Genetic and Epigenetic Reprogramming of Transposable Elements Drives ecDNA-Mediated Metastatic Prostate Cancer

Abstract

Extrachromosomal DNAs (ecDNAs), which replicate and segregate in a non-Mendelian manner, serve as vectors for accelerated tumor evolution. By integrating chromatin accessibility, whole-genome sequencing, and Hi-C-based genome topology data from a cohort of metastatic Castration-Resistant Prostate Cancer (mCRPC) cases, we show that epigenetically activated repeat DNA, amplified in ecDNAs, drive oncogene overexpression. Specifically, we identify a subgroup of mCRPCs (20%) characterized by clusters of accessible LINE1 repeat DNA elements flanking the androgen receptor (AR) gene. These LINE1 elements are co-amplified with AR and provide binding sites for prostate-lineage transcription factors, including AR, FOXA1 and HOXB13. Accessible LINE1 elements establish novel 3D chromatin interactions with the AR gene, forging a new regulatory plexus driving AR overexpression and confers resistance to androgen signaling inhibitors. Our findings indicate how tumor evolution is driven by the convergence of genetic and epigenetic alterations on repeat DNA, activating and amplifying them to allow oncogene overexpression.

Keywords: ATAC; BFB; Chromatin; HiC; LINE1; WGS; androgen receptor; castration resistant prostate cancer; ecDNA; endogenous repetitive elements; enhancer; epigenetic; epigenomic; prostate cancer; transcription factors; transposable elements; variants.

Figure 1:. Enrichment of transposable elements define mCRPC subtypes and identify clusters of accessible LINE1 elements at the AR locus.

A) Heatmap with unsupervised clustering of deviation Z-scores (Dev Zscore) for transposable element subfamilies (N=605, rows) in accessible chromatin from mCRPC samples (N=71, columns). Transposable element subgroups were identified from unsupervised clustering of transposable element enrichment patterns. DTB-135 clustered separately from the four transposable element subgroups identified. B) Upset plot displaying the number of enriched transposable element subfamilies (q<0.01; median Z-score >1) shared across the different mCRPC groups. C) Barplot showing the transposable element subfamilies uniquely enriched (q<0.01; median Z-score >1) in the different mCRPC groups (top) and allocation to transposable element families (bottom). No transposable element subfamilies were uniquely enriched in the Constant subgroup. D) Tornado plot showing the average fold enriched chromatin accessibility signal in the mCRPC groups across differentially accessible elements (q<0.01) from enriched transposable element subfamilies. The length of the elements was uniformly scaled to 500 bp with 2 Kb flanking regions. E) Barplot showing the genomic distribution of differentially accessible elements by chromosome, normalized to the number of peaks across transposable elements. F) Circos plot showing the density of accessible LINE1 elements from enriched subfamilies in 1 Mb bins for chr8 and chrX. Each track represents a single sample in the LINE1+ group. The positions of the AR and MYC oncogene are annotated.

Figure 2:. Enrichment of accessible LINE1 elements is associated with recurrent AR amplifications in mCRPC.

A) Oncoprint showing oncogenic drivers with recurring copy number alterations across 69 mCRPC samples with matched ATAC-seq and WGS. For AR, weighted copy numbers exceeding 0.9*ploidy were annotated as gains and >5*ploidy as high gains. For genes on autosomes weighted copy numbers exceeding 1.95*ploidy and 3*ploidy were annotated as gains and high gains respectively. Losses refer to a weighted copy number below 1.1 and biallelic loss below 0.5*ploidy. The barplots (right) summarize the results of the statistical comparisons (Fisher exact test, line represents P=0.05) and proportion of recurring copy number alterations across the transposable element subgroups. B) AR weighted copy number (left) and normalized expression transcripts per million (TPM; right) across the different transposable element subgroups annotated for the identified amplicon type. Statistical comparison was performed by a Kruskal-Wallis test with Dunn post-hoc test and Benjamini-Hochberg corrections. For AR copy-number, LINE1+ vs Constant and SINE1+ P=0.006, P=0.008 respectively. For AR expression, LINE1+ vs Constant, SINE1+ and SETL, P=7*10⁻⁴ and P=0.0047 and 0.0084 respectively. C) Transcriptional subtypes based on AR and neuroendocrine signaling across transposable elements subgroups. Statistical comparison was performed by a Fisher exact test grouping AR+NE− and ARL_NE, followed by pairwise comparisons with Benjamini-Hochberg corrections, LINE1+, Constant and SINE1+ vs SETL P=0.02, P=0.03 and P=0.02 respectively. D) Barplot showing the proportion of AR ecDNA (N=11), BFB (N=7) and complex non-cyclic alterations (N=1) across the different transposable element groups as determined by AmpliconArchitect. Cases where no complex rearrangement was identified were annotated as linear. Statistical comparison of AR ecDNA and BFB amplicons between the different transposable element groups was performed using a Fisher exact test, with P=0.03 for ecDNA and P=0.07 for BFB. E-H) Overall survival with respect to ARSI treatment status across transposable element subgroups in the WCDT cohort (N=48). Solid lines represent ARSI treated patients and the dotted lines represent untreated patients. Statistical comparison was performed using the log-rank test.

Figure 3:. Accessible LINE1 elements provide AR, FOXA1, and HOXB13 binding sites in mCRPC.

A) Top 2% recurring transcriptional regulators with significant enrichment (q=0.05, log-odds ratio >1.5) over LINE1 subfamilies (N=46) as identified from ReMaP. B) GIGGLE scores for top transcriptional regulators over enriched LINE1 subfamilies identified from the ReMaP analysis. C-D) Case example of accessible L1PB1 (C) and L1MA4A elements (D) in LINE1+ samples with transcription factor binding sites as identified from ReMaP. The top track depicts the average fold-enriched chromatin accessibility signal, with the second track showing accessible LINE1 elements and AR (blue), FOXA1 (green), and HOXB13 (orange) binding sites.

Figure 4:. Accessible LINE elements are cis-regulatory elements that interact with the AR TSS.

A) Mean scaled regional contact frequency score (RCFS) for the AR amplified gene locus, spanning the transcription start site (TSS) to canonical enhancers, across the transposable element subgroups (left). The red dotted line indicates the RCFS cut-off (0.9) associated with AR ecDNA status. AR TSS (+\− 10Kb) intrachromosomal loops (right) anchored to regions containing accessible DNA elements (chrX:65–70Mb). Statistical comparison was performed using a Kruskal-Wallis test with Dunn post-hoc test and Benjamini-Hochberg corrections. RCFS in LINE1+ vs Constant and SINE1+ P=0.01 and P=0.009 respectively. TSS loops in LINE1+ vs Constant, SETL, and SINE1+ P=0.04, P=0.04, and P=0.05 respectively. B) AR TSS intrachromosomal loops anchored to regions with accessible DNA elements shared across samples. Loops anchored to accessible DNA elements containing LINE1 sequences from enriched subfamilies are shown in red, with the remainder shown in black. The black dotted line represents the cut-off for identifying recurrent interactions in each transposable element group (25%). C-D) Circos plots showing recurring (>25%) AR TSS intrachromosomal loops in LINE1+ (C) and SINE1+ (D) samples within chrX:63–71.8Mb (corresponding to cytoband q11.2-q13.1). Loops (inner track) anchored to accessible DNA elements with or without LINE1 sequences from enriched subfamilies are shown in red and black respectively. Barplot summarizes the number of recurrent loops shown in the Circos plot. The AR amplified region for the LINE1+ and SINE1+ group is marked with a black line which includes the AR gene in orange and the canonical enhancers in yellow. The outer tracks show the density of accessible LINE1 elements from enriched subfamilies in 10 kb bins, for each sample in the subgroup. E) Upset plot showing recurrent AR TSS loops (>25%) to accessible LINE1 sequences from enriched subfamilies identified in LINE1+ samples overlapping AR, FOXA1 and HOXB13 binding sites derived from ReMaP. F) Schematic overview of the proposed model where a novel regulatory plexus is established through co-amplification of LINE1 elements with the AR gene and co-opted by the AR transcription factor and its cofactors to favor AR gene overexpression.