Glioblastomas acquire myeloid-affiliated transcriptional programs via epigenetic immunoediting to elicit immune evasion.

Glioblastoma multiforme (GBM) is an aggressive brain tumor for which current immunotherapy approaches have been unsuccessful. Here, we explore the mechanisms underlying immune evasion in GBM. By serially transplanting GBM stem cells (GSCs) into immunocompetent hosts, we uncover an acquired capability of GSCs to escape immune clearance by establishing an enhanced immunosuppressive tumor microenvironment. Mechanistically, this is not elicited via genetic selection of tumor subclones, but through an epigenetic immunoediting process wherein stable transcriptional and epigenetic changes in GSCs are enforced following immune attack. These changes launch a myeloid-affiliated transcriptional program, which leads to increased recruitment of tumor-associated macrophages. Furthermore, we identify similar epigenetic and transcriptional signatures in human mesenchymal subtype GSCs. We conclude that epigenetic immunoediting may drive an acquired immune evasion program in the most aggressive mesenchymal GBM subtype by reshaping the tumor immune microenvironment.

DNMT3B PWWP mutations cause hypermethylation of heterochromatin.

The correct establishment of DNA methylation patterns is vital for mammalian development and is achieved by the de novo DNA methyltransferases DNMT3A and DNMT3B. DNMT3B localises to H3K36me3 at actively transcribing gene bodies via its PWWP domain. It also functions at heterochromatin through an unknown recruitment mechanism. Here, we find that knockout of DNMT3B causes loss of methylation predominantly at H3K9me3-marked heterochromatin and that DNMT3B PWWP domain mutations or deletion result in striking increases of methylation in H3K9me3-marked heterochromatin. Removal of the N-terminal region of DNMT3B affects its ability to methylate H3K9me3-marked regions. This region of DNMT3B directly interacts with HP1α and facilitates the bridging of DNMT3B with H3K9me3-marked nucleosomes in vitro. Our results suggest that DNMT3B is recruited to H3K9me3-marked heterochromatin in a PWWP-independent manner that is facilitated by the protein's N-terminal region through an interaction with a key heterochromatin protein. More generally, we suggest that DNMT3B plays a role in DNA methylation homeostasis at heterochromatin, a process which is disrupted in cancer, aging and Immunodeficiency, Centromeric Instability and Facial Anomalies (ICF) syndrome.

Genome-wide single-molecule analysis of long-read DNA methylation reveals heterogeneous patterns at heterochromatin that reflect nucleosome organisation.

High-throughput sequencing technology is central to our current understanding of the human methylome. The vast majority of studies use chemical conversion to analyse bulk-level patterns of DNA methylation across the genome from a population of cells. While this technology has been used to probe single-molecule methylation patterns, such analyses are limited to short reads of a few hundred basepairs. DNA methylation can also be directly detected using Nanopore sequencing which can generate reads measuring megabases in length. However, thus far these analyses have largely focused on bulk-level assessment of DNA methylation. Here, we analyse DNA methylation in single Nanopore reads from human lymphoblastoid cells, to show that bulk-level metrics underestimate large-scale heterogeneity in the methylome. We use the correlation in methylation state between neighbouring sites to quantify single-molecule heterogeneity and find that heterogeneity varies significantly across the human genome, with some regions having heterogeneous methylation patterns at the single-molecule level and others possessing more homogeneous methylation patterns. By comparing the genomic distribution of the correlation to epigenomic annotations, we find that the greatest heterogeneity in single-molecule patterns is observed within heterochromatic partially methylated domains (PMDs). In contrast, reads originating from euchromatic regions and gene bodies have more ordered DNA methylation patterns. By analysing the patterns of single molecules in more detail, we show the existence of a nucleosome-scale periodicity in DNA methylation that accounts for some of the heterogeneity we uncover in long single-molecule DNA methylation patterns. We find that this periodic structure is partially masked in bulk data and correlates with DNA accessibility as measured by nanoNOMe-seq, suggesting that it could be generated by nucleosomes. Our findings demonstrate the power of single-molecule analysis of long-read data to understand the structure of the human methylome.

Elevated FOXG1 and SOX2 in glioblastoma enforces neural stem cell identity through transcriptional control of cell cycle and epigenetic regulators.

Glioblastoma multiforme (GBM) is an aggressive brain tumor driven by cells with hallmarks of neural stem (NS) cells. GBM stem cells frequently express high levels of the transcription factors FOXG1 and SOX2. Here we show that increased expression of these factors restricts astrocyte differentiation and can trigger dedifferentiation to a proliferative NS cell state. Transcriptional targets include cell cycle and epigenetic regulators (e.g., Foxo3, Plk1, Mycn, Dnmt1, Dnmt3b, and Tet3). Foxo3 is a critical repressed downstream effector that is controlled via a conserved FOXG1/SOX2-bound cis-regulatory element. Foxo3 loss, combined with exposure to the DNA methylation inhibitor 5-azacytidine, enforces astrocyte dedifferentiation. DNA methylation profiling in differentiating astrocytes identifies changes at multiple polycomb targets, including the promoter of Foxo3 In patient-derived GBM stem cells, CRISPR/Cas9 deletion of FOXG1 does not impact proliferation in vitro; however, upon transplantation in vivo, FOXG1-null cells display increased astrocyte differentiation and up-regulate FOXO3. In contrast, SOX2 ablation attenuates proliferation, and mutant cells cannot be expanded in vitro. Thus, FOXG1 and SOX2 operate in complementary but distinct roles to fuel unconstrained self-renewal in GBM stem cells via transcriptional control of core cell cycle and epigenetic regulators.

An epigenetic predictor of death captures multi-modal measures of brain health.

Individuals of the same chronological age exhibit disparate rates of biological ageing. Consequently, a number of methodologies have been proposed to determine biological age and primarily exploit variation at the level of DNA methylation (DNAm). A novel epigenetic clock, termed 'DNAm GrimAge' has outperformed its predecessors in predicting the risk of mortality as well as many age-related morbidities. However, the association between DNAm GrimAge and cognitive or neuroimaging phenotypes remains unknown. We explore these associations in the Lothian Birth Cohort 1936 (n = 709, mean age 73 years). Higher DNAm GrimAge was strongly associated with all-cause mortality over the eighth decade (Hazard Ratio per standard deviation increase in GrimAge: 1.81, P < 2.0 × 10-16). Higher DNAm GrimAge was associated with lower age 11 IQ (ß = -0.11), lower age 73 general cognitive ability (ß = -0.18), decreased brain volume (ß = -0.25) and increased brain white matter hyperintensities (ß = 0.17). There was tentative evidence for a longitudinal association between DNAm GrimAge and cognitive decline from age 70 to 79. Sixty-nine of 137 health- and brain-related phenotypes tested were significantly associated with GrimAge. Adjusting all models for childhood intelligence attenuated to non-significance a small number of associations (12/69 associations; 6 of which were cognitive traits), but not the association with general cognitive ability (33.9% attenuation). Higher DNAm GrimAge associates with lower cognitive ability and brain vascular lesions in older age, independently of early-life cognitive ability. This epigenetic predictor of mortality associates with different measures of brain health and may aid in the prediction of age-related cognitive decline.

Activation of transcription factor circuity in 2i-induced ground state pluripotency is independent of repressive global epigenetic landscapes.

Mouse embryonic stem cells (mESCs) cultured with MEK/ERK and GSK3ß (2i) inhibitors transition to ground state pluripotency. Gene expression changes, redistribution of histone H3K27me3 profiles and global DNA hypomethylation are hallmarks of 2i exposure, but it is unclear whether epigenetic alterations are required to achieve and maintain ground state or occur as an outcome of 2i signal induced changes. Here we show that ESCs with three epitypes, WT, constitutively methylated, or hypomethylated, all undergo comparable morphological, protein expression and transcriptome changes independently of global alterations of DNA methylation levels or changes in H3K27me3 profiles. Dazl and Fkbp6 expression are induced by 2i in all three epitypes, despite exhibiting hypermethylated promoters in constitutively methylated ESCs. We identify a number of activated gene promoters that undergo 2i dependent loss of H3K27me3 in all three epitypes, however genetic and pharmaceutical inhibition experiments show that H3K27me3 is not required for their silencing in non-2i conditions. By separating and defining their contributions, our data suggest that repressive epigenetic systems play minor roles in mESC self-renewal and naïve ground state establishment by core sets of dominant pluripotency associated transcription factor networks, which operate independently from these epigenetic processes.

Variable outcome and methylation status according to CEBPA mutant type in double-mutated acute myeloid leukemia patients and the possible implications for treatment.

Although CEBPA double-mutated (CEBPADM) acute myeloid leukemia is considered to be a favorable-risk disease, relapse remains a major cause of treatment failure. Most CEBPADM patients have a classic biallelic mutant combination with an N-terminal mutation leading to production of p30 protein plus a C-terminal loss-of-function in-frame indel mutation (CEBPAClassic-DM), but approximately one-third of cases have one or more non-classic mutations, with diverse combinations reported, and there is little information on the consequences of such mutants. We evaluated outcome in a cohort of 104 CEBPADM patients, 79 CEBPAClassic-DM and 25 with non-classic mutants, and found that the latter may have poorer survival (5-year overall survival 64% vs. 46%; P=0.05), particularly post relapse (41% vs. 0%; P=0.02). However, for this analysis, all non-classic cases were grouped together, irrespective of mutant combination. As CEBPADM cases have been reported to be hypermethylated, we used methylation profiling to assess whether this could segregate the different mutants. We developed a CEBPAClassic-DM methylation signature from a preliminary cohort of 10 CEBPADM (including 8 CEBPAClassic-DM) and 30 CEBPA wild-type (CEBPAWT) samples, and independently validated the signature in 17 CEBPAClassic-DM cases. Assessment of the signature in 16 CEBPADM cases with different non-classic mutant combinations showed that only 31% had a methylation profile equivalent to CEBPAClassic-DM whereas for 69% the profile was either intermediate between CEBPAClassic-DM and CEBPAWT or equivalent to CEBPAWT These results suggest that CEBPADM cases with non-classic mutants may be functionally different from those with CEBPAClassic-DM mutants, and should not automatically be included in the same prognostic group. (AML12 is registered under ISRCTN17833622 and AML15 under ISRCTN17161961).

Analysis of active and inactive X chromosome architecture reveals the independent organization of 30 nm and large-scale chromatin structures.

Using a genetic model, we present a high-resolution chromatin fiber analysis of transcriptionally active (Xa) and inactive (Xi) X chromosomes packaged into euchromatin and facultative heterochromatin. Our results show that gene promoters have an open chromatin structure that is enhanced upon transcriptional activation but the Xa and the Xi have similar overall 30 nm chromatin fiber structures. Therefore, the formation of facultative heterochromatin is dependent on factors that act at a level above the 30 nm fiber and transcription does not alter bulk chromatin fiber structures. However, large-scale chromatin structures on Xa are decondensed compared with the Xi and transcription inhibition is sufficient to promote large-scale chromatin compaction. We show a link between transcription and large-scale chromatin packaging independent of the bulk 30 nm chromatin fiber and propose that transcription, not the global compaction of 30 nm chromatin fibers, determines the cytological appearance of large-scale chromatin structures.

Ring1B compacts chromatin structure and represses gene expression independent of histone ubiquitination.

How polycomb group proteins repress gene expression in vivo is not known. While histone-modifying activities of the polycomb repressive complexes (PRCs) have been studied extensively, in vitro data have suggested a direct activity of the PRC1 complex in compacting chromatin. Here, we investigate higher-order chromatin compaction of polycomb targets in vivo. We show that PRCs are required to maintain a compact chromatin state at Hox loci in embryonic stem cells (ESCs). There is specific decompaction in the absence of PRC2 or PRC1. This is due to a PRC1-like complex, since decompaction occurs in Ring1B null cells that still have PRC2-mediated H3K27 methylation. Moreover, we show that the ability of Ring1B to restore a compact chromatin state and to repress Hox gene expression is not dependent on its histone ubiquitination activity. We suggest that Ring1B-mediated chromatin compaction acts to directly limit transcription in vivo.

Transcription factor binding predicts histone modifications in human cell lines.

Gene expression in higher organisms is thought to be regulated by a complex network of transcription factor binding and chromatin modifications, yet the relative importance of these two factors remains a matter of debate. Here, we show that a computational approach allows surprisingly accurate prediction of histone modifications solely from knowledge of transcription factor binding both at promoters and at potential distal regulatory elements. This accuracy significantly and substantially exceeds what could be achieved by using DNA sequence as an input feature. Remarkably, we show that transcription factor binding enables strikingly accurate predictions across different cell lines. Analysis of the relative importance of specific transcription factors as predictors of specific histone marks recapitulated known interactions between transcription factors and histone modifiers. Our results demonstrate that reported associations between histone marks and gene expression may be indirect effects caused by interactions between transcription factors and histone-modifying complexes.

DNA methylation reprogramming in cancer: does it act by re-configuring the binding landscape of Polycomb repressive complexes?

DNA methylation is a repressive epigenetic mark vital for normal development. Recent studies have uncovered an unexpected role for the DNA methylome in ensuring the correct targeting of the Polycomb repressive complexes throughout the genome. Here, we discuss the implications of these findings for cancer, where DNA methylation patterns are widely reprogrammed. We speculate that cancer-associated reprogramming of the DNA methylome leads to an altered Polycomb binding landscape, influencing gene expression by multiple modes. As the Polycomb system is responsible for the regulation of genes with key roles in cell fate decisions and cell cycle regulation, DNA methylation induced Polycomb mis-targeting could directly drive carcinogenesis and disease progression.

Tissue type is a major modifier of the 5-hydroxymethylcytosine content of human genes.

The discovery of substantial amounts of 5-hydroxymethylcytosine (5hmC), formed by the oxidation of 5-methylcytosine (5mC), in various mouse tissues and human embryonic stem (ES) cells has necessitated a reevaluation of our knowledge of 5mC/5hmC patterns and functions in mammalian cells. Here, we investigate the tissue specificity of both the global levels and locus-specific distribution of 5hmC in several human tissues and cell lines. We find that global 5hmC content of normal human tissues is highly variable, does not correlate with global 5mC content, and decreases rapidly as cells from normal tissue adapt to cell culture. Using tiling microarrays to map 5hmC levels in DNA from normal human tissues, we find that 5hmC patterns are tissue specific; unsupervised hierarchical clustering based solely on 5hmC patterns groups independent biological samples by tissue type. Moreover, in agreement with previous studies, we find 5hmC associated primarily, but not exclusively, with the body of transcribed genes, and that within these genes 5hmC levels are positively correlated with transcription levels. However, using quantitative 5hmC-qPCR, we find that the absolute levels of 5hmC for any given gene are primarily determined by tissue type, gene expression having a secondary influence on 5hmC levels. That is, a gene transcribed at a similar level in several different tissues may have vastly different levels of 5hmC (>20-fold) dependent on tissue type. Our findings highlight tissue type as a major modifier of 5hmC levels in expressed genes and emphasize the importance of using quantitative analyses in the study of 5hmC levels.

Expression of a large LINE-1-driven antisense RNA is linked to epigenetic silencing of the metastasis suppressor gene TFPI-2 in cancer.

LINE-1 retrotransposons are abundant repetitive elements of viral origin, which in normal cells are kept quiescent through epigenetic mechanisms. Activation of LINE-1 occurs frequently in cancer and can enable LINE-1 mobilization but also has retrotransposition-independent consequences. We previously reported that in cancer, aberrantly active LINE-1 promoters can drive transcription of flanking unique sequences giving rise to LINE-1 chimeric transcripts (LCTs). Here, we show that one such LCT, LCT13, is a large transcript (>300 kb) running antisense to the metastasis-suppressor gene TFPI-2. We have modelled antisense RNA expression at TFPI-2 in transgenic mouse embryonic stem (ES) cells and demonstrate that antisense RNA induces silencing and deposition of repressive histone modifications implying a causal link. Consistent with this, LCT13 expression in breast and colon cancer cell lines is associated with silencing and repressive chromatin at TFPI-2. Furthermore, we detected LCT13 transcripts in 56% of colorectal tumours exhibiting reduced TFPI-2 expression. Our findings implicate activation of LINE-1 elements in subsequent epigenetic remodelling of surrounding genes, thus hinting a novel retrotransposition-independent role for LINE-1 elements in malignancy.

Lactate, a product of glycolytic metabolism, inhibits histone deacetylase activity and promotes changes in gene expression.

Chemical inhibitors of histone deacetylase (HDAC) activity are used as experimental tools to induce histone hyperacetylation and deregulate gene transcription, but it is not known whether the inhibition of HDACs plays any part in the normal physiological regulation of transcription. Using both in vitro and in vivo assays, we show that lactate, which accumulates when glycolysis exceeds the cell's aerobic metabolic capacity, is an endogenous HDAC inhibitor, deregulating transcription in an HDAC-dependent manner. Lactate is a relatively weak inhibitor (IC(50) 40 mM) compared to the established inhibitors trichostatin A and butyrate, but the genes deregulated overlap significantly with those affected by low concentrations of the more potent inhibitors. HDAC inhibition causes significant up and downregulation of genes, but genes that are associated with HDAC proteins are more likely to be upregulated and less likely to be downregulated than would be expected. Our results suggest that the primary effect of HDAC inhibition by endogenous short-chain fatty acids like lactate is to promote gene expression at genes associated with HDAC proteins. Therefore, we propose that lactate may be an important transcriptional regulator, linking the metabolic state of the cell to gene transcription.

Transcriptionally repressed genes become aberrantly methylated and distinguish tumors of different lineages in breast cancer.

Aberrant promoter hypermethylation is frequently observed in cancer. The potential for this mechanism to contribute to tumor development depends on whether the genes affected are repressed because of their methylation. Many aberrantly methylated genes play important roles in development and are bivalently marked in ES cells, suggesting that their aberrant methylation may reflect developmental processes. We investigated this possibility by analyzing promoter methylation in 19 breast cancer cell lines and 47 primary breast tumors. In cell lines, we defined 120 genes that were significantly repressed in association with methylation (SRAM). These genes allowed the unsupervised segregation of cell lines into epithelial (EPCAM+ve) and mesenchymal (EPCAM-ve) lineages. However, the methylated genes were already repressed in normal cells of the same lineage, and >90% could not be derepressed by treatment with 5-aza-2'-deoxycytidine. The tumor suppressor genes APC and CDH1 were among those methylated in a lineage-specific fashion. As predicted by the epithelial nature of most breast tumors, SRAM genes that were methylated in epithelial cell lines were frequently aberrantly methylated in primary tumors, as were genes specifically repressed in normal epithelial cells. An SRAM gene expression signature also correctly identified the rare claudin-low and metaplastic tumors as having mesenchymal characteristics. Our findings implicate aberrant DNA methylation as a marker of cell lineage rather than tumor progression and suggest that, in most cases, it does not cause the repression with which it is associated.

Germline de novo mutations in families with Mendelian cancer syndromes caused by defects in DNA repair.

DNA repair defects underlie many cancer syndromes. We tested whether de novo germline mutations (DNMs) are increased in families with germline defects in polymerase proofreading or base excision repair. A parent with a single germline POLE or POLD1 mutation, or biallelic MUTYH mutations, had 3-4 fold increased DNMs over sex-matched controls. POLE had the largest effect. The DNMs carried mutational signatures of the appropriate DNA repair deficiency. No DNM increase occurred in offspring of MUTYH heterozygous parents. Parental DNA repair defects caused about 20-150 DNMs per child, additional to the ~60 found in controls, but almost all extra DNMs occurred in non-coding regions. No increase in post-zygotic mutations was detected, excepting a child with bi-allelic MUTYH mutations who was excluded from the main analysis; she had received chemotherapy and may have undergone oligoclonal haematopoiesis. Inherited DNA repair defects associated with base pair-level mutations increase DNMs, but phenotypic consequences appear unlikely.

Cluster mean-field theory accurately predicts statistical properties of large-scale DNA methylation patterns.

The accurate establishment and maintenance of DNA methylation patterns is vital for mammalian development and disruption to these processes causes human disease. Our understanding of DNA methylation mechanisms has been facilitated by mathematical modelling, particularly stochastic simulations. Megabase-scale variation in DNA methylation patterns is observed in development, cancer and ageing and the mechanisms generating these patterns are little understood. However, the computational cost of stochastic simulations prevents them from modelling such large genomic regions. Here, we test the utility of three different mean-field models to predict summary statistics associated with large-scale DNA methylation patterns. By comparison to stochastic simulations, we show that a cluster mean-field model accurately predicts the statistical properties of steady-state DNA methylation patterns, including the mean and variance of methylation levels calculated across a system of CpG sites, as well as the covariance and correlation of methylation levels between neighbouring sites. We also demonstrate that a cluster mean-field model can be used within an approximate Bayesian computation framework to accurately infer model parameters from data. As mean-field models can be solved numerically in a few seconds, our work demonstrates their utility for understanding the processes underpinning large-scale DNA methylation patterns.

Local CpG density affects the trajectory and variance of age-associated DNA methylation changes.

BACKGROUND: DNA methylation is an epigenetic mark associated with the repression of gene promoters. Its pattern in the genome is disrupted with age and these changes can be used to statistically predict age with epigenetic clocks. Altered rates of aging inferred from these clocks are observed in human disease. However, the molecular mechanisms underpinning age-associated DNA methylation changes remain unknown. Local DNA sequence can program steady-state DNA methylation levels, but how it influences age-associated methylation changes is unknown. RESULTS: We analyze longitudinal human DNA methylation trajectories at 345,895 CpGs from 600 individuals aged between 67 and 80 to understand the factors responsible for age-associated epigenetic changes at individual CpGs. We show that changes in methylation with age occur at 182,760 loci largely independently of variation in cell type proportions. These changes are especially apparent at 8322 low CpG density loci. Using SNP data from the same individuals, we demonstrate that methylation trajectories are affected by local sequence polymorphisms at 1487 low CpG density loci. More generally, we find that low CpG density regions are particularly prone to change and do so variably between individuals in people aged over 65. This differs from the behavior of these regions in younger individuals where they predominantly lose methylation. CONCLUSIONS: Our results, which we reproduce in two independent groups of individuals, demonstrate that local DNA sequence influences age-associated DNA methylation changes in humans in vivo. We suggest that this occurs because interactions between CpGs reinforce maintenance of methylation patterns in CpG dense regions.

In Vivo Modeling of Patient Genetic Heterogeneity Identifies New Ways to Target Cholangiocarcinoma.

Intrahepatic cholangiocarcinoma (ICC) is an aggressive malignancy of the bile ducts within the liver characterized by high levels of genetic heterogeneity. In the context of such genetic variability, determining which oncogenic mutations drive ICC growth has been difficult, and developing modes of patient stratification and targeted therapies remains challenging. Here we model the interactions between rare mutations with more common driver genes and combine in silico analysis of patient data with highly multiplexed in vivo CRISPR-spCas9 screens to perform a functional in vivo study into the role genetic heterogeneity plays in driving ICC. Novel tumor suppressors were uncovered, which, when lost, cooperate with the RAS oncoprotein to drive ICC growth. Focusing on a set of driver mutations that interact with KRAS to initiate aggressive, sarcomatoid-type ICC revealed that tumor growth relies on Wnt and PI3K signaling. Pharmacologic coinhibition of Wnt and PI3K in vivo impeded ICC growth regardless of mutational profile. Therefore, Wnt and PI3K activity should be considered as a signature by which patients can be stratified for treatment independent of tumor genotype, and inhibitors of these pathways should be levied to treat ICC. SIGNIFICANCE: This work shows that, despite significant genetic heterogeneity, intrahepatic cholangiocarcinoma relies on a limited number of signaling pathways to grow, suggesting common therapeutic vulnerabilities across patients.

Modelling genetic and clinical heterogeneity in epithelial ovarian cancers.

The biology underlying early-stage epithelial ovarian cancer (EOC) development is poorly understood. Identifying biomarkers associated with early-stage disease could have a significant impact on reducing mortality. Here, we describe establishment of a three-dimensional (3D) in vitro genetic model of EOC initiation and early-stage neoplastic progression. Normal primary ovarian epithelial (POE) cells, immortalized using hTERT (immortalised ovarian epithelial [IOE] cells), were partially transformed by overexpressing the CMYC oncogene (IOE(CMYC) cells). Subsequent expression of mutant alleles of KRAS (KRAS(G12V)) or BRAF (BRAF(V600E)) created double-mutant lines (IOE(CMYC.KRAS) and IOE(CMYC.BRAF)). The transformed phenotype of IOE(CMYC) cells was further enhanced in concert with KRAS(G12V)/BRAF(V600E) expression, as in vitro analyses indicated that IOE(CMYC) cells had undergone morphological and phenotypic changes characteristic of neoplastic progression. When cultured as 3D spheroids, IOE cells underwent growth arrest, reminiscent of nonproliferative, unstimulated POE in vivo. In contrast, IOSE(CMYC+BRAF/KRAS) cells formed highly proliferative, poly-aggregate spheroid structures, showing increased expression of the Wilms tumour 1 tumourigenic marker and MIB1 proliferation marker. Transcriptomic analyses identified different gene expression profiles between the different cell lines and novel candidate genes (e.g. RGS4, CTGF and THBS1) that are somatically altered in EOCs. Gene expression signatures were compared with signatures from primary EOCs; tumours with IOE(CMYC) 'like' signatures were more likely to be high grade (P = 0.018); tumours with BRAF signatures were associated with improved relapse-free survival (P = 0.003). In conclusion, we have established in vitro 3D models of early-stage EOCs, which reflect genetic and phenotypic heterogeneity of the disease. Molecular genetic characteristics of these models correlated with molecular and clinical features of primary EOCs.