A Guide to Extrachromosomal DNA: Cancer’s Dynamic Circular Genome

Abstract

Focal amplifications of oncogenes are important cancer drivers. They can occur on chromosomes or in the context of circular extrachromosomal DNA (ecDNA). Many key features of ecDNAs were described in the 1960s to 1980s, including their “unstable” nature and their ability to confer drug resistance. With the benefit of new technologies, our understanding of ecDNAs has advanced dramatically in the last decade, both in breadth and in depth, including the remarkable discovery that ecDNAs are present in 17% of all cancers and are associated with worse patient outcomes. In this study, we present a guide to ecDNA tools and biology.

Significance:

Focal amplifications on ecDNAs are commonly found in cancer and are associated with poor patient outcomes and distinct biology. In this review, we provide a guide to ecDNA biology and available tools as well as our perspective on this rapidly evolving field.

Introduction

Extrachromosomal DNA (ecDNA) is emerging as a clinically significant phenomenon in many human cancer types. In this review, we provide historical perspective, summarize the available computational and experimental tools available for ecDNA research, and highlight the latest advances in this growing field. Finally, we provide our perspective on how our understanding of ecDNA biology may drive improvements in clinical care for patients with cancer.

Historical Perspective and Notes on Nomenclature

ecDNA was first identified in the 1960s as multicopy small, paired chromatin bodies, termed “double minutes” (DM) in metaphase chromosome spreads from patients with cancer (1, 2). Subsequent work in the 1970s and 1980s identified a number of ecDNA features that contribute to our modern understanding of ecDNA biology and relevance: i) ecDNAs contain protein-coding genes, including oncogenes (3), ii) ecDNAs are circular structures (4) which lack centromeres (5), iii) individual cells from a single culture or tumor exhibit a wide range of ecDNA copy number, ranging from zero to hundreds of ecDNA copies per cell (6), and iv) ecDNAs serve as a mechanism for the amplification of drug-resistant genes and for the evolution of drug-resistant alleles (7–9).

Although the term “DM” is still used today by some researchers, we favor the term ecDNA, as it was demonstrated that only about 30% of ecDNAs appear as paired structures in metaphase chromosome spreads across a variety of cancer cell lines (10). Notably, the designation of DM remains associated with some cell lines such as COLO320DM to denote their ecDNA+ status. ecDNAs differ from other types of circular ecDNA molecules that are found in noncancerous eukaryotic cells. These include a variety of small circular DNA particles called extrachromosomal circular DNAs (eccDNA) that have been found in many species. eccDNAs are a broad class of circular extrachromosomal structures which are distinct from the ecDNAs discussed in this review in several key respects (Supplementary Fig. S1). eccDNAs (i) are much smaller than ecDNAs [typically less than 1 kilobase (kb)], (ii) can occur from any region of the genome, without preference for oncogenes or other protein-coding genes, (iii) are found in normal (noncancerous) tissue and in many eukaryotic species, and (iv) do not convey a fitness advantage. eccDNAs are reviewed elsewhere (11).

Cytogenetic analysis of metaphase chromosomes using techniques such as Giemsa-banding remained the primary method for unbiased (i.e., not sequence-specific) ecDNA detection until the development of next-generation sequencing (NGS) technology. Several studies used these chromosome spreads to quantify the frequency of ecDNA in various cancer types, but estimates varied widely (6, 12). When our team published its first study on ecDNA, it was estimated that 1.4% of all tumor samples contained ecDNAs according to the Mitelman Database (13). Limitations in the resolution of Giemsa-banding techniques and the high proportion of cases from hematologic malignancies within the database may have contributed to the low estimate of ecDNA frequency (14–16).

The prescient studies of Schimke and others (7–9, 17–20) identified ecDNA as a mechanism of “unstable amplification” (7), raising the possibility that it might be functionally important in cancer. As the cancer biology field moved from traditional microscopy–based methods of chromosome analysis to DNA sequencing–based methods, progress on understanding ecDNAs was limited because the genomic location of amplified sequences was being inferred by their annotated chromosomal location. In 2014, our team found very high levels of ecDNAs encoding the constitutively active oncoprotein EGFRvIII in treatment-naïve glioblastoma cells. Relatively rapid and reversible loss of those ecDNAs after treatment with the EGFR inhibitor erlotinib was a mechanism for erlotinib resistance (21). This finding that ecDNA can promote dynamic resistance to targeted therapy highlighted the potential clinical relevance of ecDNA, prompting the development of new tools to detect ecDNA from NGS and to understand its frequency across cancer types (see “Computational Tools”).

Characterizing ecDNAs in the NGS Era

In the past 10 years, multiple studies have leveraged the widespread use of NGS in large tumor datasets and computational tools for ecDNA prediction by whole-genome sequencing (WGS; refs. 22, 23) and characterizing the highly accessible chromatin of ecDNA by Assay for Transposase-Accessible Chromatin using sequencing (ATAC-seq; refs. 24, 25). Studies of large tumor datasets have shown that ecDNAs are widespread among solid tumors in humans, are large (typically greater than 100 kb in size), can encode a wide range of oncogenes, are associated with poor patient outcomes, are absent from noncancerous cells, and are enriched in later stage tumors and metastases (22, 23, 26). In this review, we focus on the most recent and largest survey of ecDNA prevalence and content from 15,832 samples from 14,778 patients, published by us and our colleagues in the Genomics England consortium (23). We found that 17.1% of all tumor samples contain ecDNA with frequency varying widely by tumor type (i.e., 54.9% of liposarcomas, 49.1% of glioblastomas, and 0% of oligodendrogliomas). In the vast majority of tumor types, focal amplifications achieve higher overall copy number when encoded on ecDNA compared with chromosomal amplifications (23). These findings are in broad agreement with a previous study predicting ecDNA status from more than 3,000 tumors from The Cancer Genome Atlas (TCGA) and the Pan-Cancer Analysis of Whole Genomes which identified ecDNAs in 14.3% of all tumors (22).

ecDNA content varies widely between tumor types and between individual tumors. Oncogenes frequently encoded on ecDNA include MDM2, EGFR, ERBB2, CCND1, and MYC (23). The pattern of oncogenes amplified on ecDNA varies by tissue type, suggesting a prominent role for selection during tumor development and progression (22, 23). Whereas most ecDNAs contain at least one oncogene, an ecDNA often includes immunomodulatory genes and regulatory elements, which may be present with or without an oncogene on the same ecDNA (23). The ecDNA-encoded immunomodulatory genes encompass a wide range of immune-related functions; among the most frequently amplified genes in this category are PAK1, GAB2, and LRRC32 (23). The regulatory ecDNAs included promoters, distal enhancers, and long noncoding RNAs (23). Although most ecDNAs contain sequences from a single chromosome, approximately 10% of ecDNAs arise from multiple chromosomes (23). Furthermore, approximately 25% of ecDNA+ tumors in the TCGA dataset contain more than one “species” of ecDNA encoding different oncogenes, regulatory elements, and immunomodulatory genes on distinct ecDNA molecules (27).

The ecDNA Toolbox

Computational Tools

Metaphase DNA fluorescence in situ hybridization (FISH) remains the gold standard method for ecDNA detection. Metaphase FISH provides single-molecule resolution and can be analyzed with computational tools such as EcSeg (28). There are two major challenges to using metaphase FISH for ecDNA detection in tumors: (i) cells must be actively cycling, and therefore the method is not amenable to use with fixed cells, and (ii) preexisting knowledge of the amplified gene is required in order to use the correct probe. The development of new computational tools to detect ecDNA from genomic data allows for study of ecDNA frequency and content from large publicly available datasets such as TCGA and from individual tumor samples without the need for tissue culture or for prior knowledge of the amplified region. There are now numerous available tools for detection of circular genomic structures from sequencing data which can provide not only the ecDNA status but also detailed sequence information about the ecDNA structure. In this study, we focus on a curated list of tools available specifically for detecting ecDNA from existing datasets (i.e., without ecDNA enrichment; Fig. 1). There are also numerous tools available to detect the smaller eccDNAs which may also work for ecDNA detection, but these are beyond our current focus. For scientists looking to assess other aspects of ecDNA biology, like sequence heterogeneity within a sample, there are additional tools that require ecDNA enrichment: exonuclease digestion followed by rolling-circle amplification (29, 30) and CRISPR-CATCH (31), which are beyond the scope of this review.

Figure 1.

Summary of computational and experimental tools for the study of ecDNA. [Created in BioRender. Weiser, N. (2025) https://BioRender.com/j26f674.]

There are multiple tools for graph-based exploration of complex structural variants using the breakpoints derived from paired-end or split reads in tumor sample WGS data, including JaBbA and LINX (32, 33). However, in this review, we focus on a set of tools combined in a workflow called the AmpliconSuite pipeline (Luebeck and colleagues, bioRxiv 2024.05.06.592768). AmpliconSuite uses short-read WGS to identify amplified regions (seed regions) using CNVKit followed by AmpliconArchitect (AA) to reconstruct the focal amplifications and then AmpliconClassifier to identify the amplifications as ecDNA, breakage–fusion–bridge (BFB), linear, or complex (34–36). When referring to AA results, we are referring to the entire AmpliconSuite pipeline. Benchmarking AA results on cultured cell lines with corresponding metaphase FISH data, the estimated sensitivity and positive predictive value for ecDNA are 83% and 85%, respectively (22). Notably, because actual tumors vary in their purity and may contain a mix of extrachromosomal and chromosomal amplifications of the same or similar regions, the performance of AA in patient samples is likely to be lower than for cultured cell lines. In addition, once-circular amplicons that have integrated into chromosomes to form homogeneously staining regions (HSR), as in the COLO320HSR cell line, are still predicted by AA to be ecDNAs (24, 34); therefore, we consider the results of AA to reflect an amplicon’s historic status and recommend metaphase FISH validation of AA results when possible. Due to the complexity of many focal amplifications, analysis of short-read WGS with AA may not be able to unambiguously reconstruct some amplicons. In such cases, AA has been combined with optical mapping to improve reconstruction accuracy (37). These approaches are complemented by a recent technology OM2BFB that can distinguish ecDNAs from BFB-based amplifications (38).

The advent and popularization of long-read sequencing promises to be advantageous for investigations of complex genomic rearrangements such as those often present in ecDNAs and other classes of focal amplifications. There are two newly developed methods for analyzing Oxford Nanopore Technologies long-read WGS for ecDNA analysis: Complete Reconstruction of Amplifications with Long reads (CoRAL; ref. 39) and Deconvolve Extrachromosomal Circular DNA Isoforms from Long-read data (Decoil; ref. 40). Both approaches output cyclical paths for downstream analysis.

One exciting development in the field is the development of new computational tools that aim to detect ecDNA from other types of genomic data. Machine-learning approaches have been used to generate tools for ecDNA detection from whole-exome sequencing (41), which seems to work well in validated ecDNA+ samples, although it is not clear whether this approach is able to distinguish ecDNAs from other classes of focal amplifications. Other approaches aim to exploit the enhanced chromatin accessibility of ecDNA in order to detect its presence from bulk ATAC-seq (42, 43). More recently, a droplet-based approach utilizing Hi-C technology has identified ecDNA in a cancer cell line, allowing for simultaneous analysis of structural variants and chromatin structure at single-cell resolution (44).

Experimental Systems

Cell Lines

There are many publicly available ecDNA+ cell lines. AmpliconRepository (www.ampliconrepository.org) is a new, freely available platform that reports AA predictions of ecDNA status and content from numerous datasets, including the Cancer Cell Line Encyclopedia. Although most predicted ecDNA+ cell lines do not have an analogous HSR-containing line, there are a small number of near-isogenic cell line pairs that we and others have explored in order to understand differences between focal amplifications on ecDNAs versus chromosomes. These include COLO320DM/HSR (colorectal adenocarcinoma, MYC-amplified; ref. 45); GBM39ec/HSR (glioblastoma, EGFR-amplified; ref. 10), PC3-DM/HSR (prostatic adenocarcinoma, MYC-amplified; ref. 46), and STA-NB-10/dmin/hsr (neuroblastoma, MYCN-amplified; ref. 47). These are thought to reflect the occasional integration of ecDNAs into chromosomes at the non-native locus (i.e., ectopic HSRs) in subclonal populations (18, 46–48). Thus, the cell line pairs contain very similar amplicons in a near-identical genetic background. Notably, the integrated sequence can differ in biologically significant ways from the originating ecDNAs. For example, in the COLO320 cell line pair, the ecDNA amplicon includes two isoforms of MYC: the first isoform has two copies of a fusion of the PVT1 promoter and exon 1 with the second and third exons of MYC, and the second isoform is single copy of the full-length canonical MYC (45). Interestingly, the amplicon in COLO320HSR cells, which is thought to have arisen from the COLO320DM cell line, includes mostly canonical MYC without the PVT1 fusion (18, 45, 48).

When working with new cell lines, we recommend initial validation of the ecDNA status by the gold standard method of metaphase FISH to confirm the ecDNA status and establish the frequency of ecDNA+ cells. In our experience, the initial culture obtained from public repositories can contain a mix of ecDNA and HSR cells, particularly in the case of COLO320DM. Depending on experimental needs, single-cell expansions can be helpful to establish a pure ecDNA+ or HSR line. In our experience, the purity and stability of the ecDNA status is very high after single-cell expansion.

We sometimes observe rare, spontaneous integration events in our own established ecDNA+ cell lines. Therefore, we advocate for periodic examination of ecDNA status by metaphase FISH to confirm the stability the of ecDNA status in cell lines that have undergone serial passage. When integration events are observed at a reasonable frequency, we have successfully performed single-cell expansions to generate pure populations of ecDNA+ and HSR+ cells to generate new cell line pairs such as in the PC3 line (46).

Systems for Inducing ecDNA

There are several recently developed methods to induce ecDNA, allowing the investigation of ecDNA biogenesis and its role in early tumorigenesis: Cre-loxP–based ecDNA engineering in vitro and in vivo, CRISPR-C, and drug-induced chromothripsis (Fig. 1). The most recent strategy for inducing ecDNA in both cultured cancer cells and primary mouse cells utilizes Cre recombinase to circularize a region of interest (ROI) flanked by two loxP sites in the same orientation (49). This circularization generates a functional fluorescent reporter, allowing for the detection and sorting of ecDNA+ cells after treatment with Cre. The introduction of an antibiotic resistance gene within the circularization cassette allows for selection of cells that maintain the ecDNA. This strategy has been used successfully to generate both MDM2- and MYC-encoding ecDNAs, and the engineered ecDNAs recapitulate well-established features of naturally occurring ecDNAs, including copy-number amplification, open chromatin, and clustering in interphase cells (21, 24, 45, 49, 50). Importantly, this system has been used to induce ecDNAs in immunocompetent mice, a significant advance over previously established xenograft strategies (49).

Whereas utilization of a Cre-based system for inducing ecDNA will be a powerful tool for understanding the contribution of ecDNA to tumor development and evolution, the use of exogenous recombinase precludes investigation of the endogenous mechanisms for ecDNA formation. CRISPR-C is an alternative method for inducing ecDNA from a ROI, using Cas9 in complex with guide RNAs flanking the ROI (51, 52). Interestingly, the induction of ecDNA in this system is deleterious to cells in the absence of a selective pressure to promote ecDNA retention (i.e., methotrexate treatment to facilitate retention of induced ecDNA encoding DHFR; ref. 52). The CRISPR-C system establishes that the synchronous generation of two cuts on the same chromosome is sufficient for ecDNA biogenesis, although ecDNAs are only detected in a fraction of cells (∼5–15; refs. 52, 53). Treatment with methotrexate results in a dose-dependent copy number increase of DHFR ecDNAs generated by CRISPR-C, suggesting that these ecDNAs are functional (52). Further studies are needed to better understand the mechanisms involved in ecDNA formation, maintenance, and repair.

In primary mouse cells, ecDNA generation by the Cre-based method is sufficient to immortalize the cells, and the ecDNAs are maintained through serial passages without the addition of a selective pressure (49). In cancer cells, however, both Cre-based and CRISPR-C–based methods of ecDNA generation require a selective pressure to maintain the ecDNAs over serial passages, likely because the cancer cells are already so well-adapted to cell culture that the induced ecDNA does not provide any growth advantage under normal culturing conditions (49, 52, 53). Previous studies have shown that the use of selective pressure alone, without any directed DNA cutting or recombination, is also sufficient to induce ecDNA (54). Treatment of multiple cell lines (HeLa, HT-29, and 293T cells) with methotrexate drives DHFR amplification in both ecDNAs and HSRs. Further drug treatment of HSR-containing clones resulted in chromothriptic events and subsequent ecDNA generation, highlighting the dynamic nature of focal amplifications (54). ecDNAs can also arise from other genomic rearrangements such as translocation bridges and BFBs (55, 56).

Features of ecDNA+ Cancers

The finding that ecDNA+ cancers have worse outcomes compared with patients with chromosomally encoded focal amplifications suggests that ecDNAs harbor unique biology that drives tumor progression and/or escape from therapy (22, 23). In the past 10 years, we and others have unveiled many aspects of ecDNA biology that may contribute to poor patient outcomes (Fig. 2). In this study, we highlight some of the most recent and significant findings. For a more comprehensive review of ecDNA biology, please see refs. (57, 58).

Figure 2.

Hallmarks of ecDNA+ cancers. [Created in BioRender. Weiser, N. (2025) https://BioRender.com/x64h263.]

ecDNA Formation

Recent experimental and correlative studies indicate that there may be a diversity of mechanisms by which ecDNA forms, including two double-stranded DNA breaks in cis, by breaking off from translocation–bridge amplifications or breakage–fusion bridges, by chromosome shattering (chromothripsis), or secondary to mitotic errors (53–56, 59–62). Importantly, ecDNA-based amplification usually occurs in the context of tumor-suppressor loss, most frequently TP53, although this seems to be tissue context–dependent, with other tumor-suppressor losses occurring more frequently in different tissue types (23). Furthermore, our study of the development of ecDNA during the transition of Barrett’s esophagus to esophageal cancer suggested that mutation or loss of TP53 loss precedes ecDNA-based amplification during the process of transformation (35).

Experimental studies have interrogated the specific pathways involved in ecDNA formation using both the CRISPR-C and chromothripsis approaches to induce ecDNA as described above (53, 54). Our study using the CRISPR-C approach to induce two DNA breaks in cis revealed that both nonhomologous end joining and microhomology-mediated end joining participate in ecDNA formation (53). The chromothripsis method for ecDNA biogenesis is dependent on PARP and DNA-dependent protein kinase, suggesting a dependence on nonhomologous end joining for ecDNA formation (54, 63).

Tumor Heterogeneity and Tumor Evolution

Because ecDNAs segregate asymmetrically to daughter cells, ecDNA copy number within a tumor is highly heterogeneous and facilitates rapid adaptation to stresses, including metabolic perturbations and treatment with targeted therapies (21, 52). ecDNA status itself is also not uniform within tumors. In the Genomics England study, multiple regions were sampled from the tumors of 578 patients. In more than 60% of the ecDNA+ tumors with multiple samples, ecDNAs were only detected in a subset of samples (23). Experimental evidence shows that ecDNAs can integrate into chromosomes, usually at non-native loci to form ectopic HSRs, and that HSRs can be shattered to generate ecDNAs (18, 21, 49, 54). Although the precise mechanism for ecDNA integration has yet to be established, double-stranded breaks on the ecDNA seem to promote integration into chromosomes (54, 64).

Although the precise consequences of ecDNA copy-number heterogeneity within a tumor are largely unknown, a recent study suggests that ecDNAs encoding MYC contribute to tumor heterogeneity in small cell lung cancer (65). This study generated cell lines from a patient with extensive disease, sampling from the primary tumor and three metastatic sites. Whereas all four cell lines shared similar MYC amplicons, the primary tumor and adrenal metastases were primarily MYC ecDNA+ with higher copy number than the liver and lymph node metastases which were primarily MYC HSR+. Assessment of the levels of expression of the lineage-defining transcription factors ASCL1 and NEUROD1 also showed heterogeneity between the four cell lines. These findings suggest that the heterogeneity of ecDNA copy number and ecDNA status within a single tumor may drive heterogeneity of cell states or even metastatic potential (65). In neuroblastoma and medulloblastoma, ecDNAs also drive heterogeneity of transcriptional programs and cell states within a tumor (66, 67).

Altered Transcriptional Landscapes in Space and Time

ecDNAs achieve higher levels of transcription compared with chromosomal amplifications (10, 24). There are multiple biological features that seem to drive this high transcriptional output, including highly accessible chromatin on ecDNA (24) and rewiring of enhancer–promoter interactions in cis, as circular topology generates new enhancer–promoter interactions, and in trans through ecDNA hubs (45, 50, 66, 68, 69).

Our recent studies reveal that the high transcriptional activity of ecDNA extends beyond the mere magnitude of oncogene expression. Analysis of nascent transcription from ecDNA versus chromosomal amplicons in the COLO320 near-isogenic cell line pair revealed that transcriptional upregulation is widespread throughout the ecDNA amplicon, including noncoding, antisense, and intergenic transcripts, some of which have not been previously annotated (46). Furthermore, oncogenes on ecDNAs, but not on corresponding HSRs, remain transcriptionally active at the onset of mitosis (27).

Our recent study also revealed an unexpected role for ecDNA transcription in driving coinheritance of ecDNA in daughter cells (27). To explore cosegregation, we leveraged the observation that some cell lines and 25% of primary tumors contain more than one ecDNA species (27). For example, the SNU16 cell line contains one ecDNA encoding MYC (ecDNA-A) and a separate ecDNA encoding FGFR2 (ecDNA-B; ref. 45). Using multiple orthogonal methods, we found that in cells with two or more ecDNA species, the copy numbers of the different ecDNAs are correlated with each other, indicating that ecDNAs cosegregate such that daughter cells that inherit a large number of copies of ecDNA-A are likely to also inherit a large number of copies of ecDNA-B. When transcription is inhibited by treatment with triptolide, but not other transcriptional inhibitors such as 5,6-dichlorobenzimidazole 1-β-d-ribofuranoside (DRB) or actinomycin D, the cosegregation of distinct ecDNA species is disrupted (27). This finding suggests that the persistence of ecDNA transcription into early mitosis may facilitate the cosegregation of ecDNAs into daughter cells, thus allowing ecDNAs to flout Mendel’s law of independent assortment.

Replication Stress and Genome Instability

Although early studies proposed that ecDNAs were maintained by de novo formation rather than by replication and inheritance into daughter cells (70), subsequent work indicated that ecDNAs do in fact replicate and likely only replicate once per cycle (27, 71, 72). Our recent study showed that the high level of transcription from ecDNAs induces significant replication stress on ecDNAs, as ecDNAs are enriched for markers of replication stress such as single-stranded DNA and phosphorylated RPA2 compared with chromosomal amplifications. Furthermore, replication fork speed is significantly reduced in the ecDNA+ COLO320DM cell line compared with the near-isogenic COLO320HSR line; slow replication fork progression is yet another hallmark of replication stress (46). The high degree of replication stress on ecDNA is rescued by treatment with the transcriptional inhibitor triptolide, indicating that transcription–replication conflict arising from concurrent transcription and replication on ecDNA is the source of the elevated replication stress (46). We found that ecDNAs also accumulate significantly more DNA damage compared with chromosomal amplicons and that the DNA damage is dependent on DNA replication (46), consistent with the finding that replication stress is a major source of endogenous DNA damage (73).

Our finding that DNA damage is increased on ecDNA compared with chromosomal amplifications is consistent with a growing body of evidence highlighting the importance of ecDNA as a mutagenic driver, rather than merely a symptom of underlying genomic instability. A recent analysis comparing transcriptional profiles of ecDNA+ versus ecDNA– tumors from the TCGA dataset showed that ecDNA+ tumors persistently upregulate specific DNA damage repair pathways compared with ecDNA– tumors, particularly genes involved in nonclassic nonhomologous end joining and homology-directed repair (74). Interestingly, analysis of the ecDNA amplicons from the Genomics England consortium showed that the ecDNAs themselves are significantly enriched for the SBS3 mutational signature (homologous recombination deficiency; ref. 23). The homologous recombination deficiency signature seems to arise after ecDNA formation, as the associated mutations were observed on only a subset of ecDNA copies within the sample (23). ecDNAs are also enriched for clustered mutations generated by APOBEC3 (23, 75). Data from urothelial carcinomas suggest that APOBEC3-mediated mutagenesis occurs concurrently with or shortly after ecDNA biogenesis, raising the possibility that it may play a direct role in ecDNA biogenesis as well as ecDNA evolution (76). A possible role for APOBEC3-mediated mutagenesis in ecDNA biogenesis is supported by previous studies showing co-occurence of chromothripsis and clustered APOBEC3-mediated mutations (kataegis) and that ecDNAs can arise from HSRs that undergo chromothripsis (54, 77).

Looking to the Future—Implications of ecDNA in the Clinic

Prognostic Value of ecDNA

As the association between ecDNA and poor patient outcomes has become clear, one critical question has been the extent to which ecDNA is a driver of tumor progression versus an indicator of a highly unstable genome without a direct role in tumor evolution. Indeed, given the wide range of cancer types and diversity of ecDNA cargoes, it is certainly possible that some ecDNAs are more bystanders to tumorigenesis rather than participants. However, high copy amplification of such elements suggests that they are being selected for tumor growth, and multiple recent studies provide strong evidence that ecDNAs can play a role in cancer initiation and progression. Taken together, these findings suggest that the detection of ecDNA in patient tumors may be clinically actionable.

The recent development of an inducible ecDNA system in both cultured human cells and mouse primary cells provides important evidence that ecDNAs can in fact drive cancer initiation (49). The induction of MDM2-encoding ecDNAs was sufficient to immortalize primary mouse cells and to drive oncogenesis in the setting of overexpression of a second oncogenic driver (49). These findings suggest that ecDNAs contribute to early tumorigenesis and are consistent with our recent study investigating the ecDNA status in biopsies of patients with Barrett’s esophagus, a precancerous metaplasia that gives rise to esophageal adenocarcinoma. Analyzing WGS data with AA, we found that ecDNA is present in high-grade dysplasia and is associated with the eventual development of invasive cancer (35). Based on these findings, we ​​speculate that the presence of ecDNA may be a useful prognostic marker to predict whether a precancerous lesion is likely to undergo cancerous transformation.

Another potential clinical use for ecDNA detection is in the prediction of therapeutic response. Experimental evidence suggests that ecDNAs facilitate escape from targeted therapy against the amplified oncoprotein by rapid modulation of copy number and/or by integration into chromosomes (21, 27, 46, 52, 78). ecDNAs have also been implicated in resistance to non-targeted chemotherapy. For example, in a study of urothelial carcinoma, ecDNAs encoding CCND1 increased in copy number in patients undergoing chemotherapy, suggestive of positive selection for high ecDNA copy number and a role in chemotherapy resistance (76). In patient-derived xenografts from small cell lung cancer, ecDNA-mediated amplification of MYC paralogs has been associated with the development of cross-resistance to chemotherapy (79). The finding that “new” ecDNA amplifications arise after drug treatment may be a consequence of selection and suggests a mechanism for the observation of “oncogene switching” in some cancers.

ecDNA and the Immune System

Multiple studies have reported downregulation of immunomodulatory genes in ecDNA+ tumors (22, 74, 80), suggesting that ecDNAs may suppress immune infiltration or that ecDNAs are more likely to persist in the setting of immune suppression. The mechanisms by which ecDNA modulates, or responds to, the immune system remain unclear and are a critical area for further study. The recent development of immunocompetent mouse models engineered for Cre-mediated ecDNA induction will help clarify the relationship between ecDNA and the immune system (49). One intriguing possibility is that ecDNAs may serve as platforms for the amplification and selection of not only oncogenes but immunomodulatory genes as well (23, 35). In the Genomics England cohort, 34% of ecDNA+ tumors had immunomodulatory genes with ecDNA amplification. In most cases, immunomodulatory genes were coamplified with nearby oncogenes, making it difficult to ascertain their functionality. However, tumors from more than 10% of ecDNA+ patients showed ecDNA-mediated amplification of immunomodulatory genes without oncogene amplification (23). In addition, tumors with immunomodulatory ecDNAs were significantly depleted of T cells compared with tumors harboring ecDNAs with oncogenes only (without immunomodulatory genes), suggesting that immunomodulatory genes on ecDNA may play a role in immune suppression (23).

Targeting ecDNA

The prevalence of ecDNA in diverse tumor types and its association with poor patient outcomes and therapeutic escape makes the processes involved in ecDNA formation, maintenance, and function appealing therapeutic targets. Given the ability of ecDNA to evade targeted therapy by modulating copy number and its immune-cold status (23, 52, 74), however, ecDNA poses a significant therapeutic challenge. Therefore, identifying vulnerabilities generated by ecDNA is a critical priority. One can envision targeting key components that ecDNA-containing tumors uniquely or preferentially require; such therapies could be given alone or in combination with targeted, immunologic, or conventional treatments as a promising future direction. One obvious initial area of interest resides in the DNA damage and repair mechanisms that may be altered in ecDNA-containing cancers.

Our recent finding that ecDNAs experience high levels of replication stress suggested that the S-phase checkpoint, which responds to replication stress by slowing replication forks, decreasing origin firing, and preventing cell-cycle progression, is one promising therapeutic target for ecDNA+ cancers (46). Indeed, we found that ecDNA+ cells exhibit high levels of S-phase checkpoint activation, measured by quantification of phosphorylation of CHK1, a key checkpoint mediator (46). In line with this finding, ecDNA+ cells are highly sensitive to inhibition of CHK1. We tested the efficacy of CHK1 inhibition on ecDNA+ cells using a xenograft of SNU16 cells which contain FGFR2 amplified on ecDNA. When targeted with monotherapy with the anti-FGFR2 compound infigratinib, FGFR2 copy number increases to out-titrate the drug, allowing for the development of therapeutic resistance and continued tumor growth. The combination of infigratinib with anti-CHK1 therapy, however, produced a potent antitumor effect, leading to decreased FGFR2 copy number and shrinkage of the tumor (46). These findings suggest that combination therapy against CHK1 and the amplified oncoprotein could be a promising strategy for treatment of ecDNA+ cancers.

Challenges and Opportunities

The ecDNA field has made tremendous progress over the past 10 years, from demonstrating the prevalence and significance of ecDNA amplifications to advancing a promising therapeutic target for ecDNA+ cancers. There are many remaining challenges that will need to be overcome in order to truly bring this field to fruition by improving outcomes for patients with ecDNA+ cancers. These include (i) the development of new ecDNA diagnostic tests for patient tumors that are less resource-intensive than WGS and more amenable to clinical deployment, (ii) longitudinal studies to understand the role of ecDNA at every stage of tumor initiation and progression, and (iii) the expansion of ecDNA studies to new tumor types and experimental models.

The impact of ecDNA on the tumor microenvironment has emerged as a major new area for investigation. The findings that ecDNAs can encode immunomodulatory genes and are associated with relatively immune-cold tumors suggest that ecDNAs may play an immunosuppressive role in cancer (23). Understanding the mechanisms by which ecDNA may regulate the immune system, both innate and adaptive, may reveal new therapeutic opportunities for ecDNA-driven cancers. The recent development of a system for inducing ecDNAs in immune-competent mice will be a valuable tool in these investigations (49).

Our recent finding that the high degree of replication stress sensitizes ecDNA+ cells to inhibition of CHK1 indicates that the unique biology of ecDNAs can create therapeutic vulnerabilities for ecDNA-driven cancers (46). As future investigations uncover the mechanisms driving this biology, we anticipate a broader landscape of potential targets that may be accessed with new chemistry-first approaches (27, 81) to exploit unique features of ecDNA biology, including how it transcribes, replicates, repairs, and segregates during cell division, as well as how it suppresses the immune system.

Supplementary Material

Figure S1

Summary of key differences between ecDNA and eccDNA

Authors’ Disclosures

H.Y. Chang reports grants from Cancer Research UK and the NCI during the conduct of the study and other support from Amgen, Accent Therapeutics, Boundless Bio, Cartography Bio, Orbital Therapeutics, Arsenal Bio, nChroma, and Exai Bio outside the submitted work; in addition, H.Y. Chang has a patent for ecDNA licensed to Boundless Bio. P.S. Mischel reports other support from Boundless Bio outside the submitted work. No disclosures were reported by the other authors.