A distinct mechanism of epigenetic reprogramming silences PAX2 and initiates endometrial carcinogenesis

Abstract

Functional inactivation of tumor suppressor genes drives cancer initiation, progression, and treatment responses. Most tumor suppressor genes are inactivated through 1 of 2 well-characterized mechanisms: DNA-level mutations, such as point mutations or deletions, and promoter DNA hypermethylation. Here, we report a distinct third mechanism of tumor suppressor inactivation based on alterations to the histone rather than DNA code. We demonstrated that PAX2 is an endometrial tumor suppressor recurrently inactivated by a distinct epigenetic reprogramming event in more than 80% of human endometrial cancers. Integrative transcriptomic, epigenomic, 3D genomic, and machine learning analyses showed that PAX2 transcriptional downregulation is associated with replacement of open/active chromatin features (H3K27ac/H3K4me3) with inaccessible/repressive chromatin features (H3K27me3) in a framework dictated by 3D genome organization. The spread of the repressive H3K27me3 signal resembled a pearl necklace, with its length modulated by cohesin loops, thereby preventing transcriptional dysregulation of neighboring genes. This mechanism, involving the loss of a promoter-proximal superenhancer, was shown to underlie transcriptional silencing of PAX2 in human endometrial cancers. Mouse and human preclinical models established PAX2 as a potent endometrial tumor suppressor. Functionally, PAX2 loss promoted endometrial carcinogenesis by rewiring the transcriptional landscape via global enhancer reprogramming. The discovery that most endometrial cancers originate from a recurring epigenetic alteration carries profound implications for their diagnosis and treatment.

Keywords: Mouse models; Obstetrics/gynecology; Oncology; Reproductive biology; Tumor suppressors.

Figure 1. Emergence of PAX2-deficient clones in endometrial epithelium is age dependent and associated with carcinogenesis.

(A) PAX2 immunolocalization of endometrial tissue section from younger (18–25 y/o) patient group. No PAX2-deficient clones were detected across entire specimen; representative region shown. Scale bar: 200 μm. (B) EC from 63 y/o patient showing complete loss of PAX2, which occurs in 80% of EC. Residual normal (non-neoplastic) gland in lower left corner underscores striking and complete loss of PAX2 expression in EC. Scale bar: 200 μm. (C) Endometrial tissue section from older (44–45 y/o) patient group. Dashed red circle highlights single gland in entire specimen with PAX2 loss; only portion of section shown. Right panel, magnification of boxed area showing complete (clonal) loss in all cells of the gland. Scale bars: 100 μm. (D) Parts of whole plots show cases with PAX2 protein loss among younger (n = 27) and older (n = 32) patients. P value per 2-sided Fisher’s exact test. (E) PAX2 expression in normal proliferative endometrium and loss in most (>80%) ECs of grades 1–3. G1, grade 1; G2, grade 2; G3, grade 3. Scale bars: 100 μm. (F) Box-and-whisker plots of PAX2 protein expression levels per H-scores in normal endometrium (n = 8) and ECs (grade 1, n = 45; grade 2, n = 37; grade 3, n = 33). ****P < 0.0001 per Dunnett’s multiple-comparison test. (G) Western blot analysis of human EC cell line panel (n = 13) with same PAX2 monoclonal antibody used for immunolocalization. Only 2/13 lines (AN3CA and EI) expressed normal levels of PAX2, consistent with the observed loss in approximately 80% of primary EC. (H) PAX2 mRNA expression levels across human EC lines per qPCR (n = 3, mean ± SEM).

Figure 2. PAX2 protein loss is due to transcriptional silencing specific to PAX2 locus.

(A) Top panels: PAX2-deficient EIN. Single gland of residual normal endometrium serves as internal positive control for PAX2 expression. EIN glands show complete loss of PAX2 protein. Scale bars: 200 μm. Bottom panel: break-apart FISH for PAX2 locus with flanking BAC probes 162 kbp 5′ (Spectrum Orange) and 188 kbp 3′ (Spectrum Green) from PAX2 gene body in a PAX2-deficient gland. No absent or physically separate orange and green signals are evident. White dashed line demarcates epithelial/stromal boundary. EIN from n = 12 patients analyzed with similar results. (B) Immunolocalization and RNA ISH of PAX2 loss of expression in serial sections. Top panels: normal human endometrium with single isolated PAX2-deficient gland. Bottom panels: EIN with diffuse PAX2 protein loss. Single entrapped normal (non-neoplastic) gland expressing PAX2 protein (internal positive control). n = 6 normal endometria with PAX2-null clones and n = 6 EIN with diffuse PAX2 loss were analyzed, with similar results. Scale bars: 50 μm. (C) Expression of individual genes adjacent to PAX2 locus across EC lines per RNA-seq. The y axis shows mRNA abundance as log₂ transcripts/million (TPM). Both PAX2-expressing EC lines are indicated with green bars. (D and E) Targeted methyl-Seq of PAX2 (230 kbp) and MLH1 (100 kbp) coding and flanking genomic regions. CpG islands per UCSC Genome Browser (GRCh37/hg19) shown for both loci (54). Integrated Genomics Viewer shows methylation peaks across both loci. Cell lines highlighted in blue are silenced for the respective locus (PAX2 or MLH1). (D) Methyl-seq of PAX2. Neither large- nor small-scale methylation events correlated with silencing. (E) Methyl-Seq of MLH1. Silencing correlated with strong methylation signal in single CpG island known to account for MLH1 silencing in EC.

Figure 3. Reversal of PAX2 silencing by CRISPRa.

(A) CRISPRa strategy targeting dCas9-VPR to PAX2 locus with sgRNA (created in BioRender). TSS, transcription start site. (B) All-in-one lentiviral construct with sgRNA and dCas9-VPR (created in BioRender). (C) Relative positions of 4 sgRNAs to TSS (arrow). Start ATG codon in first exon (unfilled rectangle) is shown relative to 1,598 bp PAX2 proximal-promoter enhancer region (blue hatched rectangle). The 20 bp sgRNA 4 is 133 bp 3′ of the TSS. (D) PAX2 mRNA expression by qPCR in EC lines following CRISPRa with nontargeting control (NTC) and PAX2-specific sgRNAs (n = 3, mean ± SEM, multiple t tests). FC, fold change. (E) Western blot analysis of PAX2 expression in Ishikawa cells after CRISPRa with 4 sgRNAs. (F) qPCR of PAX2 mRNA in Ishikawa cells following CRISPRa with NTC and sgRNA 4 (n = 3, mean ± SEM, multiple t tests). (G) Western analysis of PAX2 expression in EC lines subjected to CRISPRa with sgRNA guide 4. (H) Cell proliferation following CRISPRa in Ishikawa (ISK) and HEC-1-B by live-cell imaging (n = 3, mean ± SEM, unpaired, 2-tailed t test). For all panels, *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001.

Figure 4. PAX2 silencing is associated with loss of promoter-proximal active enhancer and gain of facultative heterochromatin features.

Promoter-proximal enhancer region is chr10:102504680–102506278 (1,598 bp) (GRCh37/hg19). (A) ATAC-Seq analysis of PAX2 locus in AN3CA, Ishikawa, KLE, and RL95-2 cells. AN3CA cells were PAX2+ (nonsilenced), whereas Ishikawa, KLE, and RL95-2 cells were PAX2– (silenced). (B) Comprehensive transcriptomic and epigenetic (H3K27ac, H3K4me3, and H3K27me3) profiling of PAX2+ (AN3CA) and PAX2– (Ishikawa) cells. (C) Predicted RAD21 ChIA-PET analysis of Ishikawa cells using ChIPr with varying PET interaction strengths, focusing on an H3K27me3-enriched region surrounding PAX2. The top panel shows the schematic of the ChIPr pipeline. Interaction strengths are represented by depth values ranging from ≥3 to ≥10.DNN, deep neural networks.

Figure 5. Cohesin-mediated 3D genome organization and focal PAX2 silencing in EC.

(A) Comprehensive transcriptomic and epigenetic (ATAC-Seq, H3K27ac, H3K4me3, and H3K27me3) profiling of PAX2+ and PAX2– PDX models of EC. (B) ATAC-Seq analysis of PAX2 locus in primary human EC, including 1 PAX2+ (patient tumor 1) and 2 PAX2– (patient tumors 2 and 3) specimens. Patient tumors 2-1 and 2-2 are technical replicates. (C) Pearl necklace model of PAX2 transcriptional silencing.

Figure 6. PAX2 KD and reexpression alter enhancer profiles per H3K27ac ChIP-Seq.

(A and B) Venn diagrams of H3K27ac ChIP-Seq; enhancer peaks in AN3CA (A) (scrambled and PAX2 KD) and Ishikawa cells (B) (control and PAX2 reexpressed) in the promoter and distal regions. (C) In AN3CA cells, scrambled shRNA peaks were exclusively identified in scrambled shRNA-treated cells. Overlapping peaks were common between scrambled and PAX2 KD cells, and shRNA PAX2 peaks were found in PAX2 KD cells. (D) In Ishikawa cells, empty vector (pLVX) peaks were exclusively present in pLVX-treated cells, overlapping peaks were common between pLVX- and PAX2-expressing cells (pLVX-PAX2), and PAX2 reexpression peaks were identified in pLVX-PAX2 cells.

Figure 7. Pax2 is an EC tumor suppressor synergizing with Pten in vivo.

(A) Uteri at 12 months of age. Scale bars: 2 mm. (B) Pax2/Pten mouse with distended abdomen due to tumorous uterus and ascites. This phenotype was not observed in single knockouts. (C) Distinct EC histotypes in Pax2/Pten females, as shown by H&E staining. Scale bars: 50 μm. (D) p63 immunostaining in mucinous and squamous EC confirming squamous differentiation. (E) PTEN and PAX2 immunostaining confirming endometrial-specific ablation in invasive Pax2/Pten EC. Scale bars: 50 μm. (F) Distant metastases from Pax2/Pten EC, as shown by H&E staining. Scale bars: 50 μm. (G) Survival analysis of Pax2/Pten (n = 50), Pten (n = 25), Pax2 (n = 25), and littermate control (n = 25) mice. ****P < 0.0001 per log-rank test. (H) Uterine weights at necropsy; the x axis shows number of animals per genotype. ****P < 0.0001, 1-way ANOVA, Tukey’s multiple-comparison test.

Figure 8. scRNA-seq reveals that inactivation of Pax2 loss correlates with reduction of Esr1 and Pgr expression in mouse endometrium.

Studies were performed with Pax2-mosaic uteri at 8 weeks of age. (A) Uniform Manifold Approximation and Projection (UMAP) visualization showing cells from Pax2 mouse uterus (n = 2) clustered into 15 distinct subpopulations based on established lineage markers. (B) Stacked violin plots showing expression of gene signatures associated with known uterine cell types, facilitating identification of lineages within clusters. (C and D) Volcano plots showing DEGs (P < 0.05) in Pax2-KO cluster compared with both (C) luminal epithelial (LE) and (D) glandular epithelial (GE) cell clusters. Vertical dotted lines represent log₂ fold change threshold of ±1, while the horizontal dotted line represents a P value threshold of 0.05. Selected genes are shown. (E and F) UMAP plots of Esr1 (E) and Pgr (F). (G) Immunofluorescence staining for PAX2, Erα (Esr1), and PR-A/B (Pgr) in control and Pax2 mouse uterine adjacent sections (n = 3). GE, glandular epithelium; LE, luminal epithelium. Scale bars: 50 μm. (H) Comparison of relative fluorescence intensities of PAX2, ERα, and PR-A/B between PAX2+ and PAX2– cells. Data are shown as mean ± SEM (n = 3); ****P < 0.0001, multiple 2-tailed t tests.