A comprehensively characterized cell line panel highly representative of clinical ovarian high-grade serous carcinomas

Recent literature suggests that most widely used ovarian cancer (OVCA) cell models do not recapitulate the molecular features of clinical tumors. To address this limitation, we generated 18 cell lines and 3 corresponding patient-derived xenografts predominantly from high-grade serous carcinoma (HGSOC) peritoneal effusions. Comprehensive genomic characterization and comparison of each model to its parental tumor demonstrated a high degree of molecular similarity. Our characterization included whole exome-sequencing and copy number profiling for cell lines, xenografts, and matched non-malignant tissues, and DNA methylation, gene expression, and spectral karyotyping for a subset of specimens. Compared to the Cancer Genome Atlas (TCGA), our models more closely resembled HGSOC than any other tumor type, justifying their validity as OVCA models. Our meticulously characterized models provide a crucial resource for the OVCA research community that will advance translational findings and ultimately lead to clinical applications.


INTRODUCTION
Ovarian cancer (OVCA) ranks as the fifth deadliest cancer affecting women in the United States [1]. The dismal prognosis is attributable to a lack of early detection methods, effective treatment strategies, and the current inability to overcome acquired drug resistance [2]. High-grade serous ovarian carcinoma (HGSOC) is the most common and aggressive OVCA subtype [3]. Although comprehensive genomic characterization of HGSOC tumors has provided a catalogue of somatic alterations, their biological relevance and therapeutic potential remain to be thoroughly investigated [4]. An understanding of how these and prospectively discovered genetic alterations contribute to ovarian tumorigenesis is imperative for the development of new therapies targeting tumor biology [2].
To evaluate the biological significance of molecular alterations in OVCA tumorigenesis, clinically and genetically characterized cell models that recapitulate clinical OVCA are required [5,6]. Without representative cell models, in vitro derived findings have limited potential to translate to in vivo systems in the early stages of preclinical OVCA research [6][7][8].
Recently, Domcke et al. compared genomic profiles of clinical HGSOC tumors with commonly used HGSOC cell lines and revealed striking dissimilarity [8]. Only 12 infrequently used cell lines were classified as suitable HGSOC models [8], which is a relatively small number compared to other cancer types [9][10][11]. Moreover, this number is too small to capture the broad spectrum of heterogeneity observed in HGSOC [4,5,7,12,13]. This study and other recent reports have raised awareness that additional comprehensively characterized, readily available, and representative HGSOC models are needed to facilitate translational OVCA research [6,8,[14][15][16][17].
To address this need, we sought to generate a new panel of genomically characterized OVCA cell lines that accurately model clinical tumors. Unlike another recent group [18], we focused on generating HGSOC cell lines from malignant ascites, a common form and presentation of OVCA [19,20], using standard media and growth conditions. We created 18 cell lines and 3 patient derived xenografts that display a high concordance in copy number, methylation, gene expression, and mutational profiles with the tumors they were derived from. These models are also highly similar to HGSOC tumors from the Cancer Genome Atlas (TCGA) cohort. Most importantly, we offer these comprehensively characterized, early passage cell lines and xenografts to the research community in hopes of providing a new resource to facilitate advances in OVCA research that will ultimately lead to improved patient outcomes.

Establishment of comprehensively characterized OVCA cell lines
We generated and characterized 18 cell lines, 15 patient-matched lymphoblastoid lines, and 3 patientderived xenografts from 18 treatment naïve OVCA tumor ascites, providing a new panel of suitable models for preclinical research. We achieved a success rate of 41% (18/44 malignant ascites were established as cell lines) with standard media and culture conditions. Of these 18 lines, 6 (33%) were capable of growth as tumor xenografts in mice: HCC5006, HCC5012, HCC5023, HCC5044, HCC5048, and HCC5076 (HCC5012X, HCC5023X, and HCC5048X were genomically characterized, Table 1). Our method enabled us to obtain, enrich and cryopreserve large numbers of original tumor cells for comparison with the corresponding cell lines. Clinical information and culture characteristics for each model are summarized in Table 1 and Supplementary Table  1. Figure 1 illustrates a malignant effusion sample after hemolysis, ascites tumor enrichment, and representative cell lines generated. For each line we have determined a short tandem repeat (STR) profile to enable cell line identification (Supplementary Table 2). Expression levels of EPCAM and the characteristic OVCA markers MUC16, WT1, and PAX8 were high in tumors, cell lines and xenografts, demonstrating that our models maintain an epithelial phenotype and typical OVCA marker expression (Supplementary Figure  1) [21][22][23][24]. Furthermore, serum CA125 was detected at high levels in all patients except for HCC5030, and mRNA levels of its gene (MUC16) in corresponding models were also high, suggesting plasma CA125 in xenograft bearing mice could be used as an indicator of therapeutic efficacy [23,24] (Supplementary Table 3). We proceeded with comprehensive genomic characterization to 1) assess how well cell lines represent corresponding parental tumors, 2) assess similarity to HGSOC tumors in The Cancer Genome Atlas (TCGA) [4], and 3) enable researchers to select lines with the most appropriate genetic backgrounds. All but one line (HCC5075) exhibited a strong resemblance to clinical HGSOC (Table 1, Figure 1).

Cell lines and xenografts maintain the genomic landscape of patient-matched tumors
To determine whether the cell lines and xenografts we established were representative of the tumors they were derived from, we assessed concordance in their genomic profiles. We generated mutation and copy number profiles for every specimen except for two tumors and one lymphoblast line with insufficient material for exome-sequencing, as these data provide the most unique individual signatures (Supplementary Tables 4-6). Clustering and correlation analyses of single nucleotide  Figure 4A). Chromosomal instability and copy number alterations (CNAs) are hallmark features of HGSOC [14,25], and besides mutational profiles, copy number profiles were anticipated to provide the most unique signature for each tumor case. Genome-wide CNA profiles revealed 11/15 tumor-cell line pairs showed very strong concordance, with correlation coefficients, ρ >0.6 (Pearson test, p<0.0001, Figure Table 8). Collectively, our multi-dimensional genome-wide profiling supports the conclusion that the cell lines and xenografts we generated are highly representative of the tumors they were derived from. The high genomic instability, low mutational burden, specific mutations and expression of typical OVCA  markers observed in our cell lines were highly consistent with HGSOC, with the exception of one case [14,[21][22][23][24][25]. HCC5075 had pathological and molecular features consistent with a low-grade serous ovarian cancer (Table 1) [14,25]. Additionally, one case (HCC5032) was confirmed to be of endometrial origin upon detailed histopathological review after manuscript submission (Table 1).

Established cell lines and xenografts faithfully model HGSOC tumors
We next investigated how well our models represented the TCGA's HGSOC cohort. Mutational counts and the fraction of genome altered by CNAs (FGA) in cell lines, xenografts, and tumors were comparable to those of TCGA tumors ( Figure 3A-3B, two-tailed student's t-test, p>0.05). Our cell lines and tumors had lower mutational loads (i.e. mutations/megabase) than TCGA tumors ( Figure 3C, Student's t-test, p = 0.004 and p=0.003, respectively) although the range observed in TCGA tumors was large. Despite OVCA tumors having few recurrent mutations besides TP53 [4], we observed HGSOC-characteristic mutations among the cell lines and xenografts we derived (Table 1, Supplementary Tables  4-5). TP53 mutations were identified in 16/18 cases, and BRCA1, BRCA2, CSMD3, CDK12, NF1, RB1, and MECOM mutations were also detected among the models, illustrating the spectrum of genetic heterogeneity we captured. Furthermore, we observed significant positive correlations between our cell lines and xenografts with the mean copy number (ρ range: 0.12-0.62), methylation (ρ range: 0.39-0.70), and gene expression profiles (ρ range: 0.43-0.61) of TCGA OVCA tumors (Pearson test, p<0.0001, Supplementary Figure 6). Notably, the magnitudes of the positive correlation coefficients we observed across the various cell lines/xenografts and data dimensions were similar to those observed for the CCLE lines classified as good models of HGSOC by Domcke et al. [8]. We also observed an impressive concordance in global CNA patterns between our models and the TCGA tumors ( Figure 3D, Supplementary Figure 7). GISTICbased analysis of high magnitude CNAs [26] identified Our models exhibit low mutation loads and high levels of genomic instability which indicates that their genomic landscapes are typical of HGSOC tumors [8,14,27,28].
Furthermore, we characterized the response of 8 cell lines to standard chemotherapeutics used to treat OVCA, cisplatin and paclitaxel. Adherent cell lines were assessed because they generate reproducible results in dose-response assays. The range in cell line sensitivities was consistent with those reported in the Supplementary Material of Ince et al. [18] (Supplementary Table 11).
There was no correlation observed between in vitro cell line response to cisplatin or paclitaxel and clinical patient response (data not shown). However, this is not surprising given that cell lines are excellent vehicles for predicting response to relevant targeted therapies, but less predictive for response to non-targeted chemotherapeutic agents, as has been described before [5].

UTSW ovarian cell lines are more similar to OVCA than several other TCGA tumor types
Given the known issues of cross-contamination and poor histological annotation of cell lines from different fraction of genome altered (FGA) was compared between the 18 UTSW cell lines, 11 UTSW tumors, 3 UTSW xenografts, and 583 TCGA HGSOC tumors. No significant differences in the extent of FGA between UTSW samples and TCGA tumors were found (Student's t-test, p>0.05). B., C. Comparison of somatic mutation counts and frequencies in UTSW samples versus TCGA tumors. Only functional somatic mutations (i.e. those predicted to have a biological effect on protein function) were considered. We observed no significant differences in the number of somatic mutations detected between UTSW samples and the TCGA tumors (B, Student's t-test, p>0.05). The UTSW cell lines and tumors had slightly lower mutational frequencies (ie. mutations per megabase) of DNA compared to the TCGA tumors (C, Student's t-test, p<0.05), however, given the large number of samples profiled in the TCGA cohort (n=316), the variability in mutational load of TCGA tumors is much larger than the UTSW samples. D. Comparison of the mean copy number profiles for UTSW cell lines and tumors with the TCGA HGSOC tumors for chromosomes 5 and 8. Tissue-matched TCGA non-malignant profiles are plotted as a copy-neutral (e.g. diploid) reference (black). UTSW cell line copy number patterns (blue) resemble the UTSW tumors (red), and are highly concordant with those of the TCGA tumors (green). Each plotted dot represents the copy number for an individual gene. Supplementary Figure 7 illustrates copy number patterns for UTSW samples and TCGA HGSOCs for all autosomal chromosomes. tissue origins [5,6,16,29], we next aimed to determine whether the lines we created were more strongly correlated to OVCA tumors than to other TCGA tumor types. We again investigated the correlations between gene copy number, methylation and expression profiles of our cell lines with the mean profiles of various TCGA tumor types. We also compared the mutation counts observed in our cell lines to other TCGA tumor types by considering somatic mutations predicted to have a functional impact. These analyses demonstrated that our cell lines most closely resemble OVCA compared to several other epithelial tumor types in each genomic dimension we assessed (Figure 4).

DISCUSSION
Cancer cell lines and patient-derived xenografts are invaluable tools for studying tumor biology and are widely used to test new anticancer drugs in a preclinical setting [5,9,10,17,[29][30][31]. However, it is now appreciated that for cell lines and xenografts to be clinically relevant and reliable cancer models, they must be well characterized and recapitulate the genomic landscapes typical of clinical tumors [5-8, 29, 30]. A number of reports have articulated the need for new OVCA models to improve translational research, raising numerous concerns with currently available models including: misidentification and/or contamination, lack of resemblance to clinical tumors, lack of molecular characterization, and lack of availability of the most suitable models [6-8, 15-18, 32, 33]. A revealing study by Domcke et al. found that only 12 of 47 OVCA cell lines are suitable HGSOC models, and that they are used in only 1% of studies reporting on HGSOC [8]. In their report, the authors proposed that newly generated, well-characterized HGSOC cell lines from treatment naïve patients could greatly benefit the OVCA research community by improving the value of preclinical discoveries [8].
To address this need, we generated a new panel of comprehensively characterized models which are highly representative of clinical HGSOC tumors with the exception of one case (HCC5075 was TP53 wild-type, KRAS mutant, and had low genomic instability which are features are consistent with low-grade serous ovarian carcinoma [14,25,27,34,35].). Further histopathological characterization after manuscript submission also revealed that one case (HCC5032) was an endometrial tumor, although its genomic landscape was strongly correlated to those of HGSOC. We have provided multi-faceted genomic evidence to demonstrate that the models we have generated genuinely resemble their parental primary tumors and clinical HGSOCs. Improving on a recently described panel of OVCA cell lines and xenografts [18], the models we have generated: i) include corresponding non-malignant lymphoblastoid cell lines (as sources of constitutional DNA), ii) can be cultured with simple media and growth conditions that are economical and highly amenable for high-throughput drug screening, and iii) include additional dimensions of genomic characterization (namely, wholeexome sequencing, genome-wide DNA methylation, and SKY profiling). Importantly, we have cryopreserved numerous vials of each cell line at low passages and we have deposited the genomic data to share with the research community (GEO Accession ID: GSE71525, SRA BioProject ID: PRJNA291290). The patient-derived xenografts we have generated will also be available for use as in vivo models of HGSOC. Finally, we have also cryopreserved multiple vials of the original tumor cell pellets to enable experiments using the primary tumor cells in addition to the corresponding cell lines or xenografts.
Our study sets an unprecedented standard for generating comprehensively characterized cancer cell lines to meet current research needs. In the era of precision medicine [36], it is essential to select representative experimental models that recapitulate the molecular features of the tumors being studied. Our preclinical models will be useful not only for studying HGSOC tumor biology, but also for assessing the efficacy of anticancer agents because of the ease with which they can be cultured and their known mutational profiles. Our new HGSOC panel is a much-needed invaluable tool that will enhance the translation potential of in vitro findings [15,29], ultimately leading to improved therapies and outcomes for OVCA patients.

MATERIALS AND METHODS
Detailed methods describing OVCA model establishment and genomic profiling and analyses can be found in the Supplementary Material. Cell lines and patient-derived xenografts were generated from malignant peritoneal effusions of 18 treatment naïve OVCA patients with informed patient consent at the University of Texas Southwestern (UTSW). Subsequent pathological review revealed that one tumour (HCC5032) was of endometrial origin after manuscript submission (Table 1). Patient matched lymphoblastoid lines for 15 cases were generated from peripheral blood mononuclear cells, cultured ascites fluids, or mesothelial cells from malignant effusions. Cell line sensitivities to cisplatin and paclitaxel were determined using CellTiter 96 ® Aqueous One Solution Cell Proliferation Assays (Promega) performed according to the manufacturer's instructions.
Genomic profiling including genome-wide copy number (Affymetrix SNP 6 arrays), gene expression (Illumina HT-12v4 BeadChips), DNA methylation (Illumina HM450K arrays) and exome-sequencing (SureSelect Target Enrichment System for Illumina Paired-End Sequencing) was conducted on genomic DNA or RNA extracted from tumors, cell lines, xenografts, or constitutional DNA from lymphoblastoid lines. All genomic data are available at the Gene Expression Omnibus (GSE71525) and Sequence Read Archive (PRJNA291290). Spectral karyotyping (SKY) was performed on a subset of samples as previously described [37]. Genomic data for the TCGA OVCA cohort was obtained from the TCGA Data Portal, the TCGA OVCA publication page (https://tcga-data.nci.nih.gov/docs/ publications/ov_2011/) [4], cBioPortal [38,39], and pancancer TCGA data from the Cancer Genomics Browser (CGB) [40]. Detailed materials and methods, including statistical analyses are provided in the accompanying Supplementary Material.