High level co-expression of MECOM, PAX8, SOX17 and WT1 characterizes HGSCs compared to other tumor types (Reddy et al., 2021). To ask whether this lineage-restricted expression pattern is recapitulated in benign tissues, we systematically compared expression patterns of these factors in the Genotype-Tissue Expression (GTEx) and the Cancer Genome Atlas (TCGA) databases. K-means clustering based on MECOM, PAX8, SOX17 and WT1 mRNA expression in 21 normal tissues revealed 5 main clusters (Figure 1A). A distinct cluster consisting of fallopian tube and uterus tissues was characterized by MECOM, PAX8, SOX17 and WT1 expression (MTF^high cluster). A majority 57% (12/21; MTF^low cluster) of normal tissue types did not highly express any of these four transcription factors and 33% (7/21; M^high, M&P^high cluster and SOX17^high cluster) expressed just one or two candidate MTFs. WT1 was expressed in both benign fallopian tube and uteri, but in tumors, was only observed in HGSC and not uterine adenocarcinoma (Figure 1A). Positively correlated expression between these four factors is indicative of a core regulatory circuitry characterized by high-level expression of each factor. Fallopian tube tissues showed the greatest correlations of MECOM, PAX8, SOX17 and WT1 expression compared to all other normal tissue types (Ranked 1/21, Pearson correlations, Figure 1B, Supplementary Table 1). The top 3 tumor tissues from TCGA were represented by gynecological malignancies including CESC (1/27), UCS (2/27) and HGSC (3/27) (Figure 1B, Supplementary Table 2).

Epithelial ovarian cancers comprise multiple histologic subtypes, with diverse cells-of-origin (Cochrane et al., 2017). MECOM, PAX8, SOX17 and WT1 protein expression was quantified in high-grade serous (n = 181-200), clear cell (n = 27-28), endometrioid (n = 18) and mucinous (n = 9) ovarian cancers via immunohistochemistry. Benign ovarian cortex (n = 25) and fallopian tube (n = 12-14) tissues were also evaluated as controls. As previously reported, PAX8 (0.44-0.82 mean positivity rate) and SOX17 (0.10-0.99 mean positivity rate) expression was pervasive in serous, endometrioid and clear cell tumors, with heterogenous expression in mucinous tumors (Figure 1C, Supplementary Table 3) (Dinh et al., 2021; Lin et al., 2022). MECOM (0.44-0.82 mean positivity rate) exhibited a similar expression pattern to PAX8 and SOX17 (Supplementary Table 3). 96% of HGSCs (0.68 mean positivity rate; standard deviation, 0.27), and 86% of fallopian tube tissues (0.19 mean positivity rate; standard deviation, 0.13) expressed WT1, but this marker was less commonly expressed in non-high-grade serous subtypes (Acs et al., 2004; Köbel et al., 2008; Shimizu et al., 2000) (Table 1). A majority of FTSECs (8/11) and HGSCs (142/149) co-stained for all 4 transcription factors (Figure 1D, Supplementary Table 3). Together, these data indicate that co-expression of MECOM, PAX8, SOX17 and WT1 is a unique characteristic of HGSC and fallopian tube precursor tissues.

Super-enhancer association is a prominent feature of MTFs that contributes to the maintained high expression of MTFs in a lineage-enriched manner (Bradner et al., 2017; Chapuy et al., 2013; Chen et al., 2020; Eliades et al., 2018; Jiang et al., 2020; Sanda et al., 2012). We interrogated global landscapes of the active chromatin marked by H3K27ac in FTSECs and HGSCs, including ChIP-seq and newly generated CUT&TAG profiles of primary tissues, cell lines and ex vivo cultures (Corona et al., 2020). MECOM, PAX8, SOX17 and WT1 were associated with super-enhancers in FTSECs and HGSCs (Figure 2A). Inspection of constituent enhancers revealed a marked remodeling of the SOX17 super-enhancer in HGSCs and an acquired acetylation of one of the MECOM promoters (Figure 2A). High-throughput chromosome conformation capture (Hi-C) was conducted in 2 FTSEC (FT246 and FT282) and 2 HGSC (KURAMOCHI and OVCAR4) cell lines to measure the contact loops associated with enhancers in these regions. Hi-C loops were often anchored to constituent enhancers that were part of the super-enhancer regions in FTSECs and HGSCs (Figure 2C). These data suggest that the MECOM, PAX8, SOX17 and WT1 are super-enhancer associated TFs in FTSECs and HGSCs that are likely contributing to the maintained lineage-enriched expression of these TFs from FTSEC to HGSC development.

Cancer cells are often addicted to the expression of MTFs and MTF depletion leads to tumor cell death (Chapuy et al., 2013; Jiang et al., 2020; Witwicki et al., 2018). TF expression was potently depleted in 2 FTSEC (FT246 and FT282) and 2 HGSC (KURAMOCHI and OVCAR4) cell lines using siRNA (Supplementary Table 4). MECOM, PAX8 and SOX17 were indispensable for HGSC survival as previously reported (Figure 3A & 3B) (Reddy et al., 2021). WT1 was also necessary for HGSC survival. By contrast, FTSECs were dependent on PAX8 and WT1 but not MECOM and SOX17 (Figure 3A & 3B).

Cyclin-dependent kinase 7/12/13 inhibitors are hypothesized to have anticancer effects via disruption of the transcriptional initiation machinery that in enriched in super-enhancers regulating MTFs and other lineage-specific genes (Chapuy et al., 2013; Chen et al., 2020; Jiang et al., 2020). OVCAR4 cells were sensitive to THZ1 (CDK7 inhibitor) and THZ531 (CDK7/12 inhibitor) treatment as previously reported (THZ1 IC₅₀ = 44.93nM; n=4, THZ531 IC₅₀ = 63.42nM; n=4), a trend which we replicated in KURAMOCHI cells (THZ1 IC₅₀ = 14nM; n=5, THZ531 IC₅₀ = 48.90nM; n = 2)(Figure 3C) (Reddy et al., 2021). FT246 (THZ1 IC₅₀ = 319.70nM; n = 3, THZ531 IC₅₀ =5033nM, n = 3) and FT282 (THZ1 IC₅₀ = 183.9nM; n = 3, THZ531 IC₅₀ = 4135nM; n =3) cells were 8 and 82 times more resistant for THZ1 and THZ531 treatment respectively, suggesting a tumor-specific mechanism for the anti-proliferative effect of these drugs (Figure 3C). Expression of MECOM, PAX8 and SOX17 was downregulated by 30-75% (P < 0.007, multiple comparison test) in HGSCs treated with short-term (6 hour) treatment of THZ1 at IC₅₀. By contrast, in FTSECs, expression of all four TFs was unaffected by exposure to IC₅₀ THZ1 treatment (Figure 3D). Context-specific inhibition of MTFs was also observed for THZ531, in which the expression of MECOM and PAX8 was 60% lower in OVCAR4 compared to expression in FT282 upon THZ531 treatment at IC₅₀. WT1 was not sensitive to THZ1 or THZ531 treatment in either FTSECs nor HGSCs. These data suggest that the antiproliferative effects of THZ1 and THZ531 in HGSC cells may be due to tumor-specific inhibition of MECOM, PAX8 and SOX17 expression by these drugs.

We dissected the DNA binding pattern of MECOM, PAX8, SOX17 and WT1 in FT282 and KURAMOCHI cells using Cleavage Under Targets and Release using Nuclease (CUT&RUN) (Skene et al., 2018). All four TFs co-occupied the regulatory elements within each TF’s loci in both cellular contexts, further corroborating the involvement of these TFs in autoregulation plus expression of the other MTFs (Figure 4A). MTFs are expected to co-occupy active enhancer regions to control global gene expression programs, consistent with this MECOM, SOX17 and WT1 binding was enriched within PAX8 bound regions in both cellular contexts (Figure 4B & Figure 4C). These regions were associated with active enhancer regions marked by H3K27ac, but did not coincide with heterochromatin regions marked by H3K27me3 (Figure 4B & Figure 4C). In a set intersection analysis of all available TF peaks, the proportion of TF peaks co-bound by at least two master transcription factors was 11% higher in HGSC compared to FTSEC (54%, 32,122 out of 59,349 for FT282; 65%, 37,724 out of 58,000 peaks for KURAMOCHI) (Figure 4D). Furthermore, the number of regions co-bound by all 4 TFs was around 50% higher in KURAMOCHI compared to FT282 (14,547 and 10,588 peaks respectively). Based on genomic and histone context, we annotated the TF peaks as binding within promoters, typical enhancers, super-enhancers, typical methylated regions, and super-methylated regions. Regions bound by all 4 TFs were most frequently defined as typical enhancers (63.7% for FT282; 45.7% for KURAMOCHI), or super-enhancers (10.9% for FT282; 8.95% for KURAMOCHI) and were rarely methylated (typical methylated regions - 0.9% for FT282; 0.72% for KURAMOCHI; super-methylated regions - 0.22% for FT282; 0.082% for KURAMOCHI)(Figure 4E). These data indicate that MECOM, PAX8, SOX17 and WT1 preferentially co-occupy active enhancers across the genome of FTSECs and HGSCs alike.

At the transcriptional level, PAX8 knock-down led to the depletion of MECOM, SOX17 and WT1 expression in both FTSEC and HGSC lines (Figure 4F, Supplementary Table 4). WT1 knock-down also led to the depletion of MECOM, PAX8 and SOX17. Depletion of SOX17 resulted in downregulation of WT1 and MECOM expression, and a reproducible trend towards up-regulation of PAX8 in HGSC models (Figure 4F, Supplementary Table 4). MECOM depletion resulted in upregulated SOX17 expression in HGSC cells as previously reported (Reddy et al., 2021) and also upregulated WT1 expression in HGSC cells (Figure 4F, Supplementary Table 4). In summary, patterns of cross-regulation between these transcription factors is altered in HGSCs compared to FTSECs, with PAX8 playing a dominant role in both contexts but MECOM and SOX17 participating more in the circuitry in tumor cells compared to precursor cells (Figure 4G).

Comparison of TF CUT&RUN binding sites between FT282 and KURAMOCHI showed that for each factor, only 15-23% of peaks overlap between the two cellular contexts (Figure 5A). For example - 18,488 PAX8 peaks were specific to FTSECs, 20,717 were specific to HGSC and 14,544 were in common to the two contexts. In FTSECs, FTSEC-specific and common PAX8 peaks exhibited higher local chromatin activity (H3K27ac positivity) and were more accessible than the HGSC-specific peaks. Remarkably, SOX17, MECOM and WT1 co-localized to the FTSEC-specific and common PAX8 bound regions and were depleted at HGSC-specific peaks (Figure 5B). The inverse pattern was observed for HGSC-specific PAX8 peaks, which were active and more accessible in HGSC cells compared to in FTSECs, and co-occupied by WT1, SOX17 and MECOM in HGSC specifically. TF binding sites specific to FT282 or KURAMOCHI were enriched in typical enhancers or super-enhancers, but not promoters (Figure 5C).

We sought to contextualize these multi-bound TF binding sites against the background of active enhancers gained or lost during HGSC development. Enhancers specific to FT282 or KURAMOCHI were more likely to be bound by TFs compared to the null distribution (P < 1.0×10⁻⁴). A large proportion (3068/6703 FT282 specific enhancers and 3017/9879 KURAMOCHI specific enhancers) were co-bound by all four TFs, confirming that context specific enhancers are characterized by the binding of MECOM, PAX8, SOX17 and WT1 (Figure 5D). Analysis to identify HGSC specific super-enhancer associated genes bound by MECOM, PAX8, SOX17 and WT1 revealed two other candidate master regulators of HGSCs, BHLEH41 and PBX1 (Figure 5F). Overall, although the core-regulatory circuitry of MECOM, PAX8, SOX17 and WT1 are maintained between FTSECs and HGSCs, HGSCs acquire tumor-specific TF DNA binding sites through extensive remodeling of cooperative binding events involving these four factors.

RNA-seq after TF knockdown were integrated with TF CUT&RUN to contrast the cistromes of MECOM, PAX8, SOX17 and WT1 in FTSECs and HGSCs. ERCC spike-in normalization was performed to enable sensitive detection of global transcriptional changes (Lovén et al., 2012). PAX8 preferentially functioned as a transcriptional repressor in tumor and normal cells alike. The dominant effect of MECOM depletion was gene down-regulation, suggesting this factor mainly functions as a transcriptional activator, whereas depletion of SOX17 was associated with a relatively equal number of up- and down-regulated differentially expressed genes (DEGs), implying this gene both activates and represses gene expression. WT1 displayed the most marked context-specific activity as in FTSECs this factor functioned almost exclusively as a transcriptional activator and dominated control of the transcriptome whereas in HGSC cells WT1 regulated ∼50% fewer genes and switched to largely function as a transcriptional repressor (Figure 6A & B, Supplementary Figure 1).

High-confidence target genes for each factor (absolute log₂FC ≥ 0.5 in both cell lines with TF-targeting siRNA versus two controls) were integrated with TF CUT&RUN to link TF binding events to differentially expressed genes within the same topologically associated domains (informed by Hi-C performed in the same cell lines) (Supplementary Tables 5 & 6, Figure 6C). Using this approach, we identified (1) direct TF target genes defined as DEGs with TF binding on the gene promoter, (2) putative TF target genes defined as TF binding within non-promoter regions within the same TAD as DEGs following TF knockdown and (3) indirect TF targets defined as DEGs with no TF binding within the same TAD. This identified 442, 1,397, 339 and 5,411 high-confidence direct or putative target genes for MECOM, PAX8, SOX17 and WT1 respectively in FTSECs and 279, 1,310, 578 and 1,494 direct or putative TF target genes in HGSCs (Supplementary Table 7). The majority of high confidence differentially expressed genes were FTSEC or HGSC specific across all TFs that were either directly, putatively or indirectly regulated (Figure 6D). Supervised clustering of differentially regulated genes revealed that 1,119 target genes repressed by WT1 in HGSCs were enriched in ribosomal RNA processing and citric acid cycle pathways. Of these, 612, 256 and 252 are either directly, putatively or indirectly regulated by WT1 respectively (Figure 6E). In addition, SOX17 acquired 501 down-regulated and 237 up-regulated directly, putatively or indirectly regulated genes that converged on cell cycle and extracellular matrix/platelet degradation pathways, respectively. BRCA1, the human tumor suppressor gene that is germline or somatically mutated in 20-25% of HGSCs (Manchana et al., 2019), is directly regulated by all four TFs in FTSECs. PAX8 and SOX17 maintain positive regulation of BRCA1 in HGSCs, but regulation by MECOM and WT1 is lost in tumors, accompanied by a concomitant loss of WT1 and MECOM binding in an intronic regulatory element (Figure 6 F,G). A tumor specific enhancer co-occupied by TFs makes a contact loop spanning a distance of 300kb to the HGSC oncogene RUNX3 (Lee et al., 2011; Nevadunsky et al., 2009)(Figure 6H). Furthermore, RUNX3 is marked by the repressive marker H3K27me3 with little TF activity in FTSECs, suggesting this region undergoes a demethylation process during tumorigenesis (Figure 6H). TF knock-down followed by RNA sequencing data confirmed that MECOM, PAX8 and SOX17 putatively regulates RUNX3 in HGSC lines (Figure 6I). Thus, the target genes of TFs are reprogrammed between FTSECs and HGSCs, resulting in the dysregulation of key pathways and genes involved in HGSC development.

Pan-cancer TCGA RNA sequence level 3 data was downloaded and curated as described previously (Reddy et al., 2021). Pan-normal GTEx RNA sequence data were downloaded and annotated from the GTEx portal (GTEx Analysis V8; https://www.gtexportal.org/home/datasets). Pearson correlation analysis was conducted by implementing the corr.test() function available from the Psych package in R.

MECOM immunohistochemistry was conducted by the Cedars-Sinai Cancer Biobank and Translational Research Core. Embedded slides were paraffinized at 72°C with EZ solution (Roche, Ventana) and MECOM antigen was retrieved with CC2 pre-diluted Tris solution for 64 minutes at 91℃. MECOM (Cell Signaling, 1:500, AB_2184098) primary antibodies were diluted in Antibody Dilution Buffer (AF1924, R&D), and treated for 32 minutes at 37℃, respectively. MECOM antibodies were detected with an Envision anti-Rabbit polymer (DAKO) for 32 minutes at room temperature. PAX8 and WT1 immunohistochemistry was conducted by the department of Pathology and Laboratory Medicine Division of Anatomic Pathology at Cedars-Sinai Medical Center. Pathology-grade, pre-diluted PAX8 (Cell Marque, no RRID available) and WT1 (Roche, no RRID available) antibodies were detected with the Ultraview DAB detection kit on the Ventana Benchmark Ultra.

Digital image analysis (Qupath) was used to determine the percent of epithelial cells in each sample that stained positive for each transcription factor (percent positivity rate), and the proportion of each tumor type that co-expresses all four factors was determined by measuring for samples with a mean positivity of 0.1 or higher.

Only human cell lines were used in this study and were regularly authenticated using STR profiling, performed at the University of Arizona Genetics Core. FT246 and FT282 cells were cultured in DMEM/F12 with 10% fetal bovine serum and 1× penicillin/streptomycin. OVCAR4 cells were cultured in RPMI 1640 supplemented with 10% bovine serum, 1× nonessential amino acid (NEAA), insulin (11.4μg/ml) and 1× penicillin/streptomycin. KURAMOCHI cells were cultured in RPMI 1640 supplemented with 10% bovine serum and 1× penicillin/streptomycin. All cells were maintained at 37℃ with 5% CO₂ and passed using standard cell culture procedures.

FT246 (RRID:CVCL_UH61), FT282 (RRID:CVCL_A4AX), OVCAR4 (no RRID available) and KURAMOCHI (no RRID available) cells were reverse transfected with two independent non-targeting siRNAs (Horizon Discovery catalogue number: D-001810-10 and one custom control pool) or pooled MECOM, PAX8, SOX17 and WT1 oligonucleotides (Horizon Discovery catalogue numbers: L-006530-02, L-003778-00, L-013028-01, L-009101-00). Cells were seeded for colony formation assays as previously described (Reddy et al., 2021). Each assay was performed at least three times, taking the average of technical triplicate wells from each set.

FT246, FT282, OVCAR4 and KURAMOCHI cells were plated in 96 well plates and treated with THZ1 (Selleck Chemicals), THZ531 (ApexBio), JQ1 (Tocris) or dimethyl sulfoxide as a vehicle control. Each drug was added at 1:3 dilutions starting from 10,000 nM to 1.5nM as triplicates and incubated for 72 hours at 37℃. Cell survival was quantified with the Promega Cell Titer Glo reagent kit following manufacturers guidelines. Each assay was performed at least three times, taking the average of technical triplicates well from each set.

FT246 and KURAMOCHI cells were reverse transfected with siRNAs for two scramble controls, MECOM, PAX8, SOX17 or WT1 as described above. Media were replaced 24 hours later. Total RNA was harvested from an equal number of cells 72 hours post reverse-transfection and spiked with an exogenous ERCC spike-in (Thermo Fisher). Non-stranded Poly-A libraries were constructed and sequenced with 150 base pair sequencing at 40 million reads on the DNB-seq DNA sequencing platform. Hisat2-2.1.0 was used for alignment with the reference genome gencodev26 + ERCC92. The count matrix of gene-level reads were generated using ht-seq count (v0.5.4.3). For DESEQ2 analysis, ERCC read counts were removed and gene expression analysis was performed with DESeq2 (version 1.24.0). For global ERCC spike-in analysis the transcripts per million were calculated from the count matrix followed by loess regression normalization using the affy package in R. Technical duplicate samples were generated on independent days.

CUT&TAG was conducted based on the CUTANA CUT&TAG protocol v1.5 (Epicypher). Isolated nuclei from primary fallopian tube secretory epithelial cells bound to concalavin A beads were incubated with primary antibody that detects H3K27ac (0.7μg, Diagenode, RRID:AB_2637079) at 4℃ overnight. On day 2 samples were incubated with secondary antibody and washed. Samples were then incubated with pAG-Tn5 for 1 hour, washed and placed in a tagmentation buffer followed by 1hr incubation at 37℃. Beads were then washed with TAPS buffer followed by incubation in SDS release buffer for 1 hour at 58℃. The reaction was quenched with SDS Quench buffer. PCR was conducted and DNA cleanup was performed with 1.3× Ampure beads. Samples were sequenced at the CSMC AGCT Genomics Core facility. Data processing was conducted following the workflow described by Zhen et al https://yezhengstat.github.io/CUTTag_tutorial/. Briefly, FASTQ files were aligned to hg38 with Bowtie. Duplicates and reads with alignment quality scores of 2 or lower were removed. Peaks were called using SEACR with the parameters set to --0.01 --norm -- stringent.

CUT&RUN was performed based on the CUTANA CUT&RUN protocol v1.6 (Epicypher). Isolated nuclei from FT282 and KURAMOCHI were bound to Concalavin A beads and incubated overnight at 4℃ with primary antibody for H3K27me3 (Cell Signaling, RRID:AB_2798370), H3K27ac (Diagenode, RRID:AB_2637079), MECOM (Cell Signaling, AB_2184098), PAX8 (Novus, RRID:AB_2283498), SOX17 (Abcam, RRID:AB_2801385), WT1 (Santa Cruz, RRID:AB_632611). Samples were washed after overnight incubation and treated with pAG-Mnase for 10 minutes at room temperature followed by activation by calcium chloride for 2 hours at 4℃. The reaction was quenched with STOP buffer and DNA was extracted with a DNA purification kit (Epicypher). Library preparation was conducted with the NEBnext Ultra II kit with the following protocol adjustments. The second incubation during the end prep was extended to 1 hour at 50 degrees and all SPRI select steps were conducted at 1.5× volume. Cycling parameters were set to parameters as instructed in the CUT&RUN protocol v1.6 (Epicypher). Processing and peak calls were made using the same pipeline as the CUT&TAG pipeline.

FT246, FT282, KURAMOCHI and OVCAR4 Hi-C were conducted and processed by Active Motif using the Arima-Hi-C kit and library prep performed with technical duplicates (Arima Genomics). Prepped libraries were sequenced to a depth of 1 billion reads. FASTQ files were processed with the Juicer software package with default parameters for the generation of intermediate .hic files. Loop and TAD calls were generated using HiCCUPs as described previously (Rao et al., 2014).

All data have been deposited into the Gene Expression Omnibus (https://www.ncbi.nlm.nih.gov/geo/). RNA-seq following transcription factor knockdown and siRNA control treatments are available accessions GSE151316 (FT282 and OVCAR4) and GSE227276 (FT246 and KURAMOCHI); CUT&RUN and ATAC-seq data are available under accession GSE227779. The Hi-C data is available under accession GSE229408.

This study did not generate any new materials.