In the 1980s, several transmission electron microscopy (TEM) studies suggested that ecDNA was circular in shape and consisted of chromosomal fibers containing nucleosomes (13–15). In 2019, Wu et al. uncovered additional details regarding the structure of ecDNA by combining ultrasound imaging, optical imaging, and WGS analysis (16), ultimately determining that ecDNA tended to be less compact and more accessible than chromosomal DNA, resulting in increased expression of ecDNA loci (16). In addition, this study suggested that ecDNA could be present in high copy numbers, leading to significant overexpression of genes, including oncogenes (16). Similarly, a WGS analysis performed by Kim et al. showed that amplification of ecDNA molecules resulted in higher expression of oncogenes compared with linear DNA (6). Therefore, ecDNA is currently characterized as circular molecules with relatively de-compacted chromatin, leading to increased levels of gene expression.

Previous studies have indicated a strong correlation between ecDNA formation and genome instability (17, 18). Recent research has shown that various molecular events associated with chromosomal damage can lead to the formation of ecDNA. These events include chromothripsis, religation of DNA, breakage-fusion-bridge (BFB) cycles, fork stalling, and template switching ( Figure 1 ).

ecDNA is involved in the development and evolution of cancer in several ways, including promoting oncogene expression by increasing copy number or increasing chromosomal interaction, and serving as a reservoir for DNA recombination by reintegrating into and being excised from chromosomes ( Figure 2 ).

ecDNA can be visualized with fluorescence microscopy, with the spatial location of focal amplifications revealed. Previous studies indicate ecDNA can be stained with the fluorescent dye DAPI (4,6-diamidino-2-phenylindole) and confirmed through genomic DNA and centromeric FISH (fluorescence in situ hybridization) probes (5). In interphase cells, FISH probes against target oncogenes generally reveal positive sites for both chromosomal and extrachromosomal genes. While during metaphase, the compact alignment of chromosomes enables unambiguous localization of specific genes within the chromosomes or ecDNA.

However, the combination of DAPI and FISH does not allow for unbiased quantification of ecDNA copy number, which is required to evaluate the inter-cell heterogeneity. To solve this problem, Turner et al. developed an image analysis software called ECdetect (5), which provides a robust, repeatable, and accurate quantification for ecDNA from DAPI-stained metaphases in a semi-automated fashion. ECdetect applies two coarse adaptive thresholds on DAPI images to precisely detect ecDNA. The initial coarse adaptive threshold recognizes the major components in the image and smaller components are discarded. Then the recognized major components are computed to detect the chromosome regions. Then, the ecDNA search region is masked and verified manually. In the search region, the second adaptive thresholding with a smaller window size that detects components with sizes between 3 pixels and 75 pixels, the size of ecDNA. Their results prove detection accuracy with the evidence that the computed result is highly correlated with visual detection (r = 0.98) in 2,572 metaphase cells.

However, several unique features of ecDNA hinder visual detection and algorithms based on conventional computer vision approaches. ecDNA is generally irregularly shaped and small-sized. Meanwhile, a typical metaphase cell image has a high noise ratio and contains a large portion of the background signal, including nuclei, and chromosomes that need to be distinguished from ecDNA. Hence, to detect ecDNA more accurately, more sophisticated tools are required. Recently, deep neural networks, specifically convolutional neural networks, demonstrated remarkable image-processing abilities on biological datasets (59). A deep neural network-based platform called ecSeg (50) has been developed for ecDNA detection. EcSeg is developed based on U-net (60), a convolutional neural network that has been long used for biomedical image segmentation. It incorporates DAPI signals and FISH data to specifically clarify the location of oncogene amplification on both ecDNA and chromosomes. With the help of ecSeg, each pixel of the metaphase cell image is classified into one of the following classes: ecDNA, chromosome, nucleus, and cytoplasm. Then, connected pixel components of ecDNA are connected to demarcate and count one single copy of ecDNA. Last, a separate post-processing step is performed to correlate objects with ecDNA and chromosomes, using FISH probe results. The test result shows ecSeg delivers better performance than ECdetect. EcSeg achieves a mean precision and recall values of 82% at the image level, while ECdetect rarely achieved recall above 50% (3). Additionally, ecSeg achieves a higher F1 score as well (50).

With the advancement of sequencing technologies, bioinformatic analyses have provided many approaches to reconstruct ecDNA from NGS data and detect ecDNA.

Circle-seq was originally designed to detect ecDNA in yeast systems (51), and has been successfully adopted by several studies on human ecDNA (23, 61). Circular DNAs are separated by concise alkaline treatment and column chromatography, then purified from linear chromosomal DNA and mitochondria DNA by DNase digestion. Then, ecDNA is further enriched by φ29 rolling circle amplification and sequenced for mapping. Another approach to purify and amplify ecDNA is Circle_finder (52), which utilizes adenosine 5’-triphosphate (ATP)–dependent exonuclease to remove linear DNA and random primers to amplify the ecDNA. Circle_finder has a bias for ecDNA with a smaller size (200-400bp) and has been proven to detect ecDNA from the serum of cancer patients (62).

Circle-Map (53) has been developed to accurately detect circular DNA in genome sequencing data. Standard short read aligners from WGS do not reliably detect split-reads in the sequencing result that indicate circle breaking points, especially when read signals are shorter than 19 base pairs. Thus, Circle-Map utilizes discordantly mapped paired-end reads to realign the soft clipped parts resulted from breakpoints, which enables a more accurate detection of circle breakpoints. The result shows very short soft clips(>4nt) can be accurately aligned with Circle-Map.

Since ecDNA may account for focal amplification in almost half of the cancer samples (5), AmpliconArchitect (55) was developed to reconstruct ecDNA amplicon (non-overlapping amplified genomic components) structures with whole-genome sequencing data(WGS). AmpliconArchitect is an extension of structural variant (SV) analyses. In this algorithm, short, pair-end reads mapped to the reference genome and a seed interval in an amplicon are input and used to search for other intervals in the amplicon. SV analysis is then conducted to build a breakpoint graph, and copy number variant (CNV) analysis is carried out by optimizing a balanced flow on the breakpoint graph. However, the construction of a breakpoint graph can be inaccurate due to the duplicated regions inside the amplicon and leads to errors in the estimation of copy number. Thus, AmpliconReconstructor (AR) algorithm is proposed to refine the breakpoint graph reconstruction, by integrating the long-range optical mapping (OM) of long DNA fragments with NGS data. AR reconstructs genomic scaffold with OM data before the identification of breakpoint graphs, and then reconstructs ecDNA with high fidelity, which can not only provide accurate quantification of ecDNA but also contribute to the understanding of the formation mechanisms of ecDNA (56).

With the advent of third-generation long-read sequencing, characterization of the full landscape of ecDNA has become possible. A newly developed algorithm, ecc_finder (54) adopts nanopore reads to identify eccDNA. ecc_finder identifies the tandem repeat patterns as candidate loci from read alignments since circular DNA will display two or more sub-read alignments in the same direction. Then it removes the reads generated from linear genome repeats according to the reference genome, which leaves true ecDNA reads and allows for the detection of ecDNA originating from repeated loci.

CRIPSR-Cas system has been widely used in sequence-specific genome editing, since Cas can be directed to a specific sequence with the help of guide RNA. This feature allows researchers to design CRISPR-based visualization tools that target specific gene sequences carried by ecDNA and highlight ecDNA with fluorescent markers. ecTag (57), a CRISPR-based visualization pipeline, first detects ecDNA breakpoint junctions with AmpliconArchitect, and then designs sgRNA that targets the breakpoint sequence. After that, the Casilio system that combines dead Cas9 labeling and Pumilio RNA then recruits multiple fluorescent protein molecules, making the visualization possible. Since the CRISPR-Cas system and Casilio system are not lethal to cells, ecTag can be used for living cell imaging. The dynamics of ecDNA during mitosis have been observed with ecTag, supporting the theory that ecDNA is not equally distributed for two daughter cells. Another CRISPR-base tool, CRISPR-CATCH (63) is originally designed for targeted cloning of genomic sequences as long as 100 kb. The tool is then adopted to enrich megabase-sized ecDNA from cell lines or patient tissues (58). This approach does not require cut or replicate of ecDNA, preserves the topological structure and molecular features of ecDNA, and allows for detailed analyses of both the genetic sequence and epigenomic landscape.

Although tumor tissue biopsy is considered the gold standard for cancer subtyping, it provides only a snapshot of the disease and is invasive and difficult to replicate. To overcome these disadvantages, liquid biopsies are gaining favor as a new diagnostic tool to complement conventional biopsies. Well-defined biomarkers used for liquid biopsy include circulating tumor cells (CTCs) (74), circulating tumor DNA (ctDNA) (75), and extracellular vesicles (EVs) (76). Also, previous studies confirm that circular DNA can pass through cell membranes much more easily than linear DNA (77), and studies have also shown that extrachromosomal circular DNA can be released into circulation from in situ tumors (62, 78).

ecDNA is a valuable tool for monitoring tumor progression and predicting tumor prognosis, as it can contribute greatly to oncogene copy number variation and is associated with drug resistance. For example, it has been suggested that longer ecDNAs are enriched in tumor samples from patients with lung cancer compared with paired controls (62). Recently, next-generation sequencing (NGS) of extrachromosomal DNA from plasma samples of 6 lung adenocarcinomas (LADs) and 10 healthy controls suggested that the frequencies of nine ecDNAs were higher in cancer samples (79). In addition, four ecDNAs were specifically expressed in LADs and were promising biomarkers for early diagnosis (79). Similarly, a recent study of LAD showed that the expression profiles of several ecDNAs were significantly different between LAD and normal tissues (73). When used to discriminate patients with LAD from healthy controls, serum levels of two ecDNAs achieved higher areas under the curve (AUCs) than serum carcinoembryonic antigen (CEA) and CYFRA21-1 (cytokeratin 19 fragments), which are among the most sensitive LAD tumor biomarkers (73).

ecDNA is associated with patient prognosis in several types of cancer, with different prognostic values for different genes. Analysis of patient samples has suggested that survival is compromised in patients with ecDNA in various cancers (6). Specifically, patients with glioblastoma have a significantly shorter relapse-free time if ecDNA is detected in primary tumor samples (42). In gastric cardia adenocarcinoma, focal amplification of ERBB2 is correlated with a better prognosis, whereas focal amplification of EGFR is associated with a worse prognosis (69). Further, the survival of ERBB2-positive patients was lower compared with ERBB2-negative patients within 2 years of diagnosis, but the trend was reversed after 2 years of survival (69). As focal amplification was mainly driven by ecDNA (56), the detection of ERBB2 and EGFR ecDNAs should be considered as potential prognostic biomarkers in gastric cardia adenocarcinoma. Furthermore, in high-grade serous ovarian cancer (HGSOC), ecDNA with DNMT1 was associated with a worse prognosis in primary and metastatic lesions, suggesting that ecDNA could be a prognostic indicator for patients with HGSOC (66). Therefore, evidence from different cancer types supports that circulating ecDNA could be developed as a prognostic biomarker in various contexts.

Several mechanisms have been proposed to explain the relationship between ecDNA and prognosis. ecDNA can actively contribute to genome remodeling and lead to important functional and clinical consequences (23). In addition, ecDNA can contribute to poor outcomes through copy number variation (80). For example, focal amplification has been shown to be associated with patient mortality (80), and the importance of oncogene amplification in cancer pathogenesis is well recognized (81). Genomic amplification can result from double-strand break events such as tandem duplication (82), break–fusion–bridge cycles (83), and chromothripsis (19).

ecDNA also contributes to resistance to targeted therapy. Oncogene amplicons carried by ecDNA can accelerate the development of intra-tumoral heterogeneity. As a result, a subpopulation of cells is more likely to express the untargeted oncogenes at levels that maximise tumor proliferation and survival, making the tumor more aggressive and resistant to previous treatments (5, 84). For example, amplification of DHFR ecDNA has been correlated with methotrexate resistance in in vitro studies (49, 85, 86), and increased copies of DHFR on ecDNA were observed in the methotrexate-resistant S-180 murine cell line. In addition, a stable amplification of DHFR was formed during the culture of S-180 cells with methotrexate (86). Similarly, ERK inhibitor-treated lung cancer or melanoma xenografts had amplification of BRAF^V600E , and this study further showed that combined treatment with RAF, MEK, and ERK inhibitors could inhibit the expansion of high copy number BRAF^V600E subclones (87). A recent study showed that MET ecDNA amplification correlated with resistance to ROS1 tyrosine kinase inhibitors both in vitro and in vivo, indicating that MET detection and targeting could be a strategy to overcome resistance to this class of drugs (72). In addition, amplification of RAB3B on ecDNA could confer cisplatin resistance by inducing autophagy in a hypopharyngeal squamous cell carcinoma (HSCC) cell line (88). Thus, increased ecDNA copy number is a common mechanism of drug resistance occurring during cancer treatment, and strategies to prevent or target amplification of oncogenes or drug resistance genes carried by ecDNAs could be used to combat adaptive responses to cancer therapies.

Studies have also shown that ecDNA can serve as a latent reservoir for genomic alterations that can be targeted for therapeutic purposes. Under changing environmental conditions, the number of amplicons in ecDNA can be naturally regulated and even reduced to zero, resulting in a complete loss of ecDNA (24, 89). Drug treatment could induce resistance by decreasing ecDNA copy number, whereas ecDNA could regain its copy number after drug withdrawal, such as has been shown for the response of EGFR ecDNA to erlotinib treatment (89). Therefore, the recovery of drug-sensitive targets on ecDNAs should also be considered as a means to overcome drug resistance.

As ecDNA plays an essential role in cancer development and drug resistance, targeting ecDNA is emerging as a new approach to cancer treatment. Hydroxyurea was able to restore vinblastine sensitivity of cancer cells by reducing drug resistance genes on ecDNA in human and hamster cell lines (90). Another study showed that low doses of hydroxyurea could reduce ecDNAs in cancer cells in some patients, resulting in a prolonged period of stable disease, although the study was limited by a small sample size (91). Hydroxyurea cannot be used as a clinical anti-cancer treatment, but new drugs could be developed based on this mechanism to reduce extrachromosomal amplicons with the aim of sensitizing cancer cells to chemotherapy.

PARP-dependent DNA repair following chromothripsis can re-integrate ecDNA into the genome, thus promoting oncogene amplification and drug resistance (36). Therefore, adding PARP inhibitors to standard chemotherapy may restrain gene amplification requiring PARP-dependent end-joining DNA repair, thus suppressing tumor progression and drug resistance (36). However, targeting PARP-dependent DNA repair could also increase genome instability or generate new genetic variations that could accelerate cancer progression. DNA repair inhibitors could thus be used as adjunct therapy to existing anti-cancer drugs, although their potential side effects should be carefully considered.

Another potential therapeutic target is ecDNA interactions. Previous studies have shown that ecDNA can interact with each other to form an ecDNA hub and induce overexpression of oncogenes, and that BET is essential for these interactions (45). Therefore, it is possible that a BET inhibitor may be able to dissociate ecDNA hubs and thus improve treatment response. Further research into the formation and maintenance of ecDNA and ecDNA hubs is needed before BET inhibitors can be used in cancer treatment.

In addition, a recent study showed that ecDNA correlates with immune evasion of cancer based on WGS and gene expression data (92). In tumors with ecDNA, the number of immune cells and the expression of antigen-presenting genes were reduced, suggesting that the presence of ecDNA may contribute to immune evasion in cancer (92). Thus, it is possible that reducing ecDNA and inducing antigen presentation pathways could restore immune surveillance against cancer. Future studies to validate and elucidate the relationship between ecDNA, the immune system, and cancer could have broad implications for various therapeutic approaches, including identifying combination therapy strategies with immunotherapeutics.

Table 3 summarizes current clinical implications of ecDNA in cancer.