Murine Oviductal High-Grade Serous Carcinomas Mirror the Genomic

Author Manuscript Published OnlineFirst on December 5, 2019; DOI: 10.1158/0008-5472.CAN-19-2558 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited. 1

Murine Oviductal High-grade Serous Carcinomas Mirror the Genomic

Alterations, Gene Expression Profiles and Immune Microenvironment

of Their Human Counterparts

Kevin McCool*1, Zachary T. Freeman*2,8, Yali Zhai*3, Rong Wu3, Kevin Hu4, Chia-Jen

Liu3, Scott A. Tomlins3,8, Eric R. Fearon3,5,6,8, Brian Magnuson7,8,

Rork Kuick7,8, and Kathleen R. Cho3,6,8

*Equal contribution

Departments of Obstetrics and Gynecology1, Unit for Laboratory Animal Medicine2, Pathology3,

Computational Medicine and Bioinformatics4, Human Genetics5, and Internal Medicine6,

University of Michigan Medical School, and Department of Biostatistics7, University of Michigan

School of Public Health, and the Rogel Cancer Center8, University of Michigan, United States

Running Title: Integrated Molecular Analysis of Murine Serous Carcinomas

Corresponding Author: Kathleen R. Cho, 1504 BSRB, 109 Zina Pitcher Place, University of Michigan, Ann Arbor, MI 48109-2200; email: [email protected]; phone: 734-615-0323.

Conflict of Interest Disclosure: S.A. Tomlins is a prior consultant and current employee at Strata Oncology and reports receiving other commercial research support from Compendia Bioscience/Life Technologies/Thermo Fisher Scientific. The other authors have no potential conflicts of interest to disclose.

Downloaded from cancerres.aacrjournals.org on September 30, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on December 5, 2019; DOI: 10.1158/0008-5472.CAN-19-2558 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited. 2

Abstract

Robust pre-clinical models of ovarian high-grade serous carcinoma (HGSC) are needed to advance our understanding of HGSC pathogenesis and to test novel strategies aimed at improving clinical outcomes for women with the disease. Genetically engineered mouse models of HGSC recapitulating the likely cell of origin (fallopian tube), underlying genetic defects, histology, and biologic behavior of human HGSCs have been developed. However, the degree to which the mouse tumors acquire the somatic genomic changes, gene expression profiles, and immune microenvironment that characterize human HGSCs remains unclear. We used integrated molecular characterization of oviductal HGSCs arising in the context of Brca1, Trp53,

Rb1, and Nf1 (BPRN) inactivation to determine whether the mouse tumors recapitulate human

HGSCs across multiple domains of molecular features. Targeted DNA sequencing showed the mouse BPRN tumors, but not endometrioid carcinoma-like tumors based on different genetic defects (e.g., Apc and Pten), acquire somatic mutations and widespread copy number alterations similar to those observed in human HGSCs. RNA sequencing showed the mouse

HGSCs most closely resemble the so-called immunoreactive and mesenchymal subsets of human HGSCs. A combined immuno-genomic analysis demonstrated the immune microenvironment of BPRN tumors models key aspects of tumor-immune dynamics in the immunoreactive and mesenchymal subtypes of human HGSC, with enrichment of immunosuppressive cell subsets such as myeloid derived suppressor cells and regulatory T cells. The findings further validate the BPRN model as a robust pre-clinical experimental platform to address current barriers to improved prevention, diagnosis, and treatment of this often lethal cancer.

Statement of Significance: The acquired gene mutations, broad genomic alterations, and gene expression and immune cell-tumor axis changes in a mouse model of oviductal serous carcinoma closely mirror those of human tubo-ovarian high-grade serous carcinoma.

Introduction

Ovarian high-grade serous carcinoma (HGSC) is the most lethal gynecologic malignancy, and fifth most common cause of cancer death of women in the United States (1). Most women with

HGSC present with advanced stage disease, and despite maximal surgical effort combined with chemotherapy, over 75% will experience recurrence and die of their disease (2). These challenges highlight the need for high-fidelity pre-clinical models to advance our understanding of HGSC pathogenesis and to test novel strategies aimed at improving prevention, early diagnosis, and treatment of HGSC. Initial genetically engineered mouse models (GEMMs) of

HGSC were based on the assumption that the tumors arise from the ovarian surface epithelium

(3,4). However, an expanding body of literature strongly supports the current view that many, if not most HGSCs arise from epithelial cells in the fallopian tube, preferentially in the tubal fimbriae (5-7). As a consequence, more recent HGSC GEMMs have been developed based on transformation of fallopian tube (oviductal) epithelial cells in the mouse (8-10). While tumors arising in these models recapitulate many histopathologic features of human HGSCs, the extent to which the HGSC GEMMs manifest the molecular features of their human tumor counterparts beyond the specific driver genetic lesions instigating tumorigenesis in the models, is unclear.

Data from The Cancer Genome Atlas (TCGA) revealed human HGSCs harbor nearly universal mutations in the TP53 tumor suppressor gene, but show recurrent somatic mutations in only a limited collection of additional key tumor suppressor genes (TSGs) such as BRCA1, BRCA2,

RB1, NF1, and CDK12 (11). We have previously established a GEMM of HGSC that uses the

Ovgp1 promoter to drive expression of a tamoxifen-inducible Cre recombinase (iCre-ERT2) to inactivate various combinations of the Brca1, Trp53, Rb1, and Nf1 TSGs selectively in epithelial cells of the murine oviduct (9). Tumors arising in this so-called BPRN model recapitulate the histopathology of human HGSC and other key features of the human disease, including progression from intraepithelial precursor lesions that closely mimic human serous tubal

intraepithelial carcinomas (STICs). While tumor development in BPRN mice is highly penetrant, tumors typically develop after a lengthy latency period of several months. The latency period likely allows each individual tumor to select for the additional genetic alterations that are needed for tumor progression and development, beyond the instigating lesions.

Integrated genomic and transcriptomic analysis of TCGA datasets has identified other key molecular features of human ovarian HGSC beyond ubiquitous TP53 mutations. Notably,

HGSCs are also characterized by structural genomic instability and a diverse range of resultant widespread copy number alterations, amplifications, deletions, and gene breakage events that may play key roles in the phenotypes of advanced cancers (11-13). Further, based on their gene expression profiles, human HGSCs can be classified into four molecular subtypes: differentiated, immunoreactive, mesenchymal, and proliferative (11). The mesenchymal subtype in particular has been associated with worse clinical outcomes, including lower rates of optimal surgical cytoreduction, higher rates of chemotherapy-resistant disease, and shorter progression- free and overall survival (14,15).

In human HGSCs, a high degree of tumor-infiltrating lymphocytes (TILs) has been associated with improved prognosis (16,17), but there are immunosuppressive factors in the tumor microenvironment that can impair TIL infiltration and activity (18). A recent study of human

HGSC samples identified significant tumor-immune microenvironment heterogeneity in tumors before adjuvant chemotherapy. Post-treatment samples showed augmentation of pre-existing

TIL responses, but no treatment-associated relief of major immune-suppressive mechanisms

(19). These findings, together with the rather disappointing clinical response of HGSC patients to currently available immune checkpoint therapies, highlights the complexity of tumor-immune dynamics in HGSC and the need for better in vivo models to clarify the key factors and mechanisms regulating the immune microenvironment in HGSC (20). There is optimism that

GEMMs can provide tractable and immunologically relevant model systems with which to better understand the immune response to human HGSC before and after therapy (21). Most extant pre-clinical studies on ovarian cancer-immune system dynamics have relied on the use of ID8 cells, a putative HGSC cell line derived from C57bl/6 mice that permits syngeneic transplantation to characterize tumor interactions with the immune microenvironment. However,

ID8 cells arose from ovarian surface epithelial cells (rather than fallopian tube epithelium) spontaneously transformed in vitro (22), and harbor none of the characteristic pathogenic mutations associated with HGSC (23). Hence, ID8 cells do not accurately model several key aspects of HGSC biology.

Herein we sought to use integrated molecular characterization of tumors arising in an autochthonous GEMM (BPRN), to determine whether the mouse tumors recapitulate human

HGSC biology across multiple domains of molecular features. We used targeted DNA sequencing to assess mutational and copy number status of genes known to be frequently altered in human HGSCs. We also performed whole transcriptome RNA sequencing (RNA-seq) to determine whether the mouse tumors more closely model any of the gene expression subtypes identified by the human TCGA. Finally, we employed a combined immuno-genomic analysis to characterize the immune microenvironment of tumors arising in the BPRN mice. Our overarching goal is to credential the BPRN model as a robust pre-clinical experimental platform to advance insights into the molecular pathogenesis of HGSC and to overcome current barriers to preventing, diagnosing, and treating this often-lethal cancer.

Methods

Genetically modified mice and animal care

Ovgp1-iCreERT2 mice carrying various modified TSG alleles and tumor induction with Tamoxifen have been described previously in detail (9,24). At necropsy, mice were grossly evaluated for

tumor extent, and, in each case, ovaries, oviducts, uteri, lungs, multiple abdominal organs, and omentum were examined microscopically. The histology of each tumor was reviewed by two or more participants with expertise in gynecologic pathology. All procedures involving mice described herein were approved by the University of Michigan's Institutional Animal Care and

Use Committee (PRO00008343).

Histopathology and immunohistochemistry

Formalin fixed and paraffin‐embedded (FFPE) tissue sections were stained with hematoxylin and eosin (H&E) for evaluation by light microscopy. To identify microscopic oviductal lesions, each oviduct and ovary were serially sectioned in their entirety. Alternate sections were either stained with H&E or retained for immunohistochemical (IHC) staining, performed according to standard methods as previously described (9,24). The primary antibodies used were: rabbit anti-

MYC, anti-PTEN, anti-FOXP3 (Cell Signaling Technology, Danvers, MA), rabbit anti‐PAX8

(Proteintech, Chicago, IL, USA), and rabbit anti-CD8 (Abcam, Cambridge, UK).

Targeted DNA Sequencing

Multiplexed barcoded libraries were made from 20ng of DNA from isolated FFPE mouse tissue using AmpliSeq Library Kit 2.0 (Life Technologies, Carlsbad, CA) library kit. Sequencing was performed on a custom targeted panel created through the Life Technologies Ampliseq service on an Ion Torrent Proton sequencer. The panel consisted of 1445 amplicons spanning 32 genes. Sequencing was performed following the manufacturer’s instructions, and each sample had an average of 1,817,149 mapped reads and 1353x coverage. Sequence data was aligned to mm10 and read counts quantitated using Ion Torrent Suite’s TMAP custom alignment and quantitation algorithm (25). Methods for data analysis are provided in the on-line supplement.

RNA Sequencing

High-throughput sequencing of cDNA (RNA-seq) was used to profile gene expression in mouse oviductal HGSC or MMMT (BPRN tumors) and endometrioid carcinomas (AP tumors). Four normal whole ovary samples (each sample representing a pool of 4 normal ovaries from 2 mice) and 4 normal oviducts (each representing a pool of 6 oviducts from 3 mice) from BPRN mice that had not been injected with TAM were also profiled. Total RNA was extracted from manually dissected frozen mouse tissue samples with adjacent H&E-stained sections as dissection guides to ensure that each tumor sample contained at least 90% tumor cells and lacked significant necrosis. RNA was extracted with miRNeasy Mini Kit (Qiagen, Germantown, MD).

Quality was checked with an Agilent 2100 Bioanalyzer. PolyA+, stranded libraries were prepared by the University of Michigan Advanced Genomics Core using Illumina TruSeq reagents and sequenced on the Illumina HiSeq 4000 platform. Samples were divided in two groups, each pooled and divided among 3 lanes. Methods for data analysis are provided in the on-line supplement.

Results

Oviductal tumors arising in BPRN mice recapitulate the clinical features of human HGSC

We previously described the histopathologic features of oviductal tumors arising in (n=90)

Tamoxifen (TAM)-treated mice with various combinations of floxed Brca1, Trp53, Rb1, Nf1, and

Pten alleles (9). These BPRN and BPP (Brca1, Trp53, Pten) mice developed oviductal lesions that closely resemble human serous tubal intraepithelial carcinomas (STICs) that progress to

“early” HGSCs (eHGSCs) that are invasive in the oviduct only, and then to HGSC or carcinosarcoma (also known as malignant mixed Müllerian tumor [MMMT], a variant of HGSC with both carcinomatous and sarcomatous components) (26,27). These latter tumors are typically grossly visible, and often obliterate the bursal cavity and adjacent ovary. A subset of the BPRN and BPP mice acquired widespread metastases and/or ascites. We have now assessed tumor development in well over 200 mice, including those reported in the original

study. This larger group includes 83 TAM-treated mice with various combinations of floxed

Brca1, Trp53, Rb1, and Nf1 alleles followed to humane endpoints, and the characteristics of these mice are shown in Figure 1. Importantly, BPRN mice carrying homozygous floxed (or occasionally Brca1del) alleles of all four TSGs (n=18) demonstrated complete penetrance of tumor development, and all but three had advanced stage disease (HGSC or MMMT with ovarian and/or peritoneal involvement) when euthanized. In contrast, BPR mice with homozygous floxed Brca1, Trp53, and Rb1 alleles (n=13) were more likely than homozygous floxed BPRN mice to have only eHGSC or STIC (8/13 vs. 1/18, P = 0.001, two-tailed Fisher’s exact test) and less likely (2/13 vs. 10/18, P = 0.032) to have peritoneal metastases at humane endpoints. The BPR mice took a median of 20 weeks longer to reach humane endpoints than the BPRN mice (76 vs 56 wk, P = 0.0068, logrank test, Supplementary Figure S1). BPR mice were more likely to also have non-oviductal tumors (e.g., lymphoma) than BPRN mice at the time of euthanasia (10/13 vs. 5/18, P = 0.011). Notably, BPRN mice carrying only one floxed allele of each TSG (BPRN-Het, n=8) also acquired oviductal tumors that frequently metastasized to ovaries and peritoneum, with a median survival of 78 wk. Despite comparable survival to the BPR mice (Supplementary Figure S1), the BPRN-Het mice had fewer non- oviductal tumors (1/8 vs 10/13, P = 0.008). The roughly year or more required for tumor development and progression in homozygous BPRN, BPR, and BPN (homozygous flox Brca1,

Trp53 and Nf1; n=2) and BPRN-Het mice, contrasts with our prior observations that TAM- treated Ovgp1-iCreERT2 mice carrying homozygous flox alleles of Brca1, Trp53, and Pten (BPP mice) develop oviductal STIC lesions or early HGSCs with very short latency, i.e., within 4 wk post-TAM (9). By 32 wk post-TAM, all BPP mice have well-established HGSC or MMMT and most have metastatic disease. Thus, the behavior of the various mouse HGSC models resembles that of human HGSC, supporting the use of our model system to test effects of specific genetic alterations of interest.

Somatic mutations and DNA copy number variation in BPRN HGSCs are similar to human

HGSCs

The long latency and time for disease progression in BPRN, BPR, and BPN mice, and the observation that BPRN-Het mice also develop oviductal tumors, suggest tumor initiation and/or progression requires acquisition of additional somatic alterations. As noted above, HGSC development and progression in BPP mice is much more rapid and occurs within a similar timeframe to what we observed in Ovgp1-iCreERT2;Apcfl/fl;Ptenfl/fl (AP) or Ovgp1- iCreERT2;Apcfl/fl;Ptenfl/fl;Arid1afl/fl (APA) mice. As we have reported previously, TAM-treated AP mice rapidly develop oviductal tumors that closely resemble human ovarian endometrioid carcinomas rather than HGSC (24), as do APA mice. To test if mice with different engineered genetic defects develop tumors with different patterns of acquired DNA alterations, and whether murine HGSCs acquire somatic mutations in homologs of genes recurrently mutated in human

HGSCs, we designed a custom Ampliseq panel targeting the complete coding sequence of the mouse homologs of the 32 most significantly mutated (n=9) and amplified/deleted genes in human HGSCs based on TCGA data (Figure 2) (11). While TP53 is mutated in nearly all

HGSCs, the other 8 recurrently mutated genes are mutated in a relatively small fraction (~6% or less) of tumors as reported by the TCGA. Some of these, such as RB1 and NF1, are often inactivated by other mechanisms including deletions and genome breakage events (11,12). The targeted sequencing approach allowed us to simultaneously assay both copy number alterations in tumor DNA inferred from read count data, as well as interrogate single nucleotide variants to assess for coding mutation. These 32 genes are distributed across 15 of 19 mouse autosomes, and thus provide a snapshot of copy number variation across the mouse genome in regions known to be altered in human HGSC. Barcoded, mxNGS libraries were generated from

DNA isolated from each of 94 microdissected tissue samples, and sequencing was performed on an

Ion Torrent Proton sequencer as described (28). Sequence data are publicly available in the SRA,

Bioproject PRJNA560039. Samples included 14 normal tissues, and tumors isolated from 36

BPRN, 3 BPR, 4 BPN, 10 BPP, 4 AP and 4 APA mice. A few of the BPRN tumors were obtained from mice carrying a Cre-inducible Trp53 mutant allele (LSL-R172H) in addition to the Trp53 flox allele. In some mice, DNA was isolated from tumor present at more than one site (e.g., bilateral oviductal tumors, or oviductal tumor and bulky metastasis). In three mice with carcinosarcomas

DNA was separately isolated from epithelial and sarcomatous portions of the tumor. A subset of samples included matched normal (non-neoplastic) tissue from BPRN or BPR mice, either from tail before TAM-treatment, or liver harvested at necropsy. Consistent with findings in human tumors, we found only a few point mutations in the mouse tumors (all from BPRN or BPN mice, summarized in Supplementary Table 1), suggesting recurrent point mutations in the genes analyzed are not a principal mechanism of HGSC carcinogenesis in the mouse, akin to findings in humans. Variant analysis identified nonsynonymous single nucleotide variants (SNVs) in

Csmd3, Crebbp, Mettl17, and Brca2. One tumor had a small frameshift deletion of Pten, while another was shown to have a subclone with a stop-gain Pten mutation. Sanger sequencing confirmed the presence of these somatic mutations in the mouse tumors and their absence in matched normal tissues, and loss of PTEN expression was observed in both tumors that acquired somatic Pten alterations (Supplementary Figure S2). As expected, all five Trp53

R172H mutations under the control of a Lox-Stop-Lox system were detected using the targeted sequencing panel.

Copy number estimates (log-transformed) for the 32 sequenced genes in the 94 tissue samples are shown in Figure 2, in which the tumor samples are grouped by genotype. Normal tissues are also grouped together. As expected, very low (background) levels of copy number alterations were observed in normal tissues, and normal tissues from three mice carrying one recombined Brca1 allele in the germline showed the expected deletion. In contrast, tumors arising in BPRN, BPR, and BPN mice demonstrated widespread copy number alterations distributed across all 15 autosomes sampled by the targeted panel. Deletion of Cre-targeted sequences was generally

evident by loss of genomic signal observed in the affected portion of each floxed allele. In tumors from mice carrying only one floxed copy of a targeted TSG, loss of wild-type copies of

DNA corresponding to the floxed sequences was usually evident. Examples of CNA analyses of two BPRN tumors and a normal tissue sample are shown in the left panels of Figure 3. High- level amplification of Myc was observed in the tumor shown in the left upper panel, and deep

(homozygous) deletion of Pten was present in the tumor represented in the left center panel.

Immunohistochemical staining of the tumor tissues from which DNA was isolated confirmed overexpression of MYC and loss of PTEN respectively (Figure 3, right panels).

We downloaded segment files from Broad GDAC interfaces that held log2-transformed gene copy number estimates based on SNP arrays for 579 human ovarian HGSCs [TCGA’s OV project, (11)] and 517 uterine endometrial carcinomas, 404 of which were of the endometrioid subtype and 113 of which were serous carcinomas [TCGA’s UCEC project, (29)]. Notably, human uterine endometrioid carcinomas (UECs) and serous carcinomas (USCs) are very similar to their ovarian endometrioid and HGSC counterparts based on their histopathology, clinical behavior, and characteristic genetic alterations (29-31). We called human and mouse samples to have gain at a gene if log2(copy_number/2) estimates were greater than 0.2, and to have loss if they were less than -0.2. We found that HGSCs from BPRN mice had a higher fraction of genes altered (FGA) for the 32 genes than HGSCs from BPP mice (medians of 0.55 and 0.31, respectively), as well as endometrioid-like carcinomas arising in AP/APA mice

(median 0.20), and were nearly as high as human ovarian HGSCs (median 0.63) and higher than human USCs (median 0.50), while human UECs had much lower FGA, as expected

(median 0.06, Figure 4A). Metastases from BPRN tumors had median FGA of 0.56. For many genes such as Mecom, Csmd3, Pik3ca, Myc, Kras, Irf2bp2, and Mcl1, HGSCs from BPRN mice and human HGSCs showed frequent copy gains (>40%), while shared losses were common for homologs of Map2k3, Nf1, Erbb2, and Trp53 (>40%) despite two of these genes being

commonly recombined in tumors from BPRN mice (Figure 4B). We also observed recurrent somatic Pten genomic deletions in BPRN tumors. These deletions, in conjunction with the frameshift and stop-gain mutations identified in two tumors, confirm an important role for somatic Pten inactivation in HGSC pathogenesis in the mouse and highlight the complementary nature of DNA sequencing and copy number profiling of the mouse tumors. We classified genes in both human and BPRN mouse HGSCs as having more gains than losses (or losses than gains) if P-values for the likelihood ratio tests of the gains and losses being equal gave P <

.01, and found 13 genes agreeing on more frequent gains and 5 on losses (Figure 4B-D). There were however, exceptions, such as BRCA2 being commonly lost in humans (53%) but never in mice who instead commonly had gains (72%). Loss of Brca2 in the mouse BPRN tumors would be unexpected, since virtually all of the mouse tumors have Cre-mediated inactivation of Brca1 and inactivating mutations of Brca1 and Brca2 are expected to be mutually exclusive (32).

Another example is TERT, which is commonly gained in human HGSCs (53%), but more often lost in BPRN tumors (50%, Fig 4C-D). This discordance may be at least partially explained by differences in telomere biology in humans and mice. While there is strong selection in human tumors for alterations that upregulate telomerase activity to help maintain telomere length, this selective pressure is not present in mouse tumors, given the lengthy telomeres in somatic cells of the laboratory mouse (33). All the data and analyses are given in our GEO series

GSE135590. We conclude that combined inactivation of TSGs in the BPRN, BPR, or BPN mice initiates a tumor phenotype characterized by high levels of copy number variation. The murine and human tumors share copy number alterations affecting many key genes, broadly distributed throughout the genome, which likely reflects a key biologic process underlying HGSC development in both mice and humans.

Transcriptional signatures of BPRN tumors mirror those of human HGSCs, especially the immunoreactive and mesenchymal subtypes

Whole transcriptome RNA-seq was employed to compare gene expression in four different groups of mouse tissue samples including 4 normal ovary, 4 normal oviduct, 6 oviductal endometrioid carcinomas from AP mice, and 21 oviductal HGSCs or MMMTs from 19 BPRN and 2 BPR mice (hereafter collectively referred to as BPRN tumors). We selected a subset of

18372 more highly expressed genes (of 48526), and computed principal components, leading to two BPRN samples being discarded for quality concerns. We compared groups between the remaining 33 samples using edgeR (release 3.7)(34,35), with likelihood ratio (LR) tests between pairs of groups. The renormalized data showed good separation of the four groups and greater variation within tumor groups, and particularly in BPRN tumors, as expected (Supplementary

Figure S3 A-C). EdgeR LR tests and ANOVA models on log-transformed data both found thousands of gene expression differences for our comparisons of pairs of groups at significance levels of P < 0.001, for which only 18.4 genes per comparison were expected by chance, so that false discovery rates for such comparisons are below 1% (Supplementary Figure S3D). BPRN tumors gave more differences when compared to normal oviducts (Od) than did AP tumors, partly due to their greater sample size, because the distance between centroids for the two comparisons was approximately the same (Supplementary Figure S3C, D). The genes found differentially expressed in BPRN vs AP tumors by ANOVA and edgeR when demanding P <

0.001 and that fold-changes were greater than 1.5-fold (in either direction) had very large overlaps, as expected (Supplementary Figure S3E).

To investigate the similarity of gene expression patterns in the murine tumors to human tumors we obtained processed TCGA array data from Broad GDAC Firehose interfaces

(https://gdac.broadinstitute.org/, “huex 1.0st_v2” platform) for 8 samples of human fallopian tube and 311 HGSCs that also had both mutation and copy number data available. From the HGSC samples, we selected 266 samples that had been previously found to cluster into 4 subgroups based on unified gene expression profiles computed by TCGA: 57 differentiated, 59

immunoreactive, 75 mesenchymal, and 75 proliferative (11,36). We joined log-transformed human data to mouse edgeR results using just 1-to-1 best homologs, leading to 13753 homologous genes. We subtracted the mean of normal samples (fallopian tube and oviduct) for each gene from the value for each tumor for the gene, and computed the (Pearson) correlation of these differences between every human and mouse sample to compare the average correlations between mouse BPRN and AP groups and each of the four human groups. We found average correlations between mouse BPRN tumors (n=19) and human HGSCs to be larger than correlations between mouse AP endometrioid carcinomas (n=6) and human HGSCs

(r = 0.31 vs 0.18, P < 10-300), and BPRN tumors gave larger average correlations than AP tumors for all four subgroups of human tumors (P < 10-300 for all 4 comparisons), so that BPRN tumors resemble human HGSCs more than AP mouse tumors, as expected (Figure 5A, B).

When correlations for each human sample were averaged over mouse samples in the two mouse groups only 5 of the 266 human HGSCs gave larger average correlation to AP endometrioid carcinomas than to BPRN tumors (Figure 5B). Average correlations in the mouse

BPRN group was higher in human immunoreactive and mesenchymal groups (r = 0.34 for both), than for differentiated or proliferative groups (r = 0.28 and 0.27, P < 10-300 for all 4 comparisons,

Figure 5A). Differential gene expression analysis (asking that P <0.001 and fold-changes >1.5 in either direction) showed that mouse BPRN vs Od gene selections agreed with human HGSC vs normal fallopian tube gene selections (n=311) for 648 up and 540 down genes with few disagreements (P = 5 x 10-270, Supplementary Table 2). Examples of 5-fold increases in both data sets were Ccnb2, Cxcl10, Top2a, Ccne1, Bub1, and Mki67. Detailed results for all transcript abundance analyses are available as supplements to our GEO series GSE135590.

We performed an additional analysis restricted to the subset of 1573 genes that are differentially expressed in both mouse and human tumors (p<0.001, FC>1.5; Mouse tumor vs normal oviduct, and human TCGA tumors vs normal fallopian tube). This analysis results in mean

correlations for BPRN to human tumors ranging from 0.666 (proliferative) to 0.713

(mesenchymal), while correlations for the AP tumors are lower, ranging from .518 for mesenchymal, to .548 for immunoreactive (Supplementary Figure S4A-C). Further, we show that if we first average values across groups of samples, correlations increase to 0.78 for BPRN tumors versus both immunoreactive and mesenchymal groups of human tumors

(Supplementary Figure S4D). Average correlations computed between human samples show that differences in correlations between human groups are not much larger than differences between average correlations for mouse vs. human groups, when using differentially expressed genes, or the larger set of genes (Supplementary Figure S4E, F). Based on this additional analysis limited to genes that are differentially expressed between tumors and normal fallopian tube, our observation that the BPRN tumors correlate more strongly with the immunoreactive and mesenchymal subsets of human HGSCs compared to the proliferative and differentiated subsets remains robust.

We tested our selected gene lists for enrichments in the 50 Hallmark gene sets from MSigDB

(http://www.broadinstitute.org/gsea/msigdb/index.jsp) and identified 27 pathways with FDR <.05 that were upregulated in BPRN tumors relative to normal oviducts, whereas 18 pathways were upregulated in human HGSCs relative to normal fallopian tube. Notably, there was significant overlap of pathways upregulated in both human and mouse tumors, and the top 4 pathways upregulated in human HGSCs (E2F_TARGETS, G2M_CHECKPOINT, MTORC1_SIGNALING, and MYC_TARGETS) were among the top 5 pathways upregulated in the mouse tumors (Table

1).

In addition, we obtained array data for 246 human serous and 20 ovarian endometrioid carcinomas from GEO series GSE9891 (37) and compared the log-transformed data between the two groups with two-sample T-tests (Supplementary Figure S5A). Asking which genes with

P < 0.001 and fold-change greater than 1.5-fold also had similar differences for mouse BPRN

HGSCs vs AP endometrioid carcinomas showed that 91 of 202 up genes and 85 of 225 down genes were also selected in the mouse (P = 2 x 10-27, Supplementary Figure S5B). Thus our

RNA data indicate that the mouse BPRN and AP tumors largely recapitulate the transcriptional differences between their human ovarian serous and endometrioid carcinoma counterparts, respectively.

Mouse BPRN tumors manifest an immunosuppressive tumor microenvironment like human HGSCs

There is considerable clinical interest in understanding the role of the host immune system in modulating the development of HGSC and each patient’s response to treatment. Increased

CD3+ tumor-infiltrating lymphocytes (TILs) at the time of diagnosis confers an 8-fold improvement in the 5-year survival rate (38). However, experience to date with immune checkpoint inhibitors as single agents targeting PD-1/PD-L1 or CTLA-4 have been rather disappointing, given poor overall response rates and few long-term responses (2), and there are currently no approved ovarian cancer-specific immunotherapy agents. Collectively these observations suggest that our understanding of tumor-immune dynamics in HGSC is incomplete. Hence, we sought to explore the immune cell infiltration of BPRN tumors, which arise in the context of an intact immune system. Charoentong and colleagues utilized RNA-seq data from TCGA samples to perform a pan-cancer analysis of the relative infiltration of 28 types of immune cells based on the prevalence of transcripts in bulk tumor material for genes that are associated with a given immune cell type (39). To expand this methodology to the murine context, we computed the mean log2-expression of mouse homologs for genes contained in each of these previously defined gene sets representing the 28 immune cell types, a straightforward method used previously (40) and refer to these values as meta-genes for each cell type. Between 11 and 71 genes were used to estimate the meta-gene for each cell type.

For our mouse expression data, these immune cell meta-gene expression values are represented in Figure 6A. The BPRN tumors have higher levels than AP tumors for 19 of the meta-genes, and lower levels for 3 others (monocytes and 2 CD56 NK cell subsets), using P

<0.01), though some BPRN tumors have lower levels of particular meta-genes that were higher on average. In contrast, AP tumors and normal oviducts had lower overall abundance of immune cell signatures and less sample-to sample heterogeneity. In BPRN tumors the largest fold changes compared to AP tumors were for myeloid-derived suppressor cells (MDSCs), activated CD4 T cells, and regulatory T cells (Tregs). Using the human HGSC TCGA data, we found the immunoreactive and mesenchymal groups of tumors had much higher average immune cell meta-gene expression than the differentiated or proliferative groups (Figure 6B).

Within the immunoreactive and mesenchymal human groups, MDSCs and Tregs had the largest fold changes relative to the differentiated and proliferative groups, though there was variability among samples, as was also true for BPRN mouse tumors. Recurrent human HGSC samples also had immune infiltrate values similar to the immunoreactive and mesenchymal groups.

Using just the 28 immune cell meta-genes rather than all the genes, we again centered at the mean of normal oviduct (mouse) and fallopian tube (human) and computed the correlations between mouse and human tumor samples (Supplementary Figure S6A). BPRN mouse tumors again had higher correlations to immunoreactive (I) and mesenchymal (M) human groups (r =

0.63 and 0.65) than to differentiated (D) and proliferative (P) subgroups (r = 0.55 and 0.44), while AP tumors had the reverse, with higher correlations to D and P groups (r = 0.44 and 0.47) than to I and M groups (r = 0.26 and 0.26, all 8 p-values less than 10-90, Supplementary Figure

S6 B,C). When looking at the correlations between meta-genes across samples, we found that the immunosuppressive subsets, Tregs and MDSCs, were highly correlated with each other and with macrophages and NK T cells in both human and mouse (Supplementary Figure S7).

Recent work by Zhang et al. demonstrated that human HGSCs display distinct patterns of tumor lymphocyte infiltration, showing sparse (negative), stromal or combined epithelial-stromal lymphocyte infiltration (41). To complement the RNA-seq data from bulk tumor, we used IHC to characterize lymphocyte infiltration in the 19 HGSCs and MMMTs from BPRN mice profiled by

RNA-seq. We focused on CD8+ T cells as they are abundant in human HGSC samples and thought to be anti-tumor in function, as well as immunosuppressive FoxP3+ Treg cells due to our immunogenomic analysis suggesting an important functional role for Tregs. We observed a range of T cell infiltration in the tumors examined, with some tumors showing few if any CD8+ or

FoxP3+ cells, and others showing either or both cell types in the tumor, in intervening stroma, or in both. We did not attempt to quantify TILs due to regional variability within tumors and lack of suitable antibodies for some key markers. Supplementary Figure S8 shows representative images confirming that BPRN tumors show variable patterns of TIL infiltration as reported in human HGSCs, and general concordance of the presence of CD8 and FoxP3 positive cells with the meta-gene expression values for these immune cell types in individual tumors. These data provide further evidence that the BPRN tumors model key aspects of tumor-immune dynamics observed in human HGSC.

Discussion

Robust and well-credentialed models are needed to establish key factors and mechanisms in the initiation and progression of HGSC. Such models will likely also be critical to efforts to develop more effective early detection and prevention strategies. We report here that mouse tumors developing in the context of engineered deletion of key TSGs - specifically in fallopian tube epithelium - model HGSC with biological and molecular features that closely mirror their human tumor counterparts. We also demonstrated similarity between the mouse and human tumors for multiple molecular features, including acquired DNA copy number and mutations,

RNA expression, and tumor-immune microenvironment phenotype. Our integrated analysis

validates our model as a powerful platform that can be used to enhance our understanding of this lethal disease and aid future efforts to develop more effective treatment, screening, and prevention strategies.

Genomic data indicates human HGSCs are associated with a relatively limited number of shared initiating genetic driver events. Instead, human HGSCs are characterized by acquisition of a diverse collection of structural genomic changes such as copy number alterations, tandem duplications and other complex rearrangements. Our mouse models of HGSC, like their human counterparts, acquire numerous structural genomic alterations as the tumors develop and progress following inactivation of specific combinations of TSGs. Loss of Brca1 and Trp53 is not sufficient for this widespread genomic instability phenotype, as the tumors arising in BPP mice (i.e. with concomitant loss of Brca1, Trp53 and Pten) display relatively low copy number variability. Our findings contrast with those of Perets and co-workers, who reported that engineered deletion of Brca2, Trp53 and Pten in the oviductal epithelium resulted in tumors with frequent copy number changes (8). It is possible that differences between our BPP tumors and those described by Perets et al. may reflect different genetic backgrounds of the two model systems or the different strategies (Ovgp1 vs. Pax8 promoter) used to drive Cre expression in oviductal epithelial cells. Also, we did not test effects of Brca2 in combination with Trp53 and

Pten inactivation in Ovgp1-iCreERT2 mice.

Our comprehensive RNA-seq analysis showed HGSCs from human and mouse share similar patterns of gene expression. Moreover, our observation that BPRN tumors preferentially model certain human HGSC expression subtypes (i.e. immunoreactive and mesenchymal) suggests a link between loss of specific TSGs with the development of specific expression profiles.

Consistent with this, George et al. observed an association with the immunoreactive tumor subtype of HGSC in patients with germline BRCA1 mutations (42). As noted earlier,

mesenchymal subtype tumors have been associated with poorer clinical outcomes including lower overall survival. Our BPRN model has the potential to enhance efforts to understand the development and aggressive clinical behavior of this poor prognostic subgroup. Additional work is needed to determine if other engineered genotype combinations will result in mouse tumors with gene expression profiles that more closely mimic the proliferative and differentiated subtypes.

Our immunogenomic analysis further highlights the utility of our model in studies of the tumor- immune microenvironment. Importantly, human HGSCs show substantial immune cell infiltration, albeit with heterogeneity in the nature and quantity of infiltrating immune cells, and this was also observed in tumors from BPRN mice. HGSCs arising in BPRN mice express markers indicating much greater immune cell infiltration than endometrioid-like carcinomas arising in the context of Apc and Pten inactivation. Because tumors in these mice develop in the context of a fully functional immune system, our BPRN model provides a useful experimental tool to evaluate tumor-immune dynamics during HGSC development. Our finding that a subset of mouse BPRN tumors are enriched with immunosuppressive cell types that are similarly observed in human samples suggests HGSC progression may be driven by the tumor’s innate capacity to recruit these elements and/or exclude anti-tumoral immune cells during development and progression. Such mechanisms, combined with the overall low anti-tumor immune cell infiltration, low tumor mutational burden, and low number of tumor neoantigens typically seen in HGSC, might likely contribute to the poor clinical utility of immune checkpoint blockade in human patients observed thus far. Our model provides a tractable system with which to explore varied strategies that might increase immune cell infiltration and enhance the efficacy of available immunotherapies.

A potential limitation of our models is the development of non-oviductal tumors (e.g., lymphoma and sarcoma) in a substantial subset of mice, particularly those with the BPR genotype. This may reflect low rates of stochastic inactivation of the floxed Trp53 or other TSG alleles in tissues outside of the oviduct, as the mice age. Notably, non-oviductal tumors were observed in a much smaller fraction of BPRN mice, especially those carrying only one floxed allele of each TSG. In addition, our engineering strategy results in the simultaneous inactivation of multiple TSGs, which is unlikely to occur in human tumors. Nevertheless, our results argue strongly that the oviductal tumors in BPRN mice have many similarities to human HGSCs. Another enigmatic issue is the prolonged latency following inactivation of the multiple TSG alleles and oviductal tumor development in BPRN mice compared to BPP, AP and APA mice, as noted previously

(9,24). This finding suggests that Pten and/or Apc inactivation are strong and rapid drivers of oviductal epithelial transformation, compared to Brca1, Trp53, Rb1, and Nf1. However, we would emphasize that prolonged latency likely allows accumulation of the additional cooperating genomic changes that are required for HGSC development. The long latency in the mouse is not incongruent with the fact that human HGSCs usually arise later in life. Indeed, the age distribution for newly diagnosed patients with HGSC is older than for those with non-serous types of ovarian cancer, peaking in the seventh vs. fifth decade of life, respectively (43). More comprehensive approaches such as whole genome or whole exome sequencing will likely help identify the acquired genetic alterations beyond those identified by our targeted sequencing strategy. In summary, we have shown here that inactivation of specific combinations of TSGs in murine fallopian tube epithelium results in tumors that bear striking similarities to human HGSC with respect to somatic mutations, widespread DNA CNAs, RNA expression patterns and tumor- immune microenvironment.

Acknowledgements

The authors would like to acknowledge technical assistance from Stephanie Schulman and

Yisheng Wang. The research reported in this paper was supported by the National Cancer

Institute of the National Institutes of Health under award numbers P30CA046592 (R. Kuick, B.

Magnuson, E. Fearon and K. Cho), R01CA196619 (K. Cho, Y. Zhai, R. Wu, S. Tomlins, R.

Kuick) and R01CA226756 (K. Cho, E. Fearon, Y. Zhai, R. Wu, B. Magnuson, and R. Kuick).

References

1. Siegel RL, Miller KD, Jemal A. Cancer statistics, 2019. CA Cancer J Clin 2019;69:7-34

2. Lheureux S, Braunstein M, Oza AM. Epithelial ovarian cancer: Evolution of management

in the era of precision medicine. CA Cancer J Clin 2019

3. Connolly DC, Bao R, Nikitin AY, Stephens KC, Poole TW, Hua X, et al. Female Mice

Chimeric for Expression of the Simian Virus 40 TAg under Control of the MISIIR

Promoter Develop Epithelial Ovarian Cancer. Cancer Res 2003;63:1389-97

4. Flesken-Nikitin A, Choi KC, Eng JP, Shmidt EN, Nikitin AY. Induction of carcinogenesis

by concurrent inactivation of p53 and Rb1 in the mouse ovarian surface epithelium.

Cancer Res 2003;63:3459-63

5. Kindelberger DW, Lee Y, Miron A, Hirsch MS, Feltmate C, Medeiros F, et al.

Intraepithelial carcinoma of the fimbria and pelvic serous carcinoma: Evidence for a

causal relationship. Am J Surg Pathol 2007;31:161-9

6. Lee Y, Miron A, Drapkin R, Nucci MR, Medeiros F, Saleemuddin A, et al. A candidate

precursor to serous carcinoma that originates in the distal fallopian tube. J Pathol

2007;211:26-35

7. Piek JM, van Diest PJ, Zweemer RP, Jansen JW, Poort-Keesom RJ, Menko FH, et al.

Dysplastic changes in prophylactically removed Fallopian tubes of women predisposed

to developing ovarian cancer. J Pathol 2001;195:451-6

8. Perets R, Wyant GA, Muto KW, Bijron JG, Poole BB, Chin KT, et al. Transformation of

the fallopian tube secretory epithelium leads to high-grade serous ovarian cancer in

brca;tp53;pten models. Cancer Cell 2013;24:751-65

9. Zhai Y, Wu R, Kuick R, Sessine MS, Schulman S, Green M, et al. High-grade serous

carcinomas arise in the mouse oviduct via defects linked to the human disease. J Pathol

2017;243:16-25

10. Kim J, Coffey DM, Creighton CJ, Yu Z, Hawkins SM, Matzuk MM. High-grade serous

ovarian cancer arises from fallopian tube in a mouse model. Proc Natl Acad Sci U S A

2012

11. Cancer Genome Atlas Research Network. Integrated genomic analyses of ovarian

carcinoma. Nature 2011;474:609-15

12. Patch AM, Christie EL, Etemadmoghadam D, Garsed DW, George J, Fereday S, et al.

Whole-genome characterization of chemoresistant ovarian cancer. Nature

2015;521:489-94

13. Macintyre G, Goranova TE, De Silva D, Ennis D, Piskorz AM, Eldridge M, et al. Copy

number signatures and mutational processes in ovarian carcinoma. Nat Genet

2018;50:1262-70

14. Shilpi A, Kandpal M, Ji Y, Seagle BL, Shahabi S, Davuluri RV. Platform-Independent

Classification System to Predict Molecular Subtypes of High-Grade Serous Ovarian

Carcinoma. JCO Clinical Cancer Informatics 2019;3:1-9

15. Wang C, Armasu SM, Kalli KR, Maurer MJ, Heinzen EP, Keeney GL, et al. Pooled

Clustering of High-Grade Serous Ovarian Cancer Gene Expression Leads to Novel

Consensus Subtypes Associated with Survival and Surgical Outcomes. Clin Cancer Res

2017;23:4077-85

16. Pinto MP, Balmaceda C, Bravo ML, Kato S, Villarroel A, Owen GI, et al. Patient

inflammatory status and CD4+/CD8+ intraepithelial tumor lymphocyte infiltration are

predictors of outcomes in high-grade serous ovarian cancer. Gynecol Oncol

2018;151:10-7

17. Darb-Esfahani S, Kolaschinski I, Trillsch F, Mahner S, Concin N, Vergote I, et al.

Morphology and tumour-infiltrating lymphocytes in high-stage, high-grade serous ovarian

carcinoma correlated with long-term survival. Histopathology 2018;73:1002-12

18. Rodriguez GM, Galpin KJC, McCloskey CW, Vanderhyden BC. The Tumor

Microenvironment of Epithelial Ovarian Cancer and Its Influence on Response to

Immunotherapy. Cancers (Basel) 2018; doi: 10.3390/cancers1008024210

19. Lo CS, Sanii S, Kroeger DR, Milne K, Talhouk A, Chiu DS, et al. Neoadjuvant

Chemotherapy of Ovarian Cancer Results in Three Patterns of Tumor-Infiltrating

Lymphocyte Response with Distinct Implications for Immunotherapy. Clin Cancer Res

2017;23:925-34

20. Cojocaru E, Parkinson CA, Brenton JD. Personalising Treatment for High-Grade Serous

Ovarian Carcinoma. Clin Oncol (R Coll Radiol) 2018;30:515-24

21. McCloskey CW, Rodriguez GM, Galpin KJC, Vanderhyden BC. Ovarian Cancer

Immunotherapy: Preclinical Models and Emerging Therapeutics. Cancers (Basel)

2018;10

22. Roby KF, Taylor CC, Sweetwood JP, Cheng Y, Pace JL, Tawfik O, et al. Development of

a syngeneic mouse model for events related to ovarian cancer. Carcinogenesis

2000;21:585-91

23. Walton J, Blagih J, Ennis D, Leung E, Dowson S, Farquharson M, et al. CRISPR/Cas9-

Mediated Trp53 and Brca2 Knockout to Generate Improved Murine Models of Ovarian

High-Grade Serous Carcinoma. Cancer Res 2016;76:6118-29

24. Wu R, Zhai Y, Kuick R, Karnezis AN, Garcia P, Naseem A, et al. Impact of oviductal

versus ovarian epithelial cell of origin on ovarian endometrioid carcinoma phenotype in

the mouse. J Pathol 2016;240:341-51

25. Li H, Durbin R. Fast and accurate long-read alignment with Burrows-Wheeler transform.

Bioinformatics 2010;26:589-95

26. Brustmann H. Ovarian carcinosarcoma associated with bilateral tubal intraepithelial

carcinoma: a case report. Int J Gynecol Pathol 2013;32:384-9

27. Ardighieri L, Mori L, Conzadori S, Bugatti M, Falchetti M, Donzelli CM, et al. Identical

TP53 mutations in pelvic carcinosarcomas and associated serous tubal intraepithelial

carcinomas provide evidence of their clonal relationship. Virchows Arch 2016;469:61-9

28. Lazo de la Vega L, McHugh JB, Cani AK, Kunder K, Walocko FM, Liu CJ, et al.

Comprehensive Molecular Profiling of Olfactory Neuroblastoma Identifies Potentially

Targetable FGFR3 Amplifications. Mol Cancer Res 2017;15:1551-7

29. Cancer Genome Atlas Research Network, Kandoth C, Schultz N, Cherniack AD, Akbani

R, Liu Y, et al. Integrated genomic characterization of endometrial carcinoma. Nature

2013;497:67-73

30. Bell DW, Ellenson LH. Molecular Genetics of Endometrial Carcinoma. Annu Rev Pathol

2019;14:339-67

31. Cho KR, Shih IM. Ovarian Cancer. Annu Rev Pathol Mech Dis 2009;4:287-313

32. Hauke J, Hahnen E, Schneider S, Reuss A, Richters L, Kommoss S, et al. Deleterious

somatic variants in 473 consecutive individuals with ovarian cancer: results of the

observational AGO-TR1 study (NCT02222883). J Med Genet 2019

doi:10.1136/jmedgenet-2018-105930

33. Artandi SE, DePinho RA. Telomeres and telomerase in cancer. Carcinogenesis

2010;31:9-18

34. Robinson MD, Smyth GK. Moderated statistical tests for assessing differences in tag

abundance. Bioinformatics 2007;23:2881-7

35. McCarthy DJ, Chen Y, Smyth GK. Differential expression analysis of multifactor RNA-

Seq experiments with respect to biological variation. Nucleic Acids Res 2012;40:4288-97

36. Verhaak RG, Tamayo P, Yang JY, Hubbard D, Zhang H, Creighton CJ, et al.

Prognostically relevant gene signatures of high-grade serous ovarian carcinoma. J Clin

Invest 2013;123:517-25

37. Tothill RW, Tinker AV, George J, Brown R, Fox SB, Lade S, et al. Novel molecular

subtypes of serous and endometrioid ovarian cancer linked to clinical outcome. Clin

Cancer Res 2008;14:5198-208

38. Zhang L, Conejo-Garcia JR, Katsaros D, Gimotty PA, Massobrio M, Regnani G, et al.

Intratumoral T cells, recurrence, and survival in epithelial ovarian cancer. N Engl J Med

2003;348:203-13

39. Charoentong P, Finotello F, Angelova M, Mayer C, Efremova M, Rieder D, et al. Pan-

cancer Immunogenomic Analyses Reveal Genotype-Immunophenotype Relationships

and Predictors of Response to Checkpoint Blockade. Cell Rep 2017;18:248-62

40. Faruki H, Mayhew GM, Serody JS, Hayes DN, Perou CM, Lai-Goldman M. Lung

Adenocarcinoma and Squamous Cell Carcinoma Gene Expression Subtypes

Demonstrate Significant Differences in Tumor Immune Landscape. J Thorac Oncol

2017;12:943-53

41. Zhang AW, McPherson A, Milne K, Kroeger DR, Hamilton PT, Miranda A, et al.

Interfaces of Malignant and Immunologic Clonal Dynamics in Ovarian Cancer. Cell

2018;173:1755-69 e22

42. George J, Alsop K, Etemadmoghadam D, Hondow H, Mikeska T, Dobrovic A, et al.

Nonequivalent gene expression and copy number alterations in high-grade serous

ovarian cancers with BRCA1 and BRCA2 mutations. Clin Cancer Res 2013;19:3474-84

43. Torre LA, Trabert B, DeSantis CE, Miller KD, Samimi G, Runowicz CD, et al. Ovarian

cancer statistics, 2018. CA Cancer J Clin 2018;68:284-96

Table 1. Gene set enrichment analysis of Hallmark Gene Sets from MSigDB

Top 20 gene sets for up genes in BPRN mouse tumors vs normal oviduct

# of estimated those FDR, based # of genes on genes we actual P- permutations observed/ Title on list selected value (Q-value) expected HALLMARK_E2F_TARGETS 196 152 9.81E-88 0 5.251 HALLMARK_G2M_CHECKPOINT 189 116 1.58E-49 0 4.156 HALLMARK_MYC_TARGETS_V1 186 102 1.62E-37 0 3.713 HALLMARK_TNFA_SIGNALING_VIA_NFKB 190 84 9.93E-23 0 2.993 HALLMARK_MTORC1_SIGNALING 197 80 6.05E-19 0 2.75 HALLMARK_INFLAMMATORY_RESPONSE 175 66 5.18E-14 0 2.554 HALLMARK_HYPOXIA 185 66 1.06E-12 0 2.416 HALLMARK_ALLOGRAFT_REJECTION 167 60 8.07E-12 0 2.433 HALLMARK_MYC_TARGETS_V2 56 29 8.47E-11 0 3.506 HALLMARK_MITOTIC_SPINDLE 194 61 2.61E-09 0 2.129 HALLMARK_EPITHELIAL_MESENCHYMAL_TRANSITION 182 57 1.01E-08 0 2.121 HALLMARK_IL6_JAK_STAT3_SIGNALING 79 32 2.23E-08 0 2.743 HALLMARK_GLYCOLYSIS 190 55 3.46E-07 0 1.96 HALLMARK_APOPTOSIS 155 47 5.61E-07 0 2.053 HALLMARK_COMPLEMENT 170 50 6.90E-07 0 1.991 HALLMARK_KRAS_SIGNALING_UP 195 54 2.01E-06 0 1.875 HALLMARK_INTERFERON_GAMMA_RESPONSE 177 49 5.93E-06 0 1.874 HALLMARK_UNFOLDED_PROTEIN_RESPONSE 110 30 0.00047 0.0017 1.847 HALLMARK_CHOLESTEROL_HOMEOSTASIS 68 21 0.00056 0.0016 2.091 HALLMARK_IL2_STAT5_SIGNALING 191 45 0.00080 0.025 1.595

Top 20 gene sets for up genes in human HGSCs vs normal Fallopian tube (TCGA)

HALLMARK_E2F_TARGETS 198 126 5.18E-73 0 5.97 HALLMARK_G2M_CHECKPOINT 198 108 4.63E-53 0 5.117 HALLMARK_MTORC1_SIGNALING 200 77 2.87E-25 0 3.612 HALLMARK_MYC_TARGETS_V1 197 74 1.45E-23 0 3.524 HALLMARK_MITOTIC_SPINDLE 199 55 1.72E-11 0 2.593 HALLMARK_UNFOLDED_PROTEIN_RESPONSE 113 32 1.53E-07 0 2.657 HALLMARK_MYC_TARGETS_V2 57 19 3.63E-06 0 3.127 HALLMARK_OXIDATIVE_PHOSPHORYLATION 196 42 7.47E-06 0 2.01 HALLMARK_INTERFERON_ALPHA_RESPONSE 94 24 3.56E-05 0.001111 2.395 HALLMARK_DNA_REPAIR 146 32 5.50E-05 0.001 2.056 HALLMARK_GLYCOLYSIS 200 38 0.00029 0.003636 1.783 HALLMARK_PROTEIN_SECRETION 95 21 0.00088 0.003333 2.074 HALLMARK_UV_RESPONSE_UP 158 29 0.00247 0.006923 1.722 HALLMARK_INTERFERON_GAMMA_RESPONSE 196 34 0.00293 0.007143 1.627 HALLMARK_CHOLESTEROL_HOMEOSTASIS 73 16 0.00378 0.01 2.056 HALLMARK_SPERMATOGENESIS 134 23 0.01443 0.03625 1.61 HALLMARK_PI3K_AKT_MTOR_SIGNALING 105 19 0.01461 0.035294 1.698 HALLMARK_FATTY_ACID_METABOLISM 157 26 0.0152 0.033333 1.554 HALLMARK_ANDROGEN_RESPONSE 100 16 0.0637 0.127895 1.501 HALLMARK_ADIPOGENESIS 198 28 0.07334 0.149 1.327

Figure Legends

Figure 1. Summary of oviductal tumor phenotype in TAM-treated Ovgp1-iCreERT2 mice carrying various engineered Brca1, Trp53, Rb1, and/or Nf1 alleles, and followed to humane endpoints. Data from 83 mice with varying genotypes are shown. Weeks post-TAM indicates the time points at which mice reached humane endpoints. L = left and R= right oviduct. Other (non-oviductal) tumors found at necropsy included lung adenomas (LA), lymphomas (Ly), salivary gland tumors (Sg), squamous cell carcinomas on the skin (Sk), thyroid tumors (Th), and uterine sarcoma (UT). Brackets indicate BPRN, BPR, and BPN mice with homozygous floxed alleles for all TSGs, and BPRN-Het mice with only one floxed allele of each TSG. Representative photomicrographs are provided for each type of oviductal lesion identified (MMMT, HGSC, eHGSC, and STIC). Regions in black boxes show higher magnification photomicrographs of corresponding regions in yellow boxes.

Figure 2. Somatic copy number alteration (CNA) profiling of mouse GEMM samples. The heatmap at left side of figure shows the gene-level log2 copy number ratio (vs. a composite of all normal samples) for the 32 genes shown at top, across 94 samples. Data from the floxed portions of TSGs targeted in the respective GEMMs are shown separately. Genes are sorted in order of genomic position and samples are clustered by genotype as indicated in the color key at figure bottom. Normal tissues are also grouped together. Gains and losses are indicated according to the color scale. The histologic type of each sample is also indicated in the color key at figure bottom, and include HGSC (high-grade serous carcinoma), MMMT (malignant mixed Mullerian tumor/carcinosarcoma, Endo CA (oviductal endometrioid-like carcinoma), Met (metastasis), and Normal (non-neoplastic tissue).

Figure 3. Individual somatic CNA profiles from representative mouse GEMM samples. Data from two tumors and one normal tissue sample are shown. For each sample, individual amplicon-level log2 copy number ratio is plotted for each targeted gene (points), with the gene- level estimate indicated by the thick black bar (wild-type and floxed portions of GEMM targeted genes were considered separately). Mouse chromosome (chr) location of targeted genes is shown above the top left plot, with data colored by gene. Confirmation of expected floxed deletions are indicated, along with selected chr level gains and focal amplifications (amp) and

deletions (del). Immunohistochemistry demonstrated elevated expression of MYC in tumor 13085 (upper right panel), and loss of PTEN expression in tumor 14433 (lower right panel).

Figure 4. Copy number estimates for 32 homologous genes in mice and humans. A) Fraction of the genes that had log2(copy_number/2) > 0.2 or < -0.2. Medians, and 25th and 75th percentiles are shown by red bars. BPRN mouse tumors had greater fraction than BPP tumors, P = 1.6x10-6 by two-sample T-test. Human HGSCs are from the TCGA ovarian cancer project and USC and UEC are serous and endometrioid carcinomas from the TCGA uterine cancer project. B) Fraction of samples with gains and losses for each gene. Chromosome numbers for human and mouse are shown. The 3 by 3 table shows genes gained more than lost, or vice versa, in the first two strips of the heatmap, as judged by the likelihood ratio (LR) test asking if gains and losses had equal probability gave P < 0.01, where agreement between BPRN mice and human HGSCs was substantial (P = 0.0035, Mantel-Haenszel Chi-square test of association). C, D) Plots of fraction of samples with gains and losses for each gene in human HGSCs and mouse BPRN tumors. Pearson correlations are shown in the upper left of each panel. Red data points had LR test with P < 0.01 and more gains than losses in mice, while green ones had more losses than gains.

Figure 5: Correlation of RNA expression in human and mouse HGSCs. A) Human HGSCs centered at human fallopian tube gene values show human subgroups with larger average correlations to mouse BPRN than AP (Apc/Pten) tumors centered at the mean mouse oviduct (Od) genes. Correlations for 17 recurrent human tumor samples and 8 normal fallopian tube samples are also shown. BPRN tumors gave larger correlations than Apc/Pten tumors for every human subgroup, and human correlations to BPRN tumors gave larger correlations for the immunoreactive and mesenchymal subgroups than the differentiated and proliferative subgroups, as shown by the summary statistics. T-statistics show how many standard errors the means differ by, since it was difficult to compute the exact P-values in some cases (where we show P<1E-300). B) Averaging correlations over the two groups of mouse tumors shows that just 5 of the 266 human HGSCs gave larger average correlation to Apc/Pten tumors than BPRN tumors.

Figure 6: Estimated relative immune cell subset abundances in tumors in mouse (A) and human (B) HGSCs. We averaged log2-expression of lists of genes for each cell type provided in Charoenteng et al. (32) to obtain 28 cell type meta-gene values. Ratios for mice are anti-

logarithms of such values after subtracting the midpoint of the BPRN and AP tumor means for that cell type, while human data was similarly centered at the median of all 266 primary tumors. P-values for mice are from an ANOVA model with terms for 4 groups.

Figure 2 Alleles

epytoneG ygolotsiH Apc Arid1a Alg8 Acvr2b Brca1 Trp53 Rb1 Nf1 Pten Pax8 Zmynd8 Mecom Pik3ca Mcl1 Nras Mycl1 Pdap1 Brca2 Kras Ccne1 Irf2bp2 Fat3 Smarcb1 Gabra6 Map2k3 Trp53 Nf1 Cdk12 Erbb2 Brca1 Id4 Tert Mettl17 Rb1 Csmd3 Myc Itgb7 Crebbp Pten fl Trp53 Nf1 fl Brca1 fl Rb1 fl Pten fl Paired 3867-mt 14286-lot 3867-lot 2611-mt 2611-rot 13238-rot 2611-lot 14390-rot-S 14154-lot 14396-rot 14396-lot 14390-rot-E 2979-rot 14526-mt 14104-lot 14280 -rot 3694-lot 14526-rot 14174-lot 14280-lot 8234-lot 13179-rot 13179-lot 14397-lot 14397-rot 3807-lot 2163-lot 13576-rot 13268-rot 14150-lot 13604-mt 13604-rot 13735-lot 2796-mt 2796-rot 14474-rot 2942-rot 13311-rot 13085-rot 14085-lot 14109-rot 13375-rot 14806-rot 14399-mt 14399-rot 14059-lot 14433-lot-E 14433-lot-S 14433-mt 14120-rot 2519-rot 2519-lot 2285-lot 2016-rot 3695-rot 14270-lot 2405-rot 2027-lot-E 2027-lot-S 2405-lot 14721-rot 15071-rot 14943-lot 15189-lot 15191-rot 15073-rot 15194-rot 15217-lot 15213-rot 15195-lot 15195-rot 15194-lot 14794-rot 14595-rot 14593-rot 14596-rot 10830-lot 10879-lot 10847-lot 10676-rot 3867n 14154t 2611n 14154n 14104t 14104n 8234n 13604n 2796n 14085t 14399t 14433n 2519n 2405n

Copy # Mouse Histology Genotype Alleles Paired HGSC BPRN Brca1 flox Nf1 flox 4 Unpaired MMMT BPR Brca1 del Pten flox BPN Trp53 flox Apc flox 2 Endo CA. Met. BPP Trp53 LSL-R172H Arid1a flox Rb1 flox wild type 1/2 Normal APA AP

A 1 C Fraction of samples with gains T-test 1 P = 1.6x10-6 r = 0.58, P = .00054 Mecom Author Manuscript Published OnlineFirst on0.8 December 5, 2019; DOI: 10.1158/0008-5472.CAN-19-2558 Pik3ca Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.0.8 Myc

0.6 Mcl1 Csmd3 Zmynd8 0.6 Ccne1 Tert Irf2bp2 Kras Id4 0.4 Mycl Alg8 Pdap1 0.4 Human Human HGSC Itgb7 Nras 0.2 Fat3

Fraction Fraction of genes altered for 32 genes Pten 0.2 0 Brca2 Rb1 USC UEC HGSC Human BPRN Mets BPP APA/AP Normal Human Human 0

2 3 3 1 1 1 7 1 3 5 6 5 8 8 0 0.2 0.4 0.6 0.8 1 11 11 B 20 13 12 19 22 17 17 17 17 17 17 14 13 12 16 10 Human chr#

2 2 3 3 3 3 4 5 5 6 7 7 8 9 9 Mice BPRN 11 11 11 11 11 11 11

10 13 13 14 14 15 15 15 16 19 Mouse chr#

D Fraction of samples with losses MOCEM MOCEM 1 PAX8 ZMYND8 PIK3CA MCL1 NRAS MYCL PDAP1 BRCA2 KRAS CCNE1 ALG8 IRF2BP2 FAT3 ACVR2B SMARCB1 GABRA6 MAP2K3 TP53 NF1 CDK12 ERBB2 BRCA1 ID4 TERT METTL17 RB1 CSMD3 MYC ITGB7 CREBBP PTEN Gain Human HGSC r = 0.45, Loss (N=579) P = .0092 Gain Mouse BPRN Map2k3 Loss (N=50) 0.8 Nf1 Gain Mouse BPP Brca1 Loss (N=11) Cdk12 Erbb2 Trp53 Gain Human USC Loss (N=113) Smarcb1 0.6 Gain Human UEC Rb1 Loss (N=404) Brca2 Gain Mouse APA/AP Loss (N=8) Crebbp Gain Mouse normals Human Human HGSC 0.4 Loss (N=14) Gabra6 Fraction of samples with gains or losses Pten 1.0 0.8 0.6 0.4 0.2 0.0 0.2 0.4 0.6 0.8 1.0 Human TCGA HGSC project 0.2 more more Observed / gains neither losses total Expected Expected Tert Id4 BPRN more gains 13 0 1 14 7.88 5.25 1.65 0.19 Alg8 mouse neither 2 2 6 10 Mycl tumors more losses 3 0 5 8 4.50 3.00 0.67 1.67 0 total 18 2 12 32 0 0.2 0.4 0.6 0.8 1 Mantel-Haenszel Chi-Square P = .0035 Mice BPRN

Downloaded from cancerres.aacrjournals.org on September 30, 2021. © 2019 American Association for Cancer Research. Figure 5 Author Manuscript Published OnlineFirst on December 5, 2019; DOI: 10.1158/0008-5472.CAN-19-2558 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited. A Correlation of log-transformed difference to normal Fallopian tube 0.04 0.08 0.12 0.16 0.20 0.24 0.28 0.32 0.36 0.40 Mouse sample BPRN_13604T BPRN_2611T BPRN_14396T BPRN_14576T BPRN_14085T BPRN_14104T BPRN_14526T BPRN_14174T BPRN_13179T BPRN_3807T BPRN_14397T BPRN_14474T BPRN_13085T BPRN_13576T BPRN_2163T BPRN_14399T BPRN_14433T BPRN_2285T BPRN_2519T ApcPten_10830 ApcPten_10847 ApcPten_10868 ApcPten_10870 ApcPten_10875 ApcPten_10879

Fallo- pian Summary statistics Mean Correlation BPRN vs Apc/Pten

Recur- rence Tumor groups BPRN Apc/Pten P-value T-statistic Differentiated (D) 0.284 0.194 <1E-300 59.8 Immunoreactive (I) 0.336 0.192 <1E-300 96.8 Mesenchymal (M) 0.339 0.158 <1E-300 137.4 Proliferative (P) 0.267 0.190 <1E-300 58.2 BPRN Apc/Pten BPRN Apc/Pten Comparison P-value P-value T-statistic T-statistic D vs. I <1E-300 0.38 49.9 0.9 D vs. M <1E-300 4E-89 56.1 20.3 Proliferative D vs. P 5E-68 0.04 17.6 2.1 I vs. M 0.001 4E-83 3.3 19.6 I vs. P <1E-300 0.23 71.1 1.2 M vs. P <1E-300 3E-83 79.3 19.6

B Average correlation of each human tumor Mesenchymal to mice tumors, 266 TCGA samples. 0.40 Differentiated 0.35 Immunoreactive Mesenchymal 0.30 Proliferative 266 Primary serous tumors in 4 clusters from TCGA 0.25

Immunoreactive 0.20

0.15

Correlation to Apc/Pten Apc/Pten tumors Correlation to 0.10

Differentiated 0.05 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40

Correlation to BPRN tumors

Z Z

Figure 6 Ov Od BPRN AP A

Fold- P-value Change BPRN BPRN Cell Type vs AP vs AP Ov1 Ov2 Ov3 Ov4 Od1 Od2 Od3 Od4 HGSC_13085T HGSC_13179T HGSC_13576T HGSC_13604T HGSC_14104T HGSC_14397T HGSC_14474T HGSC_14576T HGSC_2163T HGSC_2611T MMMT_14085T MMMT_14174T MMMT_14396T MMMT_14399T MMMT_14433T MMMT_14526T MMMT_2285T MMMT_2519T MMMT_3807T ApcPten_10830T ApcPten_10847T ApcPten_10868T ApcPten_10870T ApcPten_10875T ApcPten_10879T Activated CD8 T 0.15 1.28 Central memory CD8 T 3.7E-07 2.24 Effector memory CD8 T 8.3E-04 1.38 Activated CD4 T 4.4E-10 3.03 Central memory CD4 T 7.5E-07 1.52 Effector memory CD4 T 3.0E-09 1.40 T follicular helper 7.1E-07 1.91 Gamma delta T 2.0E-07 1.57 Type 1 T helper 4.2E-05 1.54 Type 17 T helper 0.29 0.91 Type 2 T helper 5.9E-07 1.45 Regulatory T 2.8E-07 2.72 Activated B 0.059 1.37 Immature B 5.1E-03 1.71 Memory B 7.9E-03 1.30 Natural killer 9.0E-08 1.66 CD56bright NK 1.3E-04 0.75 CD56dim NK 6.1E-04 0.70 MDSC 2.8E-09 3.96 NK T 1.7E-10 2.13 Activated dendritic 4.0E-06 1.71 Plasmacytoid dendritic 0.029 1.12 Immature dendritic 0.43 0.94 Macrophage 2.6E-03 1.47 Eosinophil 2.6E-03 1.31 Mast 6.7E-06 2.45 Monocyte 1.8E-07 0.62 Neutrophil 0.15 1.20 Ratio to the midpoint of the BPRN and ApcPten means.

0.5 0.71 1 1.41 2 Ttest FC B 266 primary human serous tumors in 4 clusters from TCGA Recur- Fallo- I+M vs I+M / Differentiated Immunoreactive Mesenchymal Proliferative rence pian D+P D+P Activated CD8 T 1.1E-27 1.31 Central memory CD8 T 7.7E-46 1.25 Effector memory CD8 T 4.0E-32 1.33 Activated CD4 T 4.7E-20 1.30 Central memory CD4 T 6.7E-26 1.14 Effector memory CD4 T 6.4E-17 1.08 T follicular helper 2.2E-28 1.16 Gamma delta T 2.3E-17 1.12 Type 1 T helper 4.9E-36 1.15 Type 17 T helper 1.8E-06 1.06 Type 2 T helper 3.0E-05 1.06 Regulatory T 1.9E-45 1.48 Activated B 3.6E-18 1.23 Immature B 9.1E-19 1.18 Memory B 1.6E-07 1.07 Natural killer 1.1E-28 1.16 CD56bright NK 1.1E-05 1.05 CD56dim NK 7.2E-03 1.04 MDSC 7.2E-40 1.61 NK T 6.8E-31 1.13 Activated dendritic 8.8E-27 1.18 Plasmacytoid dendritic 2.7E-15 1.08 Immature dendritic 1.4E-06 1.06 Macrophage 1.0E-27 1.20 Eosinophil 5.4E-08 1.09 Mast 3.9E-21 1.18 Monocyte 3.5E-07 1.06 Neutrophil 2.9E-03 1.05 Ratio to the median of primary tumors.

0.5 0.71 1.411 2

Murine Oviductal High-Grade Serous Carcinomas Mirror the Genomic Alterations, Gene Expression Profiles, and Immune Microenvironment of Their Human Counterparts

Kevin McCool, Zachary T Freeman, Yali Zhai, et al.

Cancer Res Published OnlineFirst December 5, 2019.

Updated version Access the most recent version of this article at: doi:10.1158/0008-5472.CAN-19-2558

Supplementary Access the most recent supplemental material at: Material http://cancerres.aacrjournals.org/content/suppl/2019/12/05/0008-5472.CAN-19-2558.DC1

Author Author manuscripts have been peer reviewed and accepted for publication but have not yet been Manuscript edited.

E-mail alerts Sign up to receive free email-alerts related to this article or journal.

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Subscriptions Department at [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://cancerres.aacrjournals.org/content/early/2019/12/05/0008-5472.CAN-19-2558. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.