The paucity of genetically informed, immunocompetent tumor models impedes evaluation of conventional, targeted, and immune therapies. By engineering mouse fallopian tube epithelial organoids using lentiviral gene transduction and/or CRISPR/Cas9 mutagenesis, we generated multiple high-grade serous tubo-ovarian cancer (HGSC) models exhibiting mutational combinations seen in patients with HGSC. Detailed analysis of homologous recombination (HR)–proficient (Trp53−/−;Ccne1OE;Akt2OE;KrasOE), HR-deficient (Trp53−/−;Brca1−/−;MycOE), and unclassified (Trp53−/−;Pten−/−;Nf1−/−) organoids revealed differences in in vitro properties (proliferation, differentiation, and “secretome”), copy-number aberrations, and tumorigenicity. Tumorigenic organoids had variable sensitivity to HGSC chemotherapeutics, and evoked distinct immune microenvironments that could be modulated by neutralizing organoid-produced chemokines/cytokines. These findings enabled development of a chemotherapy/immunotherapy regimen that yielded durable, T cell–dependent responses in Trp53−/−;Ccne1OE;Akt2OE;Kras HGSC; in contrast, Trp53−/−;Pten−/−;Nf1−/− tumors failed to respond. Mouse and human HGSC models showed genotype-dependent similarities in chemosensitivity, secretome, and immune microenvironment. Genotype-informed, syngeneic organoid models could provide a platform for the rapid evaluation of tumor biology and therapeutics.
The lack of genetically informed, diverse, immunocompetent models poses a major barrier to therapeutic development for many malignancies. Using engineered fallopian tube organoids to study the cell-autonomous and cell-nonautonomous effects of specific combinations of mutations found in HGSC, we suggest an effective combination treatment for the currently intractable CCNE1-amplified subgroup.
This article is highlighted in the In This Issue feature, p. 211
The past 30 years of cancer research have yielded remarkable therapeutic advances along two main fronts (1, 2). “Targeted therapies” were developed against oncogenic “driver” tyrosine and serine/threonine kinases or key downstream signaling components (3). Concomitantly, powerful “immune therapies” emerged, including “immune checkpoint inhibitors” (e.g., anti-CTLA4, anti–PD-1, and anti–PD-L1; ref. 4). These new modalities complement or replace conventional chemoradiotherapy and are lifesaving for some patients. Nevertheless, most patients with metastatic solid tumors still succumb to their disease.
Targeted and immune therapies were developed in parallel, usually using distinct experimental systems. Even today, targeted agents are typically tested against cancer cell lines/cell-derived xenografts, patient-derived xenografts (PDX), or, more recently, human tumor spheroids/organoids. Such models are of human origin, have relevant mutational/epigenetic events, and sometimes retain a degree of tumor heterogeneity, but they do not allow evaluation of antitumor immune responses. PDXs can be established in “humanized” mice, but approximately 30% of human/mouse growth factors, cytokines, and chemokines fail to interact with the cognate receptor(s) in the other species, imposing intrinsic limits on “humanization” (5). Immune therapies, in contrast, are usually tested against syngeneic mouse tumors (6). These models (e.g., B16, CT26, and MC38) are mainly carcinogen-induced, arise from unknown, irrelevant, or not the most relevant cell-of-origin, and often lack key causal mutations found in the cognate human disease. Some targeted agents/immune therapies have been evaluated in genetically engineered mouse models (GEMM), which harbor disease-relevant genetic abnormalities and have intact immune systems (7). For most malignancies, however, only a few mutational combinations are generated, limiting the diversity of the human disease that can be analyzed. Most GEMMs introduce cancer-associated defects simultaneously into all epithelia in a target tissue; in contrast, real-world tumors initiate clonally and expand and progress in a sea of predominantly normal cells. Transplantable GEMM-derived models (e.g., Yum/Yummer melanoma cells; ref. 8) have been generated, but they have the same truncal mutations and limited genetic diversity.
The tumor genotype, in the context of the cell-of-origin, determines its susceptibility to conventional and targeted therapies, intrinsic immunogenicity (e.g., by neoantigens, altered surface expression of MHC class I molecules, and/or ligands for activating/inhibitory receptors on immune cells), and the spectrum of cytokines and chemokines (“secretome”) produced (9–11). Secretome components, in turn, recruit immune cells to the tumor microenvironment (TME). Save for mutation-selective agents (e.g., RASG12C inhibitors and osimertinib for mutant EGFR), targeted and conventional agents affect cells in the TME in addition to tumor cells (12, 13). A suite of immunocompetent mouse models bearing tumors with genetic defects seen in patient neoplasms might simulate tumor biology with greater fidelity and facilitate development of novel therapeutic agents or combinations of existing drugs.
We developed such a platform for the most common and deadly form of ovarian epithelial cancer, high-grade serous tubo-ovarian cancer (HGSC; ref. 14). Patients with HGSC usually present at an advanced stage with bulky metastatic spread throughout the peritoneum, although some have more discrete tumor deposits. Current treatment includes surgical “debulking” and platinum/taxane-based chemotherapy and often results in complete responses (CR). Unfortunately, disease almost always recurs, eventually in drug-resistant form. Despite the recent addition of antiangiogenics (Avastin) and PARP inhibitors (PARP-I) to the HGSC armamentarium, survival has improved only marginally in the past 30 years, and most (70%–90%) patients die from their disease (15). Clearly, better therapies are needed for this deadly malignancy.
Much is known about HGSC pathogenesis. Despite its appellation, HGSC most often initiates with mutation, deletion, or silencing of TP53 in fallopian tube epithelium (FTE), not the ovary. The Cancer Genome Atlas (TCGA) reveals additional pathogenic single-nucleotide variants (SNV), but HGSC is primarily a disease of copy-number abnormalities (CNA), including amplifications, deletions, and more complex chromosomal rearrangements, which affect multiple genes and pathways (16). The most clinically useful molecular classification groups HGSCs by homologous recombination (HR) status. Defects in known HR genes, including BRCA1, BRCA2, RAD51, or other Fanconi Anemia genes, occur in approximately 40% of cases; another approximately 15% to 20% of cases have PTEN loss or EMSY amplification and are probably HR-deficient (17). Defective HR confers sensitivity to platinum agents (the mainstay of HGSC therapy), and some (but not all) of these defects confer PARP-I responsiveness (18, 19). The remaining approximately 40% of tumors are HR-proficient, respond poorly to current therapy, and result in shorter survival (20). CCNE1 amplification, found in approximately 20% of HGSCs, is notorious for causing chemoresistance and poor outcome (21); hence, there is a particular need to develop new therapeutic strategies for these tumors. Despite this impressive progress in delineating the molecular anatomy of HGSC, how specific combinations of mutations determine the transformed phenotype, including the tumor transcriptome, secretome, antitumor immunity, and therapy response, remains poorly understood.
The paucity of genetically relevant, immunocompetent models of HGSC poses a major barrier to addressing such issues. Many studies have used cancer cell lines, most of which (including the most frequently used) lack the characteristic genomic abnormalities of HGSC (22). Human HGSC organoids have been derived (23, 24), but although organoids have been cocultured with immune cells (25–27), such systems cannot fully simulate the antitumor response. ID8 cells have been the primary model for studying the host immune response to HGSC, but these cells originate from ovarian surface epithelium (OSE) and have wild-type (WT) Trp53 (16, 28). GEMMs that use FTE-selective promoters/enhancers to direct mutational events have been developed (29, 30), but these involve artificial alterations (e.g., SV40 large T antigen expression) or relatively rare mutational combinations (e.g., Brca1/Pten/Trp53). Also, most are on mixed-strain backgrounds, which impedes some tumor immunology studies. Notably, no immunocompetent models for CCNE1-amplified HGSC have been reported.
Exploiting our mouse FTE organoid culture system (31), combined with viral-based overexpression and CRISPR/Cas9 mutagenesis, we developed multiple new syngeneic models of HGSC. We demonstrate the utility of this platform for uncovering cellular genotype/phenotype relationships, complementation groups for tumorigenicity, the effect of tumor genotype on drug sensitivity, secretome, CNAs, tumor immune landscape, and metastatic spread, and, most importantly, the rational development of a highly effective therapeutic combination for Ccne1-overexpressing HGSC using existing combinations of FDA-approved drugs.
FTE Organoid–Based Platform for HGSC
Most HGSCs initiate from the distal fallopian tube (fimbria), which mainly comprises secretory (PAX8+) and ciliated (acetyl-α-tubulin+) epithelia (32, 33). The initial event (except in germline carriers of mutations in BRCA1/2 or other predisposing genes) is mutation of TP53 in a PAX8+ cell, which, together with other defects, evokes the precursor lesion serous tubal intraepithelial carcinoma. Additional SNVs/CNAs confer invasive potential and promote metastasis to the ovarian surface, peritoneum, and distal organs (34, 35).
We used mouse FTE organoids (31) to model this complex biology. Briefly, fimbrial cells from Trp53f/f (or, where indicated, Brca1f/f:Trp53f/f mice) were seeded in Matrigel and cultured in defined media. Cyst-like organoids formed from single PAX8+ cells, a mixture of secretory and ciliated cells was seen after 6 days of culture, and tube-like epithelial folds developed by 10 days (ref. 29; Supplementary Fig. S1). After expansion, floxed alleles were excised by infection with adenovirus-Cre (Ad-Cre), yielding parental Trp53−/− organoids or, where indicated, compound mutants (all in C57BL6/J background). Additional genetic changes were introduced by lentiviral or retroviral gene transduction to model overexpression and/or by CRISPR/Cas9 mutagenesis to model deletions or mutations (Supplementary Fig. S2). Models were tested in cellular assays or transferred to 2-D cultures for larger-scale studies. Tumorigenesis was assessed by orthotopic injection into the ovarian bursa (for details, see Methods). Our current collection of models is summarized in Supplementary Table S1. To evaluate the utility of this platform for simulating HGSC pathogenesis and therapeutics, we performed detailed studies on representative examples of HR-proficient, HR-deficient, and unclassified subgroups.
Trp53−/−; Brca1−/−;MycOE FTE Organoids Give Rise to HGSC-Like Tumors
BRCA1/2 alterations are found in approximately 20% of HGSCs (36), so we chose Trp53−/−/Brca1−/− models to represent the HR-deficient subgroup (Fig. 1A and B; Supplementary Table S1). We infected Trp53f/f;Brca1f/f FTE with Ad-Cre, picked single organoids, and confirmed deletion of the relevant loci (Fig. 1B; Supplementary Fig. S2A). Neither Trp53 nor Brca1 deletion alone or in combination altered organoid morphology or ciliated cell differentiation (Fig. 1C; Supplementary Fig. S2B and S2C), although Trp53−/−;Brca1−/− organoids were significantly larger than their parental counterparts, most likely because of enhanced proliferation (assessed by Ki-67 staining). MYC is amplified in approximately 40% of HGSCs and often co-occurs with Brca1 alterations (Fig. 1A). Overexpression of Myc in Trp53−/−/Brca1−/− organoids further increased proliferation and organoid size, while impeding ciliary differentiation (Fig. 1B and C; Supplementary Fig. S2A and S2B). Orthotopic injection of Trp53−/− or Trp53−/−/Brca1−/− FTE cells (2 × 106) did not result in tumors within the 6-month observation period. In contrast, Trp53−/−;Brca1−/−;MycOE organoid cells evoked ovarian masses and omental metastases, resulting in death of all injected mice within 4 months. These tumors expressed HSGC markers, including PAX8, cytokeratin 7 (CK7), P16, and Wilms Tumor 1 (WT1), and were strongly Ki-67+ (Fig. 1D–F). Hence, whereas compound BRCA1/TP53 deficiency is insufficient to cause HGSC, superimposing high MYC levels (or PTEN and/or NF1 deficiency; Supplementary Table S1) results in a highly invasive, metastatic, lethal malignancy.
Trp53−/−;Pten−/−;Nf1−/− FTE Organoids Also Cause HGSC-Like Tumors
NF1 deficiency, due to NF1 mutation/deletion, is seen in approximately 12% of HGSCs (37). PTEN loss also occurs fairly frequently (∼7%) and is associated with poor prognosis (ref. 38; Fig. 1A). We therefore assessed the effects of PTEN, NF1, or compound PTEN/NF1 deficiency on Trp53−/− FTE. Using lentiviral transduction, a single-guide RNA (sgRNA) targeting Pten exon 2 was introduced into Trp53−/− organoids, three clones with biallelic deletion were identified and expanded, and PTEN deficiency was confirmed by immunoblotting. An analogous strategy was used to target Nf1 exon 2 in Trp53−/− or Trp53−/−;Pten−/− organoids (Fig. 1B; Supplementary Fig. S2D). As expected, PTEN deficiency increased AKT (pAKT) and mTOR (pS6 and pS6K) activation, whereas loss of NF1 led to increased pERK1/2 (Supplementary Fig. S2D). Pten−/− organoids showed increased proliferation and organoid size, filled lumens, and decreased ciliary differentiation. Nf1 deletion decreased ciliary differentiation and altered organoid shape, but proliferation and luminal integrity were unaffected. Triple-deleted (Trp53−/−;Pten−/−;Nf1−/−) and Trp53−/−;Pten−/− organoids behaved similarly in these assays (Fig. 1C; Supplementary Fig. S2E).
We also tested the tumorigenicity of Trp53−/−;Pten−/−, Trp53−/−;Nf1−/−, and Trp53−/−;Pten−/−;Nf1−/− organoid cells (at least two clones each). Some double mutant–injected mice (8/20 mice with Trp53−/−;Pten−/− and 8/24 mice with Trp53−/−;Nf1−/−) developed tumors within 6 months, but Trp53−/−;Pten−/−;Nf1−/− cells showed more rapid and penetrant (28/30) tumorigenesis and also caused tumors more rapidly than Trp53−/−;Brca1−/−;MycOE cells (Fig. 1D and E; Supplementary Table S1). Trp53−/−;Pten−/−;Nf1−/− tumors metastasized to the omentum and produced more ascites than Trp53−/−;Pten−/−, Trp53−/−;Nf1−/−, Trp53−/−;Brca1−/−;MycOE, or Ccne1OE tumors (Fig. 1D). Trp53−/−;Pten−/−;Nf1−/− tumor–bearing mice expressed HGSC markers and had shorter life spans than double knockouts (Fig. 1E and F).
AKT2 and/or KRAS Cooperate with CCNE1 to Cause HGSC
Amplification or gain of CCNE1, encoding the cell-cycle regulator cyclin E1, is the best-characterized driver of HR-proficient HGSCs and accounts for approximately 20% of cases (17, 20). AKT2 and KRAS amplification occurs in 8% and 16%, respectively, of HGSCs, and co-occurs with CCNE1 amplification (Fig. 1A). To model CCNE1 amplification (CCNE1amp) alone or with KRASamp and/or AKT2amp, Ccne1, Akt2, and/or Kras were overexpressed (OE) sequentially in Trp53−/− FTE using lentiviruses harboring different selection markers (Fig. 1B). Overexpression/increased activation of each protein was confirmed by immunoblotting (Supplementary Fig. S2F). Organoid diameter/morphology was not affected significantly by CCNE1, AKT2, or KRAS overexpression alone (compared with parental Trp53−/− organoids). However, CCNE1, but not AKT2 or KRAS, overexpression significantly increased proliferation (Supplementary Fig. S2G). This increase was probably offset by a comparable increase in cell death, accounting for the lack of alteration of organoid size; notably, CCND1 overexpression had analogous effects on MCF10A mammary organoids (39). Superimposing Akt2OE or KrasOE on Trp53−/−;Ccne1OE organoids further enhanced proliferation and increased organoid diameter, lumen filling, and organoid disorganization, which was even more pronounced in quadruple mutants. Ccne1OE alone did not alter ciliary differentiation, but ciliated cells were virtually undetectable in triple- and quadruple-mutant organoids (Fig. 1C; Supplementary Fig. S2G). Trp53−/−;Ccne1OE cells did not give rise to tumors by 6 months after orthotopic injection. In contrast, Trp53−/−;Ccne1OE;Akt2OE, Trp53−/−;Ccne1OE;KrasOE, and Trp53−/−;Ccne1OE;Akt2OE;KrasOE cells formed large, palpable ovarian tumor masses and massive omental metastasis, and death occurred within 2 months of injection (Fig. 1D and E). There was no apparent difference in tumor formation by each triple mutant, but quadruple mutants showed accelerated tumorigenesis and displayed histologic and IHC features of high-grade, poorly differentiated, invasive carcinoma (Fig. 1E and F).
Hence, several combinations of genetic abnormalities seen in human HGSC give rise to lethal ovarian cancers in immunocompetent mice and can be used to assign complementation groups for in vitro properties (proliferation, differentiation, and organoid morphology) and tumorigenic capacity. Other combinations of genetic abnormalities reported by TCGA also give rise to HGSC in mice (Supplementary Table S1).
Organoid Genotype Affects Genome Stability, Drug Sensitivity, and Secretome
HGSC is characterized by widespread CNAs/aneuploidies, which have been assigned to seven “copy-number signatures” associated with distinct mutational processes and driver abnormalities (40). We used shallow (2×) whole-genome sequencing (sWGS) to assess the CN status of our models. WT and Trp53−/− organoids (two clones each) showed normal diploid profiles, whereas two independent Trp53−/−;Brca1−/− organoid clones exhibited gains of mouse chromosome 5 (Supplementary Fig. S3A). Trp53−/−;Brca1−/−;MycOE organoids (two independent clones) showed additional, but distinct CNAs. Trp53−/−;Ccne1OE;Akt2OE;KrasOE (two independent clones) andTrp53−/−;Pten−/−;Nf1−/− organoids (one clone assessed at different times) also had multiple CNAs. Notably, the two Trp53−/−;Ccne1OE; Akt2OE;KrasOE clones had some shared and other distinct CNAs, whereas a Trp53−/−;Pten−/−;Nf1−/− organoid clone assessed at different times had a stable, although markedly aneuploid, genome (Fig. 2A).
Next, we tested these models for sensitivity to FDA-approved drugs and investigational/experimental agents for HGSC (Fig. 2B; Supplementary Fig. S3B). Organoids were triturated into small clumps, dissociated into single cells, and dispensed into 96-well Matrigel-precoated plates (see Methods). Each agent was added at various doses, and cell viability was assessed 5 days later. As expected, Trp53−/−;Brca1−/−;MycOE cells showed increased sensitivity to PARP-Is (Fig. 2B), although differential sensitivity varied for individual PARP-Is and was less than that seen in conventional ovarian cancer cell lines (41). Brca1-deleted cells showed slightly increased sensitivity to carboplatin, although there was substantial overlap with the other mutants. Comparison of Trp53−/−;Brca1−/− and Trp53−/−;Brca1−/−;MycOE organoids showed that MYC overexpression reduces PARP-I and/or platinum sensitivity (Supplementary Fig. S3C). Trp53−/−;Ccne1OE;Akt2OE;KrasOE organoids were more sensitive to gemcitabine than the other models, consistent with the increased replication stress caused by CCNE1 overexpression (23). In contrast, and unexpectedly, Trp53−/−;Pten−/−;Nf1−/− cells showed enhanced susceptibility to paclitaxel, and comparisons to Trp53−/−;Pten−/− and Trp53−/−;Nf1−/− cells attributed this difference to NF1 deficiency (Fig. 2B; Supplementary Fig. S3D). Trp53−/−;Brca1−/−;MycOE and Trp53−/−; Pten−/−;Nf1−/− organoids had increased sensitivity to the ATR inhibitor BAY1895344, whereas chloroquine, which inhibits endosomal acidification and is often used as an autophagy inhibitor, was differentially toxic for all genotypes (Trp53−/−;Pten−/−;Nf1−/− > Trp53−/−;Ccne1OE;Akt2OE;Kras > Trp53−/−;Brca1−/−;MycOE). All genotypes showed comparable sensitivity to the CDK7 inhibitor YKI-5-1241, the CDK7/9 inhibitor PHA767491, and the CDK2/7/9 inhibitor seliciclib (Supplementary Fig. S3B).
We also used Luminex technology to assay organoid-conditioned media (Fig. 2C). Notably, engineered organoids secreted a complex mixture of chemokines, cytokines, and growth factors, and their secretome was genotype-dependent. As these factors could help initiate immune-cell immigration and/or survival, these results raised the possibility that, as demonstrated below, tumor genotype instructs the TME.
Ovarian Tumors with Different Genotypes Have Distinct Transcriptomes
We used RNA sequencing (RNA-seq) to analyze the transcriptomes of tumors (four each) of each genotype and normal fallopian tube. Unsupervised hierarchical clustering showed clear separation between tumor and normal samples and between each model, with Trp53−/−;Ccne1OE;Akt2OE;KrasOE tumors showing the greatest difference from the others (Fig. 3A). Pathway analysis revealed that, compared with normal fallopian tube, tumor transcriptomes were enriched primarily for Kyoto Encyclopedia of Genes and Genomes (KEGG) gene sets associated with the immune response (e.g., cytokine/cytokine receptor interaction, chemokine signaling pathway, antigen processing and presentation, Leishmania infection, Toll-like receptor signaling pathway, etc.) and, to a lesser extent, for processes related to proliferation (e.g., DNA replication, cell cycle, ribosome, etc.). Hallmark gene sets associated with inflammatory/immune (allograft rejection, TNFα signaling, IFNγ response, IFNα response, complement signaling, etc.) and proliferative (G2–M-phase checkpoint, MYC targets, KRAS signaling, mTORC1 signaling, etc.) processes and multiple oncogenic gene sets were also enriched (Fig. 3B and C). Pairwise comparisons revealed significant differences between tumors, comporting with their distinct genotypes. For example, compared with Trp53−/−;Ccne1OE;Akt2OE;KrasOE models, Trp53−/−;Pten−/−;Nf1−/− tumors showed lower expression of “PTEN down” and of “MEK up,” “KRAS up,” and “EGFR up” genes; these findings likely reflect stronger RAS–ERK pathway activation in KRAS-overexpressing, compared with NF1-deficient, cells. In contrast, Trp53−/−;Pten−/−;Nf1−/− tumors showed enrichment for “KRAS up” and “AKT up” gene sets compared with their Trp53−/−;Brca1−/−;MycOE counterparts (Fig. 3C)
We also examined chemokine, cytokine, and hematopoietic growth factor gene expression in each type of tumor (Fig. 3D). Most ILs were expressed at low/undetectable levels in all models, as were many chemokines, whereas IL15, IL16, IL18, IL33, and IL34 were expressed significantly in all tumors (as in their cognate organoids; Fig. 2C). Lif, IL1b, Csf1 (MCSF), and, to a lesser extent, Tnfα were expressed at higher levels in Trp53−/−;Pten−/−;Nf1−/− and Trp53−/−;Ccne1OE;Akt2OE tumors, compared with Tp53−/−;Brca1−/−;MycOE tumors. Some chemokines (e.g., Cxcl12 and Cxcl16) were expressed at similar, high levels in all models. Others showed genotype-specific differences; for example, Ccl2 and Ccl5 were expressed most highly in Trp53−/−;Ccne1OE;Akt2OE;KrasOE tumors, at intermediate levels in Trp53−/−;Pten−/−;Nf1−/− tumors, and at lower levels in Trp53−/−;Brca1−/−;MycOE tumors. Cxcl1 levels were higher in Trp53−/−;Brca1−/−;MycOE and Trp53−/−;Pten−/−;Nf1−/− tumors. Ccl6–9 were expressed in most models, although at generally lower levels in Trp53−/−;Brca1−/−;MycOE tumors. In contrast, Cxcl9 was expressed at highest levels in Trp53−/−;Brca1−/−;MycOE tumors and at lowest levels in Trp53−/−;Ccne1OE; Akt2OE, KrasOE tumors, whereas Cxcl10 was expressed at higher levels in the latter. Vegfa and Tgfb1 transcripts were high in all of the models.
Comparing the organoid secretome with the tumor transcriptome suggested specific cytokines/chemokines that initiate and help to maintain the TME (e.g., CCL2, CCL5, and CXCL10 for Trp53−/−;Ccne1OE;Akt2OE;KrasOE tumors; MCSF, CXCL1, and CXCL9 for Trp53−/−;Brca1−/−;MycOE tumors; and MCSF, CXCL1, CCL2, and VEGFA for Trp53−/−;Pten−/−;Nf1−/− tumors). CCL2, CCL5, and CXCL10 were also detected at high levels in the serum of tumor-bearing Trp53−/−;Ccne1OE;Akt2OE;KrasOE mice (213.6 ± 56.39 pg/mL). Other factors might contribute to TME initiation, but were no longer expressed at high levels in tumors themselves (e.g., G-CSF/Csf2 in Trp53−/−;Brca1−/−;MycOE and Trp53−/−;Pten−/−;Nf1−/− tumors). Some presumably emanate primarily from tumor-infiltrating immune cells, rather than cancer cells themselves (e.g., GMSCF/Csf3 and CXCL5 in Trp53−/−;Brca1−/−;MycOE). We tested some of these predictions by perturbation experiments, as described below.
The HGSC Microenvironment Depends on Tumor Genotype
Given their markedly different secretomes, we suspected that organoids of different genotype might elicit distinct TMEs. To test this hypothesis, we assayed Trp53−/−;Brca1−/−;MycOE, Trp53−/−;Pten−/−;Nf1−/−, and Trp53−/−;Ccne1OE;Akt2OE;KrasOE tumors by flow cytometry using lymphoid and myeloid marker panels (Supplementary Fig. S4A and S4B). Levels of CD45+ immune cells (compared with CD45− tumor/stromal cells) were approximately 2-fold higher in Trp53−/−;Ccne1OE;Akt2OE;KrasOE and Trp53−/−; Brca1−/−;MycOE tumors than in Trp53−/−;Pten−/−;Nf1−/− tumors (Supplementary Fig. S4C). None of the models had many tumor-associated B (CD19+), natural killer (NK; NK1.1+), or NKT (NK1.1+CD3+) cells (Fig. 4A; Supplementary Fig. S4C).
Nevertheless, the composition of the CD45+ population in tumors with different genotypes differed substantially (Fig. 4A and B). Trp53−/−;Pten−/−;Nf1−/− tumors had a predominant macrophage (CD11b+F4/80+) population, smaller numbers of myeloid dendritic cells (mDC; CD11b+CD11C+), granulocytic myeloid-derived suppressor cells (g-MDSC; CD11b+Ly6CloLy6Ghi), and monocytic myeloid-derived suppressor cells (m-MDSC; CD11b+Ly6GloLy6Chi), and sparse T lymphocytes (CD3+ cells). Given their lower fraction of total CD45+ cells (Supplementary Fig. S4C), absolute T-cell number in Trp53−/−;Pten−/−;Nf1−/− tumors was even lower compared with the other models. The macrophages in Trp53−/−;Pten−/−;Nf1−/− tumors had greater “M2-like” character, with high percentages of CD11b+F4/80+ cells expressing CD206 and a lower percentage of iNOS+ cells (Fig. 4B); most, however, coexpressed M1 and M2 markers, consistent with an “M0-like” state (42, 43). Immunofluorescence (IF) staining provided direct confirmation of higher CD3+ cell infiltration into Trp53−/−;Brca1−/−;MycOE and Trp53−/−;Ccne1OE;Akt2OE;KrasOE tumors than into Trp53−/−;Pten−/−;Nf1−/− tumors, and highest levels of Ly6G+ cells in Trp53−/−;Ccne1OE;Akt2OE;KrasOE tumors (Supplementary Fig. S4D and S4E).
Trp53−/−;Ccne1OE;Akt2OE;KrasOE tumors were more inflamed, exhibiting infiltration with macrophages, mDCs, g-MDSCs, and T lymphocytes (Fig. 4A and B). Nearly half of the CD4+ T cells in these tumors were T regulatory cells (Treg; CD25+FOXP3+), though, whereas most CD8+ cells showed “exhaustion” markers (TIM3+ and PD1+). Tumor-associated macrophages (TAM) expressed “M1-like” (MHCII+ and iNOS+) and “M2-like” (CD206+) markers, although the former predominated.
Finally, Trp53−/−;Brca1−/−;MycOE tumors had large percentages of macrophages and lower fractions of g-MDSCs, m-MDSCs, and mDCs. Unlike in the other models, CD4+ and CD8+ T cells in Trp53−/−;Brca1−/−;MycOE tumors were predominantly (>60%) CD44+ and strongly CTLA4+ and PD-1+ (Fig. 4B; Supplementary Fig. S4C), suggesting activation. Compared with cognate cells in Trp53−/−;Ccne1OE;Akt2OE;KrasOE tumors, CD8+ cells in Trp53−/−;Brca1−/−;MycOE tumors showed less TIM3 positivity, suggestive of less exhaustion, and there were fewer Tregs. Trp53−/−;Brca1−/−;MycOE tumors had more a balanced population of Th1 (Tbet+) and Th2 (GATA3+) cells (Th1/Th2, 0.7), whereas the other models mostly had Th1 cells (Th1/Th2, 2.4 in Trp53−/−;Pten−/−;Nf1−/− and Th1/Th2, 3.4 in Trp53−/− in Ccne1OE;Akt2OE;KrasOE). Trp53−/−;Brca1−/−;MycOE macrophages also had more M1-like character (%CD206/%iNos, 1.3) than did the other models (%CD206/%iNOS, 0.2 in Ccne1OE;Akt2OE;KrasOE and %CD206/%iNOS, 0.6 in Trp53−/−;Pten−/−;Nf1−/−).
PD-L1 expression was also genotype-dependent. In all models, approximately 40% to 45% of m-MDSCs were PD-L1+. In Ccne1OE;Akt2OE;KrasOE tumors, 45% of g-MDSCs also expressed PD-L1, whereas expression on g-MDSCs was lower in Trp53−/−;Brca1−/−;MycOE (25%) and Trp53−/−;Pten−/−;Nf1−/− (14%) tumors. In contrast, approximately 60% of Trp53−/−;Pten−/−; Nf1−/− TAMs were PD-L1+. Trp53−/−;Pten−/−;Nf1−/−, Ccne1OE;Akt2OE; KrasOE, and Trp53−/−;Brca1−/−;MycOE tumors showed PD-L1 expression on 2%, 8%, and 5% of CD45− cells (malignant cells, respectively).
To explore whether specific organoid-produced cytokines/chemokines elicit particular features of the TME, we focused on Trp53−/−;Ccne1OE;Akt2OE;KrasOE tumors and CCL2, CCL5, CXCL10, and GM-CSF; these proteins and their cognate transcripts were detected at high levels in organoid-conditioned media and tumors, respectively. First, we evaluated their effects on immune cell migration in vitro (Fig. 4C). Bulk CD45+ cells or CD3+ cells were purified from tumors and placed in the top well of a Transwell chamber. Conditioned media from Trp53−/−;Ccne1OE;Akt2OE;KrasOE organoids were placed in the bottom chamber with or without appropriate neutralizing antibodies. Anti-CXCL10 or anti-CCL5 blocked T-cell migration, whereas anti–GM-CSF and, to a lesser extent, anti-CCL5 blocked the migration of total CD11b+ cells. GM-CSF was the prime mediator of macrophage chemotaxis, whereas CCL5 and CCL2 were contributory. GM-CSF or CCL5 blockade impaired g-MDSC migration.
We also tested the effects of neutralizing these cytokines/chemokines on TME development. Trp53−/−;Ccne1OE;Akt2OE;KrasOE organoid cells were injected orthotopically (day 0), followed by neutralizing antibody injections at days 8 and 11 (Fig. 4D). Consistent with the in vitro chemotaxis assays, CXCL10 blockade resulted in fewer T cells in the TME, whereas GM-CSF blockade resulted in decreases in TAMs and g-MDSCs, compared with isotype control treatment. These results argue that CXCL10 and GM-CSF are important drivers of T-cell, macrophage, and g-MDSC immigration into the Trp53−/−;Ccne1OE;Akt2OE;KrasOE TME, respectively. Although nominal decreases were observed, anti-CCL2 or anti-CCL5 did not significantly reduce tumor-associated T-cell or myeloid-cell infiltration compared with isotype control–injected mice (Fig. 4D). Combination effects were not, however, excluded. Indeed, a complex mix of immune- modulatory factors, acting in concert, probably sculpt the microenvironment of these tumors.
Rationally Derived Combination Therapy Yields Durable CRs in Trp53−/−;Ccne1OE;AktOE;KrasOE HGSC
We next assessed the utility of our platform for developing HGSC therapies. To enable rapid clinical translation, we focused on CCNE1-overexpressing tumors, given their limited response to current therapies and poor prognosis, and on FDA-approved drugs. Consistent with our in vitro findings, gemcitabine administration to mice with Trp53−/−;Ccne1OE;Akt2OE;KrasOE tumors reduced, but did not eliminate, disease burden (Supplementary Fig. S5A and S5B). As in other tumor models (44–46), gemcitabine also decreased g-MDSCs (CD11b+Ly6CloLY6Ghi) in the TME, but other cell populations, most notably Tregs (CD24+CD25+FOXP3+) and T cells expressing exhaustion markers (TIM3/PD1), were unchanged (Supplementary Fig. S5C).
Given these data, we designed a regimen to attack tumor cells while normalizing the TME (Fig. 5A): gemcitabine to decrease tumor cells and g-MDSCs, anti-CTLA4 antibodies to target Tregs (47), and anti–PD-L1 antibodies to reactivate exhausted CD8 cells (48, 49). This combination (GCP) produced CRs in 10 of 10 treated mice (Fig. 5B–E). Treatment was stopped after day 35 (Fig. 5A), yet tumors failed to recur over a 60-day observation period (Fig. 5D; Supplementary Fig. S5C). Gemcitabine plus anti–PD-L1 (but not anti-CTLA4) evoked a greater decrease in tumor burden and ascites than gemcitabine alone, but no CRs. Gemcitabine/anti-CTLA4 reduced ascites, but did not measurably diminish tumor burden (Fig. 5C; Supplementary Fig. S5D and S5E). Upon therapy cessation, tumors recurred in all mice in the two-drug combination groups, leading to their rapid demise (Fig. 5D). Histologic analysis of GCP-treated animals after eight cycles revealed normal fat abutting minimal amounts of residual tumor in the injected bursae; in contrast, considerable tumor remained in mice treated with gemcitabine/anti–PD-L1 or gemcitabine/anti-CTLA4 (Fig. 5E). Multicolor IF confirmed that Ly6G+ cells were decreased in mice treated with gemcitabine, alone or in combination with anti-CTLA4 and/or anti–PD-L1. Only tumors from GCP-treated mice showed significantly increased T-cell (CD3+) infiltration, which included CD4+ and CD8+ T cells (Fig. 5F). These mice also showed increases in granzyme B+ (cytolytic) cells, and decreased numbers of TAMs and Tregs (Supplementary Fig. S5F and S5G).
The durability of the responses, and the attendant T-cell influx, prompted us to ask whether GCP responses were T cell–dependent. To this end, we depleted CD4+ and/or CD8+ T cells and reassessed efficacy. To enhance our ability to monitor tumors, we transduced Trp53−/−;Ccne1OE;Akt2OE;KrasOE organoids with a luciferase-expressing lentivirus prior to implantation; control experiments showed that luciferase-expressing and parental tumors behaved similarly (Supplementary Fig. S6A). Depletion of the expected T-cell population was confirmed by flow cytometry of peripheral blood (Supplementary Fig. S6B and S6C). Notably, CD4- or CD8-cell depletion impaired the response to GCP, whereas tumors from mice lacking CD4 and CD8 T cells actually grew faster in the presence of therapy than did tumors in PBS-treated mice with intact immune systems (Fig. 5G and H). GCP-treated, CD8- or CD4+CD8-depleted tumor-bearing mice had survival times similar to PBS-treated mice with intact immune systems. CD4 depletion impaired survival in the combination-treated group, but to a lesser extent (Fig. 5I).
Therapeutic Efficacy Is Tumor Genotype–Specific
To ask whether GCP efficacy was specific for Trp53−/−; Ccne1OE;Akt2OE;KrasOE tumor–bearing mice, we tested the regimen in Trp53−/−;Pten−/−;Nf1−/− tumor–bearing mice. Remarkably, the latter were completely refractory to the GCP regimen, as measured by tumor burden and percentage of mice with ascites (Fig. 6A). We also tested the effects of single-agent paclitaxel. Both models showed some response, but, as predicted by our in vitro experiments, Trp53−/−;Pten−/−;Nf1−/− tumor–bearing mice experienced more regression than those with Trp53−/−;Ccne1OE;Akt2OE;KrasOE tumors (Fig. 6B–D). These differences in clinical parameters were accompanied by differences in survival. Although single-agent paclitaxel did not result in CRs, it did evoke potentially beneficial changes in the TME, including an influx of CD44+ CD4 and CD8 T cells, less evidence of exhaustion, and decreases in g-MDSCs and TAMs, with those remaining showing a more M1-like phenotype (Supplementary Fig. S7A–S7C).
Similarities between Ccne1OE Models and Human CCNE1amp HGSC
To further evaluate the translational potential of our mouse organoid platform, we established organoid lines from 9 patients with HGSC who had undergone genomic profiling. These included examples of the major alterations engineered into our mouse organoids (Supplementary Fig. S8A), including CCNE1Amp, BRCA1 or BRCA2 mutation (BRCAMut), and NF1 deletion (NF1Del). Different human organoids had distinct morphologies, but all expressed PAX8, were highly proliferative, and stained positively for the DNA-damage marker γH2A.x (Fig. 7A). Although they showed a range of sensitivities, the CCNE1Amp lines were more sensitive to gemcitabine than the BRCAMut or NF1Del organoids. We noticed that the one CCNE1Amp organoid (HGS-3.1), tested by Kopper and colleagues, also showed profound gemcitabine hypersensitivity (24). In contrast, NF1Del organoids were resistant to gemcitabine, but more sensitive to paclitaxel, while, as expected, BRCAMut organoids were more sensitive to olaparib (Fig. 7B). Hence, human HGSC organoids showed a pattern of drug sensitivities similar to our mouse models.
We next compared the secretomes of CCNEAmp and BRCA1Mut organoids. Except for G-CSF, which was expressed at lower levels in Trp53−/−;Ccne1OE;Akt2OE;KrasOE than in Trp53−/−;Brca1−/−;MycOE organoids, the cytokines, chemokines, and growth factors detected in both species followed similar patterns in each (Fig. 7C). Of particular note, GM-CSF and CXCL10, which are functionally important for myeloid (g-MDSCs and macrophages) and T-lymphocyte recruitment, respectively, to Trp53−/−;Ccne1OE;Akt2OE;KrasOE tumors (Fig. 4D), were significantly higher in CCNE1Amp organoids compared with their BRCA1Mut counterparts. Likewise, CCL2 and CCL5, which are required for chemotaxis in Transwell assays and had nominal, although not significant, effects in tumors, had a similar secretion pattern in mouse and human HGSC organoids, as did the angiogenic growth factor VEGF (Figs. 4C and D and 7C).
We did not have ready access to large numbers of HGSC cases for prospective characterization of their genomic abnormalities and TME. Instead, we inferred the immune landscape in tumors of different genotypes by applying the quanTIseq (50) and CIBERSORT (51) algorithms to TCGA data; CIBERSORT was implemented in “abs mode” to allow intrasample comparison between cell types and intersample comparisons of the same cell type. HGSC cases were grouped as TP53−/−;PTEN−/−;NF1−/− (TPN), TP53−/−;CCNEamp/OE;KRASamp/OE (TCK), TP53−/−;CCNEamp/OE;AKT2OE;KRASamp/OE (TCAK), or TP53−/−;BRCA1−/−;MYCamp/OE (TBM) tumors, based on copy number and RNA-seq profiles (for details, see Methods). The TCK group (which includes TCAK tumors) was included to increase sample size and because multiplex IHC showed that KRAS overexpression was primarily responsible for the major features of the Trp53−/−;Ccne1OE;Akt2OE;KrasOE TME (Supplementary Fig. S8B).
Consistent with the immune phenotypes observed in mouse organoid–derived tumors, quanTIseq revealed that CCNE1 (TCAK and TCK) and BRCA1 (TBM) tumors had nominally more total immune cells than the TPN tumors, although the differences did not reach statistical significance. There also was a trend toward increased CD8 cells and significantly higher levels of Tregs in TCAK and TCK tumors, as well as a trend toward more neutrophils (likely g-MDSCs) in TCK and TCAK tumors, compared with those of other genotypes. Also as in the mouse models, TBM tumors tended to have more monocytes, whereas all tumors tended to have a mixture of M1 and M2 macrophages (but predominantly the latter). Notably, TCK and TCAK tumors had similar inferred TMEs, comporting with the dominant effect of KRAS in the organoid-derived tumors (Fig. 7D; Supplementary Fig. S8B). The total immune cells and tumor-associated T cells predicted by CIBERSORT were similar to quanTIseq inferences, although the myeloid cell predictions differed from those inferred by quanTIseq and those found in our mouse models (Supplementary Fig. S8C).
Finally, we performed IHC for CD8, FOXP3, and CD68 on seven primary HGSC samples for which we had genotype data available (Supplementary Fig. S8D). Consistent with our mouse models and the quanTIseq/CIBERSORT analyses, the CCNE1Amp;KRASAmp tumor showed significantly more T-cell infiltration than the CCNE1Amp;AKT2Amp tumor and the non-CCNE1amp samples (Fig. 7E and F). Tregs (FOXP3+) also were significantly higher in the CCNE1Amp;KRASAmp tumor. Although one of the two BRCAMut tumors (HGSC7) also showed a higher number of Tregs, we did not observe higher CD8+ T-cell infiltration in BRCA1/2 tumors, which might reflect the co-occurring PTEN deletions in these two cases. Then BRCA1/2mut, PTENmut, and NF1mut tumors had more macrophages (CD68+) than the CCNE1Amp tumors, again consistent with the cognate mouse models. We could not obtain consistent, reliable staining for g-MDSC/neutrophils and therefore could not test whether CCNE1amp led to greater immigration of these cells in humans as in mice. Overall, although additional, more detailed analyses are clearly needed, these results indicate significant similarity between the phenotypes of our mouse models and human HGSC.
Like most solid tumors, HGSC is genetically complex and heterogeneous, yet, with the exception of PARP-Is for BRCA-mutant tumors, current therapy for HGSC (as for most other neoplasms) is genotype-agnostic. Perhaps unsurprisingly, approximately 20% of patients with HGSC experience minimal or no clinical benefit from this uniform approach, and, even of those who initially respond, nearly all relapse and die (9). Rational development of genotype-informed therapies is impeded by a paucity of relevant experimental systems. Our FTE organoid–based system remedies these deficiencies, enabling analysis of the effects of specific genetic aberrations on in vitro properties (proliferation, differentiation, morphology, genome stability, drug sensitivity, and secretome), assignment of complementation groups for tumorigenicity, assessment and perturbation of the TME, and evaluation of drug therapies (Supplementary Fig. S1). Organoid-derived tumors are derived from the “correct” cell-of-origin and are formed in the relevant anatomic location surrounded by normal host cells. We demonstrate the utility of this platform by developing a specific combination regimen that evokes durable CRs in mice bearing Ccne1OE tumors, but has no activity against Trp53−/−;Pten−/−;Nf1−/− tumors. The latter tumors, in contrast, are more sensitive to paclitaxel (Fig. 6D). Analysis of human HGSC organoids, primary tumors, and TCGA data reveals similarities between our mouse models and the human disease (Fig. 7). In concert, our results argue strongly against therapeutic approaches that treat HGSC as a single entity and support the development of new, genotype-informed strategies.
The HGSC cell-of-origin remains controversial. Transcriptomic (33, 52, 53), proteomic (54), epigenomic (52), and mouse modeling (31, 55, 56) data suggest that at least some cases initiate in OSE, but most often HGSC initiates in FTE (33, 57, 58). Consequently, we focused our models on FTE organoids. Others have reported that orthotopic injection of 105 Trp53−/−;Brca1−/−;MycOE OSE cells also yields HGSC-like tumors, which kill recipients within 50 days (59). In contrast, mice injected with more FTE cells (2 × 106) of the same genotype survive for 70 to 150 days (Fig. 1E), consistent with our finding that the cell-of-origin influences HGSC biology (31). Our platform can be adapted easily to model OSE-derived HGSC, as well as other cancers for which mouse organoids can be cultured/engineered (60–62). Indeed, while our article was in review, others reported that FTE (“oviductal” in their article) and OSE organoids engineered with the same genetic abnormalities could give rise to HGSC, although OSE-derived tumors could only be established orthotopically after an initial subcutaneous passage. This study was restricted to Trp53−/−;Brca1−/+;Pten−/− and Trp53−/−;Brca1−/+;Nf1−/− combinations, which are not frequently seen in human HGSC, and used organoids from B6 × 129 mice, precluding transplantation into immunocompetent recipients and analysis of the TME (56).
Human HGSC is profoundly aneuploid, featuring amplifications, deletions, and complex rearrangements. Importantly, our engineered organoids also are aneuploid (Fig. 2A). Recent computational analyses identified recurrent patterns of abnormalities in human HGSC and defined seven specific CN “signatures” (40). That report noted correlations between specific driver genes/signaling aberrations and particular signatures, but did not establish a causal relationship. Although we analyzed relatively few engineered organoids, our results suggest that different drivers cause distinct patterns of CNAs. Future, expanded studies will ask whether mouse CN signatures also exist, potentially reflecting interspecies conservation of mutational processes, whether aneuploidy affects the antitumor immune response, and whether the aneuploid genome is, at least in part, sculpted by the host TME.
Tumorigenic organoids showed several expected, but other unanticipated, sensitivities to small-molecule inhibitors/drugs. In line with previous studies of human ovarian cancer cell lines, and Trp53−/−;Brca1−/−;MycOE OSE–derived cells, Brca1-mutant FTE-derived tumor organoids showed increased PARP-I sensitivity. Hypersensitivity was less in Trp53−/−;Brca1−/−;MycOE FTE organoids than in conventional Brca-mutant cell lines (63); however, a comparison of Trp53−/−;Brca1−/− and Trp53−/−;Brca1−/−;MycOE organoids showed that MYC confers relative PARP-I (and platinum) resistance (Supplementary Fig. S2C). Hence, MYC could be an important biomarker for PARP-I/platinum resistance in patients with HGSC, as suggested previously (64). Although there was a class-specific increase in PARP-I sensitivity in Trp53−/−;Brca1−/−;MycOE organoids, the extent of hypersensitivity differed for individual PARP-Is. Our models could be used to elicit the mechanistic basis for such differences, as well as their respective effects on the TME. ATR inhibitors also showed increased efficacy against Trp53−/−;Brca1−/−;MycOE organoids, in accord with the HR deficiency conferred by BRCA1 deficiency, whereas the sensitivity of Trp53−/−;Pten−/−;Nf1−/− cells to ATR inhibition comports with the reported role for nuclear PTEN in HR (65, 66). The mechanisms underlying genotype-dependent differences in paclitaxel (for Trp53−/−;Pten−/−;Nf1−/− cells) and chloroquine (for Trp53−/−;Pten−/−;Nf1−/− > Trp53−/−;Ccne1OE;Akt2OE;Kras) sensitivity are less clear. Comparison of Trp53−/−;Pten−/− and Trp53−/−;Nf1−/− organoids implicates NF1 deficiency as the main cause of increased paclitaxel sensitivity (Supplementary Fig. S3D); notably, NF1 associates with microtubules (28, 67, 68), the target of paclitaxel. KRAS-mutant cells require autophagy for survival (69), whereas PTEN deficiency or AKT2 overexpression, by increasing mTOR activity, should suppress autophagy. Conceivably, increased RAS activity, combined with lower basal autophagy because of increased mTOR, sensitizes FTE cells to further autophagy inhibition. Regardless, these differences emphasize the value of genotype-defined models for developing new therapies and identifying biomarkers and mechanisms of resistance. Although we tested a small number of agents, our models can be adapted to high-throughput drug screens or genetic perturbations (e.g., CRISPR/Cas9 screens). Furthermore, the genotype-specific sensitivities that we observed suggest that only certain patient subsets will respond to standard-of-care single agents or combinations. For example, combining paclitaxel with platinum, a practice developed empirically (70, 71), might benefit only patients with NF1-deficient tumors; others might simply incur taxane-based toxicity.
Human HGSC also has a complex TME, with differences in infiltrating immune cells and tumor-associated chemokines/cytokines associated with prognosis (72, 73). As in many other malignancies, intratumor CD8+ cells and high CD8+/Treg ratio correlate with improved survival, whereas high levels of Tregs are a negative prognostic sign (74–77). Intratumor T cells have been associated with expression of CXCL9, CXCL10, CCL5, CCL21, and/or CCL22, whereas high VEGF levels inversely correlate with T-cell infiltration (77–79). A large pan-cancer genomic analysis indicated that high levels of CCL5 RNA and protein (by IHC) correlate with intratumor CD8 cells in HGSC and other solid tumors (80). CCL5 and CXCL9 also correlated in this analysis, and dual expression of these chemokines was associated with better prognosis. Interestingly, ovarian cancers with high intratumor CD8+ cells, but low CCL5 RNA, had high levels of CXCL9 (81). In contrast, high levels of TAMs, particularly M2-like TAMs, and MDSCs correlate with poor outcome (82, 83). Aside from describing greater T-cell infiltration and better prognosis in BRCA-mutant tumors (72, 84, 85), examining the regulatory mechanism of specific immune-regulatory molecules (e.g., silencing of CCL5; ref. 81), and a very recent report correlating mutational signature 3 (which is associated with HR deficiency) and immune score with response to combined PARP-I/anti–PD-1 treatment, previous studies have been tumor genotype–agnostic. Yet the three mouse models that we examined in detail displayed major differences in TME, associated with major differences in chemokine/cytokine/growth factor expression (Figs. 2–4). Furthermore, perturbation experiments clearly identified specific secreted factors that influence TME (and likely tumor) development (Fig. 4C and D). As tumor genotype also affects response to targeted and conventional agents, understanding how genotypic differences direct host immune responses could aid in therapy development. Our ability to manipulate tumors (e.g., by further engineering of chemokine/cytokine genes) and host immune cells (e.g., by depletion studies and injection of tumorigenic organoid cells into various knockout backgrounds), and to study tumors over time, can provide insights into how the TME develops and responds to therapy.
Earlier reports noted differences in tumor immune infiltrates in other systems (10, 86) and implicated MYC, KRAS, mTOR, YAP, and β-catenin signaling in cancer cells (11). Many of these studies used syngeneic tumor models, GEMMs, or GEMM-derived cell lines, and some pointed to specific cytokines/chemokines as the cause of differences in the TME. Nevertheless, the extent to which TME responses are “hard-wired” by specific oncogenic defects has remained unclear. For example, PTEN deficiency leads to impaired T-cell infiltration owing to immunosuppressive myeloid cells in mouse prostate cancer (87) and melanoma (88) models. But, whereas CXCL5 (mouse)/CXCL6 (human) are implicated in myeloid cell immigration in prostate cancer, CCL2 and VEGF are the apparent culprits in melanoma. We also observed increased myeloid cells in Trp53−/−;Pten−/−;Nf1−/− HGSC, along with increased levels of CCL2 and VEGF. However, MCSF1 and CXCL1 might also contribute to myeloid infiltration in this model, whereas CXCL5 is not elevated and is unlikely to play a role. These findings, and many others (10, 11, 89), argue that cellular context (e.g., cell-of-origin and cooperating mutations) might be as important as specific oncogenic abnormalities for determining the ultimate TME and antitumor immune response. Our ability to rapidly engineer FTE organoids with all major combinations of genetic defects as seen in HGSC positions us to address this important issue.
Attempts to manage HGSC with immune therapy have not been very fruitful. Single-agent anti-CTLA4 or anti–PD-1/PD-L1 yielded only modest results, with response rates of 10% to 15% (90, 91). Combining anti–PD-1 and anti-CTLA4 increases response rate to 34%, but the clinical data are very immature (92). Our Brca1-mutant mouse model shows greater T-cell infiltration, as does BRCA1-mutant HGSC (Fig. 7D); such tumors might show a better response to immune checkpoint inhibition, alone or in combination with PARP-Is (59, 84, 93). However, these responses are rarely durable, and whether other tumor genotypes confer more or less sensitivity is not clear. A major advantage of our organoid platform is its ability to rapidly suggest and credential potential therapies. Our chemo-immunotherapy combination of three approved drugs, gemcitabine, anti–PD-L1, and anti-CTLA4, led to durable, T cell–dependent CRs in a highly aggressive, CCNE1-overexpressing HGSC model. It will be critical to test this combination in models in which Ccne1 overexpression is accompanied by other frequently co-occurring genetic defects (e.g., Mecom and/or Myc overexpression), as well as to develop and test combination immunotherapies with paclitaxel in PTEN/NF1-deficient HGSC; the changes in the TME that ensue following paclitaxel treatment already suggest several potential combination strategies.
A major consideration for any animal model is the extent to which it represents the cognate human disease. Although much more detailed studies are warranted, initial indications reveal similarities between our mouse models and human HGSC organoids in drug response (Fig. 7B), secretome (Fig. 7C), and TME (Fig. 7E and F; Supplementary Fig. S8). However, the latter analyses were limited by relatively small sample size, contradictory predictions of myeloid populations by quanTIseq and CIBERSORT, and lack of well-defined, consensus IHC/IF markers for identifying tumor-infiltrating myeloid cell subsets by IF/IHC (94, 95).
In conclusion, our ability to rapidly generate multiple, genetically defined, complex HGSC organoid models should facilitate studies of the diversity and host response of this disease. Our models also suggest a genomics-informed, rationally based combination treatment for CCNE1-amplified HGSC, and suggest new interventional strategies for other genomic subgroups of this highly complex disease.
Organoid Culture and Engineering
FTE organoids from Trp53f/f or Trp53f/f;Brca1f/f mice were established as described previously (31). Cultures were checked monthly for Mycoplasma by PCR. Organoid cells were collected by using cold Cultrex Organoid Harvesting Solution (Stem Cell Technologies) to dissolve Matrigel, following the manufacturer's instructions. Trp53f/f female or Trp53f/f;Brca1f/f organoids were dissociated into single cells as described previously (96) and infected with 105 pfu Adenovirus-CMV-Cre (Vector Development Laboratory, Baylor College of Medicine, Houston, TX) by “spinoculation” at 37°C for 1 hour. Cell pellets were recovered and seeded into Matrigel in media containing nutlin-3 to enrich for Trp53−/− organoids. Organoids were released 7 days later, and multiple clones were picked and expanded. Deleted clones were identified by PCR (97). Primer sequences are provided in Supplementary Data.
Mouse Ccne1 and Akt2 were cloned into pLV-EF1a-IRES-Neo (Addgene, #85139), with neomycin resistance or pLV-EF1a-IRES-Blast (Addgene, #64832) with blasticidin resistance genes, respectively. For Myc overexpression, we used the vector, MSCV-transgene-PGK-Puro IRES-GFP, purchased from Addgene (#75124). Mouse Kras was cloned into pMSCV-IRES-mCherry (Addgene, #52114). Successful gene insertion was confirmed by Sanger sequencing. Pten (agatcgttagcagaaacaaa) or Nf1 (ctcgtcgaagcggctgacca) sgRNA sequences were designed with the CRISPR design tool (http://crispr.mit.edu/) and inserted into LentiCRISPR v2 (Addgene, #52961). For virus production, lentiviral vectors were cotransfected with psPAX2 and pMD2.G into HEK293T cells at a ratio of 10:7.5:2.5, or retroviral vectors were cotransfected with pVPack and VSV-G into HEK293T cells at a ratio of 10:6.5:3.5. All transfections were performed by using Lipofectamine 2000 Transfection Reagent (Thermo Fisher Scientific), according to the manufacturer's instructions. Media were changed 8 hours after transfection, and viral supernatants were collected 48 hours later by passage through a 0.45-mm filter, aliquoted, and stored at −80°C.
Organoids were dissociated into single cells and “spinoculated” with lentiviruses/retroviruses, also as described previously (96). Briefly, viral supernatants were added to cells in 48-well plates, centrifuged at 600 × g at 37°C for 60 minutes, incubated at 37°C for another 6 to 8 hours, collected, and reseeded in Matrigel-containing media. Infected organoids were selected 72 hours after viral transduction with G418 (Thermo Fisher Scientific, 10131027) or blasticidin (Sigma, 15205), as indicated. Gene deletion and/or overexpression were assessed by PCR or immunoblotting. At least two independent clones of each genotype were used for experiments.
For human HGSC organoid cultures, samples were obtained from the University Health Network Tissue Bank (Toronto, Ontario, Canada) with written informed consent. All studies were conducted in accordance with recognized ethical guidelines (e.g., Declaration of Helsinki, CIOMS, Belmont Report, and U.S. Common Rule) and research ethics board approval [equivalent to institutional review board (IRB) in the United States]. Tumor cells were isolated from fresh surgical material or ascites, as described previously (98, 99). HGSC cells of the indicated genotypes were thawed, seeded in Matrigel, and cultured in human organoid growth medium, composed of: Ad+++ AdDMEM/F12 (Invitrogen); HEPES (Thermo Fisher Scientific, 100× diluted); and penicillin/streptomycin and GlutaMAX, each 100× diluted (Life Technologies), supplemented with B27 (Invitrogen, 50× diluted), N2 supplement (Thermo Fisher Scientific, 100× diluted), 1.25 mmol/L N-acetylcysteine (Sigma), 50 ng/mL EGF (Thermo Fisher Scientific), 500 ng/mL RSPO1 (PeproTech) or R-spondin-1 conditioned medium (25%–50%, v/v), WNT3a conditioned medium (0%–25%, v/v), 100 ng/mL Noggin (PeproTech), 10 nmol/L 17-β Estradiol (Sigma), 50 ng/mL EGF, 10 ng/mL FGF10, 0.5 μmol/L A83-01 (Thermo Fisher Scientific), 50 ng/mL human recombinant Heregulin-beta 1, and 10 μmol/L Forskolin (both from Stemcell Technologies). For the first 3 days after thawing, media were also supplemented with 10 μmol/L Y-27632 (Sigma-Aldrich).
Cytokine, chemokine, and growth factor levels in 72-hour conditioned media from organoid cultures were profiled using services at Eve Technologies. The Mouse Cytokine Array/Chemokine Array 31-Plex (MD31) panel included: eotaxin (CCL11), G-CSF, GM-CSF, IFNγ, IL1α, IL1β, IL2, IL3, IL4, IL5, IL6, IL7, IL9, IL10, IL12 (p40), IL12 (p70), IL13, IL15, IL17A, IP10, KC (CXCL1), LIF, LIX (CXCL5), MCP1 (CCL2), M-CSF, MIG (CXCL9), MIP1α (CCL3), MIP1β (CCL4), MIP2 (CXCL2), RANTES (CCL5), TNFα, and VEGF. The Human Cytokine Array/Chemokine Array 48-Plex (HD48) included: sCD40L, EGF, eotaxin, FGF2, Flt3 ligand, fractalkine, G-CSF, GM-CSF, GROα, IFNα2, IFNγ, IL1α, IL1β, IL1ra, IL2, IL3, IL4, IL5, IL6, IL7, IL8, IL9, IL10, IL12 (p40), IL12 (p70), IL13, IL15, IL17A, IL17E/IL25, IL17F, IL18, IL22, IL27, IP10, MCP1, MCP3, M-CSF, MDC (CCL22), MIG, MIP1α, MIP1β, PDGFAA, PDGFAB/BB, RANTES, TGFα, TNFα, TNFβ, and VEGFA.
Drug Sensitivity Assays
Organoids were seeded into 96-well plates at 1,000 cells/well (day 0). The indicated concentrations of rucaparib (Selleckchem, S1098), niraparib (MCE, HY-10619), olaparib (Selleckchem, S1060), gemcitabine (MCE, HY-B0003), doxorubicin (Sigma, D1515), paclitaxel (Selleckchem, S1150), carboplatin (Sigma, 1096407), seliciclib (MCE, HY-30237), PHA767491(Sigma, PZ0178), BAY1895344 (Selleckchem, S8666), chloroquine (Selleckchem, S4157), and YKL-5-124 [a gift from Dr. Kwok-kin Wong, NYU Grossman School of Medicine (NYUGSoM), NYU Langone Health, New York, NY] were added on the day following seeding (day 1). Media were changed and fresh drug was added on day 3. Cell viability was assessed on day 5 by adding 10 μL PrestoBlue and incubating for 30 minutes in 37°C. Fluorescence was measured in a FlexStation 3 Multi-Mode Microplate Reader (BOSTONind). Results were normalized to DMSO controls, and IC50 values were determined using GraphPad Prism 7.
To assess tumor-infiltrating myeloid cell migration, CD45+ cells were isolated from Trp53−/−;Ccne1OE;Akt2OE;Kras tumors by using CD45 MicroBeads (Miltenyi Biotec, 130-052-301). Cell culture inserts (8-μm pore size) were placed into 24-well plates, and 5 × 105 CD45+ cells were added into each insert. Trp53−/−;Ccne1OE;Akt2OE;Kras organoid conditioned medium (500 μL) with or without anti-CCL5 (1 μg/mL), anti-CCL2 (1 μg/mL), or anti-GM-CSF (1 μg/mL) was added to the bottom chamber. For T-cell migration assays, tumor-infiltrating T cells were purified by using the EasySep Mouse T Cell Isolation Kit (Stemcell Technologies, catalog # 19851), 5 × 105 purified cells were added into inserts (pore size = 3 μm), and conditioned medium with or without anti-CXCL10 (1 μg/mL) or anti-CCL5 (1 μg/mL) was added to the bottom chamber. After incubation for 24 hours at 37°C, inserts were removed, and cells that had migrated to each bottom well were collected, stained with the LIVE/DEAD Fixable Blue Dead Cell Stain Kit (Thermo Fisher Scientific, L23105) and the indicated cell surface markers, and quantified by flow cytometry. Each antibody was tested in triplicate.
Trp53f/f female mice were obtained from Dr. Kwok-kin Wong and Trp53f/f;Brca1f/f mice (97) were provided by Dr. Richard Possemato (both NYUGSoM, NYU Langone Health, New York, NY). Female C57BL/6 mice (6–8 weeks old) were purchased from Charles River Laboratories. All animal experiments were approved by, and conducted in accordance with the procedures of, the Institutional Animal Care and Use Committee at NYUGSoM (protocol no. 170602).
For orthotopic tumorigenicity assays, organoid cell pellets were collected and injected into ovarian bursae, as described previously (31). Briefly, mice were anesthetized by intraperitoneal injection of xylazine (10 mg/kg) and ketamine (50 mg/kg), shaved, and cleaned with betadine. A dorsal incision above the ovary was made, followed by incision of the peritoneal cavity. The ovary was externalized and, using an insulin syringe with a 31-G needle, 2 × 106 cells in 50 μL PBS/Matrigel (1:1 v/v) were injected through the ovarian fat pad into the bursa. Injected ovaries were returned to the peritoneal cavity, and incisions were sealed with wound clips. Mice that developed tumors were euthanized at the indicated time(s), or for survival experiments they were monitored until death or upon veterinary recommendation. Where indicated, mice received intraperitoneal injections of gemcitabine (50 mg/kg), paclitaxel (40 mg/kg), anti-CTLA4 (50 μg, clone 9H10, BioXCell), and/or anti–PD-L1 (50 μg, clone 4H2, BioXCell), beginning 8 days after cell implantation. Dosing was repeated every 3 days, as indicated. Control mice were injected with PBS or isotype control antibody (clone LTF-2, BioXCell).
CD4+ and/or CD8+ T cells were depleted by intraperitoneal injection of 200 μg of InVivoMAb anti-mouse CD4 (clone GK1.5, BioXCell) and/or InVivoMAb anti-mouse CD8α (clone 2.43, BioXCell), respectively, 1 week after cell implantation. Injections were repeated every 3 days (100). Other mice received isotype control antibody (clone LTF-2, BioXCell). Depletion of the appropriate lymphoid population was confirmed by flow cytometry of peripheral blood and reassessed every 2 weeks for the duration of the study. Peripheral blood was collected from tail veins into heparinized microhematocrit capillary tubes, centrifuged, and prepared for flow cytometry by lysing red blood cells in ACK buffer, followed by serial washes in RPMI. For cytokine neutralizations, mice were injected intraperitoneally with 50 μg anti-CXCL10 (clone 134013, Thermo Fisher Scientific), 50 μg anti-CCL5 (clone 53405, Thermo Fisher Scientific), 100 μg anti-CCL2 (clone 2H5, BioXCell), 100 μg anti-GM-CSF (clone MP1-22E9, BioXCell), or isotype control IgG (100 μg), as indicated, 1 week after implantation of Trp53−/−;Ccne1OE;Akt2OE;Kras organoid cells, as above. Antibody injections were repeated every 3 days, and tumors were collected 2 days after the final injection and analyzed by flow cytometry.
Mice were injected with 150 mg/kg D-luciferin Firefly (PerkinElmer, part no., #122799), and luminescence was assessed 15 minutes later using a PerkinElmer IVIS Lumina III Imaging System. Images were analyzed with Living Image Software 4.7.3.
Tumors were minced, chopped, and digested with Gentle Collagenase, 0.012% Dispase (w/v), and DNaseI (Stemcell Technologies) at 37°C for 1 hour. Single-cell suspensions were obtained by passage through a strainer (70 μm), washed in FACS buffer (PBS with 5% FBS), incubated with LIVE/DEAD Fixable Zombie Yellow Fixable Viability Kit (BioLegend, 423104) for 30 minutes, and blocked with anti-CD16/32 (BioLegend, clone 93) for 5 minutes on ice. Primary fluorophore-conjugated antibodies were added, and samples were incubated on ice for 45 minutes. FOXP3 Fixation/Permeabilization Buffer Set (BioLegend) was used for intracellular markers, according to the manufacturer's instructions. Antibodies for flow cytometry are listed in Supplementary Table S2. Flow cytometry was performed on an LSR II flow cytometer at the Flow Cytometry Core of the PCC Precision Immunology Shared Resource and analyzed with FlowJo software. Organoids cultured 6 days after infection with MSCV-Kras-mCherry were collected and digested as above, passed through a strainer (70 μm) to obtain single-cell suspensions, centrifuged at 1,000 × g for 5 minutes, and resuspended in PBS containing 2% FBS, 10 μmol/L Y-27632 (Stemcell Technologies Inc.), and DAPI (1 μg/mL). FACS was performed immediately on a MoFloTM XDP, and mCherryhi and mCherryneg cells were seeded at 5,000 cells/well.
Histology, IF, and IHC
Mouse tumor tissues were fixed in 4% paraformaldehyde (PFA) for 48 hours, processed, and embedded for standard histology, IHC, and IF. Clinical molecular profiling results were used to identify appropriate HGSC cases. Formalin-fixed, paraffin-embedded tissue blocks were retrieved from institutional archives under IRB approval (study # i16-01086). Sections (5 μm) were deparaffinized, rehydrated, stained with H&E, or subjected to antigen retrieval (citrate) at 120°C in a pressure cooker for 15 minutes. For IHC, endogenous peroxidase activity was quenched in 3% H2O2 in methanol for 15 minutes, and sections were blocked in 0.5% BSA in PBS for 1 hour. Primary antibodies were added overnight at 4°C, then slides were washed in PBS three times for 10 minutes, incubated with secondary antibodies for 1 hour at room temperature, and washed again. Antigens were detected by using the HRP Polymer Detection Kit and DAB Peroxidase Substrate (34002, Life Technologies). For IF, after antigen retrieval, slides were washed in PBS three times for 10 minutes and then blocked in 0.5% BSA in PBS for 1 hour. Primary antibodies were incubated at 4°C overnight, and sections were washed in PBS (three times, 10 minutes each), followed by incubation with Alexa Fluor secondary antibodies, as indicated. Washed slides were mounted with Prolong Gold Antifade Mountant (Thermo Fisher Scientific, P36930). For IF, organoids were released from Matrigel (as above), transferred to a μ-Slide 8 Well Glass Bottom (Ibidi), fixed in 4% PFA (pH 7.4) for 20 minutes, permeabilized in 1% Triton X-100 in PBS, and blocked in PBS, 1% BSA, 3% normal goat serum, and 0.2% Triton X-100. After overnight incubation with primary antibody at 4°C, organoids were washed three times for 10 minutes in PBS and incubated at room temperature with the appropriate Alexa Fluor secondary antibody (1:200). Organoids were washed with PBS and mounted using ProLong Gold Antifade (Molecular Probes, Invitrogen). Antibodies for IHC/IF are described in Supplementary Table S3. IHC slides were scanned using a Leica SCN400 F whole-slide scanner. IF images were taken with a ZEISS LSM 700 confocal microscope.
Cell pellets were resuspended in SDS lysis buffer (50 mmol/L Tris-HCl pH 7.5, 100 mmol/L NaCl, 1 mmol/L EDTA, 1% SDS, and 2 mmol/L Na3VO4), supplemented with protease (40 μg/mL PMSF, 2 μg/mL antipain, 2 μg/mL pepstatin A, 20 μg/mL leupeptin, and 20 μg/mL aprotinin) and phosphatase (10 mmol/L NaF, 1 mmol/L Na3VO4, 10 mmol/L β-glycerophosphate, and 10 mmol/L sodium pyrophosphate) inhibitors. Total lysate was resolved by SDS-PAGE, followed by transfer to Immobilon-FL PVDF Membranes (Millipore). Membranes were blocked in 1% BSA/TBS containing 0.1% Tween20 for 30 minutes and incubated for 1 hour in blocking buffer containing the indicated antibodies (Supplementary Table S3), followed by IRDye-conjugated Secondary Antibodies (LI-COR). Images were obtained using a LI-COR ODYSSEY CLx Quantitative IR Fluorescent Detection System.
RNA Extraction and Sequencing
Tumors were lysed in TRIzol, and RNA was extracted using the miRNeasy Mini Kit (Qiagen) according to the manufacturer's instructions. RNA-seq was performed by the PCC Genome Technology Center Shared Resource (GTC). Libraries were prepared using the Illumina TruSeq Stranded Total RNA Sample Preparation Kit and sequenced on an Illumina NovaSeq 6000 using 150 bp paired-end reads. Sequencing results were demultiplexed and converted to FASTQ format using Illumina bcl2fastq software. Average read pairs/sample were 35.4 million. Data were processed by the PCC Applied Bioinformatics Laboratories Shared Resource (ABL). Briefly, reads were adapter and quality trimmed with Trimmomatic (101) and then aligned to the mouse genome (build mm10/GRCm38) using the splice-aware STAR aligner (102). The featureCounts program (103) was utilized to generate counts for each gene, based on how many aligned reads overlapped its exons. Counts were normalized and tested for differential expression, using negative binomial generalized linear models implemented by the DESeq2 R package (104). For pairwise differential expression analysis between tumor groups, normal fallopian tube samples were not included in the model. Statistical analysis and visualization of gene sets were performed using the fgsea (105) and clusterProfiler R packages (106).
Organoid DNA was extracted using the DNeasy Blood & Tissue Kit (Qiagen), according to the manufacturer's instructions. Libraries were prepared using the Nextera DNA Flex Library Kit (Illumina, 96rxn kit, catalog no., 20025520). To save on costs, the manufacturer's protocol was miniaturized by reducing reactions to one fourth of the recommended volumes. Following PCR amplification (five cycles total), water (38 μL) was added to the amplified material (12.5 μL) to increase the volume to 50 μL for the final 1 × Ampure XP Bead Cleanup (Beckman Coulter, #A63882). Library DNA was evaluated on an Agilent TapeStation 2200 with high-sensitivity DNA screen tape to verify library size of approximately 50 bp, and each library was quantified by qPCR using the Kapa-Roche Library Quant Kit (Illumina, catalog no., KK4824) and a Bio-Rad CFX384 real-time PCR system. Libraries were run on half of an SP300 flow cell (paired-end 150 dual indexing run) using the Illumina NovaSeq 6000 System. Sequencing reads were adapter and quality trimmed with Trimmomatic (101) and then aligned to the mouse reference genome (build mm10/GRCm38) using the Burrows-Wheeler Aligner with the BWA-MEM algorithm (107). Low-confidence mappings (mapping quality < 10) and duplicate reads were removed using Sambamba (108). Further local indel realignment and base quality score recalibration were performed using the Genome Analysis Toolkit (109). The average coverage ranged from 1.5 × to 2.2 ×. Copy-number profiles were calculated using Control-FREEC (110) with a fixed window size of 50 kb.
CIBERSORT and quanTIseq Analyses
The immune cell constitution of TCGA samples was inferred by downloading TCGA-OV RNA-seq and CNV data (HTSeq - Counts) from the Genomic Data Commons (GDC) data portal, and normalizing RNA-seq reads to transcripts per million (TPM). Samples with a CNV score of −2 or with a score of −1 and a TPM value within the bottom 33% of all samples were defined as having PTEN, NF1, or BRCA1 loss, respectively. Samples with a CNV score of 2 or with a score of 1 and TPM value within the top 33% of all samples were defined as having CCNE1, AKT2, KRAS, or MYC gain, respectively. Samples were then identified as TPN, TCK/TCAK, or TBM on the basis of the gain/loss status of each gene. To avoid ambiguity, we excluded samples belonging to more than one genotype group. Tumor-infiltrating immune cells were inferred using quanTIseq (50) and CIBERSORT (51) in abs. mode. For deconvolution, we used TIMER2.0 (73) with TPM data as input. To compare groups of samples, we first performed t tests of the abundance of each cell population, and then adjusted P values for multiple comparisons by the Benjamini–Hochberg procedure.
Quantification and Statistical Analysis
Bioinformatic analyses were performed in R (version 3.5.1). All other statistical analyses were performed using GraphPad Prism. Statistical tests used, sample sizes (n), and P values are displayed in the figures and figure legends. P < 0.05 was considered significant.
RNA-seq data have been deposited in the Gene Expression Omnibus database under the accession code GSE147276. sWGS data were deposited in Sequence Read Archive (SRA) under BioProject accession number PRJNA613661. All other data supporting the findings of this study are available within the article or the Supplementary Data, or from the corresponding author upon request.
S. Iyer reports grants from Ludwig Fund for Cancer Research (postdoctoral fellowship) during the conduct of the study and other from AstraZeneca (employment with AstraZeneca as of 26 May 2020 and received no funding from AstraZeneca for the submitted work) outside the submitted work. C.J.R. Foster reports grants from NIH during the conduct of the study. D.A. Levine reports grants from Department of Defense, The Honorable Tina Brozman Foundation, and V Foundation for Cancer Research during the conduct of the study, and Splash Pharmaceuticals, personal fees from Merck and Tesaro/GlaxoSmithKline outside the submitted work, as well as has a patent for patent application US20130078319A1, detection of ovarian cancer issued, and is founder, Nirova BioSense, Inc. B.G. Neel reports grants from The Mary Kay Foundation during the conduct of the study, personal fees and other from Navire Pharma (equity in the company), Northern Biologics, Inc (equity in the company), and Arvinas, Inc. (equity in the company), personal fees from Drinker Biddle & Reath (expert witness for J&J in talc litigation), and other from Moderna, Inc (wife held shares), Regeneron (wife holds shares), Amgen (wife holds shares), and Gilead (wife held shares) outside the submitted work. No disclosures were reported by the other authors.
S. Zhang: Formal analysis, supervision, validation, investigation, writing-original draft, project administration, writing-review and editing. S. Iyer: Formal analysis, validation, investigation, writing-original draft. H. Ran: Validation, investigation. I. Dolgalev: Data curation, validation. S. Gu: Data curation, formal analysis. W. Wei: Formal analysis, validation. C.J.R. Foster: Validation. C.A. Loomis: Validation. N. Olvera: Validation. F. Dao: Resources, validation. D.A. Levine: Resources. R.A. Weinberg: Resources, supervision. B.G. Neel: Supervision, project administration, writing-review and editing.
We thank the PCC Experimental Pathology, Precision Immunology, Microscopy, GTC, Preclinical Imaging Laboratory, and ABL shared resources for technical support, and Dr. Justin Mastroianni (PCC) for assistance with IVIS imaging. We thank Drs. Kwan Ho Tang, Mitchell Geer, and Carmine Fedele (Neel laboratory) and Drs. Myles Brown and Xiaole Shirley Liu (Dana-Farber Cancer Institute) for helpful advice and discussions. Work on this project was supported by grants MOP-191992 from the Canadian Institutes for Health Research and 02-20 from the Mary Kay Foundation to B.G. Neel. S. Zhang was supported by a postdoctoral fellowship from the Ovarian Cancer Research Fund Alliance. S. Iyer was supported by the Ludwig Fund for Cancer Research. D.A. Levine and F. Dao were supported by The Honorable Tina Brozman Foundation and DOD CDMRP award W81XWH-19-1-0232PCC. D.A. Levine and N. Olvera were supported by The V Foundation for Cancer Research and DOD CDMRP award W81XWH-15-1-0429. Shared resources were supported by P30 CA01687. R.A. Weinberg was funded by grants from the NIH (R01 CA0784561 and P01 CA080111), Samuel Waxman Cancer Research Foundation, Breast Cancer Research Foundation, and Ludwig Fund for Cancer Research.