DNA hypomethylation silences anti-tumor immune genes in early prostate cancer and CTCs

Abstract

Cancer is characterized by hypomethylation-associated silencing of large chromatin domains, whose contribution to tumorigenesis is uncertain. Through high-resolution genome-wide single-cell DNA methylation sequencing, we identify 40 core domains that are uniformly hypomethylated from the earliest detectable stages of prostate malignancy through metastatic circulating tumor cells (CTCs). Nested among these repressive domains are smaller loci with preserved methylation that escape silencing and are enriched for cell proliferation genes. Transcriptionally silenced genes within the core hypomethylated domains are enriched for immune-related genes; prominent among these is a single gene cluster harboring all five CD1 genes that present lipid antigens to NKT cells and four IFI16-related interferon-inducible genes implicated in innate immunity. The re-expression of CD1 or IFI16 murine orthologs in immuno-competent mice abrogates tumorigenesis, accompanied by the activation of anti-tumor immunity. Thus, early epigenetic changes may shape tumorigenesis, targeting co-located genes within defined chromosomal loci. Hypomethylation domains are detectable in blood specimens enriched for CTCs.

Keywords: CD1A; DNA hypomethylation; IFI16; NKT cells; circulating tumor cells; immune surveillance; lipid antigens; prostate cancer; single-cell sequencing; tumorigenesis.

Figure 1.. Partially Methylated Domains (PMDs) and Preserved Methylation Islands (PMIs) in single metastatic prostate cancer cells.

(A) Schematic of CTC enrichment (10⁴-fold leukocyte depletion), and paired DNA methylation sequencing (nucleus) and RNA-seq (cytoplasm) from individual prostate CTCs. (B) Confirmation of CTC identity using stringent RNA expression thresholding of prostatic lineage and epithelial versus leukocyte markers. Maximum log10 (RPM) expression of epithelial (KRT7, KRT8, KRT18, KRT19, EPCAM) and prostatic markers (AR, KLK3, FOLH1, AMACR) are plotted against leukocyte markers (CD45, CD16, CD37, CD53, CD7, CD66b). Only confirmed CTCs without WBC contamination (red crosses) were used in analyses. (C) Representative DNA copy number variation (CNV) analysis in individual CTCs from two patients, compared with a diploid normal prostate epithelial cell (HPrEC) and a healthy donor-derived leukocyte. Single-cell DNA methylation sequencing data was used to infer DNA copy number. (D) IGV representation (hg19) of DNA methylation spanning chromosome 8, showing extensive PMDs (yellow) across 37 individual CTCs from four patients (GU114, GU216, GU181 and GURa15), and 17 cells from prostate cancer cell lines (LNCaP, PC3, VCaP, 22Rv1). As controls, 4 normal bulk prostate tissues (N.P.), 36 cells from two prostate epithelial cell lines (HPrEC, BPH-1) and normal leukocytes (WBCs) are shown. Normal methylation level (blue). (E-F) Higher resolution of chromosome 8 in IGV, showing precise PMD boundaries shared across individual CTCs and prostate cancer cell lines (panel E), with magnified view of the nested PMI, bracketing a few genes, with precise boundaries of preserved methylation flanked by profound hypomethylation (panel F). (G-H) Components of coding genes and classes of repeats differentially enriched in PMDs versus PMIs (panel G), with differences among subtypes of repeats (panel H). ns, not significant; *P<0.05; **P<0.01, assessed by permutation test.

Figure 2.. Acquired chromatin marks in prostate cancer PMDs and nomination of shared core PMDs.

(A) Differential enrichment of chromatin marks within prostate cancer PMDs and PMIs. Annotated chromatin marks from ChIP-seq dataset of PC3 cells in ENCODE (https://www.encodeproject.org/). ns, not significant; *P<0.05; **P<0.01, assessed by permutation test. (B) Line plots showing differential enrichment of silencing chromatin marks at PMDs across the genome in prostate cancer cells (LNCaP; 3 biological replicates, red lines), compared with cultured benign prostatic hyperplasia cells (BPH-1; 2 biological replicates, green lines) and normal prostate epithelial cells (HPrEC; 2 biological replicates, blue lines). Across the genome, prostate cancer cells acquire H3K27me3, with highest levels at the boundaries of PMDs (left panel), whereas H3K9me3 enrichment towards the center of PMDs is not altered between cancer and non-transformed prostate cells (right panel). (C) Boxplot showing enrichment of Cut and Run signal for H3K27me3, but not H3K9me3, across prostate cancer PMDs between LNCaP cells and non-transformed cell lines (HPrEC and BPH-1). Pvalue, one-tailed Student’s t-test. (D) IGV track showing representative cancer-associated PMD (DNA hypomethylation: yellow), with pronounced enrichment of H3K27me3 at PMD borders in cancer cells (LNCaP: red) versus non-transformed cells (HPrEC: blue, BPH-1: green), whereas PMD-centered H3K9me3 occupancy is unaltered. (E) Inter- and intra-patient heterogeneity of PMDs among single CTCs from four prostate cancer patients (red) and single cells from prostate cancer cell lines. Mean Jaccard index indicates heterogeneity, with higher mean score indicating less heterogeneity among samples. Error bar, mean with 95% confidence interval (CI). (F-G) IGV representation of total PMDs and core PMDs at chromosome 3 locus, across 8 sample sources (4 patients and 4 prostate cancer cell lines). Total PMDs (blue) are the union of PMDs defined in each sample source, while core PMDs (black) are shared across all 8 sample sources (panel F); representation of PMDs from the single-cell components of an individual sample source (22 CTCs from patient GU181) showing a core PMD shared across all sample sources (black) and neighboring non-core PMDs that are shared by >90% CTCs in this patient, but not across different sample sources (panel G). See Figure S2D and Methods for criteria in core PMD and PMI designation.

Figure 3.. Demethylation of core PMDs during early prostate tumorigenesis suppresses immune-related genes, while core PMIs spare proliferation genes.

(A) Schematic showing prostate tumor microdissection, single nucleus isolation and single-cell DNA methylation sequencing. (B) Ranking of methylation level at 40 core PMDs (red dots) among all 1,496 total PMDs, as a function of timeline from normal prostate, to localized (GS6; GS8) and metastatic cancer (CTCs), showing early demethylation of core PMDs. Within normal prostate, all 40 core PMDs have methylation level >75%, and 31 are hypomethylated as early as GS6. (C) Quantitation of demethylation as a function of Gleason Score (GS). Demethylation of core PMDs (red curve) precedes that of other PMDs (magenta) within microdissected prostate tumor cells and in CTCs. In contrast, core PMIs nested between PMDs (blue) show minimal DNA methylation changes during tumorigenesis. Error bar, mean with SEM. Statistical analysis of DNA methylation curves utilizing longitudinal linear mixed effects model, by which tumor progression x methylation domains was tested. (D) Quantitation of demethylation as a function of GS in TCGA prostate cancer methylation array data, showing early and progressive loss of methylation of core PMDs (red curve), with an attenuated trend for other PMDs (magenta). The core PMIs (blue) display stable DNA methylation pattern during prostate tumorigenesis. Statistical analysis as for panel C. (E-F) Gene set enrichment analysis (GSEA) of genes residing within core PMDs and downregulated in primary prostate cancer (E), and of genes residing within core PMIs with gene expression preserved (up-regulated and not significantly changed) in primary prostate cancer (F), compared with normal prostate. (FDR <0.1; two-tailed Student’s t-test with FDR correction).

Figure 4.. Correlation of DNA demethylation at the CD1A-IFI16 locus with accumulation of chromatin silencing marks and reduced gene expression.

(A) IGV of single-cell DNA methylation at the CD1A-IF16 genomic locus, including five lipid antigen presentation and four interferon inducible genes. Tumor cells (37 single CTCs from four prostate cancer patients (red) and 17 single cells from four prostate cancer cell lines (green)) exhibit marked hypomethylation at this locus (shaded yellow), while normal samples (4 bulk normal prostate tissues, 37 single cells from normal prostate cell lines and leukocytes (blue)) show a preserved DNA methylation (shaded blue). (B) Heatmap (upper panel; hypomethylation shaded yellow) and matched quantitative scatter plots (lower panel) of single-cell DNA methylation levels within all 1,496 prostate cancer PMDs, showing progression from normal prostate to localized prostate cancer (GS6, GS8) and metastatic CTCs. The CD1A-IFI16 locus (dashed vertical red line) shows early and profound demethylation, starting at GS6, with its rank number across all PMDs at each tumor stage shown in parentheses (red). (C) IGV screenshot of single-cell DNA methylation data showing progressive demethylation of CD1A-IFI16 locus (box with red dashed line) from normal prostate cells to localized (GS6 and GS8) and metastatic prostate cancer (CTCs). Heterogeneity of hypomethylation (shaded yellow) across single cells is evident at GS6, becoming more prevalent at GS8, and uniform in CTCs . (D) Plots showing suppressed expression of lipid antigen presentation and interferon inducible genes within the CD1A-IFI16 locus, during transition from normal prostate to low-grade GS6, with persistent silencing in higher grade GS7, 8 and 9 cancers (TCGA dataset). Error bar, mean with SEM. (E) Analysis of 33 different tumor types (TCGA) for DNA methylation differences at core prostate cancer PMDs, compared with corresponding normal tissues. 30 of 35 (86%) evaluable PMDs are hypomethylated across all tumor types (red circles), with the CD1A-IFI16 locus having the strongest hypomethylation. (F) Histograms of DNA methylation level within 100kb windows (200bp offsets) across the genome in normal prostate cells (BPH-1), following 5-azacytidine treatment (days 1 and 5), compared with DMSO control. (G) Quantitation of H3K27me3-related fluorescence intensity within single-cell nuclei (confocal microscopy). Error bar, mean with SEM. P-value, two-tailed Student’s t-test. (H) Sequential reduction in CD1d protein expression in normal prostate cells (BPH-1) treated with 5-azacytidine, compared with DMSO control. Representative flow cytometry (left panel); median fluorescence intensity (right panel). Error bar, mean with SEM. P-value, two tailed Student’s t-test. (I-J) Western blot showing reduced H3K27 trimethylation in 22Rv1 cells treated with EZH2 inhibitor GSK126 for 6 days (panel H); qPCR of genes within the CD1A-IFI16 cluster show induced expression (panel I), while non-PMD resident control genes (PP1A, HPRT and β-actin) remain unchanged. P-value, Tukey’s multiple comparison tests, where GSK126 treatment conditions (red bars) were compared to controls (blue bar). n.s. not significant; ****P<0.0001.

Figure 5.. Restoring expression of genes within CD1A-IFI16 syntenic locus abrogates tumorigenesis in an immunocompetent mouse prostate cancer model.

(A) Plots quantifying Cd1d1 and Ifi204 mRNA in the murine prostate tumor cell line Myc-CaP, which have silenced the syntenic genes (blue), compared to normal prostate cells from 4 isogenic mice FVB (orange). Ectopic expression of murine Cd1d1 (CD1D ortholog, green) and Ifi204 (IFI16 ortholog, red) is comparable to that of normal prostate. Error bar, mean with SEM. (B) Overexpression (OE) of Cd1d1 or Ifi204 in Myc-CaP cells does not alter in vitro proliferation compared with controls. Error bar, mean with SD. (C) Overexpression of either Cd1d1 (green) or Ifi204 (red) in Myc-CaP cells (mCherry-luciferase tagged) suppresses tumorigenesis in isogenic immunocompetent FVB mice. Mock-transfected control tumors are shown as control (blue). Tumor size quantified by luciferase imaging (representative images). Error bar, mean with SEM. (D) Myc-CaP cells engineered as in (C) show no difference in tumor growth in immune-deficient NSG mice. Error bar, mean with SEM. (E) Flow cytometry of Cd1d-restored Myc-CaP tumors in FVB mice, showing recruitment of CD1d-restricted NKT cells (marked by α-GalCer CD1d Tetramer) and activated NKT cells (marked by CD69), compared with controls. Error bar, mean with SD. (F) Flow cytometry of Ifi204-restored Myc-CaP tumors in FVB mice, showing unaltered infiltration of total CD4⁺ and CD8⁺ T cells, but reduced immune infiltration by PD-1⁺ CD8⁺ T cells and increased presence of TNFα⁺ CD8⁺ T cells, compared with controls. Error bar, mean with SD. P-values, two-tailed Student’s t-test; ns, not significant.

Figure 6.. Detection of CTC-derived DNA hypomethylation in blood specimens using Nanopore sequencing.

(A) IGV screenshot showing concordance of DNA hypomethylation measurements between Oxford Nanopore native sequencing of bulk VCaP cells [B], compared with Illumina bisulfite sequencing of three single VCaP cells (#1, #2, #3). DNA methylation across entire chromosome 4 is shown (hypomethylation in shaded yellow). (B) Scatter plot showing high Pearson correlation (r=0.81) between Nanopore native sequencing and Illumina bisulfite sequencing. (C-D) Mathematical modeling showing minimal precision using short reads (average 5 CpG sites per read) for detection of hypomethylated DNA domains. Modest improvement in detection is provided by interrogating predetermined PMDs, instead of whole genome (panel C). Significantly improved precision is predicted using Nanopore long read sequencing (10 or 50 CpGs per read). Highest predicted accuracy by combining Nanopore long reads (>10 CpG sites per read) with selected analysis-predetermined PMD regions (panel D). (E) Schematic of microfluidic CTC enrichment (followed by direct Nanopore sequencing of bulk cells (approximatly 0.1% CTC purity). HMW, high molecular weight. (F-G) Scatter plot quantitation of hypomethylation signal by Nanopore sequencing, comparing leukocyte-depleted blood samples from patients with either metastatic (panel F) or localized prostate cancer before surgical resection or radiation therapy (panel G), versus healthy age-matched male donors (HDs). Error bar denotes mean with SEM. P-value assessed by two-tailed Student’s t-test. Dotted lines indicate thresholds of hypomethylation signal that encompass all healthy donors tested, with the fraction of cancer patients with hypomethylation signal above that threshold considered positive.