Abstract Prostate cancer is a complex disease that affects millions of men each year. Recent research has revealed that the hormone-dependent nature of prostate cancer stems from the high expression and genetic amplification of the androgen receptor (AR) and may be driven by extrachromosomal DNA (ecDNA). Identifying ecDNA from other modes of amplification (e.g. HSR via breakage-fusion-bridge cycles) using whole-genome sequencing (WGS) data remains challenging due to the complexity of copy number and structural variation in cancer. This is compounded in FFPE where this sample type is often at risk of nucleic acid fragmentation, DNA crosslinks, and deamination further complicating variant discovery. In this study we investigated if structural variants and linkage patterns derived from two Dovetail linked read chemistries, LinkPrep™ (a Tn5-based method compatible with cells and fresh-frozen tissue) and Dovetail®-FFPE (an MNase-HiC-based method compatible with FFPE), could predict the presence of ecDNA. We benchmarked our approach using four cancer cell lines with known ecDNA or HSR, then applied this technique to clinical metastatic castration-resistant prostate cancer (mCRPC) cases stored in FFPE. We generated libraries at ∼30x coverage and utilized BWA for alignment, Hi-C breakfinder for structural variants, DeepSomatic for SNV and InDel detection, Purple for CNVs, and an in-house script for ecDNA detection. AmpliconArchitect was used on orthogonal shotgun WGS to predict ecDNA. By analyzing the long-range linkage from amplicons to elsewhere in the genome, our method correctly identified the presence (or absence) of ecDNAs in all four cell lines and made ecDNA predictions across the samples with AR amplicons among the clinical mCRPC FFPE cohort. Two analytical methods were employed to predict ecDNA presence: a novel Z-score-based method that quantifies the observed difference from expected contact distribution across the amplicon (ecDNA 5.7-fold increase vs. HSR) and a log2-ratio test comparing intra- vs. inter-chromosomal amplicon-linked contacts (ecDNA 3-fold increase vs. HSR). Both methods were found to be robust down to 5x coverage. Finally, for samples with suspected integrated amplicon(s), the percentage of all amplicons integrated at each site is estimated through site-specific vs. total amplicon linkage. In addition to ecDNA prediction, Dovetail libraries detect variants across the full range of oncogenic driver alterations, including SNVs/InDels, copy number, and structural variants. Comparing against a fresh frozen, WGS-derived truth set for the mCRPC cohort, and Dovetail®-FFPE libraries successfully recalled homozygous PTEN deletions, TMPRSS2::ERG fusions, and oncogenic mutations to BRCA2, TP53, and KMT2C genes. Our findings show that Dovetail libraries provide researchers with a powerful platform for comprehensive genetic variation discovery in both fresh-frozen and FFPE tissues. Citation Format: Alexander Fortuna, Natalie Fredriksson, Mital Bhakta, Rahul Mannan, Fengyun Su, Rui Wang, Dan Robinson, Yi-Mi Wu, Xuhong Cao, Arul M. Chinnaiyan, Lisa Munding, J. Zachary Sanborn. Leveraging linked-reads to predict extrachromosomal DNA in advanced prostate cancer abstract. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 2690.
Fortuna et al. (Fri,) studied this question.