Abstract
Robust preclinical 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 preclinical experimental platform to address current barriers to improved prevention, diagnosis, and treatment of this often lethal cancer.
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-related 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 preclinical 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 (GEMM) 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 (TSG) 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 (STIC). 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). Furthermore, 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 (TIL) 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. Posttreatment samples showed augmentation of preexisting TIL responses, but no treatment-associated relief of major immune-suppressive mechanisms (19). These findings, together with the rather disappointing clinical response of patients with HGSC 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 preclinical 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 preclinical 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.
Materials and 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 IHC
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 described previously (9, 24). The following primary antibodies were used: rabbit anti-MYC, anti-PTEN, anti-FOXP3 (Cell Signaling Technology), rabbit anti‐PAX8 (ProteinTech), and rabbit anti-CD8 (Abcam).
Targeted DNA sequencing
Multiplexed barcoded libraries were made from 20 ng DNA isolated from FFPE mouse tissue using AmpliSeq Library Kit 2.0 (Life Technologies). 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 1,445 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 1,353× 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 online 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 tamoxifen 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). 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 three lanes. Methods for data analysis are provided in the online supplement.
Results
Oviductal tumors arising in BPRN mice recapitulate the clinical features of human HGSCs
We previously described the histopathologic features of oviductal tumors arising in (n = 90) tamoxifen-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 STICs that progress to “early” HGSCs (eHGSC) 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; refs. 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 tamoxifen-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 Fig. 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 eHGSCs or STICs (8/13 vs. 1/18, P = 0.001, two-tailed Fisher 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 weeks, P = 0.0068, log-rank test, Supplementary Fig. S1). BPR mice were more likely to also have nonoviductal 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 weeks. Despite comparable survival to the BPR mice (Supplementary Fig. S1), the BPRN-Het mice had fewer nonoviductal 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 tamoxifen-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, that is, within 4 weeks post-tamoxifen (9). By 32 weeks post-tamoxifen, 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, tamoxifen-treated AP mice rapidly develop oviductal tumors that closely resemble human ovarian endometrioid carcinomas rather than HGSC (24), as do APA mice. To test whether 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 (Fig. 2; ref. 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 (SNV) to assess for coding mutations. 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 previously (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 (nonneoplastic) tissue from BPRN or BPR mice, either from tail before tamoxifen 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 S1), 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 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 Fig. 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 Fig. 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 panel of Fig. 3. High-level amplification of Myc was observed in the tumor shown in the top left panel, and deep (homozygous) deletion of Pten was present in the tumor represented in the left center panel. IHC staining of the tumor tissues from which DNA was isolated confirmed overexpression of MYC and loss of PTEN, respectively (Fig. 3, right).
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; ref. 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; ref. 29). Notably, human uterine endometrioid carcinomas (UEC) and serous carcinomas (USC) 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, Fig. 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 (Fig. 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 < 0.01, and found 13 genes agreeing on more frequent gains and 5 on losses (Fig. 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, because 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 and 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 18,372 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; refs. 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 Fig. S3A–S3C). 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 Fig. S3D). BPRN tumors gave more differences when compared with 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 Fig. S3C and S3D). The genes found differentially expressed in BPRN versus 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 Fig. 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 13,753 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 (Fig. 5A and 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 (Fig. 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, Fig. 5A). Differential gene expression analysis (asking that P <0.001 and fold-changes >1.5 in either direction) showed that mouse BPRN versus Od gene selections agreed with human HGSC versus normal fallopian tube gene selections (n = 311) for 648 upregulated and 540 downregulated genes with few disagreements (P = 5 × 10−270, Supplementary Table S2). 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 1,573 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 0.518 for mesenchymal, to 0.548 for immunoreactive (Supplementary Fig. S4A–S4C). Furthermore, 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 Fig. 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 versus human groups, when using differentially expressed genes, or the larger set of genes (Supplementary Fig. S4E and S4F). On the basis of 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 with 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 < 0.05 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 four pathways upregulated in human HGSCs (E2F_TARGETS, G2M_CHECKPOINT, MTORC1_SIGNALING, and MYC_TARGETS) were among the top five pathways upregulated in the mouse tumors (Table 1).
Top 20 gene sets for upregulated genes in BPRN mouse tumors versus normal oviduct . | |||||
---|---|---|---|---|---|
Title . | # of Genes on list . | # of Genes we selected . | Actual P . | Estimated FDR, based on permutations (Q-value) . | Observed/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 upregulated 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 |
Top 20 gene sets for upregulated genes in BPRN mouse tumors versus normal oviduct . | |||||
---|---|---|---|---|---|
Title . | # of Genes on list . | # of Genes we selected . | Actual P . | Estimated FDR, based on permutations (Q-value) . | Observed/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 upregulated 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 |
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 Fig. S5A). Asking which genes with P < 0.001 and fold-change greater than 1.5-fold also had similar differences for mouse BPRN HGSCs versus AP endometrioid carcinomas showed that 91 of 202 upregulated genes and 85 of 225 downregulated genes were also selected in the mouse (P = 2 × 10−27, Supplementary Fig. 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+ 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 Fig. 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 with AP tumors were for myeloid-derived suppressor cells (MDSC), 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 (Fig. 6B). Within the immunoreactive and mesenchymal human groups, MDSCs and Tregs had the largest fold changes relative to the differentiated and proliferative groups, although 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 Fig. 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 Fig. S6B and S6C). 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 Fig. S7).
Recent work by Zhang and colleagues 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 antitumor 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+ 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 indicate 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 colleagues, 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 and colleagues 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 and colleagues 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 whether 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 antitumoral immune cells during development and progression. Such mechanisms, combined with the overall low antitumor 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 nonoviductal 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, nonoviductal 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 with 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 nonserous types of ovarian cancer, peaking in the seventh versus 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.
Disclosure of Potential Conflicts of Interest
S.A. Tomlins was a prior consultant and now a current employee at Strata Oncology. No potential conflicts of interest were disclosed by the other authors.
Authors' Contributions
Conception and design: K.W. McCool, Z.T. Freeman, Y. Zhai, S.A. Tomlins, E.R. Fearon, B. Magnuson, R. Kuick, K.R. Cho
Development of methodology: K.W. McCool, Z.T. Freeman, Y. Zhai, R. Wu, S.A. Tomlins, B. Magnuson, K.R. Cho
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): K.W. McCool, Y. Zhai, C.-J. Liu, B. Magnuson
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): K.W. McCool, Z.T. Freeman, K. Hu, B. Magnuson, R. Kuick, K.R. Cho
Writing, review, and/or revision of the manuscript: K.W. McCool, Z.T. Freeman, Y. Zhai, K. Hu, S.A. Tomlins, E.R. Fearon, B. Magnuson, R. Kuick, K.R. Cho
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Zhai, R. Wu, K. Hu, B. Magnuson, K.R. Cho
Study supervision: K.R. Cho
Acknowledgments
The authors would like to acknowledge technical assistance from Stephanie Schulman and Yisheng Wang. The research reported in this paper was supported by the NCI of the NIH under award numbers P30CA046592 (to R. Kuick, B. Magnuson, E. Fearon, and K. Cho), R01CA196619 (to K. Cho, Y. Zhai, R. Wu, S. Tomlins, and R. Kuick), and R01CA226756 (to K. Cho, E. Fearon, Y. Zhai, R. Wu, B. Magnuson, and R. Kuick).
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.