Most Commonly Mutated Genes in High-Grade Serous Ovarian Carcinoma Are Nonessential for Ovarian Surface Epithelial Stem Cell Transformation

SUMMARY

High-grade serous ovarian carcinoma (HGSOC) is the fifth leading cause of cancer-related deaths of women in the United States. Disease-associated mutations have been identified by the Cancer Genome Atlas Research Network. However, aside from mutations in TP53 or the RB1 pathway that are common in HGSOC, the contributions of mutation combinations are unclear. Here, we report CRISPR mutagenesis of 20 putative HGSOC driver genes to identify combinatorial disruptions of genes that transform either ovarian surface epithelium stem cells (OSE-SCs) or non-stem cells (OSE-NSs). Our results support the OSE-SC theory of HGSOC initiation and suggest that most commonly mutated genes in HGSOC have no effect on OSE-SC transformation initiation. Our results indicate that disruption of TP53 and PTEN, combined with RB1 disruption, constitutes a core set of mutations driving efficient transformation in vitro. The combined data may contribute to more accurate modeling of HGSOC development.

Graphical Abstract

In Brief

Yamulla and colleagues perform combinatorial CRISPR mutagenesis of 20 genes commonly disrupted in high-grade serous ovarian carcinoma. They find combinations driving or enhancing efficient transformation of surface epithelium stem cells, with a core set being Tp53, Pten, and Rb1/Cdkn2a. Several genes either do not promote or inhibit transformation, including Brca2.

INTRODUCTION

Ovarian cancer is a complex disease consisting of several distinct subtypes that differ in progression, prognosis, cell of origin, and genetic alterations (Bowtell, 2010). It is the fifth leading cause of cancer-related female deaths in the western world (Siegel et al., 2020). High-grade serous ovarian carcinoma (HGSOC) is the most common (about 70%) and lethal subtype of ovarian cancer, in part due to its propensity to metastasize and to relapse following chemotherapy (Auersperg, 2013). Additionally, HGSOC screening methodologies are inefficient, typically resulting in late-stage diagnosis. The rarity of early-stage HGSOC detection has complicated ascertainment of the cell of origin, initiating mutations, and the identification of precursor lesions (Auersperg, 2013; Feeley and Wells, 2001).

Much progress has been made regarding the genetic etiologies of HGSOC. The Cancer Genome Atlas (TCGA) completed a comprehensive genomic analysis of patient tumor samples and commonly dysregulated genes and pathways (Table 1; Cancer Genome Atlas Research Network, 2011). Common mutations and deletions of genes are of particular interest, as they may drive HGSOC initiation and development. Several of the putative TCGA driver genes have been thoroughly investigated. For instance, TP53 (Trp53 in mice) is mutated or inactivated in nearly all tumors (Cancer Genome Atlas Research Network, 2011) and has been validated as a crucial driver of carcinogenesis in mouse models (Bobbs et al., 2015; Harlan and Nikitin, 2015; Kim et al., 2018). However, most of the recurrently altered HGSOC driver genes are mutated or deleted in a smaller subset of tumors and have not been validated experimentally in animal models or cell transformation paradigms. Several genes, including WWOX, LRP1B, CDKN2A, and PTEN, exist near fragile sites in the genome and therefore may be mutated simply as a consequence of genome instability rather than cancer initiation (Hess et al., 2019).

Table 1.

Alteration Frequency of Minilibrary Target Genes in HGSOC

Gene Name	Mutations	Amplifications	Deletions	Total	% Disrupted
Trp53	303	2	1	306	96.20
Brca1	37	1	1	39	12.03
Brca2	34	3	2	39	11.39
Csmd3	19	47	1	67	6.33
Nf1	12	1	24	37	11.39
Fat3	18	8	1	27	6.01
Gabra6	6	3	1	10	2.22
Rb1	6	1	25	32	9.81
Apc	7	2	3	12	3.16
Lrp1b	13	6	13	32	8.23
Prim2	0	6	2	8	0.63
Cdkn2a	0	1	7	8	2.22
Crebbp	7	3	10	20	5.38
Wwox	0	2	14	16	4.43
Ankrd11	4	1	10	15	4.43
Map2k4	1	0	11	12	3.80
Fancm	2	3	0	5	0.63
Fancd2	1	3	0	4	0.32
Rad51c	0	2	0	2	0.00
Pten	2	2	21	25	7.28

Putative HGSOC driver genes were derived primarily from the Cancer Genome Atlas Research Network. Most genes, except for FANCM and APC, were found to be significantly mutated or deleted in HGSOC tumors by TCGA.

Although TP53 is mutated in nearly all HGSOC cases, experiments with mouse models indicate that a Trp53 mutation alone is insufficient for HGSOC initiation; rather, multiple mutations appear to be required, consistent with the multi-hit hypothesis of cancer (Flesken-Nikitin et al., 2003; Knudson, 1971). For example, concurrent inactivation of Trp53 and Brca1, both commonly mutated in HGSOC, could not drive transformation in mice (Xing and Orsulic, 2006). However, activation of Myc along with disruption of both Trp53 and Brca1 did initiate HGSOC. It was also reported that disruption of Trp53 or Rb1 alone in the ovarian surface epithelium (OSE) caused neoplasms in only 4/31 and 1/21 mice, respectively, but simultaneous mutations in both genes caused 100% cancer incidence after a median 227 days (Flesken-Nikitin et al., 2003). Given the substantial numbers of commonly mutated genes identified by TCGA, there are a myriad of possible TCGA driver gene combinations, but very little data regarding how these different combinations could affect transformation efficiency of putative HGSOC “cells of origin.”

There is a growing consensus that HGSOC may have several places of origin, such as OSE, tubal epithelium (TE), and peritoneal serosa (Hao et al., 2017; Harlan and Nikitin, 2015; Kim et al., 2018; Lawrenson et al., 2019; Zhang et al., 2019). The OSE is a flat to cuboidal cell monolayer that overlies the ovary and was originally proposed as the HGSOC putative cell type of origin due to a correlation with tumor localization and the observation that a greater number of ovulatory cycles correlates with increased cancer incidence (Auersperg, 2013; Fathalla, 1971, 2013; Okamura et al., 2006). Research has suggested that repeated cycles of follicular rupture, OSE damage, inflammation, and repair may trigger oncogenic transformation of OSE (Katabuchi and Okamura, 2003). Inclusion cysts, or entrapment of OSE within the ovarian stroma, may also facilitate OSE transformation by exposing it to high concentrations of hormones, growth factors, and inflammatory cytokines that are not present at the ovarian surface. OSE within inclusion cysts has been previously shown to express HGSOC markers like PAX8 (Auersperg, 2013). Importantly, the OSE has been experimentally shown to transform into HGSOC-like neoplasms (Flesken-Nikitin et al., 2003; Kim et al., 2018; Zhang et al., 2019).

In mice, HGSOC can initiate from OSE stem cells (OSE-SCs) (Flesken-Nikitin et al., 2013). Such cells are marked by high aldehyde dehydrogenase (ALDH) activity and transform more readily than OSE non-SCs (OSE-NSs) characterized by low ALDH activity. OSE-SCs were found to have increased colony-forming potential in primary culture, greater sphere formation capacity in vitro, and an increased ability to proliferate in culture before undergoing senescence (Flesken-Nikitin et al., 2013). The cells also express multipotency markers and readily transform following combined knockout of Trp53 and Rb1 in mice. Thus, the ALDH⁺ OSE-SC subpopulation is a candidate originating source of HGSOC.

Here, we sought to identify and functionally validate combinations of putative driver genes in HGSOC and the transformation susceptibility of different ovarian epithelial cell types to combinations of mutations. The 20 candidate genes tested were primarily those that were identified by TCGA as commonly mutated in HGSOC (Table 1). Random sets of mutations were induced by infection of mouse OSE-SCs and OSE-NSs with a minilibrary of lentiviruses encoding Cas9 and CRISPR guide RNAs directed against the candidate driver genes. We found that OSE-SCs transform more efficiently than OSE-NSs and that only a fraction of commonly mutated HGSOC genes contribute to transformation in vitro. In addition to Trp53 and Rb1, mutation of Pten was found to be centrally important for the transformation of mouse OSE in vitro. We also report unusual transformation-enhancing mutation combinations and propose a model of core OSE-SC transformation requirements.

RESULTS

Strategy for Screening Candidate HGSOC Suppressors and Construction of a Validated CRISPR-Based Lentiviral Minilibrary

We adapted a strategy (Figure 1), analogous to that described by Zender et al. (2006), (2010), to validate candidate tumor suppressors in a sensitized cell type. Sensitized cells do not undergo transformation in culture or in vivo upon transfer to a mouse host but can do so when an additional gene or combination of genes are disrupted. Because TP53 is mutated in nearly all HGSOCs but Trp53 mutagenesis alone does not cause spontaneous ovarian tumors in mice (Donehower, 1996, 2014; Flesken-Nikitin et al., 2003), we decided to use Trp53^+/− cells in an inbred mouse strain background as the sensitized platform. Accordingly, we created and validated a new null Trp53 allele in strain FVB/NJ as a source of ovarian surface epithelial cells (Figures S1A–S1D) for all screening experiments. Strain FVB/NJ was chosen because it is neither susceptible nor resistant to spontaneous ovarian lesions (Huang et al., 2008; Mahler et al., 1996).

Figure 1. Strategy for Identifying HGSOC Tumor Suppressor Combinations.

A total of 60 constructs were made in the vector LentiCRISPRv2, constituting the “minilibrary.” OSE-NSs or OSE-SCs were transduced with functionally validated LentiCRISPRs and then plated in soft agar. Individual transformants/colonies were isolated and individually cultured. Genome-integrated LentiCRISPRs from each transformant were identified by sequencing, and overrepresented combinations were later validated in directed soft agar transformation assays. OSE-SCs, ovarian surface epithelium stem cells; OSE-NSs, ovarian surface epithelium non-stem cells; NGS, next-generation sequencing.

We generated a list of 20 potential HGSOC driver genes, largely corresponding to the most commonly mutated or deleted genes according to TCGA. We also included FANCM due to the putative role of Fanconi-anemia-related genes in HGSOC development and APC because of its critical role in canonical Wnt signaling and association with type I ovarian carcinoma (Auersperg, 2013; Bowtell, 2010; Cancer Genome Atlas Research Network, 2011; Yamulla et al., 2014; Table 1). We hypothesized that if these genes were true tumor suppressors, then mutating them alone or in combination with Trp53 and/or other gene mutations would drive transformation in the relevant cell type. We next constructed a lentiviral CRISPR (version 2; hereafter called “LentiCRISPR”) (Sanjana et al., 2014; Shalem et al., 2014) minilibrary targeting the 20 putative HGSOC tumor drivers, infected the potential cells of origin at a high multiplicity of infection (MOI), and assessed which vectors were overrepresented in transformed cells (see STAR Methods; Figure 1). The library contained 3 constructs per gene, with the single guide RNAs (sgRNAs) in each vector targeting the earliest possible exon of each gene to increase the likelihood of causing loss-of-function indel mutations by error-prone non-homologous end joining (NHEJ) repair. We tested the minilibrary for gene editing efficiency in pilot assays before its use in HGSOC driver screening (Figure S2; Table S1) and designed new guides to replace vectors that did not efficiently mutate targets (Figure S3; Table S2).

Random Combinatorial Mutagenesis of Candidate HGSOC Driver Genes in OSE-SCs and OSE-NSs

To assess transformation frequency in vitro, we infected 30,000 fluorescence-activated cell sorting (FACS)-processed Trp53^+/− OSE-SCs and OSE-NSs (ALDH⁺ and ALDH⁻, respectively; Figure S4)witheither the LentiCRISPR minilibrary ora GFP-containing lentivirus, both at an MOI of ~7 Figures S5 and S6). The cells were then plated in soft agar to allow for assessment of adhesion-independent growth, a hallmark of carcinogenesis and transformation (hereafter, formation of adhesion-independent colonies will be referred to as “transformation,” and individual colonies will be referred to as “transformants”) (Borowicz et al., 2014; Hamburger and Salmon, 1977; Horibata et al., 2015; de Larco and Todaro, 1978; Puck et al., 1956; Roberts et al., 1985). Given an MOI of 7 and 3 LentiCRISPRs per gene, each of the 20 genes represented in the minilibrary should be targeted in 30.2% (~9,000) of transduced cells. Control experiments supported this estimate and that there were no technical biases, including the appearance of a control GFP LentiCRISPR at an expected frequency of 12% or transformants (STAR Methods; Figures 3A, 3B, and S7). Assuming random integration, any two genes would be co-targeted in 9% of samples, three genes in 2.7% of cells, and so on.

Figure 3. Identification of Genome-Integrated LentiCRISPRs and Overrepresented Target Gene Combinations in OSE-SCs.

(A) Genome-integration and hierarchical clustering of LentiCRISPRv2 constructs in OSE-SC samples. Hierarchical clustering was performed on both sample similarity and gene targeting.

(B) Overall percent gene targeting and co-targeting frequency. Significance (p < 0.05) for single integration overall was assessed using χ² (degrees of freedom [df] = 19) and is indicated with an asterisk. The heatmap displays co-integration frequency of each gene present on the x axis with a gene shown on the y axis.

(C) Overrepresentation of co-targeted genes in sample subgroups. Over-or underrepresentation was determined using χ² analyses. χ² values corresponding to p ≤ 0.05 (df = 19) are colored in green. Red coloration indicates that co-integration may have occurred by chance and that the p ≥ 0.05.

No colonies were observed in either non-transduced cells or cells infected with the GFP-containing vector (Figures S1E, S1F, and S6). Thus, heterozygosity for Trp53 alone, or the process of infection, did not enable transformation of either cell type. However, transformants were observed in cells transduced with the entire library, albeit at low frequency despite an efficient infection frequency, as judged by control infections with GFP lentivirus and LentiCRISPRv2 serial dilution assays (Figures 2A and S5). OSE-SCs were transformed (formed colonies) much more efficiently than OSE-NSs or unsorted OSE (0.66%, 0.02%, and 0.21% of total plated cells, respectively; Figure 2A). That unsorted OSE cells produced 10.5-fold more colonies than OSE-NSs is likely due to its OSE-SC subpopulation (Figures 2A and S4A). To rule out the possibility that OSE-SCs are simply infected by lentiviruses more efficiently than OSE-NSs, we transduced each cell type with equal concentrations of mCherry lentivirus and scored for mCherry expression in each population. OSE-SCs and OSE-NSs were transduced at similar rates (72.9% and 85.6% mCherry expression, respectively), suggesting that transformation efficiency differences were indeed cell-type specific (Figures 2B and 2C).

Figure 2. OSE-SCs (ALDH⁺) Transform More Frequently Than OSE-NSs (ALDH⁻) Despite SimilarViral Transduction Rates.

(A) Percent transformation of OSE-SCs, OSE-NSs, and unsorted OSE cells following LentiCRISPRv2 minilibrary transduction. OSE-SCs transformed more frequently than OSE-NSs and unsorted OSE cells. Unsorted OSE cells transformed more frequently than OSE-NSs (5 technical replicates, Standard error of the mean [SEM] error bars).

(B and C) FUGW-mCherry (mCherry-expressing lentivirus) transduction efficiency in OSE-SCs and OSE-NSs detected by flow cytometry. Flow cytometry was used to count mCherry⁺ cells following transduction with equal concentrations of FUGW-mCherry lentivirus. Percentages indicate the percentage of total cells that are mCherry⁺. The dark gray lines represent cell counts of untransduced cells. The red line represents cell counts of FUGW-mCherry-transduced cells. OSE-SCs and OSE-NSs gained mCherry fluorescence at similar rates following lentiviral transduction.

The rate of OSE-SC transformation (0.66%) was approximately the frequency expected for a particular combination of 4 targeted genes (0.8%), suggesting either that there is indeed one major 4-gene mutation combination that is essential for transformation or that there are more combinations but not all are 100% effective in transformation.

Enrichment of Mutated Gene Combinations in Transformed OSE-SC samples

Next, we determined whether particular minilibrary constructs and combinations thereof were over-represented in the transformants. To do this, we isolated all individual colonies and cultured them as transformed, adherent, clonal cell lines. The LentiCRISPR vectors were identical save for their unique 20-bp Cas9 guide sequences that served as molecular barcodes (Table S3). We therefore identified all genome-integrated LentiCRISPRs from genomic DNA isolated from each transformant line by using a next-generation-sequencing-based approach (Figure 1). Three LentiCRISPRs per target gene were included in our minilibrary to control for unintended effects of any single construct. Overrepresented integration of only one of three gene-targeting constructs would indicate potential spurious technical errors, such as excessively high titer or off-target effects. Our sequencing dataset revealed genome integration of all three LentiCRISPRs corresponding to target genes found in more than 30% of colonies, suggesting that gene targeting was not a consequence of technical issues (STAR Methods).

Trp53 Loss Alone Is Necessary but Insufficient for OSE-SC Transformation

Nearly all (96%) OSE-SC colonies contained LentiCRISPRs targeting Trp53, consistent with human tumor samples that almost universally harbor TP53 mutations (Figures 3A and 3B; χ² = 19, p < 0.05). As expected given the high MOI, most OSE-SC transformants (88 of 92) also harbored LentiCRIPSRs targeting other genes. The most common genome-integrated vectors corresponded to Pten (76%), Rb1 (68%), Cdkn2a (55%), Fancd2 (49%), Wwox (45%), Gabra6 (49%), Fat3 (39%), Apc (26%), Crebbp (26%), and Brca2 (24%) (Figures 3A, 3B, and S8B). The rarity of transformants lacking Trp53 LentiCRISPRs (4%) suggests that heterozygosity of Trp53 is insufficient for transformation, even in the context of other mutations that can synergize with partial Trp53 loss (Figures 3A and 3B). Frequent co-targeting of Trp53 also supports previous assertions that inactivating mutations in Trp53 are crucial transformation precursors but are alone insufficient for OSE transformation (Flesken-Nikitin et al., 2003).

LentiCRISPRs Targeting Trp53, Rb1, Cdkn2a, and Pten Are Overrepresented in OSE-SC Transformants

In addition to Trp53, most colonies (74%) had LentiCRISPRs targeting either Rb1 or Pten, and nearly half of the colonies (45%) had LentiCRISPRs targeting both (Figures 3A and 3B). Most Pten− colonies (hereafter, we will refer to transformants bearing particular LentiCRISPRs by the corresponding gene symbol followed by “−”; e.g., Pten−, with the caveat that the actual target gene may not have been mutated to a null state) were also Rb1− (70%), suggesting that concurrent inactivation of these two genes facilitates transformation (Figure 3B). This observation supports significant co-occurrence of RB1 and PTEN mutations in HGSOC and also in many other cancers, such as metastatic prostate cancer, lipomas, and astrocytomas (Cancer Genome Atlas Research Network, 2011; Cerami et al., 2012; Chow et al., 2011; Filtz et al., 2015; Gao et al., 2013; Hamid et al., 2019). Although Rb1 was not targeted in all Pten− colonies, 88% of Pten- colonies lacking a Rb1 LentiCRISPR had at least one LentiCRISPR targeting Cdkn2a (Figure 3A), which encodes p14ARF and p16INK4a. These are important regulators of TRP53 and RB1, respectively (Li et al., 2011; Nielsen et al., 1998; Zhao et al., 2016). Only 3% of Pten− colonies in our dataset had no Rb1 or Cdkn2a LentiCRISPRs, suggesting that either direct or indirect disruption of Rb1 is important in Pten− colonies (Figure 3A).

Mutations in Trp53, Rb1, Pten, and Cdkn2a Function Synergistically to Promote Transformation

To validate and dissect the apparent major contributions of Trp53, Pten, Rb1, and Cdkn2a in suppressing OSE-SC transformation, we performed additional infections with vectors corresponding to combinations of just these four genes (three LentiCRISPRs per gene). Confirming our observations from the whole library screen (see above), mutating the second allele of Trp53 alone in Trp53^+/− OSE-SCs was inefficient in transformation (Figures 4A, S9, and S10A). However, co-mutagenesis with combinations of Rb1, Cdkn2a, and Pten significantly enhanced transformation efficiency (colony number) and colony size (Figures 4A, S9, and S10A). Specifically, we observed that targeting Trp53 alongside Pten and Rb1, Pten and Cdkn2a, or Pten, Rb1 and Cdkn2a led to greater transformation efficiencies and larger colonies (Figures 4A, S9, and S10A). The most efficient transformation and largest colony sizes occurred when Trp53, Rb1, Cdkn2a, and Pten were targeted simultaneously (Figures 4A, S9, and S10A). There was no significant increase in transformation efficiency when Trp53 was co-targeted with only 1 of the other 3 genes, although there was a significant increase in the size of Trp53−/Rb1− and Trp53−/Cdkn2a− colonies (Figure S10A). These results indicate that a deficiency of Trp53, Rb1, and Pten constitutes a core state for efficient transformation of OSE-SCs in vitro.

Figure 4. Targeted OSE-SC-Transformation Assay and Validation of Overrepresented LentiCRISPR Combinations.

A baseline level of adhesion-independent growth was first assessed y induction of specific “core mutations” by LentiCRISPRv2 targeting. Additional minilibrary target genes were then mutated (using LentiCRISPRv2) alongside core mutations to assess whether they act synergistically to promote adhesion independent growth. ANOVA post hoc analyses were used to assess differences between means. The p values for group means greater than the baseline are shown. Colony counts that are significantly lower than baseline rates (p < 0.05) are labeled with an obelisk (†). SEM error bars. n = 6 for all cases unless otherwise specified.

(A) Targeted transduction of Trp53, Rb1, Pten, and Cdkn2a LentiCRISPRs in OSE-SCs. Significantly greater rates of OSE-SC transformation versus OSE-SC transduced with Trp53 LentiCRISPRs alone occurred when all four genes were targeted together or by mutagenesis of Trp53, Rb1 (or Cdkn2a), and Pten.

B) Targeted transduction of Trp53, Cdkn2a, and Pten LentiCRISPRs plus putative transformation enhancers. Only the addition of Ankrd11 or Wwox LentiCRISPRs to Trp53, Cdkn2a, and Pten LentiCRISPRs significantly enhanced colony formation versus Trp53−/Cdkn2a−/Pten− OSE-SCs. The addition of Fancd2 or Rad51c significantly decreased colony count.

(C) Targeted transduction of Brca2 LentiCRISPRs and Brca2-associated LentiCRISPRs. Fancd2 mutations functioned synergistically with Brca2, Trp53, and Rb1 mutations to significantly enhance colony formation versus Trp53−/Rb1−/Brca2− or Trp53−/Rb1− OSE-SCs.

(D) Targeted mutagenesis of Rad51c LentiCRISPRs and Rad51c-associated LentiCRISPRs. Fat3 mutagenesis alongside Trp53, Cdkn2a, Pten, and Rad51c significantly increased colony count compared to Trp53−/Rb1−/Pten−/Rad51c− OSE-SCs.

Disruption of Ankrd11 or Wwox Can Further Enhance Transformation of Trp53−/Cdkn2a−/Pten− OSE-SCs

Interestingly, certain cohorts of genes were co-targeted at significant frequencies in Trp53−/Rb1−/Pten− (Cdkn2a, Gabra6, Fancd2, Wwox, Fat3, Apc, Crebbp, Lrp1b, Fancm, and Csmd3; χ² = 19, p < 0.05) or Trp53−/Cdkn2a−/Pten− colonies (Rb1, Gabra6, Fancd2, Wwox, Fat3, Apc, Lrp1b, Rad51c, and Fancm; χ² = 19, p < 0.05), despite vectors for each being underrepresented overall (Figures 3B and 3C). To explore whether mutations in these genes influenced transformation, we performed additional OSE-SC infections in which individual genes were mutated along with the core combinations of Trp53, Cdkn2a (or Rb1), and Pten. Most (Lrp1b, Fancm, Crebbp, Rad51c, Fat3, Apc, Fancd2, and Gabra6) did not enhance transformation rates, and two (Fancd2 and Rad51c) actually decreased transformation rates (Figures 4B and S11). However, significant increases in transformation frequency were observed when LentiCRISPRs targeting Ankrd11 or Wwox were added to the core of Trp53, Cdkn2a, and Pten (Figure 4B). No significant increase in colony size was noted following mutagenesis of any TCGA driver genes other than Trp53, Rb1, Cdkn2a, and Pten (Figure S10B). We surmised that the overall underrepresented target genes that had no (or a negative) effect on OSE-SC transformation in targeted experiments, but that were often present in cells with high numbers of other vectors, were technical artifacts. Indeed, an association between underrepresented target genes and the number of genes targeted per colony was observed (Figure S12). For instance, most samples with LentiCRISPRs targeting Ankrd11, Apc, Lrp1b, Brca1, Nf1, Fancm, Fancd2, and Map2k4 occurred in colonies with 8 or more targeted genes (Figure S12). Such clones also generally contained common LentiCRISPRs for Trp53, Rb1, Pten, and Cdkn2a (Figures 3A and 3B), suggesting that many or all of these lower frequency “hits” are unrelated to transformation but that perhaps the parental cell was particularly susceptible to viral infection.

Brca2 Disruption Deters Transformation of Trp53−/Rb1− OSE-SCs but Enhances Transformation when Co-mutated with Trp53, Rb1, and Fancd2

Notably, LentiCRISPRs for Brca2 were underrepresented overall in OSE-SC colonies, despite the association of mutations in this gene with familial HGSOC (Figures 3A and B; Cancer Genome Atlas Research Network, 2011; Risch et al., 2006; Walsh, 2015). Brca2 deficiency is cell lethal in the absence of other mutations, causing replication stress, mitotic abnormalities, 53BP1 activation, and G1 arrest (Feng and Jasin, 2017; Zhu et al., 2015). However, most cancer cells develop mechanisms to overcome this G1 arrest through rescuing mutations in other genes like Trp53 (Feng and Jasin, 2017). In our screen, 95% of Brca2− colonies were also Trp53−/Rb1− (Figures 3A and 3B). Many Brca2− colonies also contained LentiCRISPRs targeting Fancd2 (82%), Wwox (77%), and Crebbp (45%) (Figure 3C). Constructs targeting Trp53, Fancd2, and Wwox, in particular, were overrepresented alongside Brca2 and Rb1 (χ² (19); p < 0.05). We hypothesized, therefore, that many of these co-targeted genes are necessary for efficient Brca2− colony growth. Targeted LentiCRISPR co-infections revealed that a Brca2 mutation significantly reduced transformation rates of cells also targeted for Trp53 and Rb1, which is in agreement with previous reports (Figures 4C and S13; Feng and Jasin, 2017; Zhu et al., 2015). However, we found that the addition of Fancd2 LentiCRISPRs to constructs targeting Trp53, Rb1, and Brca2 rescued the detrimental effects of single mutations in either Brca2 or Fancd2 on transformation rate and significantly increased colony size (Figures 4C, S10C, and S13). Trp53−/Rb1−/Fancd2−/Brca2− cells had 3-fold more colonies than Trp53−/Rb1− OSE-SCs (Figures 4C and S13). Despite co-targeting with Brca2 in 48% of cases, neither Pten nor Wwox LentiCRISPR transductions significantly increased in number or size relative to Trp53−/Rb1− or Trp53−/Rb1−/Brca2− colonies (Figures 4C, S10C, and S13). We did, however, find that Crebbp targeting restored Trp53−/Rb1−/Brca2− colony formation to the level observed for Trp53−/Rb1− colonies and significantly increased colony size (Figures 4C and S10C). These results suggest that multiple concurrent driver mutations are necessary to overcome growth-deterring effects of Brca2 mutagenesis.

Rad51c Synergizes with Fat3 to Promote Transformation

Like Brca2, Rad51c is involved in DNA double-strand break repair (Somyajit et al., 2010). Loss of Rad51c has also been shown to be detrimental to cell growth, so synergistic mutations may be required for efficient adhesion independent growth of Rad51c− OSE-SCs (Kuznetsov et al., 2009). Although Rad51c targeting was detrimental to the transformation of Trp53−/Cdkn2a−/Pten− OSE-SCs (Figure 4B), we observed that Rad51c was frequently co-mutated with Gabra6 (100%), Wwox (89%), and Fat3 (95%) in our screen (Figure 3B). We therefore performed targeted infections of LentiCRISPRs targeting Gabra6, Wwox, and Fat3 alongside Rad51c and core mutations in Trp53, Cdkn2a, and Pten. We observed significantly increased transformation rates following concurrent mutagenesis of Rad51c and Fat3, but not Wwox or Gabra6, in Trp53−/Cdkn2a−/Pten− colonies (Figures 4D and S14). However, only Trp53−/Cdkn2a−/Pten−/Rad51c−/Gabra6− colonies were significantly larger than Trp53−/Cdkn2a−/Pten− or Trp53−/Cdkn2a−/Pten−/Rad51c− colonies (Figure S10D). These results suggest that some HGSOC-associated mutations, like those in Rad51c, can promote transformation only with additional synergistic mutations to overcome synthetic lethality.

LentiCRISPR Integration Patterns in OSE-NS Transformants

OSE-NSs transformed much less efficiently than OSE-SCs, yielding only 11 colonies of the 30,000 cells infected (Figure 5). This small sample size preempted meaningful statistical analyses of target gene overrepresentation. Nevertheless, like OSE-SCs, most (10/11) clones contained LentiCRISPRs targeting Trp53, Rb1, Pten, and Cdkn2a. Other targeted genes included Nf1 (91%), Crebbp (82%), Brca2 (55%), Brca1 (46%), and Wwox (46%).

Figure 5. Identification of Genome-Integrated LentiCRISPRs Overrepresented Target Gene Combinations in OSE-NSs.

(A) Percent gene-targeting frequency in all 11 OSE-NS colonies obtained.

(B) Genome integration and hierarchical clustering of LentiCRISPRv2 constructs in OSE-SC samples. The binary color scale shows whether a gene is targeted by at least one LentiCRISPR in each individual sample. Light gray indicates that a given gene was not targeted, and dark gray indicates that a gene was targeted by at least one LentiCRISPRv2 construct. Hierarchical clustering was performed on both sample similarity and gene targeting, resulting in several clusters of co-targeted genes and similar transformants.

DISCUSSION

Carcinogenesis typically requires multiple genetic events. Tumor sequencing has not only supported this tenet but also informed the constellations of mutations that are commonly present and thus likely contributing to cancer formation and progression. However, the requirements for tumor initiation can be complex in terms of gene combinations and susceptible cells; for HGSOC, precursor lesions have not been conclusively identified, and patient tumor samples have an average of 46 mutations (Cancer Genome Atlas Research Network, 2011). Modeling of cancer driver events suggest that only 5 to 8 driver mutations may be necessary for initiation (Pon and Marra, 2015; Stratton et al., 2009), so most of the mutations present in late-stage tumors are irrelevant to initiating events. Although contributions of some putative driver genes have been assessed in vitro and in vivo (Bobbs et al., 2015; Harlan and Nikitin, 2015; Hasan et al., 2015; Kim et al., 2018; Zhang et al., 2019), such directed approaches cannot assess the myriad mutation combinations of commonly altered genes in HGSOC. Our screen sought to experimentally define the combinations of commonly mutated genes that drive the transformation of ovarian epithelial cells.

The issue of HGSOC initiation is further complicated by uncertainty regarding HGSOC cells of origin. We focused on OSE, but there are others, such as the distal fallopian TE, which can be transformed (Corzo et al., 2017; Karst et al., 2011; Labidi-Galy et al., 2017; Perets et al., 2013; Piek et al., 2001; Sherman-Baust et al., 2014). The presence of serous tubal intraepithelial carcinomas (STICs) in patients with HGSOC, and frequent presence therein of identical TP53 mutations to those in concurrent HGSOC tumors, led to the proposal that HGSOC can initiate in the TE (Kindelberger et al., 2007; Piek et al., 2001). The TE also shares several well-characterized markers of HGSOC that have not been observed in OSE (Perets and Drapkin, 2016). For example, secretory cells in the TE express PAX8, which is present in HGSOC but not in untransformed OSE (Adler et al., 2015; Ozcan et al., 2011; Tacha et al., 2011). Monolayers of PAX8-positive TE cells can also form lesions that express “p53 signatures” that are often associated with mutant TP53 but generally have a low proliferative index and lack cellular atypia (Lee et al., 2007; Leonhardt et al., 2011). Additionally, the distal TE and HGSOC have similarities in their gene expression landscapes but so does OSE-SC in many cases (Hao et al., 2017; Lawrenson et al., 2019). Hence, these data support the suggestion that all three cell types (OSE-SC, OSE-NS, and TE) have the ability to transform into HGSOC (Kim et al., 2018; Neel et al., 2018; Zhang et al., 2019). These data are supported by clinical evidence in which ~1/2 of HGSOCs can be explained by STIC origin (Auersperg et al., 2008; Kindelberger et al., 2007; Piek et al., 2001).

We focused on the OSE because it consists of a single cell type (unlike the heterogeneous TE) that is readily isolated and cultured (Auersperg et al., 2001; Flesken-Nikitin et al., 2003) and transforms in rodents at locations consistent with HGSOC in humans (Auersperg et al., 2008; Flesken-Nikitin et al., 2003; Godwin et al., 1992; Kim et al., 2018; Orsulic et al., 2002; Scully, 1999; Testa et al., 1994). Similarities between OSE-sourced tumors and HGSOC have been demonstrated both genomically and histopathologically (Flesken-Nikitin et al., 2003, 2013; Hao et al., 2017; Matz et al., 2017; Rosen et al., 2009; Zhang et al., 2019). However, recent evidence has suggested that only a small subpopulation of the OSE (i.e., OSE-SC) is particularly cancer prone (Flesken-Nikitin et al., 2013). Such cells have been shown to possess stem-cell-like characteristics, such as the ability to replace OSE lost during ovulation and expression of stem cell markers (Flesken-Nikitin et al., 2013). That study also found that Trp53 and Rb1 knockouts in OSE-SCs result in more tumors at lower latency than in OSE-NSs.

Our results support the theory that the ovarian hilum, which is the putative location of OSE-SC and a transition zone between the OSE, TE, and mesothelium, is particularly prone to transformation. We found that our 20-gene lentiCRISPR library transformed OSE-SCs 41-fold more frequently than OSE-NSs. These data support a long-standing suspicion that stem cells within transition zones are especially prone to carcinogenesis (Ferraro et al., 2010; Greene et al., 1991; McNairn and Guasch, 2011). Seidman (2015) demonstrated that most STICs occur in close proximity to the tubal-peritoneal junction (Seidman, 2015). Additionally, the tubal-peritoneal junction, like the ovarian hilum region, contains LEF1-expressing cells, and patients with higher LEF1 expression have a poorer 5-year survival (Schmoeckel et al., 2017). It is possible, therefore, that HGSOC lesions arise in either OSE-SCs or TE stem cells located in transitional zones.

Based on the most common mutation combinations we observed in transformed OSE-SCs, we developed a model of events for OSE-SC transformation initiation (Figure 6). We propose that functional loss of only 3 of the 20 genes assessed in this study, Trp53, Rb1, and Pten, are necessary for efficient OSE-SC transformation (Figures 3A, 3C, 4A, and S10A). Knockout of these three genes in the OSE has been previously shown to cause development of both low-grade and high-grade serous carcinoma in mouse models (Shi et al., 2020). Mutation of Cdkn2a could partially compensate for Rb1 disruption because Cdkn2a encodes p16(INK4a) and p14(ARF), which are known regulators of Rb1 and Trp53, respectively (Nielsen et al., 1998). Our model also suggests that Trp53 and/or Rb1 disruption can cause low-efficiency transformation without Pten mutations, but colony size was much smaller. The greatest colony size and quantity were observed in Trp53−/Rb1−/Cdkn2a−/Pten− colonies, suggesting additive effects of mutations in each gene.

Figure 6. Model of Mutations Necessary for Efficient In Vitro OSE-SC Transformation.

Random mutagenesis assays and targeted experiments revealed minimal requirements for adhesion-independent growth and mutations that enhance transformation. The blue box contains genes that are minimal requirements for transformation or cause transformation at low efficiency. The addition of mutations shown in the yellow box cause significant degrees of transformation. The addition of further mutations in genes shown in the green box allow for the highest rates of transformation. Genes listed in the red box inhibit transformation. However, two exceptions exist. Brca2 and Fancd2 (marked with a single asterisk) co-mutagenesis alongside Trp53 and Rb1 result in efficient OSE-SC transformation. Similarly, Rad51c (marked with two asterisks) and Fat3 plus Trp53, Cdkn2a, and Pten caused efficient transformation.

Our proposition that Trp53 and Rb1 mutations are core minimal OSE transformation requirements is consistent with observations that single knockouts of Trp53, Rb1, or Brca1/2 did not efficiently drive carcinogenesis or pathologic changes, but combined knockouts of Trp53 and Rb1, or these two plus either Brca1 or Brca2, caused the formation of tumors histopathologically similar to HGSOC (Szabova et al., 2012; Flesken-Nikitin et al., 2013). These and other models support findings that TP53 mutations are nearly ubiquitous in ovarian carcinoma and that the RB1 pathway is dysregulated in 67% of HGSOC tumors (Cancer Genome Atlas Research Network, 2011; Chien et al., 2015; Flesken-Nikitin et al., 2003; Karst et al., 2011).

Our observation that Pten disruption significantly increases transformation frequency and colony size is also consistent with data from genetically modified mouse models. HGSOC-like tumor development in mouse models without Pten mutations have longer latencies than mice that do (Perets et al., 2013; Zhai et al., 2017). Tumors with PTEN mutations are more aggressive and have a worse prognoses (Martins et al., 2014). PTEN and RB1 mutations also significantly co-occur in HGSOC, suggesting that they may act synergistically (Cancer Genome Atlas Research Network, 2011; Cerami et al., 2012; Gao et al., 2013).

In addition to identifying a core set of transformation-enhancing mutations, our data suggest that mutating two other TCGA driver genes can further enhance OSE-SC transformation susceptibility (Figure 6). Given a core set of mutations in Trp53, Cdkn2a, and Pten, additional disruption of Ankrd11 or Wwox significantly promoted adhesion-independent growth. Interestingly, genomic analyses of tumors from a mouse model deficient for Trp53, Brca1, Brca2, and Pten revealed deletions in both Ankrd11 and Wwox, suggesting that they may play a role in tumor initiation or progression (Perets et al., 2013). Both Ankrd11 and Wwox have also been implicated in Trp53-related pathways. ANKRD11 is a putative tumor suppressor that interacts with TP53 and promotes its transcription factor activity and binding to the CDKN1A promoter (Lim et al., 2012; Neilsen et al., 2008; Noll et al., 2012). WWOX greatly influences the response of TP53 to genotoxic stress, and Wwox mRNA inhibition abolishes TRP53-dependent apoptosis (Chang et al., 2001; Schrock and Huebner, 2015). Mutations in Wwox and Ankrd11 in our project may therefore contribute to further dysregulation of Trp53 or may promote transformation in the presence of non-null mutations in Trp53 induced by LentiCRISPR mutagenesis.

Several genes assessed in our study, namely Brca2, Fancd2, Prim2, and Rad51c, appeared to deter OSE-SC adhesion-independent growth when singly mutated alongside core disruption of Trp53, Rb1, and/or Pten. However, combined disruptions of subsets of these genes actually functioned synergistically to enhance OSE-SC transformation. Namely, co-mutation of Brca2 and Fancd2 or Rad51c and Fat3, in addition to the core mutations, increased in colony growth. The growth-deterring effects of Brca2, Fancd2, or Rad51c single mutations we observed have also been documented by others and are likely related to accumulated DNA damage and consequent checkpoint activation (Feng and Jasin, 2017; Hinz et al., 2003; Kondrashova et al., 2017; Kuznetsov et al., 2009; Thompson et al., 2017; Tian et al., 2017). We speculate that concurrent mutagenesis of these potentially growth-deterring genes may be necessary to rescue proliferation.

A fundamental aspect of our project was not only to identify and validate combinations of transformation-associated genes but also to assess whether commonly mutated HGSOC genes are simply passengers or unnecessary for transformation initiation. For many genes in our study, there was no evidence for their involvement in OSE transformation. They include Apc, Crebbp, Fancm, Nf1, Csmd3, Map2k4, Brca1, and Prim2. Although mutations in these genes are associated with developed tumors, sequencing data alone aew unable to distinguish whether mutations occurred during cancer initiation or were later events. Our study only assessed transformation initiation, so it is possible that these genes facilitate later stages of carcinogenesis. Alternatively, many genes that are not involved in transformation initiation may be passenger mutations or a consequence of the genomic instability associated with HGSOC tumor cells.

Our study has elucidated core mutations in putative tumor suppressors necessary for efficient transformation of OSE-SCs, shedding light on transformation mechanisms in the process. Importantly, however, the screening method we used did not address the potential roles of oncogene activation in HGSOC initiation. It may be informative to perform transformation screens that simultaneously combine tumor suppressor knockouts (as we have done here) with activation of candidate oncogenes by using methods such as CRISPRa (Konermann et al., 2015). Additionally, we emphasized loss-of-function Trp53 alleles as a baseline mutational state, but gain-of-function alleles are known to be involved in tumorigenesis (Kim and Lozano, 2018), and specific alleles could be incorporated into the starting OSE. Future screens may also consider rarer mutations that may synergize with the top 20 HGSOC-associated putative tumor suppressors addressed here. It will also be important to model the various mutation combinations in vivo to assess transformation in the context microenvironmental factors important for HGSOC development. Others have modeled the omental metastases common in HGSOC by intraperitoneal (i.p.) injection of cells bearing multiple driver mutations and have noted similar microenvironmental factors between mouse HGSOC models and human disease (Maniati et al., 2020). The results obtained here will form the basis for future experiments that address such issues.

To date, TCGA has generated comprehensive genomic and transcriptomic data for 33 cancer types by using more than 20,000 primary tumor samples (Gao et al., 2019). These comprehensive datasets report genes that are significantly mutated or commonly deleted in different diseases, but the roles of many of these genes in cancer initiation or downstream biology are unclear. Combinatorial screening of these genes, such as that we have applied here, in the proper cellular paradigms, may help disentangle steps of neoplastic transformation in multiple types of cancer.

STAR★METHODS

RESOURCE AVAILABILITY

Lead Contact

Requests for resources should be directed to the Lead Contact, John Schimenti (JCS92@Cornell.edu)

Materials Availability

All LentiCRISPRv2 plasmids, transformed cell lines, and mice generated in this study are available upon request. Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, John Schimenti (JCS92@Cornell.edu)

Data and Code Availability

All reasonable requests for raw data, including flow cytometry data, viral titer assays, transformant counts, next generation sequencing data, and mouse genotyping data, are available upon request to the lead contact. No new code was produced for this manuscript. However, software and algorithms utilized for this manuscript are listed in the Key Resources Table.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Rabbit polyclonal anti-TRP53	Cell Signaling	Cat#9282
Rabbit polyclonal anti-β-actin	Abcam	Cat#ab8227
Goat anti rabbit HRP	Cell Signaling	Cat#7074S
Rat monoclonal anti CK8 (TROMA-1)	University of Iowa Developmental Hybridoma Bank	Cat#AB_531826
Alexa Fluor 488 goat polyclonal anti rat antibody	Thermo Fisher	Cat#A-11006
Bacterial and Virus Strains
One Shot Stbl3 competent E. coli	ThermoFisher	Cat#C737303
LentiCRISPRv2	Addgene	Cat#52961
FUGW	Addgene	Cat#14883
Biological Samples
Trp53+/− FVB/NJ Mouse Ovaries	This paper	N/A
Chemicals, Peptides, and Recombinant Proteins
Cas9 mRNA	TriLink	Cat#L-7206-20
Phosphate-Buffered Saline	Thermo Fisher	Cat#10010023
0.25% trypsin EDTA solution	Thermo Fisher	Cat#25200072
DMEM	VWR	Cat#10-017-CM
Fetal Bovine Serum (FBS)	Atlanta Biologicals	Cat#S11050H
Collagenase	Sigma-Aldrich	Cat#10269638001
Dispase	Sigma-Aldrich	Cat#10269638001
DNaseI	Sigma-Aldrich	Cat#11284932001
Bovine Serum Albumin (BSA)	Sigma-Aldrich	Cat#A9418
Gelatin from porcine skin	Sigma Aldrich	Cat#G9136-10MG
Hams F12	Thermo Fisher	Cat#11320033
Hydrocortisone	Sigma-Aldrich	Cat#H4001
Insulin-transferrin-sodium selenite	Sigma-Aldrich	Cat#11074547001
Non essential amino acids (NEAA)	Thermo Fisher	Cat#11140050
Glutamate	Thermo Fisher	Cat#25030081
Sodium pyruvate	Thermo Fisher	Cat#11360070
Penicillin-streptomycin	Thermo Fisher	Cat#15140122
RIPA buffer	Sigma Aldrich	Cat#R0278-50ML
4-15% gradient polyacrylamide gels	BioRad	Cat#4561083EDU
Nitrocellulose membranes	Thermo Fisher	Cat#88018
BSMBI restriction enzyme	New England Biolabs	Cat#R0580S
T4 DNA ligase	New England Biolabs	Cat#M0202S
Ampicillin	Sigma-Aldrich	Cat#A0166
LB broth	Sigma-Aldrich	Cat#L3147
Methanol	Sigma-Aldrich	Cat#322415-1L
Goat serum	Sigma-Aldrich	Cat#NS02L
Coverslip mounting media	Vector Laboratories	Cat#H-1000
TransIT-LT1 transfection reagent	Mirus	Cat#MIR 2305
Amicon Ultra-15 columns	Millipore	Cat#UFC903024
0.45um syringe filters	Thermo Fisher	Cat# 725-2545
Puromycin	Sigma-Aldrich	Cat#8833-10MG
Agarose	VWR	Cat#VWRVN605-500G
Critical Commercial Assays
MEGAshortscript T7 Transcription kit	Thermo Fisher	Cat#AM1334
MinElute Columns	QIAGEN	Cat#28004
BCA assay	Thermo Fisher	Cat#23227
GeneJet Plasmid Miniprep kit	Thermo Fisher	Cat#K0502
GeneJet Plasmid Midiprep kit	Thermo Fisher	Cat#K0481
ALDEFLUOR detection kit	StemCell Technologies	Cat#01700
Agencourt DNAdvance DNA isolation kit	Beckman Coulter	Cat#A48705
QiAquick PCR purification kit	QIAGEN	Cat#28104
Surveyor mutagenesis assay	Integrated DNA Technologies	Cat#706020
Cell Biolabs Inc 96 Well Cell Transformation Assay	Cell Biolabs	Cat#CBA-135
Miseq Reagent Kit v2	illumina	Cat#MS-102-2001
Experimental Models: Cell Lines
HEK293T	ATCC	CRL-3216
HELA	ATCC	CCL-2
OSN2	Alexander Nikitin	Corney et al., 2007
Experimental Models: Organisms/Strains
FVB/NJ Mice	Jackson Laboratories	Stock#001800, MGI:2163709
Oligonucleotides
See Tables S1–S3
Recombinant DNA
60 unique LentiCRISPRv2 constructs	This paper	N/A
LentiCRISPR v2	Addgene	Cat#52961
PsPax2	Addgene	Cat#12260
VSV-G	Addgene	Cat#8454
FUGW	Addgene	Cat#14883
Software and Algorithms
BWA MEM software	arXiv:1303.3997	N/A

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Cells and culture

Mouse Embryonic Fibroblasts (MEFs) were generated using previously described methodology (Todaro and Green, 1963). Briefly, a mouse breeding was set up between two FVB/NJ Trp53^+/− parents. Embryos (both male and female) were isolated from pregnant mice at 13 day post-coitum (dpc) and washed in PBS (Thermo Fisher; Cat#: 10010023). Working with one embryo at a time, each embryo was placed in a clean Petri dish with 0.25% trypsin EDTA solution (Thermo Fisher; Cat#: 25200072). A small biopsy was collected for later genotyping. Sterile scalpel blades were used to mince tissue until it was able to be maneuvered with a pipette. Minced tissue incubated for 15 minutes at 37°C in trypsin solution. DMEM (VWR; Cat#: 10-017-CM) with 10% FBS (Atlanta Biologicals; Cat#: S11050H) was then used to inactivate trypsin. MEFs were plated on 10cm plates in DMEM with 10% FBS and expanded until passage 2.

We cultured OSE, OSE-SC and OSE-NS primary mouse cells following previously described methodology (Flesken-Nikitin et al., 2013). Briefly, mouse ovaries were isolated from Trp53 heterozygous FVB/NJ adult (12 week old) females and placed into phosphate-buffered saline (PBS) without Ca²⁺ or Mg²⁺ on ice. Ovaries were washed three times with PBS under a laminar flow hood, and a sterile scalpel blade was used to separate ovaries from the bursa. OSE was separated from the ovary via treatment with a digestion buffer consisting of collagenase (Sigma-Aldrich; Cat#: 10269638001), dispase (Sigma-Aldrich; Cat#: 10269638001), DNaseI (Sigma-Aldrich; Cat#: 11284932001) and Bovine Serum Albumin (BSA) (Sigma-Aldrich; Cat#: A9418). We finally added the pellet from each ovary to 2mL OSE medium on gelatin-coated (Sigma-Aldrich; Cat#: G9136-10MG) 24 well culture plates (Corning Costar; Cat#: CLS3527-100EA). Epithelial lineage of isolated cells was determined via detection of CK8 expression using immunofluorescent microscopy (Figures S4B–S4D).

Primary OSE cultures and the mouse OSN2 cell line (Flesken-Nikitin et al., 2003) were maintained in culture in media containing DMEM (VWR; Cat#: 10-017-CM), Hams F12 (Thermo Fisher, Cat#: 11320033), 5% FBS (Atlanta Biologicals; Cat#: S11050H), hydrocortisone (Sigma-Aldrich; Cat#: H4001), insulin-transferrin-sodium selenite (Sigma-Aldrich; Cat#: 11074547001), non essential amino acids (NEAA) (Thermo Fisher; Cat#: 11140050), glutamate (Thermo Fisher; Cat#: 25030081), sodium pyruvate (Thermo Fisher; Cat#: 11360070), and penicillin-streptomycin (Thermo Fisher; Cat#: 15140122) on 0.2% gelatin-coated (Sigma-Aldrich; Cat#: G9136-10MG) culture plates (Corning Costar; Cat#: 07-200-83). Cells were passaged up to two times using 0.25% Trypsin-EDTA with Phenol Red (Thermo Fisher; Cat#: 25200072) to remove adherent cells from plates.

HEK293T (Human, Female, ATCC CRL-3216) and HELA (Human, Female, ATCC CCL-2) cells were cultured in DMEM (VWR; Cat#: 10-017-CM) with 10% FBS (Atlanta Biologicals; Cat#: S11050H), NEAA (Thermo Fisher; Cat#: 11140050), and sodium pyruvate (Thermo Fisher; Cat#: 11360070) on gelatin-coated (Sigma-Aldrich; Cat#: G9136-10MG) plates (Corning Costar; Cat#: 07-20083). Cells were passaged onto 10cm plates using 0.25% Trypsin-EDTA (Thermo Fisher; Cat#: 25200072) with phenol Red to remove adherent cells from plates.

Animals

All animal use was conducted under protocol (2004-0038) to J.C.S. and approved by Cornell University’s Institutional Animal Use and Care Committee. CRISPR/Cas9 gene editing was used to generate a Trp53^+/− co-isogenic mouse line in strain FVB/NJ (The Jackson Laboratory Stock#: 001800, MGI:2163709). A cloning-free overlap PCR method was used to generate the DNA template for making sgRNA (Carrington et al., 2015). The following guide RNA sequence corresponding to exon 4 of Trp53 was used: AGTGAAGCCCTCCGAGTGTC (Sanjana et al., 2014; Shalem et al., 2014). The DNA template was reverse transcribed into RNA using the MEGAshortscript T7 Transcription kit (Thermo Fisher; Cat#: AM1334), then purified using MinElute Columns (QIAGEN; Cat#: 28004). The sgRNA (50ng/uL) and Cas9 mRNA (25ng/uL, TriLink; Cat#: L-7206-20) was microinjected into the pronuclei of FVB/NJ zygotes, then transferred injected zygotes into oviducts of pseudopregnant females. A male founder carrying a 1bp insertion in exon 4 and a predicted STOP codon before the DNA binding domain of TRP53 was selected to establish a line (Figures S1A and S1B). Initial phenotyping was done after three generations, all of which were crossed to FVB/NJ animals (Figures S1C and S1D). Backcrossing was completed every three generations to FVB/NJ animals (MGI:2163709). Mice were aged until 12 weeks, and were then sacrificed to isolation of ovaries and ovarian surface epithelial populations.

METHOD DETAILS

Western blotting

Trp53^+/+, Trp53^+/−, and Trp53^−/− MEFs were treated with 10Gy irradiation to activate the p53 pathway. Equal quantities of cells from each group were pelleted, lysed with RIPA buffer (Sigma-Aldrich; Cat#: R0278-50ML), and utilized for immunoblotting experiments. Protein samples were collected from lysed cell pellets. Protein concentrations were normalized via both cell number and BCA assay (Thermo Fisher; Cat#: 23227). Samples were run through 4%–15% gradient polyacrylamide gels (BioRad; Cat#: 4561083EDU) and transferred onto nitrocellulose membranes (Thermo Fisher; Cat#: 88018). We used the Rabbit anti TRP53 primary antibody (Cell Signaling; Cat#: 9282) for detection of TRP53, along with goat anti-rabbit HRP-linked secondary antibody (Cell Signaling; Cat#: 7074S). Rabbit primary antibody was used for detection of ACTB (β-actin) (Abcam: Cat#: ab8227), along with goat anti-rabbit HRP-linked secondary antibody (Cell Signaling; Cat#: 7074S) (Figure S1C).

LentiCRISPRv2 construct design and cloning

LentiCRISPR v2 was obtained from Addgene (plasmid # 52961; http://addgene.org/52961; RRID:Addgene 52961). We designed sgRNA guides targeting the earliest possible exon of 20 TCGA driver genes (Table 1) with the intent of inducing a gene-inactivating nonsense mutation in targets. Three LentiCRISPR constructs per gene were designed and utilized for experiments in all cases. The only exception was Rad51c, which was targeted with two LentiCRISPRv2 constructs (see Minilibrary validation via sequencing section). Guides were designed using parameters described previously (Hsu et al., 2013). With few exceptions, guide sequences were chosen with an optimal off target score (> 75). If no guides with a score of 75 or greater were found, we chose a guide with the highest possible score (Table S3). Synthetic sense and antisense oligonucleotides for each guide were produced as in 25nmol quantities by Integrated DNA Technologies, such that each strand has overhangs necessary for cloning using the BSMBI restriction enzyme (New England Biolabs; Cat#: R0580S).

Target guide sequences were cloned into LentiCRISPRv2 plasmids using previously published methodology (Sanjana et al., 2014; Shalem et al., 2014). Briefly, sense and antisense oligonucleotides corresponding to each sgRNA were annealed to one another in a thermocycler (BioRad MyCycler). LentiCRISPRv2 plasmid was cut using BSMB1 restriction enzyme, and T4 DNA ligase (New England Biolabs; Cat#: M0202S) was used to ligate oligos into LentiCRISPRv2 plasmid. LentiCRISRv2 plasmids were transformed into One Shot Stbl3 chemically competent E. coli (Thermo Fisher; Cat#: C737303). Transformed bacteria were plated on LB agar plates with ampicillin (Sigma-Aldrich; Cat#: A0166) for selection, and single colonies were picked and cultured in LB broth (Sigma-Aldrich; Cat#: L3147). Plasmid was isolated from bacteria using the GeneJet Plasmid Miniprep kit (Thermo Fisher; Cat#: K0502) or the GeneJet Plasmid Midiprep kit (Thermo Fisher; Cat#: K0481).

We validated that each LentiCRISPRv2 construct contained the correct guide via Sanger sequencing (Cornell Biotechnology Resource Center) primed with the following oligonucleotide: 5′-GAGGGCCTATTTCCCATGATT-3′.

Immunofluorescent Microscopy

OSE, OSE-SC, OSE-NS, MEFs from FVB/N mice and OSN2 cells were grown on gelatin-coated glass coverslips for 24 hours and then fixed using methanol (Sigma-Aldrich; Cat#: 322415-1L). Fixed cells were washed with PBS, and then blocked using goat serum (Sigma-Aldrich; Cat#: NS02L). Primary Rat anti CK8 (TROMA1) antibody (University of Iowa Developmental Hybridoma Bank; Cat#: AB_531826) was added to coverslips overnight at 4°C, followed by PBS washes. Alexa Fluor 488 goat anti rat antibody (Thermo Fisher; Cat#: A-11006) was added for one hour using the manufacturer-recommended concentration. One drop of mounting media (Vector Laboratories; Cat#: H-1000) was used to adhere coverslips to slides, and cells were imaged using an Olympus BX51 microscope and Olympus XM10 camera at 10x magnification. GFP expression conveyed by FUGW viral transductions was detected using a BioRad ZOE Fluorescent Cell Imager at 20x.

OSE-SC and OSE-NS isolation

We used the ALDEFLUOR detection kit (StemCell Technologies; Cat#: 01700) to detect ALDH enzymatic activity in primary OSE cells. ALDEFLUOR reagent contains bodipy-aminoacetaldehyde (BAAA), which is a substrate for ALDH and can be acted upon by the enzyme. The molecule is water soluble and can pass freely into cells. ALDH converts BAAA into bodipy-aminoacetate, which is negatively charged and consequently becomes trapped within cells with ALDH activity. Intracellular BAA accumulation leads to fluorescence. In a mixed population of cells, those with the highest levels of ALDH enzyme will convert larger quantities of BAAA to BAA, making them more fluorescent than cells with lower ALDH activity. 4-diethylamino benzaldehyde (DEAB) is an inhibitor of the ALDH enzyme and can be added to a portion of ALDEFLUOR-treated cells to act as a negative control. Inhibition of ALDH prevents high levels of BAAA conversion to BAA, resulting in lower levels of fluorescence.

Cells (4x10⁶) were placed in ALDEFLUOR buffer and active reagent according to the manufacturer’s protocol. A subpopulation of ALDEFLUOR-treated cells was also treated with DEAB as a negative control. Fluorescence activated cell sorting (FACS) of ALDEFLUOR-treated cells was performed on an Aria II sorter using FACS DiVa software (BD Biosciences). The brightest 2%–5% of ALDE-FLUOR-treated cells were identified and gated electronically based on their characteristic light-scatter properties on the fluorescein isothiocyanate (FITC)-channel emission pattern after excitation with a 13–20 mW, 488-nm ellipse-shaped laser (elliptical) BD FACSAria II. The ALDH fluorescence emissions were captured simultaneously through a 515/20-nm band-pass and 505-nm long-pass filter. ALDH⁺ (OSE-SC) and ALDH⁻ (OSE-NS) OSE cells were collected in 5ml falcon tubes, cultured, and were subjected to lentiviral transduction and colony formation assays (Figure S4A).

Viral Packaging and transduction

The following vectors were used for lentivirus packaging: PsPax2 (Addgene; Cat#: 12260); VSV-G (Addgene; Cat#: 8454); LentiCRISPRv2 (Addgene; Cat#: 52961). HEK293T cells were transfected with 10ug LentiCRISPRv2, 7.5ug PsPax2 and 2.5ug VSV-G using TransIT-LT1 transfection reagent (Mirus; Cat#: MIR 2305) via manufacturer instructions. FUGW (Addgene; Cat#: 14883), a GFP-expression lentiviral construct, was also packaged separately to perform control viral transduction experiments (Figures S5, S6, and S7).

Following transfection of HEK293T cells, cell media was collected after 48 hours and 72 hours. Media was concentrated via centrifugation in Amicon Ultra-15 columns (Millipore; Cat#: UFC903024) such that final volume was 500uL. Concentrated virus was filtered using 0.45um syringe filters (Thermo Fisher; Cat#: 725-2545) and immediately added to cultured OSE in 24-well plates. 30,000 cells were transduced with lentivirus for 48 hours for all cell types and experiments. All transductions completed here include three LentiCRISPR constructs targeting each gene of interest. After transduction, viral media was replaced with OSE media.

Viral Titer, Multiplicity of Infection (MOI), and Gene Targeting Calculations

LentiCRISPRv2 constructs have a puromycin resistance gene, which allows any cell transduced with a LentiCRISPRv2 virus to survive puromycin (Sigma-Aldrich; Cat#: 8833-10MG) treatment. Puromycin survival among a population of transduced cells is therefore a function of higher viral MOI. Higher survival due to high MOI is also negatively correlated with single infection percentage (SIP), since a higher number of viral particles in solution increases the probability that a given cell will receive more than one transduction. An MOI of 7 was used in our experiments, because this was the highest level before toxicity to OSE was observed. The mathematical relationship between puromycin survival and MOI/SIP is calculated as a Poisson Distribution as described (Sanjana et al., 2014) and is summarized below:

$$\begin{gathered} \text{Where} \\ \mathrm{m}=\mathrm{MOI};\mathrm{n}=\text{Number of viral particles};\mathrm{P}(\text{survival})=\text{puromycin survival}, \end{gathered}$$

If no minilibrary genes influence cell growth, then any gene could be expected to be targeted at a random rate. The random rate of gene targeting is equal to the number of ways a cell could receive at least one of three LentiCRISPRs targeting a specific gene from 60 total, divided by the total number of possibilities. Because there are 7 functional viral particles per cell, the total number of ways to get one of 3 LentiCRISPRs is 60⁷. The number of ways that a different gene can be targeted is 57⁷. The total number of possibilities is 60⁷. Therefore, the rate of random gene targeting given a library with MOI of 7 is 30.2%. The chance of a cell receiving the single control GFP-virus is (60⁷-59⁷)/60⁷ = 0.111.

Minilibrary validation via sequencing

OSN2 cells (Trp53^+/−) were transduced with all 60 minilibrary constructs at a MOI of 7 for 48 hours, and all transduced cells were collected following brief culture. Cells were spun down, then genomic DNA was isolated from cells using the Agencourt DNAdvance DNA isolation kit (Beckman Coulter; Cat#: A48705). All LentiCRISPR target sites (except for those targeting Trp53) were amplified using PCR, and amplicons were barcoded (Table S1). Successful amplification of all LentiCRISPRv2 target sites was verified via agarose gel electrophoresis to assess amplicon size. We also performed Sanger sequencing on all PCR products to confirm that all intended regions were amplified. Following verification, all reactions were pooled into a single tube and purified using the QiAquick PCR purification kit (QIAGEN; Cat#: 28104) (Figure S2A). LentiCRISPRs targeting Trp53 were excluded from initial verification experiments because OSN2 cells lack Trp53 alleles but were later assessed using a Surveyor mutagenesis assay (Integrated DNA Technologies; Cat#: 706020).

300bp paired end sequencing was performed using Illumina MiSeq to detect indels in minilibrary target site amplicons at a read depth of 25 million reads. BWA MEM software was used for genome alignment (arXiv:1303.3997). Insertions or deletions greater than 4 base pairs in all minilibrary target sites were then tallied in transduced and untransduced control cells. Individual minilibrary constructs were considered “functional” if two-fold more indels were present in transduced cells compared to untransduced cells. We found that most constructs were functional (Figure S2B). Non-functional LentiCRISPRs (less than two-fold difference in number of indels) or those targeting Trp53 were redesigned and functionally validated using a Surveyor mutagenesis assay (Figure S3).

Surveyor Mutagenesis Assay

The Surveyor mutagenesis assay was used to validate activity of some vectors (Figure S3). OSE cells were transduced with Trp53-targeting LentiCRISPRs because OSN2 cells lack Trp53 alleles. OSN2 cells were transduced with redesigned LentiCRISPRs intended to replace non-functional constructs. Briefly, LentiCRISPR target sites were PCR-amplified in transduced (edited) and untransduced (control) OSE cells (Table S2). Amplicons from control and edited cells were mixed in equal concentrations, heated to cause separation of complementary strands, then cooled to cause re-annealing of complementary strands. If a mutation is present in the transduced cell amplicons, then heteroduplexes containing several unmatched basepairs will form as amplicons from LentiCRISPR-transduced cells try to anneal with amplicons from untransduced cells. As a control, amplicons from untransduced cells were mixed with amplicons from other untransduced cells. Heating and re-annealing of amplicons from untransduced cells are not expected to cause mismatches in re-annealed DNA, since amplicons do not contain LentiCRISPR-induced mutations and should all be identical. Surveyor nuclease recognizes mismatches in annealed amplicons and cleaves DNA at that site. Therefore, if a LentiCRISPR-induced mutation(s) is present in amplicons from transduced cells, and DNA from those amplicons are annealed to non-mutated amplicons from the same genomic region, a mismatch would occur and the site would be cut by Surveyor nuclease. Surveyor nuclease was added to both experimental and control groups, and all samples were run through a 2% agarose gel (VWR; Cat#: VWRVN605-500G). Bands unique to transduced samples indicate that mutagenesis of transduced cell target sites has occurred. Numbers displayed below gene names refers to minilibrary construct ID number. We observed unique bands in all transduced DNA samples, suggesting that all replacement LentiCRISPRs and Trp53-targeting LentiCRISPRs possess editing ability.

Efforts to functionally assess all minilibrary constructs resulted in a validated minilibrary of 60 constructs, including one construct targeting Gfp as a negative control (Figures S2 and S3). Rad51c was only targeted by two LentiCRISPRs in the finalized library due to failed validation of a third construct.

Soft Agar Assay, Colony Isolation and Imaging

Adhesion independent growth was assessed using Cell Biolabs Inc 96 Well Cell Transformation Assay (Cell Biolabs; Cat#: CBA-135). Transduced or untransduced cells were plated at a density of 3,000 cells per well in a 48 well dish and were suspended in 150ul agar/media solution. This cell quantity, well size, and volume of agar/media were constant for both minilibrary screening and targeted mutagenesis assays. Six replicated were plated following single viral transduction for targeted mutagenesis assays. Transformation was monitored for one week, and colonies were collected following the manufacturer’s instructions. In short, Cell Biolabs Inc matrix solubilization solution was added to each well of cells (Cell Biolabs; Cat#: CBA-135). Colonies were resuspended in OSE media and plated at very low density on 15cm plates. Individual, distinct colonies growing on 15cm plates were picked and cultured independently. Sterile filter paper was soaked in trypsin and was used to pick individual colonies from plates. Picked colonies were added to 24-well plates and were passaged using OSE culturing methodology. Brightfield images of culture wells were taken using a Nikon SMZ1500 microscope and Nikon Digital Sight DS-Fi1 camera system at 2x magnification. 10x and 20x images were taken using a Nikon TMS-F microscope and Moticam 2300 3.0MP Live Resolution camera system.

LentiCRISPRv2 identification via Sequencing

Individual transformants from soft agar assays were isolated and expanded in culture as adherent cell lines. 500,000 cells per transformant were grown, spun down, and lysed for DNA extraction using the Agencourt DNAdvance DNA isolation kit. We designed a single pair of primers flanking unique LentiCRISPR guide sequences to amplify all genome-integrated constructs in each sample. We used the following primers: CTTGGCTTTATATATCTTGTGGAAAGG and CGACTCGGTGCCACTTT. Illumina overhangs were also added to each primer. PCR reactions were performed on genomic DNA isolates from each individual colony, resulting in amplification of any genome-integrated LentiCRISPRv2 construct. Amplicons corresponding to individual transformants were uniquely barcoded and library prep was completed using the Miseq Reagent Kit v2 (illumina; Cat#: MS-102-2001) according to the manufacturer’s instructions. 2 × 251bp paired end sequencing was performed on pooled, uniquely barcoded amplicons.

We created a custom “genome” of guide sequences and used BWA MEM to align reads from uniquely barcoded transformants to guide sequences. Each alignment to a particular guide represents a “hit,” meaning that a particular genome-integrated guide was PCR-amplified in transformant DNA isolates, and was detected via sequencing. We calculated the average number of reads per construct and determined the average background read count. We designated any individual construct as genome-integrated if it had more aligned reads than twice the background read count. We finally performed hierarchical clustering of integration data to best visualize patterns of LentiCRISPR integration.

Mouse genotyping

To genotype mice, we followed previously described methodology to generate crude tissue lysates for PCR (Truett et al., 2000). The following forward and reverse primers were used for PCR and genotyping: TTGTTTTCCAGACTTCCTCCA and GCATTGAAAGGTCA CACGAA. PCR success was confirmed on an agarose gel, and the forward primer was re-used for Sanger sequencing.

QUANTIFICATION AND STATISTICAL ANALYSES

Percent transformation frequency for OSE-SC, OSE-NS, and OSE were plotted as means with SEM error bars. Means of five platings following a single viral transduction is shown (Figure 2A). Significance of minilibrary gene co-targeting was determined using Chi² (df = 19) analyses (Figure 3). In targeted mutagenesis assays (Figures 4 and S11), mean transformation frequencies for individual experimental groups were compared to the mean transformation frequency of a control group (baseline) using an ANOVA post hoc test.

Highlights.

• Mutations of TP53, PTEN, and RB1 are a core set for ovarian cell transformation

• Many genes commonly mutated in HGSOC do not aid transformation in vitro

• Ovarian surface epithelium stem cells transform more efficiently than non-stem cells

• Unusual combinations of gene mutations aid or suppress transformation