KDM2B promotes pancreatic cancer via Polycomb-dependent and -independent transcriptional programs

Abstract

Epigenetic mechanisms mediate heritable control of cell identity in normal cells and cancer. We sought to identify epigenetic regulators driving the pathogenesis of pancreatic ductal adenocarcinoma (PDAC), one of the most lethal human cancers. We found that KDM2B (also known as Ndy1, FBXL10, and JHDM1B), an H3K36 histone demethylase implicated in bypass of cellular senescence and somatic cell reprogramming, is markedly overexpressed in human PDAC, with levels increasing with disease grade and stage, and highest expression in metastases. KDM2B silencing abrogated tumorigenicity of PDAC cell lines exhibiting loss of epithelial differentiation, whereas KDM2B overexpression cooperated with KrasG12D to promote PDAC formation in mouse models. Gain- and loss-of-function experiments coupled to genome-wide gene expression and ChIP studies revealed that KDM2B drives tumorigenicity through 2 different transcriptional mechanisms. KDM2B repressed developmental genes through cobinding with Polycomb group (PcG) proteins at transcriptional start sites, whereas it activated a module of metabolic genes, including mediators of protein synthesis and mitochondrial function, cobound by the MYC oncogene and the histone demethylase KDM5A. These results defined epigenetic programs through which KDM2B subverts cellular differentiation and drives the pathogenesis of an aggressive subset of PDAC.

Figure 1. KDM2B is upregulated in advanced PDAC.

(A) Heat map of quantitative RT-PCR data showing relative expression of HDM family members in human PDAC cell lines versus HPDE cells. Color bar indicates relative fold change. (B) KDM2B transcript levels in a series of human PDAC specimens compared with normal pancreatic tissue, determined by RNA-seq (samples from MGH tumor bank). (C–F) IHC staining for KDM2B in human pancreatic tissues. (C) Normal (N.) adult pancreas was negative for KDM2B. Insets: higher magnification of normal duct and islet. (D) PDAC showing strong nuclear expression in neoplastic cells, whereas normal duct cells were negative. Insets: higher magnification of PDAC cells (top) and normal duct (bottom). (E) Primary PDAC showing stronger staining of the poorly differentiated component (arrows) compared with the well-differentiated glandular component (asterisks). Insets: Higher magnification of poorly differentiated (top) and glandular (bottom) elements. (F) PDAC liver metastasis showed strong staining of tumor cells, whereas normal liver parenchyma was negative. Inset: higher magnification of PDAC cells (top) and normal liver (bottom). Scale bars: 50 μm; 20 μm (insets). (G) Relative KDM2B IHC staining score for 69 primary PDAC specimens of different grades. (H) Percent distribution of KDM2B staining scores in primary and metastatic tumors. (I) Quantitative RT-PCR analysis of relative KDM2B expression in PDAC cell lines versus HPDE and HPNE cells. Cell lines were classified as quasimesenchymal (QM) and classical based on ref. . See also Supplemental Figure 1.

Figure 2. KDM2B is required for tumorigenicity of poorly differentiated PDAC cell lines.

(A and B) Human PDAC cell lines (MiaPaca and PANC1) were infected with lentiviruses expressing empty vector (shControl) or 2 different shRNAs against KDM2B (shKDM2B) and assessed by (A) in vitro growth assays and (B) subcutaneous tumorigenesis (xenograft) experiments (see Methods). (C) Growth curve showing that overexpression of a nontargeted wild-type KDM2B construct (rKDM2B) rescued the proliferative arrest induced by shKDM2B. (D) Percent growth inhibition of a panel of PDAC cell lines upon knockdown of KDM2B (9 days after seeding). Results are mean ± SD. (E–G) Transplantable murine PanIN-PDAC progression model using Ptf1α-Cre;LSL-Kras^G12D pancreatic ductal cells. Cells were transduced with lentiviruses to overexpress wild-type and JmjC domain deletion mutant (ΔJmjC) of KDM2B. EV, empty vector. (E) Schematic of model system. (F) Western blot of KDM2B expression in transduced ductal cells and in vitro growth curves. Asterisk denotes nonspecific (N.S.) band. (G) Histological images of the pancreas 4 weeks after orthotopic injection with 2 × 10⁴ cells. Control Kras^G12D cells failed to form tumors, whereas wild-type KDM2B induced poorly differentiated PDAC. Insets show high-magnification views of the boxed regions (enlarged ×4). Scale bars: 200 μm. See also Supplemental Figure 2.

Figure 3. KDM2B is required for Polycomb-mediated repression and for maintenance of a metabolic gene signature.

(A) GSEA analysis of the KDM2B-dependent transcriptome in 6 PDAC cell lines. Shown are pathways that had the same direction of change in at least 5 of 6 lines. Gene sets with P values less than 0.05 were selected for comparison across the different cell lines. (B) IPA transcription factor prediction analysis to identify alterations in transcription factor activity induced by shKDM2B in PDAC cell lines (see Methods). Shown are transcription factors with z scores greater than 2 and P overlap values less than 0.01. (C and D) KDM2B knockdown downregulated expression of KDM5A transcript (C) and protein (D). See also Supplemental Figure 3 and Supplemental Table 1.

Figure 4. KDM2B regulates 2 distinct transcriptional modules.

(A) ChIP-seq results showing overlapping gene targets among KDM2B, EZH2, and KDM5A in PANC1 cells. (B and C) Overlap of KDM2B, EZH2, and KDM5A binding in PANC1 cells and of MYC binding in mouse ES cells (B) and HeLa cells (C). (D and E) IPA of KDM2B, EZH2, and KDM5A target genes identified by Chip-seq, shown for each factor individually (D) and for the different KDM2B-cobinding modules (E). The y axis (log scale) corresponds to the binomial raw P values. See also Supplemental Figure 4 and Supplemental Table 2.

Figure 5. KDM2B negatively and positively regulates transcription through different modular associations.

(A) Read density profiles of KDM2B, EZH2, and KDM5A around the TSS of bound genes. x axis, nucleotide coordinates around the TSS; y axis, average log₂ ChIP-seq signal intensity. (B) Composite read density profiles of KDM2B, EZH2, and KDM5A binding in relation to the presence and absence of CpG islands around the TSS (axes as in A). The percentage of factor-bound genes that have CpG islands is shown. (C) Average percent methylation of cytosines in promoters of genes bound by different KDM2B modules in relation to the presence or absence of CpG islands, measured by reduced representation bisulfite sequencing. Boxes represent range between lower and upper quartiles; bands within boxes denote median; whiskers extend to the most extreme data point not more than 1.5 times the interquartile range; outliers are not shown. (D) Combinatorial binding profiles of KDM2B, EZH2, KDM5A, and H3K4me3 enrichment in PANC1 cells and MYC binding in HeLa cells. Each horizontal line represents a separate gene and its TSS. A ±2-kb window is shown for each gene. Bar shows average log₂ ChIP-seq signal intensity. Binding profile overlap is shown at left. (E and F) Effect of KDM2B knockdown on expression of genes bound by KDM2B. (E) Effect of shKDM2B on expression of genes in different binding modules: total KDM2B-bound genes (KDM2B_ALL); cobinding with EZH2, KDM5A, or MYC; or none of these factors (KDM2B_ONLY). Downregulated and upregulated genes are plotted to the left and right, respectively, of the red arrows. KDM2B-EZH2–cobound genes were primarily upregulated, and those bound by KDM2B-KDM5A or KDM2B-MYC were downregulated. OFF, inactive; ON, active. (F) Graphical representation of data in E filtered for genes exhibiting greater than 1.5-fold difference in expression upon KDM2B knockdown. See also Supplemental Figure 5.

Figure 6. Differential roles of KDM2B in chromatin regulation at activated and repressed targets.

(A) ChIP analysis of MiaPaca cells for binding of the indicated proteins to different KDM2B-EZH2 and KDM2B-KDM5A-MYC targets. (B and C) Quantitative RT-PCR analysis of expression changes of selected genes cobound by KDM2B-EZH2 or KDM2B-KDM5A-MYC in response to knockdown of (B) KDM2B or (C) EZH2, MYC, and KDM5A. Cobinding modules are represented by ovals. Note that genes cobound by KDM2B-EZH2 were upregulated upon KDM2B or EZH2 knockdown, whereas KDM2B-KDM5A-MYC–cobound genes were downregulated upon KDM2B, c-MYC, or KDM5A knockdown. (D) ChIP analysis showing the effect of KDM2B knockdown on the indicated histone marks and on binding of EZH2 and RING1B at selected KDM2B target genes in MiaPaca cells. Change in mRNA expression induced by KDM2B knockdown (RNA-seq) is also shown. Results are mean ± SD. *P < 0.05; **P < 0.01. (E) ChIP analysis showing relative fold enrichment of the indicated histone marks on selected target genes that define the different KDM2B modules in MiaPaca cells. KDM2B-EZH2–cobound and KDM2B-KDM5A–cobound genes are shown as blue and red bars, respectively. (F and G) The indicated PDAC cell lines were infected with shControl or shKDM2B lentiviruses and cultured for 36 hours in standard (25 mM) or reduced (1 mM) glucose concentrations. Attached and floating cells were analyzed (F) by Western blotting or (G) for viability (trypan blue dye exclusion method). See also Supplemental Figure 6 and Supplemental Table 3.

Figure 7. EZH2 mediates the proliferative effects of KDM2B in PDAC cell lines.

(A) Percent growth inhibition of the indicated PDAC cell lines upon knockdown of c-MYC, EZH2, or KDM5A (at 9 days after seeding). N/D; not done. (B) 10⁴ MiaPaca cells overexpressing EZH2 were infected with shKDM2B or shControl and plated on 6-cm plates. After 3 weeks, cells were stained and photographed. Efficient expression of EZH2 was demonstrated by Western blot. (C) KDM2B controls tumorigenicity via separate transcriptional repression and activation programs. KDM2B-PRC2 cobinding silences the expression of lineage specification genes (OFF) through H3K36 demethylation (red arrow) and H3K27 methylation (solid arrow) at the gene promoters. PRC1 participates in the silencing of a subset of these targets. KDM2B activates the expression of genes involved in metabolic homeostasis and protein synthesis (ON) in association with KDM5A and/or MYC. KDM2B does not regulate H3K36 or H3K27 methylation levels at these promoters, but rather is required to maintain the H3K4 mark of activated transcription (dashed arrows). me, methylation; ub, ubiquitination.