Methylation reprogramming associated with aggressive prostate cancer and ancestral disparities

Abstract

African men are disproportionately impacted by aggressive prostate cancer (PCa). The key to this disparity is both genetic and environmental factors, alluding to epigenetic modifications. However, African-inclusive prostate tumour DNA methylation studies are lacking. Assembling a multi-geo-ancestral prostate tissue cohort, including men with (57 African, 48 European, 23 Asian) or without (65 African) PCa, we interrogate for genome-wide differential methylation. Overall, methylation appears to be driven by ancestry over geography (152 southern Africa, 41 Australia). African tumours show substantial heterogeneity, with universal hypermethylation indicating more pervasive epigenetic silencing, encompassing PCa suppressor genes and enhancer-targeted binding motifs. Conversely, African tumour-associated heterochromatic hypomethylation suggests chromatin relaxation and developmental pathway activation via enhancer targets. Notably, non-prostate lineage elements appeared preferentially exploited in African tumorigenesis, with ancestry potentially influencing the extent of lineage-inappropriate activation, and tumour progression marked by repression of developmental regulators. Together, these findings point to extensive epigenetic plasticity in African tumours, with intergenic regulatory remodelling promoting genomic instability, metastatic potential and aggressive disease phenotypes.

Keywords: African Ancestry; Differential Methylation; Health Disparity; Prostate Tumours.

Figure 1. DNA methylation geo-ancestral cohort analysis using the EPICv1 and EPICv2 arrays.

(A–D) Distribution of African-confounded probe loss on EPICv1 and EPICv2 arrays across genomic features. Percentage (bars) and number (labels) of EPICv1 and EPICv2 probe sites identified to be confounded by African genomic diversity, shown relative to (A) gene regions, (B) CpG island context, (C) CpG probe type and (D) FANTOM5 enhancer overlap. Percentages are calculated relative to the total number of African-confounded probes lost per platform. (E) Scatterplot showing the correlation between methylation measurements from EPICv1 and EPICv2 for African replicate pair TAF002. (F) Principal component analysis (PCA) of a 193-sample merged and normalised EPICv1/EPICv2 dataset across the 10,000 most variable positions. Samples are annotated according to array version: EPICv1 (n = 78) and EPICv2 (n = 115); and the location from which samples were recruited: Australia (n = 41), Namibia (n = 4) and South Africa (n = 148). (G) Admixture plot (K = 3, cross-validation error = 0.524) replicated in 10/10 runs, including 40 southern Africans from the Southern African Prostate Cancer Study (SAPCS), both including and excluding for Khoe-San fractions (Jaratlerdsiri et al, 2022), and 20 Europeans (CEU) and 20 Chinese (CHB) from the Human Genome Diversity Project (HGDP) and 1000 Genomes Project (1KGP) subset of gnomAD v3.12 (Chen et al, 2023), together with our geo-ancestral cohort (n = 192, with exclusion of a single African with insufficient genome sequencing coverage). Source data are available online for this figure.

Figure 2. Differential DNA methylation analysis for African- versus non-African-derived prostate tumours.

(A) PCA for the 10,000 most variable positions for the EPICv1 “ancestry-associated discovery cohort” by ancestry: African (n = 21), European (n = 27) and Asian (n = 22), and geography: South Africa (n = 48) and Australia (n = 22). (B) DNA methylation heatmaps at ancestry-associated DMPs. Left: β-values (range 0–1) for tumour (n = 70) and normal (n = 7) prostate samples. Right: Row-scaled z-scores for tumour samples only. Top annotations indicate ancestry, geography and ISUP grade group, and rows (n = 861 CpGs) are annotated by chromatin state. Heatmaps display absolute (left) and relative (right) methylation differences at ancestry-associated loci. (C) Volcano plot representing 861 DMPs by ancestral (21 African versus 49 non-African) significance determined using linear regression (BH FDR p < 0.05, |Δβ| ≥ 20% and 40%, dashed lines), including hypermethylated (green) or hypomethylated (blue), compared with non-significant (grey) DMPs. (D) Heatmap showing gene region (green) and CpG island-related region (purple) enrichment for DMP- and DMR-related CpG sites across various contexts. (E) Heatmap showing chromatin state enrichment of DMP- and DMR-related CpG sites across various contexts. (F) Stacked bar graphs of the percent overlap of DMPs and DMRs with various CpG island-related (purple) and chromatin state (blue) contexts. (G) Venn diagrams illustrating significant DMP, DMR (as per linear regression) and annotated gene agreement between the ancestry-associated discovery (21 African, 49 non-African) and validation (35 African, 22 non-African) cohorts. (H) MSigDB (C2) enrichment for validated genes collectively annotated to hypermethylated DMPs and DMRs. Enrichment analysis was performed using a hypergeometric test, with the top 15 terms (minimum 5 genes) shown. Cancer-related terms are highlighted in grey. Point size reflects the number of genes associated with each gene set. Association strength is denoted by –log₁₀(P value). DMP differentially methylated position, DMR differentially methylated region, ExonBnd within 20 bases of an exon boundary, IGR intergenic region, ISUP International Society of Urological Pathology, MSigDB Molecular Signatures Database, PCA principal components analysis, TSS transcription start site, UTR untranslated region, |Δβ| absolute difference in mean methylation. Source data are available online for this figure.

Figure 3. African versus non-African CpG DNA methylation plots for four individual potentially unknown African-specific prostate cancer (PCa) targets.

Noteworthy significant differentially methylated positions (DMPs) between African (orange, n = 21) and non-African (blue, n = 49) PCa samples. The y axis represents sample beta values for each individual CpG probe, after covariate adjustment. The FDR significance in the difference of mean beta values between the two groups, as per a t test, is shown. Source data are available online for this figure.

Figure 4. African-specific prostate tumour versus normal tissue differential methylation.

(A) PCA for the 10,000 most variable positions for the EPICv2 African “tumour-associated discovery cohort” by cancer status: those with PCa (n = 35), and those without PCa (n = 58). (B) Boxplots of global DNA methylation levels in African prostate tumour versus normal tissue samples across CpG islands, CpG island shores, LINE-1 repetitive elements and LTRs. The box represents the interquartile range (IQR), with the bottom and top boundaries indicating the 25th and 75th percentiles, respectively. The horizontal line within the box shows the median (50th percentile). The whiskers extend from the box to the minimum and maximum values within 1.5 times the IQR, with individual points beyond the whiskers considered outliers. Each dot represents the median methylation value for each patient. The “X” indicates group mean. (C) Volcano plot representing 9501 DMPs by cancer status (35 prostate tumour versus 58 normal tissue) significance determined using linear regression (BH FDR P < 0.05, |Δβ| ≥ 30% and 40%, dashed lines), including hypermethylated (green) or hypomethylated (blue), compared with non-significant (grey) DMPs. (D) DMP cluster analysis heatmap by cancer status, ISUP grade group and chromatin state. Rows represent CpG sites (n = 9501) and columns, patients (n = 93). The methylation level is represented by a β-value between 0 (completely unmethylated, blue) and 1 (fully methylated, red). (E) Heatmap showing gene region (green) and CpG island-related region (purple) enrichment for DMP- and DMR-related CpG sites across various contexts. (F) Heatmap showing chromatin state enrichment of DMP- and DMR-related CpG sites across various contexts. (G) Stacked bar graphs of the percent overlap of DMPs and DMRs with various CpG island-related (purple) and chromatin state (blue) contexts. (H) Venn diagram illustrating shared and distinct significantly differentially methylated CpGs (BH FDR P < 0.05, |Δβ| ≥ 10%) identified in the African tumour-associated cohorts and TCGA-PRAD European cohort. (I) MSigDB (C2) enrichment for genes collectively annotated to hypermethylated DMPs and DMRs not identified within TCGA-PRAD European cohort. Enrichment analysis was performed using a hypergeometric test, with the top 15 terms (minimum 5 genes) shown. Cancer-related terms are highlighted in grey. Point size reflects the number of genes associated with each gene set. Association strength is denoted by –log10(P value). (A–I) For all tumour-associated analyses, the 35 African tumours include 7 samples reclassified as “presumably PCa” through methylation profiling. DM differentially methylated, DMP differentially methylated position, DMR differentially methylated region, ISUP International Society of Urological Pathology, LTRs long tandem repeats, MSigDB Molecular Signatures Database, PCA principal component analysis, TCGA The Cancer Genome Atlas, TSS transcription start site, UTR untranslated region, |Δβ| absolute difference in mean methylation; ****P ≤ 0.0001. Source data are available online for this figure.

Figure 5. African prostate tumour versus normal tissue gene methylation plots highlighting two individual potentially unknown African-specific CpG prostate cancer targets.

Differential methylation over noteworthy genes SLC12A9 and PYCARD, highlighting (in bold) significant potentially unknown African-specific prostate tumour versus normal CpG targets. Probes overlapping CpG islands are shown in blue, the y axis represents the mean beta value per group for each individual CpG probe, after covariate adjustment, and differential mean methylation significance between groups, determined by FDR P < 0.05, is represented by an asterisk (*). Source data are available online for this figure.

Figure 6. Proposed model of DNA methylation-driven epigenetic aberrations in African prostate tumours.

(A) Promoter hypermethylation-induced epigenetic silencing of (tumour-suppressor) genes. (B) Enhancer hypermethylation disrupts transcription factor (e.g., FOXO4) recruitment and binding dynamics, and impairs enhancer-promoter looping, ultimately altering target gene expression. Target genes of aberrantly methylated enhancers were often involved in tumour-suppressor binding motifs. (C) DNA hypomethylation of closed chromatin promotes a more permissive chromatin state, which may be aided by diminished HDAC activity, collectively contributing to genomic instability. DNMT DNA methyltransferase, HAT histone acetyltransferase, HDAC histone deacetylase, TET ten-eleven translocation. Created in BioRender. https://BioRender.com/rx2ybm9.

Figure EV1. Identifying EPICv1 and EPICv2 probes that overlap southern African polymorphic variants.

Workflow for the identification of EPIC probes (v1.0 and v2.0) overlapping African SNV and indel variants, rendering filtering resources. Consensus probe filtering resources are based on germline variants (MAF > 0.01) from 99 southern African men. indel insertion and deletion, MAF minor allele frequency, SBE single base extension, SNV single-nucleotide variant, VCF variant call format.

Figure EV2. African prostate tumours exhibit elevated methylation variability at ancestry-associated DMPs.

Standard deviation of DNA methylation β-values is shown for each ancestry group at the 31 top DMPs exhibiting significant variance differences by ancestry (Levene’s test, P < 0.05). Each panel presents within-group variability across African and non-African tumours. In 90.3% of these DMPs, African tumours displayed the highest intra-group variability, supporting the presence of elevated ancestry-associated epigenetic heterogeneity.

Figure EV3. Ancestry-associated methylation patterns at four key DMP genes and their relationship to tumour purity.

Scatterplots (left) and boxplots (right) depicting methylation levels at ancestry-associated CpG sites located in four genes of interest: CHSY1, GALM, EVC2 and SPDYA. Scatterplots show methylation levels versus tumour purity across prostate tumours (n = 70), coloured by ancestry. Boxplots show ancestry-stratified methylation levels further grouped by tumour purity quartiles (Q1-Q4). Non-African Q4 tumour samples are outlined to highlight ancestry-related differences at high tumour purity.