facilitated
M-FISH karyotyping, which identified structures not detected
by sequencing, including acentric fragments and ring chro-
mosomes. However, the key advantage of a living biobank is the
ability to perform detailed phenotypic studies on early passage
tumour cells, and here we show that ovarian cancer cells display
an unprecedented level of mitotic heterogeneity. Analysis of
established cell lines has not captured this heterogeneity, pre-
sumably because long-term cell culture selects the fitter, more
stable subclones. Indeed, clonal evolution analysis of estab-
lished colorectal cancer cells shows that despite persistent
chromosome segregation errors, specifickaryotypesaremain-
tained52, and while multipolar spindles were prevalent in the
OCMs, established ovarian cancer cell lines typically undergo
bipolar divisions. Another advantage of viable cultures is the
ability to analyse highly purified tumour fractions unfettered by
contaminating, genetically normal stromal cells and the
microenvironment. The workflow does however retain matched
tumour-associated fibroblasts and can be adapted to retain
tumour-infiltrating lymphocytes53,inturnallowingrecon-
struction of tumour-microenvironment interactions.
Consistent with the highly deviant karyotypes, mitosis in the
OCMs was often highly aberrant. Note however that most of our
analysis was performed on cells grown as monolayers. Impor-
tantly, it was recently shown that tissue architecture can influence
chromosome segregation fidelity44. Specifically, mouse epithelial
cells in 3D spheroids exhibited very low missegregation rates; but
when disaggregated and analysed in 2D, ~7% of cells displayed a
lagging chromosome, a level comparable to that displayed by the
patient-derived stromal cells analysed in this study. By contrast,
the OCM tumour cells exhibited a much higher rate of abnormal
mitoses; 52% of the mitoses we analysed were abnormal. Thus,
disrupted tissue architecture is unlikely to account for this very
high rate of chromosome missegregation. Indeed, when cultured
in 3D, OCM.66–1 exhibited a high frequency of aberrant mitoses.
Despite the high frequency of catastrophic mitoses, sufficient
daughter cells survive to yield actively proliferating cultures.
However, the doubling times are long compared with established
cell lines. Several factors contribute to this including long cell
cycle times, cell cycle blocks and apoptosis, indicating that the
prior cell division yielded a fatal genome. Nevertheless, the fact
that many cells survive following highly abnormal divisions
indicates that post-mitotic responses are severely compromised,
most likely due in large part to loss of p53 function14.
However, p53-independent mechanisms may also be defective.
For example, as well as driving proliferation and biogenesis, MYC
drives an apoptosis module that sensitises cells to mitotic
abnormalities54,55. Interrogating the apoptotic machinery in these
models is a future priority, as it may open up opportunities to
explore pro-survival inhibitors as therapeutics56.
The workflow we describe here represents a major step forward
in modelling ovarian cancer. In 36 months, we generated 76
ex vivo models from 44 patients, yielding a diverse and com-
prehensive collection, with the exemplar panel described here
providing proof of concept. By addressing the limitations asso-
ciated with established cell lines, these models better reflect the
specific diseases of individual patients, and as such the living
biobank will serve as a resource to enable discovery research, in
particular enabling a better understanding of CIN, genome evo-
lution and tumour micro-heterogeneity. The tractability of the
models in terms of drug sensitivity profiling will also provide
tools for drug discovery. Indeed, we recently showed that chemo-
naïve OCMs derived from patients with platinum-refractory
disease are sensitive to a first-in-class compound targeting PARG
when combined with a CHK1 inhibitor38. A key future priority
will be to correlate the drug sensitivity of the ex vivo cultures with
in vivo tumour behaviours, in response to both standard of care
chemotherapy and emerging agents, a process that will be
facilitated by correlating clinical outcomes with each OCM. While
the numbers here are small, initial results in terms of platinum
responses are encouraging, suggesting that models generated by
this workflow could potentially serve as predictive patient avatars.
This in turn will provide opportunities to tailor chemotherapy
choices based on phenotyping individual tumours as well as
stratifying patients for clinical trials testing new agents.
Methods
Patient samples. Research samples were obtained from the Manchester Cancer
Research Centre (MCRC) Biobank with informed patient consent obtained prior to
sample collection. The MCRC Biobank is licensed by the Human Tissue Authority
(license number: 30004) and is ethically approved as a research tissue bank by the
South Manchester Research Ethics Committee (Ref: 07/H1003/161+5). The role of
the MCRC Biobank is to distribute samples and does not endorse studies per-
formed or the interpretation of results. For more information see www.mcrc.
manchester.ac.uk/Biobank/Ethics-and-Licensing.
Cell culture. Ovarian cancer and stromal cells were cultured in OCMI media36
using a 50:50 mix of Nutrient Mixture Ham’s F12 (Sigma Aldrich) and Medium
199 (Life Technologies) was supplemented with 5% FBS (Life Science Group) or
5% Hyclone FBS (GE Healthcare), 2 mM glutamine (Sigma Aldrich), 100 U/ml
penicillin, 100 U/ml streptomycin (Sigma Aldrich), 10 mM HEPES at pH7.4,
20 µg/ml insulin, 0.01 µg/ml EGF; 0.5 µg/ml hydrocortisone, 10 µg/ml transferrin,
0.2 pg/ml Tridothyronine, 5 µg/ml o-phosphoryl ethanolamine, 8 ng/ml selenious
acid, 0.5 ng/ml 17 β-oestradiol, 5 µg/ml all trans retinoic acid, 1.75 µg/ml hypox-
anthine, 0.05 µg/ml lipoic acid, 0.05 µg/ml cholesterol, 0.012 µg/ml ascorbic acid,
0.003 µg/ml α-tocopherol phosphate; 0.025 µg/ml calciferol, 3.5 µg/ml choline
chloride, 0.33 µg/ml folic acid, 0.35 µg/ml vitamin B12, 0.08 µg/ml thiamine HCL,
4.5 µg/ml i-inositol, 0.075 µg/ml uracil, 0.125 µg/ml ribose, 0.0125 µg/ml para-
aminobenzioic acid, 1.25 mg/ml BSA, 0.085 µg/ml xanthine and 25 ng/ml cholera
toxin (all from Sigma). Taxol (Sigma Aldrich) and Nutlin-3 (Sigma Aldrich),
dissolved in DMSO, and Cisplatin (Sigma Aldrich), dissolved in 0.9% sodium
chloride, were stored below −20 °C. Nutlin-3 was used at a final concentration of
10 µM. Taxol and Cisplatin were used as described in the figure legends. Estab-
lished ovarian carcinoma cell lines COV318, COV362 (Sigma), CAOV3 (ATCC)
were cultured in DMEM, while OVCAR3 (ATCC), Kuramochi, OVSAHO,
OVMANA and OVISE (JCRB Cell Bank) were cultured in RPMI. RMG1 (JCRB
Cell Bank) were cultured in Hams-F12 media. HCT116 colon cancer cells were
from the ATCC and cultured in DMEM. All cell lines were grown with 10% foetal
bovine serum, 100 U/ml penicillin, 100 U/ml streptomycin and 2 mM glutamine,
and were maintained at 37 °C in a humidified 5% CO
2
atmosphere. OV56 (Sigma)
were cultured in DMEM/F12 as above but supplemented with 10 mg/ml insulin,
0.5 mg/ml hydrocortisone and 5% foetal bovine serum. All lines were authenticated
Fig. 10 Drug sensitivity profiling. a Rank ordered plot measuring population doubling times (time-lapse microscopy). bCell fate profiles of untreated
cultures and following exposure to paclitaxel, with each horizontal line showing the behaviour of a single cell and the columns quantitating specific cell
fates. cLine graphs using green object count (GOC) to measure nuclear proliferation of sample OCM.46 in response to increasing concentrations of
cisplatin and paclitaxel, plus corresponding IC
50
curves. dDot plots showing IC
50
values for cisplatin (rank ordered) and paclitaxel. Asterisk represents p<
0.05 for comparison of the sensitivity of OCMs 64–3Ep−and 64–3Ep+to cisplatin (one-way ANOVA; Tukey’s multiple comparison). In aand d, lines
represent mean and standard deviation from at least three biological replicates. In clines show mean and standard deviation from three technical
replicates. See also Supplementary Fig. 5. Source data for panels aand dare provided as a Source Data file.
ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14551-2
14 NATURE COMMUNICATIONS | (2020) 11:822 | https://doi.org/10.1038/s41467-020-14551-2 | www.nature.com/naturecommunications
Content courtesy of Springer Nature, terms of use apply. Rights reserved

by the Molecular Biology Core Facility at the CRUK Manchester Institute using
Promega Powerplex 21 System and periodically tested for mycoplasma.
Establishment of ex vivo models. Ascites was centrifuged (500 × gfor 10 min at
4 °C) and cell pellets pooled in HBSS (Life Technologies). Red blood cells were
removed using a red blood cell lysis buffer (Miltenyi Biotec) as per the manu-
facturer’s instructions. Tumour cells were plated into Primaria flasks containing
OCMI. Solid tumour samples were processed using a tumour dissociation kit
(Miltenyi Biotec) following manufacturer’s instructions and cells plated into
collagen-coated 12.5 cm2flasks containing OCMI. All cultures were incubated for
2–4 days at 37 °C in a humidified 5% CO
2
and 5% O
2
atmosphere. Media was
replaced every 3–4 days. Upon cell attachment, stromal cells were separated from
the mixed sample using 0.05% trypsin-EDTA, and plated in gelatin-coated flasks in
OCMI media containing 5% FBS. Once tumour cells reached 95% confluency, cells
were passaged using 0.25% Trypsin-EDTA, centrifuged in DMEM containing 20%
FBS and re-plated at a 1:2 ratio. For long-term storage, cells were frozen in
Bambanker (Wako pure chemical). Cell separation using EpCAM microbeads
(Miltenyi Biotec) was performed according to the manufacturer’s instructions.
To generate 3D organoids, 20,000 cells were plated in 40 μl Matrigel in 24-well
plate. Once solidified Advanced DMEM/F12 supplemented with 1% penicillin-
Streptomycin, 1% HEPES, 100 ng/ml Rspondin, 100 ng/ml Noggin, 50 ng/ml EGF,
10 ng/ml FGF-10, 10 ng/ml FGF2, 1x B27, 10 mmol/L Nicotinamide, 1.25 mmol/l
N-acetylcysteine, 1 μmol/l Prostaglandin, 10 μmol/l SB202190, 500 nmol/l A83-01
was added and cultured at 37 °C in a humidified 5% CO
2
and 5% O
2
atmosphere.
Media was replaced every 3–4 days.
Lentiviral transduction. AAV293T cells (Agilent Technologies) were transfected
with pLVX-myc-EmGFP-H2B, psPAX2 and pMD2.G (Addgene) using CaCl
2
(Promega) in DMEM supplemented with 10% Hyclone serum (GE Healthcare) and
incubated overnight. Virus was harvested 48 later, centrifuged, filtered then added
to tumour cells with 10 µg/ml polybrene (Sigma Aldrich) and the cells centrifuged
at 300 × g,30
°C for 2.5 h followed by overnight incubation. Puromycin (Sigma
Aldrich) (1 µg/ml) was added 48 h after transduction.
Cell biology. For immunoblotting, proteins were extracted by boiling cell pellets in
sample buffer (0.35 M Tris pH 6.8, 0.1 g/ml sodium dodecyl sulphate, 93mg/ml
dithiothreitol, 30% glycerol, 50 µg/ml bromophenol blue), resolved by SDS-PAGE,
then electroblotted onto Immobilon-P membranes (Merck Millipore). Following
blocking in 5% dried skimmed milk (Marvel) dissolved in TBST (50 mM Tris pH
7.6, 150 mM NaCl, 0.1% Tween-20), membranes were incubated overnight at 4 °C
using the following antibodies: mouse anti-p53 (DO-1) (Santa Cruz Biotechnology
cat#sc-126, 1:1000); mouse anti-p21 (F-5) (Santa Cruz Biotechnology cat#sc-6246,
1:100); rabbit anti-c-myc (Y69) (Abcam cat#ab32072, 1:3,500); sheep anti-Tao157
(1:1000). Membranes were then washed three times in TBST and incubated for at
least 1 h with appropriate horseradish-peroxidase-conjugated secondary antibodies
(Rabbit anti-sheep IgG (HL) HRP, cat#618620; Goat anti-mouse IgG (HL) HRP,
cat#G21040; Goat anti-rabbit IgG (HL) HRP, cat#G21234; all Invitrogen). After
washing in TBST, bound secondary antibodies were detected using either EZ-
Chemiluminescence Reagent (Geneflow Ltd) or Luminata Forte Western HRP
Substrate (Merck Millipore) and a Biospectrum 500 imaging system (UVP) or
ChemiDoc Touch Imaging System (BioRad). For immunofluorescence, cells were
plated on collagen- or gelatin-coated 13 mm coverslips and incubated for 48 h.
Cells were washed and fixed in 1% formaldehyde, quenched in glycine, then
incubated for 30 min at room temperature using the following primary antibodies:
rabbit anti-Mucin-16 (Merck Millipore cat#ABC240, 1:50); mouse anti-EpCAM
(VU1D9) (Cell Signaling cat#2929, 1:800); rat anti-CD44 (Calbiochem
cat#217594,1:200); rabbit anti-Vimentin (EPR3776) (Abcam cat#ab92547, 1:1,000);
mouse anti-pan cytokeratin (C-11) (Abcam cat#ab7753, 1:500); mouse anti-Pax8
(Abcam cat#ab53490, 1:100); rabbit anti-Ki67 (Abcam cat#ab15580,1:2,000); and
mouse anti-p53 (DO-1) (Santa Cruz Biotechnology cat#sc-126,1:1,000). Coverslips
were washed twice in PBS-T (PBS, 0.1% Triton X-100) and incubated with the
appropriate fluorescently conjugated secondary antibodies (Donkey anti-Rabbit
Cy3, cat#711-165-152; Donkey anti-Rat Cy3, Cat#712-165-153; Donkey anti-
Mouse Cy3, Cat#715-165-150; Donkey anti-Mouse Cy2, Cat715-225-150; all
Jackson ImmunoResearch Laboratories Inc.) for 30 min at room temperature.
Coverslips were washed in PBS-T and DNA stained for 1 min with 1 μg/ml
Hoechst 33258 (Sigma Aldrich) at room temperature. Coverslips were further
washed in PBS-T and mounted (90% glycerol, 20 mM Tris, pH 9.2) onto slides.
Slides were stored at −20 °C prior to image acquisition. Note, for analysis of spindle
poles, cells were plated on collagen or gelatin-coated 19 mm coverslips, cultured for
48 h then stained with antibodies to detect phospho-Histone-H3(S10) (Merck
Millipore cat#06-570, 1:500) and Aurora A (58 1:1000). Images were acquired using
an Axioskop2 (Zeiss, Inc.) microscope fitted with a CoolSNAP HQ camera
(Photometrics). Image analysis was conducted using Adobe Photoshop CC 2015
(Adobe Systems Inc.). For time-lapse imaging, cells were cultured on collagen-
coated 35 mm glass bottom dishes (MatTek Corp) then imaged using an inverted
microscope (Axiovert 200; Carl Zeiss, Inc.) equipped with an automated stage (PZ-
2000; Applied Scientific Instrumentation) and an environmental control chamber
(Solent Scientific), which maintained the cells at 37 °C in a humidified stream of 5%
CO
2
. Imaging was performed using a ×40 Plan NEOFLUAR objective. Shutters,
filter wheels, and point visiting were driven by MetaMorph software (MDS Ana-
lytical Technologies) and images captures using an Evolve®Delta camera (Pho-
tometrics). For flow cytometry, tumour and stromal cells were incubated in
Accutase®(Sigma) to obtain single cell populations then stained with anti-CA125
antibodies (618 F) (Biolegend cat#666902, 1:25) for 30 min at 4 °C, followed by goat
anti-mouse Alexa Fluor®488 (Molecula r Probes cat#a11029, 1:100) for 30 min at
4 °C. Cells were washed in PBS then stained with antibodies against EpCAM
PE/Cy7 (Biolegend cat#324222, 1:100), CD44 BV421 (Biolegend cat#338810,
1:100) and CD105 APC (Biolegend cat#323208, 1:100), plus Zombie yellow live
dead reagent (Biolegend cat#423103, 1:500) for 30 min at 4 °C. Samples were
analysed on a Novocyte flow cytometer (ACEA biosciences) and data analysed
using Flowjo®software (FlowJo, LLC). To analyse mitosis in 3D organoids, images
were acquired using a CSU-X1 spinning disc confocal (Yokagowa) on a Zeiss Axio-
Observer Z1 microscope with a ×40/1.3 Plan-Apochromat objective, Evolve
EMCCD camera (Photometrics), motorised XYZ stage (ASI) and an environmental
control chamber which maintained the cells at 37 °C in a humidified stream of 5%
CO
2
. The 488-nm and 561-nm lasers were controlled using an AOTF through the
laserstack (Intelligent Imaging Innovations (3I)) allowing both rapid ‘shuttering’of
the laser and attenuation of the laser power. Slidebook software (3I) was used to
capture images every 5 min over 90 μmat2μm Z-intervals. Movies were analysed
with Slidebook, and Imaris (bitplane) software.
Cell proliferation. To measure proliferation and to perform cell fate profiling, cells
expressing GFP-H2B were seeded onto collagen-coated µclear®96 well plates
(Greiner Bio-One). Drugs were added 24 h post-seeding in fresh media. Cells were
then imaged using an IncuCyte®ZOOM (Essen BioScience), capturing nine fields
of view per well, either every 1–6 h for proliferation and drug sensitivity mea-
surements, or every 10 min for mitot ic cell fate profiling. IncuCyte®ZOOM soft-
ware was used in real-time to measure confluency and green fluorescent object
count. The doubling time for each culture was calculated by performing a log2
transformation of the normalised nuclear count; these data were plotted against
time and the inverse slope of the log-phase portion of the graph calculated. For
drug sensitivity assays, the Area Under the Curve (AUC) at each drug con-
centration was calculated and plotted against drug concentration to generate dose-
response curves from which IC
50
values were calculated. For cell fate profiling,
image sequences were exported in MPEG-4 format and analysed manually. Note
that 0 h on the fate profiles represents when imaging started.
TP53 genotyping by Sanger sequencing. RNA was extracted using RNeasy Plus
Mini kit (Qiagen) and TP53 cDNA generated by RT-PCR using Superscript III One
Step RT-PCR Platinum Taq HiFi (Thermofisher). PCR products were cloned into a
pBluescript SK- vector and transformed into XL1-Blue competent cells. Plasmid
DNA was extracted using QIAprep Spin Miniprep Kit (Qiagen) and sequenced
using the following primers (5′-CAC CAG CAG CTC CTA CAC CG-3', 5'-ATG
AGC GCT GCT CAG ATA GCG-3′,5′-CGG CTC ATA GGG CAC CAC C-3′,5′-
TCT TCT TTG GCT GGG GAG AGG-3′). Tumour and stromal sequences were
aligned using Seqman Pro (DNASTAR).
Analysis of primary tumours. Formalin-fixed and paraffin-embedded (FFPE)
archival tumour blocks were analysed by immunohistochemistry by collecting 4 µm
sections on Superfrost charged slides. After drying overnight at 37 °C, samples were
processed using a Ventana Benchmark immunohistochemistry platform (Roche)
with antibodies against p53 (Dako cat#M700101-2, 1:50), Cytokeratin7 (CK7, Dako
cat#M701801-2, 1:250), PAX8 (Roche cat#06523927001, 1:100) and WT1 (Abcam
cat#ab89901, 1:100). Heat induced epitope retrieval was performed using CC1
(Roche), incubating samples at 95 °C for 36, 52, 40 and 64 min for p53, CK7, PAX8
and WT1 respectively. Antibodies were incubated at 37 °C for 32, 40, 32 and
40 min for p53, CK7, PAX8, and WT1, respectively. p53 and CK7 were detected
using Ultraview universal DAB kit (Roche), while PAX8 and WT1 were detected
using Optiview universal DAB kit (Roche), all as per manufacturer’s instructions.
Sections were counterstained using Haematoxylin II (Roche) for 12 min and bluing
reagent (Roche) for eight min, and slides imaged using a Leica DM2500 micro-
scope (Leica Microsystems), using a ×20 objective lens under brightfield and
processed using Adobe Photoshop. For TP53 genotyping, FFPE blocks were
assessed for total cellularity and the neoplastic cell content of the sample expressed
as a percentage of all nucleated cells on a Haematoxylin and Eosin (H&E) stained
slide. A neoplastic cell count of ≥10% was required before undertaking DNA
extraction. DNA extraction was performed using the cobas®DNA Sample Pre-
paration Kit (Roche). Tumour from 5 × 5 µM unstained pathology slides were
available for DNA extraction. Extracted DNA was quantified using Qubit 2.0
Fluorometer (ThermoScientific). Targeted enrichment was performed using the
GeneRead Clinically Relevant Tumour Targeted Panel V2 (Qiagen). For somatic
variants in TP53 the target read depth across all coding regions (exon 2 to 9) was a
minimum of 350×. Mutations were named according to Human Genome Variation
Society guidelines (http://www.hgvs.org/) using reference sequence NM_000546.5.
All variant calls were independently reviewed using the BAM files and a genome
browser (Integrated Genomic Viewer). At a variant allele frequency ≥4% the call
sensitivity was >90% and specificity >95% after manual review.
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14551-2 ARTICLE
NATURE COMMUNICATIONS | (2020) 11:822 | https://doi.org/10.1038/s41467-020-14551-2 | www.nature.com/naturecommunications 15
Content courtesy of Springer Nature, terms of use apply. Rights reserved

RNASeq. RNA was extracted usingRNeasy Plus Mini kit (Qiagen), quantified using a
Qubit fluorometer (Life Technologies) and quality/integrity assessed using a 2200
TapeStation (Agilent Technologies). Sequencing libraries were then generated using
the TruSeq®Stranded mRNA assay (Illumina, Inc.) according to the manufacturer’s
protocol. Adaptor indices were used to multiplex libraries, which were pooled prior to
cluster generation using a cBot instrument (Illumina, Inc.). The loaded flow-cell was
then paired-end sequenced (76 +76 cycles, plus indices) on an Illumina HiSeq4000
instrument. The output data was demultiplexed (allowing one mismatch) and BCL-
to-Fastq conversion performed using Illumina’s bcl2fastq software. Unmapped
paired-reads of 76bp were interrogated using a quality control pipeline comprising of
FastQC v0.11.3 and FastQ Screen v0.9.2 (Babraham Institute). The reads were
trimmed to remove any adaptor or poor quality sequence using Trimmomatic
v0.36;59 reads were truncated at a sliding 4 bp window, starting 5′, with a mean
quality <Q20, and removed if the final length was <35 bp. The filtered reads were
mapped to the human reference sequence analysis set (hg38/Dec. 2013/GRCh38)
from the UCSC browser, using STAR v2.4.2a60. The genome index was created using
the comprehensive Gencode v23 gene annotation61.Theflag ‘–outSAMtype BAM
Unsorted’was used for the next step. Samtools v1.462 was used to identify properly
paired reads and create the BAM format required to count read-pairs into genes using
htseq-co unt v0.6.1p163 using the flag ‘–order =name’. The gene counts for each
sample were combined using the Linux bash ‘paste’function. A header was added to
the resulting file (Gene_ID, plus sample names) and the htseq-count summary footer
lines were removed. Normalisation and differential expression analysis was performed
using DESeq2 v1.10.0 on R v3.2.364.
Exome sequencing. Genomic DNA was extracted using Purelink Genomic DNA
Mini kit (Life Technologies), quantified using a Qubit fluorometer (Life Technologies)
and normalised to a final concentration of 5ng/µl. Indexed, paired-end, sequencing
libraries were then prepared using the Nextera Rapid Capture Expanded Exome
enrichment kit (Illumina, Inc.), designed to deliver 62 Mb of expertly selected,
expanded exonic content. The exome-enriched libraries were then loaded on to a
flow-cell and clusters generated using a cBot instrument. Pooled groups of 12 mul-
tiplexed libraries were clustered over three lanes (aiming to generate a predicted
>100x coverage of the exomes). The flow-cell was then paired-end sequenced (76:76
cycles plus indices) on the Illumina HiSeq4000 instrument and the output data
demultiplexed and converted to.fastq format using Illumina’s bcl2fastq software.
Unmapped paired-reads of 101 bp were interrogated using a quality control pipeline
comprising of FastQC v0.11.3 and FastQ Screen v0.9.2 (Babraham Institute). The
reads were trimmed to remove any adaptor or poor quality sequence using Trim-
momatic v0.36;59 reads were truncated at a sliding 4 bp window, starting 5′,witha
mean quality <Q30, and removed if the final length was <50 bp. The filtered reads
were mapped to the human reference sequence analysis set (hg38/Dec. 2013/
GRCh38) from the UCSC browser65, using BWA-MEM v0.7.1566.The-Mflag was
used to flag secondary reads (multimapped). BWA enforces a minimum read map-
ping score of Q30. The mapped reads were furtherprocessedusing samtools v1.462,to
identify properly paired reads, fixmates and sorted by coordinates. Read groups were
added to each read, and duplicates flagged using Picard Tools v2.1.0 AddOrRepla-
ceReadGroups and MarkDuplicates, respectively. Variant calling using samtools
mpileup and somatic mutations were identified using Varscan 2 (v.2.4.3)67,68.
Using the matched stromal cell samples as the baseline, variants were classified as
follows. If the tumour and stromal sequences matched but do not match the reference
genome, the variant is classified as a “germline”mutation. If the tumour and stromal
sequences do not match and there is a significant difference in allele frequency, and
the stromal sequence matches the reference genome, then the variant is classified as a
“somatic”mutation. If by contrast the stromal variant is heterozygous and the tumour
variant is homozygous, then the latter is classified as a “loss of heterozygosity”(LOH)
event. SNP and indels were annotated by mapping to the Catalogue of Somatic
Mutations in Cancer (COSMIC)69. Mutations not described in COSMIC were
identified as ‘unknown’. Data interpretation was aided by use of the Integrative
Genomics Viewer (IGV)70,71.
Single-cell transcriptomics - Fluidigm. Single cells were isolated using the Flui-
digm C1 platform (Fluidigm Corporation). Medium (10–17 µm) IFCs were used in
conjunction with protocol number 100–7168 (Vers. I1) to simultaneously generate
cDNAs from the single cells. Sequencing libraries were then constructed using the
Nextera XT DNA library kit (Illumina, Inc.) according to the manufacturers’
protocol (P/N 15031942, Rev. c) and modified according to Fluidigm Protocol (P/N
100–7168, Vers. I1). Nucleic acid sequencing (76:76 cycles plus indices) was then
performed on the NextSeq500 (Illumina, Inc.) using the NextSeq500 mid-output
reagent kit (Illumina, Inc.) and the output data demultiplexed and converted to
.fastq format again using Illumina’s bcl2fastq software. A pre-processing step was
used (trimmomatic v0.36) to trim the adaptor and low-quality reads using TruSeq
adaptor. The trimmed reads were then aligned to human reference genome
GRCh38.p5 (gencode v24)61 using STAR aligner (v.2.4.2a)60. Reads aligning to
genes were then counted using HTSeq (v0.6.1.p1)63. This count matrix was then
used to analyse the dataset using statistical computing programming language, R (R
Core Team). During the analysis we started with 192 cells, 96 cells from each
fraction. For quality control of cells and genes, we used the function clean.counts()
from SCDE package72 and set the min.lib.size to 500 to filter out cells expressing
fewer than 500 genes. For gene filtering, the min.reads was set to 10 and min.
detected to 5 to retain genes with a minimum read count of 10 in at least 5 cells.
This filtering, yielded 89 stromal cells and 96 tumour cells for downstream analysis.
We then applied PCA to reduce the dimensions of the data. To refine the classi-
fication in the PCA plot we applied PAGODA, a pathway based clustering
method73. A heatmap was generated by taking 10,000 most highly variable genes.
We used Cyclone74 to classify cells to their respective cell-cycle stages based on
gene expression. Gene ontology analysis of overdisperesed genes was performed
using DAVID 6.875 and visualised using REVIGO76. The network analysis was
performed using Cytoscape 3.4.077 with the GeneMANIA app78. Physical inter-
action and pathway databases were interrogated to generate network edges.
Single-cell transcriptomics –10x Genomics. Gene expression libraries were pre-
pared from singe cells using the Chromium Controller and Single Cell 3ʹReagent Kits
v3 (10x Genomics, Inc. Pleasanton, USA) according to the manufacturer’sprotocol
(CG000183 Rev A). Briefly, nanoliter-scale Gel Beads-in-emulsion (GEMs) were
generate d by combining barcoded Ge l Beads, a mast er mix containing cells, and
partitioning oil onto a Chromium chip. Cells were delivered at a limiting dilution,
such that the majority (90–99%) of generated GEMs contain no cell, while the
remainder largely contain a single cell. The Gel Beads were then dissolved, primers
released, and any co-partitioned cells lysed. Primers containing an Illumina TruSeq
Read 1 sequencing primer, a 16-nucleotide 10x Barcode, a 12-nucleotide unique
molecular identifier (UMI) and a 30-nucleotide poly(dT) sequence were then mixed
with the cell lysate and a master mix containing reverse transcription (RT) reagents.
Incubation of the GEMs then yielded barcoded cDNA from poly-adenylated mRNA.
Following incubation, GEMs were broken and pooled fractions recovered. First-strand
cDNA was then purified from the post GEM-RT reaction mixture using silane
magnetic beads and amplified via PCR to generate sufficient mass for library con-
struction. Enzymatic fragmentation and size selection were then used to optimise the
cDNA amplicon size. Illumina P5 & P7 sequences, a sample index, and TruSeq Read
2 sequence were added via end repair, A-tailing, adaptor ligation, and PCR to yield
final Illumina-compatible sequencing libraries. The resulting sequencing libraries
comprised standard Illumina paired-end constructs flanked with P5 and
P7 sequences. The 16-bp 10x Barcode and 12 bp UMI were encoded in Read 1, while
Read 2 was used to sequence the cDNA fragment. Sample index sequences were
incorporated as the i7 index read. Paired-end sequencing (28:98) was performed on
the Illumina NextSeq500 platform using NextSeq 500/550 High Output v2.5 (150
Cycles) reagents. The.bcl sequence data were processed for QC purposes using
bcl2fastq software (v. 2.20.0.422) and the resulting .fastq files assessed using FastQC
(v. 0.11.3), FastqScreen (v. 0.9.2) and FastqStrand (v. 0.0.5) prior to pre-processing
with the CellRanger pipeline. The sequence files generated from the instrument were
then processed using 10x Genomics custom pipeline Cell Ranger v3.0.1, which gen-
erated the fastq files, aligned those files to the required genome, identified the valid
barcodes as cells, counted UMIs and reported a gene by cell count matrix in sparse
matrix format. Sequences were mapped to the prebuild hg38 genome provided with
Cell Ranger package. We used three measures to identify and remove the low-quality
cells. Namely, the library size; the number of expressed genes; and the proportion of
reads mapped to mitochondrial genes in all four samples. Cells exhibiting a library
size lower than three Median Absolute Deviations (MAD) were filtered out. Also, cells
expressing a gene count lower than three MAD were filtered out. For mitochondrial
read proportions, we filtered out the cells that displayed a percentage of reads
mapping to mitochondrial genes greater than three MAD. After this filtering two cells
still showed an outlier distribution on the library size. These two “cells”were assumed
to be muliplets and were excluded from further analyses. Raw counts of the remaining
cells were then normalised using the deconvolution-based method79 and then log-
transformed. We also filtered out the genes with average counts below 0.01 assuming
these low-abundance genes to be unreliable for statistical inference80. For visualisation
and clustering we first selected the Highly Variable Genes (HVGs). For this we first
decomposed the variance of expression in each gene to technical and biological
components and identified the genes as HVGs where the biological components were
significantly greater than zero. These HVG genes were then used to reduce the
dimensions of the dataset using PCA. T-SNE plots were then generated by taking
1–14 components of the PCA. In order to cluster cells in to putative subpopulations
we used the Dynamic Tree cut method that can combine the strength of hierarchical
clustering and partition around medoids81.Thismethodgaveus10clustersacrossall
populations.
scWGS karyotyping. Single G1 nuclei were isolated by cell sorting then processed
for sequencing using a Bravo Automated Liquid Handling Platform (Agilent
Technologies)82,83. Samples were sequenced on an Illumina NextSeq 450 at ERIBA
(Illumina). Unprocessed sequencing reads were demultiplexed using library-
specific barcodes and converted into fastq format using standard Illumina software
(bcl2fastq version 1.8.4). Demultiplexed reads were aligned to human reference
genome GRCh38 using Bowtie2 (version 2.2.4). Duplicate reads were marked and
removed using BamUtil (version 1.0.3.). Aligned sequencing reads were analysed
and curated using AneuFinder (version 1.4.0)82 using 1 Mb bins. The generation of
the heterogeneity and aneuploidy scores are defined82. The structural score is
defined as the number of copy number state transitions (within a single chro-
mosome) per Mb, then normalised to the number of cells analysed. Data presented
as circos plots were generated using Circa software (OMGenomics).
ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14551-2
16 NATURE COMMUNICATIONS | (2020) 11:822 | https://doi.org/10.1038/s41467-020-14551-2 | www.nature.com/naturecommunications
Content courtesy of Springer Nature, terms of use apply. Rights reserved

M-FISH. Cells were treated with 50 ng/ml colcemid (Sigma Aldrich) for 6 h, then
harvested. Cell pellets were incubated for 10 min at room temperature in pre-
warmed (37 °C) buffered hypotonic solution (Genial Helix) followed by 20 min on
ice. Samples were centrifuged, resuspended in freshly prepared fixative of metha-
nol:acetic acid (3:1) and incubated for 30 min. Samples were furt her centrifuged
and incubated in cold (4 °C) fixative for 10 min at room temperature, re-
centrifuged and resuspended in cold fixative, dropped onto glass slides, air dried
and stored at room temperature. Slides were then experimenter-blinded and
hybridised with the M-FISH probe kit 24XCyte (Zeiss MetaSystems) following
manufacturer’s instructions, and analysed using an Olympus BX60 microscope for
epifluorescence equipped with a Sensys CCD camera (Photometrics, USA). Images
were collected and analysed using the Genus Cytovision software (Leica). A
minimum of 25 metaphases were karyotyped for each cell line/condition.
Quantification and statistical analysis. Prism 7 (GraphPad) was used to deter-
mine doubling times, AUCs, IC
50
values and other statistical analyses.
Reporting summary. Further information on research design is available in
the Nature Research Reporting Summary linked to this article.
Data availability
Exome sequencing, RNAseq, single-cell RNAseq and scWGS karyotyping data have been
deposited at the EMBL-EBI with the following accession numbers E-MTAB-7225,E-
MTAB-7223,E-MTAB-724,E-MTAB-8559 and PRJEB28664 respectively. The data
underlying Figs. 2b, d, 3a, d, 4b–e, g, 5b, e, 6b, 7a, b, d, 8b, c, 9b, c and 10a, d, and
Supplementary Figs. 2a, c and 4a, b are provided as a Source Data file. All other data
supporting the findings of this study are available within the article, the Supplementary
information files, or the corresponding author upon request. A reporting summary for
this article is available as a Supplementary Information file.
Received: 21 November 2018; Accepted: 14 January 2020;
References
1. Ferlay, J. et al. Cancer incidence and mortality worldwide: sources, methods and
major patterns in GLOBOCAN 2012. Int. J. Cancer 136, E359–E386 (2015).
2. Labidi-Galy, S. I. et al. High grade serous ovarian carcinomas originate in the
fallopian tube. Nat. Commun. 8, 1093 (2017).
3. Lee, Y. et al. A candidate precursor to serous carcinoma that originates in the
distal fallopian tube. J. Pathol. 211,26–35 (2007).
4. Perets, R. et al. Transformation of the fallopian tube secretory epithelium leads
to high-grade serous ovarian cancer in Brca;Tp53;Pten models. Cancer Cell 24,
751–765 (2013).
5. Piek, J. M. et al. Dysplastic changes in prophylactically removed Fallopian
tubes of women predisposed to developing ovarian cancer. J. Pathol. 195,
451–456 (2001).
6. Lheureux, S., Gourley, C., Vergote, I. & Oza, A. M. Epithelial ovarian cancer.
Lancet 393, 1240–1253 (2019).
7. Bowtell, D. D. et al. Rethinking ovarian cancer II: reducing mortality from
high-grade serous ovarian cancer. Nat. Rev. Cancer 15, 668–679 (2015).
8. Ahmed, A. A. et al. Driver mutations in TP53 are ubiquitous in high grade
serous carcinoma of the ovary. J. Pathol. 221,49–56 (2010).
9. Ciriello, G. et al. Emerging landscape of oncogenic signatures across human
cancers. Nat. Genet. 45, 1127–1133 (2013).
10. Cancer Genome Atlas Research Network. Integrated genomic analyses of
ovarian carcinoma. Nature 474, 609–615 (2011).
11. Martins, F. C. et al. Combined image and genomic analysis of high-grade
serous ovarian cancer reveals PTEN loss as a common driver event and
prognostic classifier. Genome Biol. 15, 526 (2014).
12. Patch, A. M. et al. Whole-genome characterization of chemoresistant ovarian
cancer. Nature 521, 489–494 (2015).
13. McCormick, A. et al. Ovarian cancers harbor defects in nonhomologous end
joining resulting in resistance to rucaparib. Clin. Cancer Res. 23, 2050–2060
(2017).
14. Sansregret, L., Vanhaesebroeck, B. & Swanton, C. Determinants and clinical
implications of chromosomal instability in cancer. Nat. Rev. Clin. Oncol. 15,
139–150 (2018).
15. Burrell, R. A., McGranahan, N., Bartek, J. & Swanton, C. The causes and
consequences of genetic heterogeneity in cancer evolution. Nature 501,
338–345 (2013).
16. Levine, M. S. & Holland, A. J. The impact of mitotic errors on cell
proliferation and tumorigenesis. Genes Dev. 32, 620–638 (2018).
17. Negrini, S., Gorgoulis, V. G. & Halazonetis, T. D. Genomic instability–an
evolving hallmark of cancer. Nat. Rev. Mol. Cell Biol. 11, 220–228 (2010).
18. Macintyre, G. et al. Copy number signatures and mutational processes in
ovarian carcinoma. Nat. Genet. 50, 1262–1270 (2018).
19. Wang, Y. K. et al. Genomic consequences of aberrant DNA repair mechanisms
stratify ovarian cancer histotypes. Nat. Genet. 49, 856–865 (2017).
20. Christie, E. L. et al. Multiple ABCB1 transcriptional fusions in drug resistant
high-grade serous ovarian and breast cancer. Nat. Commun. 10, 1295 (2019).
21. Coleman, R. L. et al. Veliparib with first-line chemotherapy and as maintenance
therapy in ovarian cancer. N. Engl. J. Med. 381,2403–2415 (2019).
22. Golan, T. et al. Maintenance olaparib for germline BRCA-mutated metastatic
pancreatic cancer. N. Engl. J. Med. 381, 317–327 (2019).
23. Gonzalez-Martin, A. et al. Niraparib in Patients with Newly Diagnosed
Advanced Ovarian Cancer. N. Engl. J. Med. 381, 2391–2402 (2019).
24. Litton, J. K. et al. Talazoparib in patients with advanced breast cancer and a
germline BRCA mutation. N. Engl. J. Med. 379, 753–763 (2018).
25. Mateo, J. et al. DNA-repair defects and olaparib in metastatic prostate cancer.
N. Engl. J. Med. 373, 1697–1708 (2015).
26. Moore, K. et al. Maintenance olaparib in patients with newly diagnosed
advanced ovarian cancer. N. Engl. J. Med. 379, 2495–2505 (2018).
27. Robson, M. et al. Olaparib for metastatic breast cancer in patients with a
germline BRCA mutation. N. Engl. J. Med. 377, 523–533 (2017).
28. Domcke, S., Sinha, R., Levine, D. A., Sander, C. & Schultz, N. Evaluating cell
lines as tumour models by comparison of genomic profiles. Nat. Commun. 4,
2126 (2013).
29. Jamal-Hanjani, M. et al. Tracking the evolution of non-small-cell lung cancer.
N. Engl. J. Med. 376, 2109–2121 (2017).
30. Scott, C. L., Becker, M. A., Haluska, P. & Samimi, G. Patient-derived xenograft
models to improve targeted therapy in epithelial ovarian cancer treatment.
Front. Oncol. 3, 295 (2013).
31. Cybulska, P. et al. A genomically characterized collection of high-grade serous
ovarian cancer xenografts for preclinical testing. Am. J. Pathol. 188,
1120–1131 (2018).
32. Sachs, N. et al. A living biobank of breast cancer organoids captures disease
heterogeneity. Cell 172, 373–386 e310 (2018).
33. van de Wetering, M. et al. Prospective derivation of a living organoid biobank
of colorectal cancer patients. Cell 161, 933–945 (2015).
34. Kodack, D. P. et al. Primary patient-derived cancer cells and their potential for
personalized cancer patient care. Cell Rep. 21, 3298–3309 (2017).
35. Kopper, O. et al. An organoid platform for ovarian cancer captures intra- and
interpatient heterogeneity. Nat. Med. 25, 838–849 (2019).
36. Ince, T. A. et al. Characterization of twenty-five ovarian tumour cell lines that
phenocopy primary tumours. Nat. Commun. 6, 7419 (2015).
37. Strauss, R. et al. Analysis of epithelial and mesenchymal markers in ovarian cancer
reveals phenotypic heterogeneity and plasticity. PLoS One 6, e16186 (2011).
38. Pillay, N. et al. DNA replication vulnerabilities render ovarian cancer cells
sensitive to poly(ADP-Ribose) glycohydrolase inhibitors. Cancer Cell 35,
519–533 e518 (2019).
39. Meraldi, P., Draviam, V. M. & Sorger, P. K. Timing and checkpoints in the
regulation of mitotic progression. Dev. Cell 7,45–60 (2004).
40. Gascoigne, K. E. & Taylor, S. S. Cancer cells display profound intra- and
interline variation following prolonged exposure to antimitotic drugs. Cancer
Cell 14, 111–122 (2008).
41. Thompson, S. L. & Compton, D. A. Examining the link between chromosomal
instability and aneuploidy in human cells. J. Cell Biol. 180, 665–672 (2008).
42. Daum, J. R. et al. Cohesion fatigue induces chromatid separation in cells
delayed at metaphase. Curr. Biol. 21, 1018–1024 (2011).
43. Stevens, D., Gassmann, R., Oegema, K. & Desai, A. Uncoordinated loss of
chromatid cohesion is a common outcome of extended metaphase arrest.
PLoS One 6, e22969 (2011).
44. Knouse, K. A., Lopez, K. E., Bachofner, M. & Amon, A. Chromosome
segregation fidelity in epithelia requires tissue architecture. Cell 175, 200–211
e213 (2018).
45. Hill, S. J. et al. Prediction of DNA repair inhibitor response in short-term
patient-derived ovarian cancer organoids. Cancer Discov. 8, 1404–1421 (2018).
46. Norden, C. et al. The NoCut pathway links completion of cytokinesis to spindle
midzone function to prevent chromosome breakage. Cell 125,85–98 (2006).
47. Steigemann, P. et al. Aurora B-mediated abscission checkpoint protects
against tetraploidization. Cell 136, 473–484 (2009).
48. Ghani, F. I. et al. An ex-vivo culture system of ovarian cancer faithfully
recapitulating the pathological features of primary tumors. Cells 8, E644 (2019).
49. Thu, K. L. et al. A comprehensively characterized cell line panel highly
representative of clinical ovarian high-grade serous carcinomas. Oncotarget 8,
50489–50499 (2017).
50. Penner-Goeke, S. et al. The temporal dynamics of chromosome instability in
ovarian cancer cell lines and primary patient samples. PLoS Genet. 13,
e1006707 (2017).
51. Latifi, A. et al. Isolation and characterization of tumor cells from the ascites of
ovarian cancer patients: molecular phenotype of chemoresistant ovarian
tumors. PLoS One 7, e46858 (2012).
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14551-2 ARTICLE
NATURE COMMUNICATIONS | (2020) 11:822 | https://doi.org/10.1038/s41467-020-14551-2 | www.nature.com/naturecommunications 17
Content courtesy of Springer Nature, terms of use apply. Rights reserved

52. Wangsa, D. et al. The evolution of single cell-derived colorectal cancer cell lines
is dominated by the continued selection of tumor-specific genomic imbalances,
despite random chromosomal instability. Carcinogenesis 39, 993–1005 (2018).
53. Owens, G. L. et al. Ex vivo expanded tumour-infiltrating lymphocytes from
ovarian cancer patients release anti-tumour cytokines in response to
autologous primary ovarian cancer cells. Cancer Immunol. Immunother. 67,
1519–1531 (2018).
54. Topham, C. et al. MYC is a major determinant of mitotic cell fate. Cancer Cell
28, 129–140 (2015).
55. Littler, S. et al. Oncogenic MYC amplifies mitotic perturbations. Open Biol. 9,
190136 (2019).
56. Zervantonakis, I. K. et al. Systems analysis of apoptotic priming in ovarian
cancer identifies vulnerabilities and predictors of drug response. Nat.
Commun. 8, 365 (2017).
57. Westhorpe, F. G., Diez, M. A., Gurden, M. D., Tighe, A. & Taylor, S. S. Re-
evaluating the role of Tao1 in the spindle checkpoint. Chromosoma 119,
371–379 (2010).
58. Girdler, F. et al. Molecular basis of drug resistance in aurora kinases. Chem.
Biol. 15, 552–562 (2008).
59. Bolger, A. M., Lohse, M. & Usadel, B. Trimmomatic: a flexible trimmer for
Illumina sequence data. Bioinformatics 30, 2114–2120 (2014).
60. Dobin, A. et al. STAR: ultrafast universal RNA-seq aligner. Bioinformatics 29,
15–21 (2013).
61. Harrow, J. et al. GENCODE: the reference human genome annotation for The
ENCODE Project. Genome Res 22, 1760–1774 (2012).
62. Li, H. et al. The Sequence Alignment/Map format and SAMtools.
Bioinformatics 25, 2078–2079 (2009).
63. Anders, S., Pyl, P. T. & Huber, W. HTSeq–a Python framework to work with
high-throughput sequencing data. Bioinformatics 31, 166–169 (2015).
64. Love, M. I., Huber, W. & Anders, S. Moderated estimation of fold change and
dispersion for RNA-seq data with DESeq2. Genome Biol. 15, 550 (2014).
65. Tyner, C. et al. The UCSC Genome Browser database: 2017 update. Nucleic
Acids Res. 45, D626–D634 (2017).
66. Li, H. & Durbin, R. Fast and accurate long-read alignment with Burrows-
Wheeler transform. Bioinformatics 26, 589–595 (2010).
67. Koboldt, D. C., Steinberg, K. M., Larson, D. E., Wilson, R. K. & Mardis, E. R.
The next-generation sequencing revolution and its impact on genomics. Cell
155,27–38 (2013).
68. Koboldt, D. C. et al. VarScan 2: somatic mutation and copy number alteration
discovery in cancer by exome sequencing. Genome Res. 22, 568–576 (2012).
69. Forbes, S. A. et al. COSMIC: somatic cancer genetics at high-resolution.
Nucleic Acids Res. 45, D777–D783 (2017).
70. Robinson, J. T. et al. Integrative genomics viewer. Nat. Biotechnol. 29,24–26
(2011).
71. Thorvaldsdottir, H., Robinson, J. T. & Mesirov, J. P. Integrative genomics
viewer (IGV): high-performance genomics data visualization and exploration.
Brief Bioinform. 14, 178–192 (2013).
72. Kharchenko, P. V., Silberstein, L. & Scadden, D. T. Bayesian approach to
single-cell differential expression analysis. Nat. Methods 11, 740–742 (2014).
73. Fan, J. et al. Characterizing transcriptional heterogeneity through pathway and
gene set overdispersion analysis. Nat. Methods 13, 241–244 (2016).
74. Scialdone, A. et al. Computational assignment of cell-cycle stage from single-
cell transcriptome data. Methods 85,54–61 (2015).
75. Huang da, W., Sherman, B. T. & Lempicki, R. A. Systematic and integrative
analysis of large gene lists using DAVID bioinformatics resources. Nat. Protoc.
4,44–57 (2009).
76. Supek, F., Bosnjak, M., Skunca, N. & Smuc, T. REVIGO summarizes and
visualizes long lists of gene ontology terms. PLoS One 6, e21800 (2011).
77. Shannon, P. et al. Cytoscape: a software environment for integrated models of
biomolecular interaction networks. Genome Res. 13, 2498–2504 (2003).
78. Montojo, J. et al. GeneMANIA Cytoscape plugin: fast gene function
predictions on the desktop. Bioinformatics 26, 2927–2928 (2010).
79. Lun, A. T., Bach, K. & Marioni, J. C. Pooling across cells to normalize single-
cell RNA sequencing data with many zero counts. Genome Biol. 17, 75 (2016).
80. Bourgon, R., Gentleman, R. & Huber, W. Independent filtering increases
detection power for high-throughput experiments. Proc. Natl Acad. Sci. USA
107, 9546–9551 (2010).
81. Langfelder, P., Zhang, B. & Horvath, S. Defining clusters from a hierarchical
cluster tree: the Dynamic Tree Cut package for R. Bioinformatics 24, 719–720
(2008).
82. Bakker, B. et al. Single-cell sequencing reveals karyotype heterogeneity in
murine and human malignancies. Genome Biol. 17, 115 (2016).
83. van den Bos, H. et al. Single-cell whole genome sequencing reveals no
evidence for common aneuploidy in normal and Alzheimer’s disease neurons.
Genome Biol. 17, 116 (2016).
Acknowledgements
We thank the patients for their commitment to research; the MCRC Biobank for the
sample collection; and members of the Taylor lab for advice and comments on the
manuscript. We also thank Peter March and Gary Spencer for 3D microscopy and
histology expertise respectively. This research was funded by Cancer Research UK
(C1422/A19842) with additional support from the Wellcome Trust Institutional Strategic
Support Fund, NWO-TOP (91215003), the NIHR Manchester Biomedical Research
Centre, and the University of Manchester. A.G. is supported by an Irshad Akhtar
Memorial PhD Scholarship.
Author contributions
Methodology, investigation and validation by L.N., A.T., A.G., S.L., B.B., D.M., D.S. and
R.D.M.; resources by S.D., B.W.-R., A.C., G.J., R.E., C.M.G., A.H. and F.F.; formal
analysis by S.M.B., I.D., B.N., R.W. and G.J.B.; conceptualisation, funding, supervision
and writing by S.S.T.
Competing interests
The authors declare no competing interests.
Additional information
Supplementary information is available for this paper at https://doi.org/10.1038/s41467-
020-14551-2.
Correspondence and requests for materials should be addressed to S.S.T.
Peer review information Nature Communications thanks Viive Howell, Jong Kim, and
the other, anonymous, reviewer for their contribution to the peer review of this work.
Peer reviewer reports are available.
Reprints and permission information is available at http://www.nature.com/reprints
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons
Attribution 4.0 International License, which permits use, sharing,
adaptation, distribution and reproduction in any medium or format, as long as you give
appropriate credit to the original author(s) and the source, provide a link to the Creative
Commons license, and indicate if changes were made. The images or other third party
material in this article are included in the article’s Creative Commons license, unless
indicated otherwise in a credit line to the material. If material is not included in the
article’s Creative Commons license and your intended use is not permitted by statutory
regulation or exceeds the permitted use, you will need to obtain permission directly from
the copyright holder. To view a copy of this license, visit http://creativecommons.org/
licenses/by/4.0/.
© The Author(s) 2020
ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14551-2
18 NATURE COMMUNICATIONS | (2020) 11:822 | https://doi.org/10.1038/s41467-020-14551-2 | www.nature.com/naturecommunications
Content courtesy of Springer Nature, terms of use apply. Rights reserved

1.
2.
3.
4.
5.
6.
Terms and Conditions
Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center GmbH (“Springer Nature”).
Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers and authorised users (“Users”), for small-
scale personal, non-commercial use provided that all copyright, trade and service marks and other proprietary notices are maintained. By
accessing, sharing, receiving or otherwise using the Springer Nature journal content you agree to these terms of use (“Terms”). For these
purposes, Springer Nature considers academic use (by researchers and students) to be non-commercial.
These Terms are supplementary and will apply in addition to any applicable website terms and conditions, a relevant site licence or a personal
subscription. These Terms will prevail over any conflict or ambiguity with regards to the relevant terms, a site licence or a personal subscription
(to the extent of the conflict or ambiguity only). For Creative Commons-licensed articles, the terms of the Creative Commons license used will
apply.
We collect and use personal data to provide access to the Springer Nature journal content. We may also use these personal data internally within
ResearchGate and Springer Nature and as agreed share it, in an anonymised way, for purposes of tracking, analysis and reporting. We will not
otherwise disclose your personal data outside the ResearchGate or the Springer Nature group of companies unless we have your permission as
detailed in the Privacy Policy.
While Users may use the Springer Nature journal content for small scale, personal non-commercial use, it is important to note that Users may
not:
use such content for the purpose of providing other users with access on a regular or large scale basis or as a means to circumvent access
control;
use such content where to do so would be considered a criminal or statutory offence in any jurisdiction, or gives rise to civil liability, or is
otherwise unlawful;
falsely or misleadingly imply or suggest endorsement, approval , sponsorship, or association unless explicitly agreed to by Springer Nature in
writing;
use bots or other automated methods to access the content or redirect messages
override any security feature or exclusionary protocol; or
share the content in order to create substitute for Springer Nature products or services or a systematic database of Springer Nature journal
content.
In line with the restriction against commercial use, Springer Nature does not permit the creation of a product or service that creates revenue,
royalties, rent or income from our content or its inclusion as part of a paid for service or for other commercial gain. Springer Nature journal
content cannot be used for inter-library loans and librarians may not upload Springer Nature journal content on a large scale into their, or any
other, institutional repository.
These terms of use are reviewed regularly and may be amended at any time. Springer Nature is not obligated to publish any information or
content on this website and may remove it or features or functionality at our sole discretion, at any time with or without notice. Springer Nature
may revoke this licence to you at any time and remove access to any copies of the Springer Nature journal content which have been saved.
To the fullest extent permitted by law, Springer Nature makes no warranties, representations or guarantees to Users, either express or implied
with respect to the Springer nature journal content and all parties disclaim and waive any implied warranties or warranties imposed by law,
including merchantability or fitness for any particular purpose.
Please note that these rights do not automatically extend to content, data or other material published by Springer Nature that may be licensed
from third parties.
If you would like to use or distribute our Springer Nature journal content to a wider audience or on a regular basis or in any other manner not
expressly permitted by these Terms, please contact Springer Nature at
onlineservice@springernature.com