Approximately 30% of ecDNA exist in pairs within the nucleus, and thus, for a long time, they were referred to as double minutes (4). ecDNA has a complex structure, lacks centromeres, and can originate from multiple chromosomes (5). The frequency, copy number, and size of ecDNA vary greatly among different tumors (3). The prevalence of ecDNA across various cancers is shown in Figure 1 . Typically, ecDNA exists in a circular form (4). However, ecDNA is distinct from extrachromosomal circular DNA (eccDNA). eccDNA has a small molecular weight (<1 kb), does not undergo amplification, does not contain complete gene sequences or regulatory elements, does not carry mutated genes, and can appear in normal tissues (5–11). ecDNA has a large molecular weight (>100 kb), undergoes clonal selection, possesses self-replication and amplification capabilities, can contain oncogenes, regulatory elements, recombinant genes, and mutated genes, and is rare in normal tissues (5, 6, 9–12).

ecDNA can carry a variety of common oncogenes, such as MYC, MYCN, Jun, KRAS, MYCL, MDM2, epidermal growth factor receptor (EGFR), fibroblast growth factor receptor 2 (FGFR2), platelet derived growth factor receptor alpha (PDGFRA), erb-b2 receptor tyrosine kinase 2 (ERBB2), and cyclin-dependent kinase 4 (CDK4), among others (13, 14). Even when carrying the same oncogene, ecDNA can exhibit substantial variability in both size and sequence composition (15). ecDNA with distinct sequence architectures are referred to as ecDNA species. ecDNA can simultaneously carry multiple oncogenes (13), and share adjacent regulatory regions (16). In addition, there exists a class of ecDNA that does not carry oncogenes but only carries promoters, enhancers, or long noncoding RNA (lncRNA) regulatory elements, referred to as regulatory ecDNA (3). Compared to ecDNA carrying oncogenes, regulatory ecDNA have a simpler structure, smaller size, and lower copy number (3). Whole-genome sequencing of human papillomavirus mediated oropharyngeal cancer (HPVOPC) revealed that HPVOPC contains ecDNA composed of host genome and HPV-host genome hybrids, and both types of ecDNA can carry multiple oncogenes (17). Additionally, ecDNA can also carry immune regulatory genes and inflammation-related genes (3, 13). Gene function enrichment analysis confirmed that genes carried by ecDNA are often upregulated in biological processes such as cell cycle, cell division, and DNA damage, and downregulated in processes related to the immune system (18).

During mitosis, the spindle apparatus pulls the centromeres of chromosomes, guiding their alignment and equal segregation to ensure that daughter cells have identical chromosomal DNA. Multiple studies have shown that the copy number of ecDNA among tumor cells exhibits significant heterogeneity (19, 20). ecDNA does not appear to follow Mendelian inheritance during cell division, differing from chromosomal inheritance patterns (21). FISH-based methods combined with unbiased image analysis have shown that after mitosis in multiple tumor cell lines, the number of ecDNA in daughter cells follows a Gaussian distribution, and the segregation process is independent of tumor type and ecDNA species (22). Subsequently, CRISPR-based ecDNA tagging with live-cell imaging was used to dynamically track ecDNA during the cell cycle, further confirming that ecDNA undergoes random segregation during cell division (22, 23). Analysis of The Cancer Genome Atlas Program (TCGA) database showed that more than 25% of ecDNA-containing (ecDNA+) tumors contain more than two types of ecDNA, and different ecDNA species coexist at copy numbers much higher than expected by chance (14). Therefore, when cells contain multiple ecDNA species, their segregation may not be completely independent and random. Recent studies have shown that when tumor cells contain multiple ecDNA species, cooperative ecDNA species are coordinately inherited through mitotic co-segregation (14). In summary, during cell division, a single type of ecDNA segregates into daughter cells in a binomial random manner, while multiple coexisting ecDNA species can be co-segregated into daughter cells ( Figure 2 ).

The excisional model, also known as the episome model, is a simple hypothesis regarding the origin of ecDNA. The excisional model posits that ecDNA originates from double-stranded DNA breaks on chromosomes, so the resulting ecDNA usually have simple structures and low diversity (25). As early as 1988, researchers confirmed that episomes can be directly formed from deleted chromosomal fragments (26). Subsequently, a case of acute myeloid leukemia was reported in which the leukemic cells contained double minutes carrying the MYC gene, and a chromosome 8 with a deletion in the MYC region was also present, suggesting that the MYC in the double minutes originated from chromosome 8 (27). Through next-generation sequencing, single nucleotide polymorphism array, fluorescent in situ hybridization, and polymerase chain reaction-based techniques, the genomic structure and evolutionary mechanisms of seven MYC ecDNA+ tumor cell lines were analyzed, revealing that ecDNA are gradually formed through multiple steps such as amplification, recombination, and deletion of ancestral episomes from a single chromosome (28). Chromosome conformation capture (Hi-C) analysis showed that spatial proximity is not required for the generation of ecDNA, and long-distance double-stranded breaks can still efficiently form ecDNA (29). In addition, CRISPR-C technology can be used to artificially cut specific genes to construct ecDNA+ cell models (12, 22).

According to whether the chromosome can be accurately repaired after DNA fragment excision, the excisional model is divided into the scarring excisional model and the scarless excisional model. In the scarring excisional model, DNA fragment excision occurs before DNA replication, and the organism connects the chromosomal break ends through nonhomologous end joining (NHEJ) (25). In the scarless excisional model, DNA fragments are usually formed by replication fork breakage, and the missing part of the chromosome is precisely repaired through a homologous recombination-dependent DNA replication process using the normal sister chromatid as a template (25). Deep sequencing of the junctions between ecDNA and chromosomal excision scars revealed that the formation of ecDNA and chromosomal scars is independent, with NHEJ predominantly repairing chromosomal scars, while microhomology-mediated end joining (MMEJ) is more common in ecDNA circularization (29). However, some studies have shown that the breakpoints of circular amplicons often have no sequence homology or only minimal sequence homology (<5 bp), suggesting that NHEJ is the main mechanism for ecDNA formation (30).

BFB cycles were originally described by Barbara McClintock in 1939 for the fate of a dicentric chromosome during meiotic mitosis and endosperm development in maize (31). BFB events are common in tumors and are a frequent cause of increased oncogene copy number (32, 33). The BFB cycle begins with telomere loss, and chromosomes lacking telomeres or sister chromatids of telomere-deficient chromosomes after replication fuse to form dicentric chromosomes (34). During cell division, dicentric chromosomes are pulled in opposite directions by spindle fibers, forming chromosomal bridges (35). Chromosomal bridges break under mechanical tension during cell division, generating new telomere-deficient chromosomes and broken chromosomal fragments (34). In the absence of telomeres, BFB will continue to occur in subsequent generations of cells until telomeres are restored (32). Studies have shown that there is a strong overlap between oncogenes amplified by BFB cycles and those amplified by ecDNA, suggesting that chromosomal bridge fragments generated by the BFB cycle can circularize to form ecDNA (32).

Chromosomal translocations can also form dicentric chromosomes. During cell division, dicentric chromosomes separate to form chromosomal bridges, which break and circularize to form ecDNA (36). TB amplification elucidates the amplification mechanism of key oncogenes ERBB2 and cyclin D1 (CCND1) in breast cancer (36). Chromosomal translocation is the most common cause of chromosomal bridge formation in tumors, and tumors with TB amplification often exhibit loss of heterozygosity (LOH) on the bridge arm (36). In the TB amplification model, dual-LOH occurs on two chromosomal arms, so chromosomal translocation in TB amplification occurs in the G1 phase (36).

Chromothripsis refers to the occurrence of a large number of random breaks in one or several chromosomes within a short period (37). Chromosomal shattering does not occur through the accumulation of gene mutations but is a single catastrophic genomic event (38). Fanconi anaemia (FA), a model syndrome of genome instability, is caused by a deficiency in DNA interstrand crosslink repair resulting in chromosome breakage (39). Studies have shown that the FA pathway is a driving factor for chromothripsis, with the core FA complex monoubiquitinating and activating FANCI-FANCD2, which recruits the SLX4-XPF-ERCC1 endonuclease to cleave micronuclear chromosomes, triggering large-scale chromosomal shattering (40). The shattered chromosomal fragments undergo erroneous joining and assembly, leading to extensive chromosomal rearrangement or the formation of ecDNA (41). Chromothripsis is widespread in tumors, with an incidence exceeding 40% in glioblastoma, lung adenocarcinoma, osteosarcoma, and liposarcoma (42). Additionally, approximately 50% of cases with circular amplicons exhibit chromothripsis (30, 43). Some researchers also believe that chromothripsis may arise through TREX1-mediated fragmentation of dicentric chromosomes formed in telomere crisis (44). BFB cycles and chromothripsis are hallmarks of telomere crisis (45). Therefore, chromothripsis and BFB cycles may share a common origin and coexist and promote each other in the formation of ecDNA (30, 43).

In tumors, gene amplification is a measure to antagonize anti-cancer treatments by directly increasing the dosage of target proteins (58), or by activating an alternative cell-proliferation pathway (59). More commonly, genes promoting tumorigenesis and progression employ gene amplification to increase their protein level (60). Gene amplification occurs in two forms: linear amplification and circular amplification, manifested as homogeneously staining regions (HSR) and ecDNA, respectively (61). Analysis of 8,068 circularly amplified genes and 6,247 linearly amplified genes in 77 tumor samples showed that the copy number of circularly amplified genes was significantly higher than that of linearly amplified genes (15). Therefore, increased copy number of oncogenes carried by ecDNA is one of the main mechanisms for increased gene expression. In addition, engineered ecDNA can also spontaneously accumulate in primary cells and promote cell proliferation, transformation, immortalization, and drive tumor formation (62).

ecDNA can carry multiple regulatory regions, resulting in a non-linear relationship between transcriptional output and gene copy number. Transcription of ecDNA is very common (63). After normalizing the copy number of oncogenes carried by ecDNA and chromosomal oncogenes, oncogenes on ecDNA still produce more transcripts, significantly higher than those on chromosomes (15, 30). Studies have shown that ecDNA can promote oncogene transcription through non-copy number-dependent mechanisms such as increased chromatin accessibility and enhancer hijacking (64). In addition, as the copy number of ecDNA increases, the transcription level of genes carried by ecDNA also increases accordingly (20). Below, we discuss in detail the mechanisms underlying the high transcriptional activity of ecDNA.

As mentioned above, ecDNA does not follow Mendelian inheritance and undergoes random segregation during cell division. Therefore, after multiple rounds of cell division, the copy number of ecDNA in cells will exhibit significant heterogeneity ( Figure 2 ) (21, 22). When the cellular microenvironment changes, natural selection enables cells to rapidly accumulate ecDNA carrying oncogenes to cope with adverse environments (22). ecDNA+ tumors adapt faster, exhibit more pronounced intratumoral heterogeneity, and develop drug resistance earlier than BFB-amplified tumors (32). DNA methylation is an important epigenetic modification regulating gene expression (85). In multiple SCLC cell lines, the DNA methylation level in ecDNA+ cells is slightly lower than that in ecDNA− cells (76). Using nanopore sequencing technology, it was found that the methylation level of the EGFR gene promoter region on ecDNA in GBM39 cells is significantly lower than that of the same region on chromosomes (24). Therefore, ecDNA can regulate gene methylation status to achieve higher transcriptional activity.

In glioblastoma, exons 2–7 of EGFR are often deleted, resulting in the constitutively active mutant EGFRvIII (86). EGFRvIII in GBM39 cells is mainly located on ecDNA, while chromosomes usually contain full-length wild-type EGFR (24). Therefore, EGFRvIII ecDNA in glioblastoma can provide a unique selective advantage for tumor evolution (14). Studies have shown that the expression of apolipoprotein B mRNA editing enzyme catalytic polypeptide 3 (APOBEC3) is significantly higher in ecDNA+ tumors than in ecDNA− tumors, and 31% of samples containing ecDNA exhibit kyklonic events (APOBEC3 kataegis and ecDNA occurring simultaneously) (87). Among all kyklonic events, 41% overlap with known tumor driver genes, resulting in mutations in tumor driver genes (87). Thus, APOBEC3 plays an important role in ecDNA mutation and tumor evolution. In addition, HPV-host ecDNA can undergo multiple rounds of amplification and recombination, forming highly heterogeneous virus-host ecDNA, thereby promoting heterogeneity and clonal evolution in HPV-related tumors (56). The above studies indicate that ecDNA exhibits copy number heterogeneity and species heterogeneity among cells, as well as genetic and epigenetic heterogeneity with chromosomal DNA.

Studies have shown that the detection rate of ecDNA is significantly increased in tumor patients receiving chemotherapy and targeted therapy, suggesting that ecDNA may be an adaptive mechanism for tumor cells to cope with treatment pressure (3). DHFR gene amplification is the main cause of acquired methotrexate resistance (88). The amplification of the DHFR gene in the methotrexate-resistant HT29 human colon cancer cell line undergoes three stages: pre-amplification, HSR, and ecDNA, with ecDNA being the main driver of resistance (88). After methotrexate treatment of HAP1 cells containing DHFR ecDNA, the copy number of ecDNA increased in a strongly dose-dependent manner (22). In urothelial cancer, the CCND1 gene undergoes amplification through ecDNA-mediated structural variants (SVs), driving cell cycle progression and enhancing cellular adaptability under selective therapeutic pressure (89). Treatment of GBM39-EC (EGFRvIII located on ecDNA) and GBM39-HSR (EGFRvIII located on HSR) cells with the EGFR tyrosine kinase inhibitor (TKI) erlotinib showed that GBM39-EC cells resist erlotinib by reducing the copy number of EGFRvIII ecDNA, while the copy number of EGFRvIII in GBM39-HSR cells remains unchanged, and these cells remain sensitive to erlotinib (22). In glioblastoma patients receiving EGFR TKI targeted therapy, tumor cells acquire EGFR TKI resistance by eliminating EGFRvIII ecDNA, and EGFRvIII ecDNA reappear after drug withdrawal (90). Pancreatic ductal adenocarcinoma (PDAC) is usually difficult to survive in a WNT-deficient environment, and acquired WNT independence can promote PDAC progression (19). In PDAC organoid cultures under WNT-deficient conditions, MYC ecDNA+ cells can be selected to adapt to the WNT-deficient environment (19). In addition, GBM39-HSR cells are extremely sensitive to glucose deprivation, while GBM39-EC cells show no significant changes (22). The above studies indicate that ecDNA+ tumors have a stronger selective advantage and can rapidly adapt to changes in the microenvironment when facing metabolic stress or drug treatment, leading to rapid tumor progression and drug resistance.

Studies have shown that patients with ecDNA+ medulloblastoma were more than twice as likely to relapse and three times as likely to die within 5 years of diagnosis (20). After adjusting for tumor type, stage, age, sex, and genomic instability in 14,778 tumor patients (39 types), it was found that ecDNA detection was associated with tumor stage, metastasis, and shorter overall survival (3). Research indicates that reduced expression of MHC class I molecules in ecDNA+ urothelial carcinoma cells enables immune evasion from T cell attack, thereby contributing to poor prognosis in ecDNA+ urothelial cancer patients (91). Moreover, ecDNA can harbor genes regulating immune and inflammatory responses, which is associated with reduced T cell infiltration in cancer patients (3). Therefore, ecDNA may affect tumor progression and prognosis by influencing the expression of immune-related genes, such as inhibiting immune clearance of tumor cells or promoting immune evasion (3, 92). The p53 tumor suppressor protein is a transcription factor that inhibits cell division or survival in response to various stresses (93). TP53 mutations are associated with enhanced chromosomal instability, including increased amplification of oncogenes and deep deletion of tumor suppressor genes (94, 95). Studies have shown that ecDNA is the main driver of the progression of high-grade dysplasia Barrett’s esophagus (HGD) to esophageal adenocarcinoma (EAC), and TP53 mutations drive the formation of ecDNA (13). TP53 mutations are significantly enriched in ecDNA+ endometrial cancer, renal cancer, breast cancer, and medulloblastoma (3, 12, 20). Therefore, some researchers believe that the impact of ecDNA on patient survival is due to TP53 mutations (20). The overall survival rate of neuroblastoma patients with ecDNA-derived rearrangements is significantly lower than that of patients without ecDNA-derived rearrangements, and the overall survival rate of patients with MYCN ecDNA-derived rearrangements is also significantly lower than that of patients without MYCN ecDNA-derived rearrangements (96). Therefore, genomic instability caused by the integration of ecDNA into chromosomes may be another mechanism by which it affects patient survival outcomes (97). In addition, in SCLC, ecDNA is the main source of the RLF-MYCL oncogenic fusion (76). RLF-MYCL, the most common oncogenic fusion in small cell lung cancer, can accelerate transformation and proliferation of murine SCLC and increase metastatic dissemination and the diversity of metastatic sites (98). In summary, multiple studies have confirmed that ecDNA can lead to poor patient prognosis, making it another indicator for prognosis prediction. However, the biological mechanisms by which ecDNA drives tumor progression and metastasis require further study to provide more evidence for the development of future targeted therapies.

ecDNA exhibits autonomous replication and undergoes replication only once during the S phase of the cell cycle (99). ecDNA replication activates the ataxia telangiectasia mutated (ATM)-mediated DNA damage response (DDR) pathway (12). This DDR pathway is essential for ecDNA maintenance, and its inhibition would disrupt ecDNA circularization (12). Furthermore, during mitosis, ecDNA achieves segregation to daughter cell nuclei by attaching to chromosomes – a phenomenon termed “hitchhiking” or “tethering” (100). Research indicates that ecDNA achieves efficient nuclear segregation by tethering to mitotic chromosome ends, and this tethering is essential for ecDNA maintenance (101). In addition, ecDNA can form new, more complex structures through replication, amplification, and rearrangement (41, 102).

As early as 1985, research revealed that MYC amplification in human colonic carcinoma cell lines evolved from ecDNA to HSR on an X chromosome (103). Studies have shown that focal amplification of BRAF in the M249 human melanoma cell line resistant to vemurafenib and selumetinib initially exists as ecDNA and can subsequently integrate into chromosomes to form homogeneously staining regions (HSR) (104). In neuroblastoma, most genomic structural rearrangements are caused by the integration of ecDNA into chromosomes (96). For example, after chromosome 2 breaks, ecDNA containing MYCN, NBAS, and rs13028343 are formed, and the part of ecDNA containing NBAS and rs13028343 can integrate into chromosome 13, causing the doublecortin like kinase 1 (DCLK1) gene to break (96). In addition, JQ1 can induce the integration of ecDNA into chromosomes, leading to the elimination of ecDNA in cells (102).

ecDNA usually carries oncogenes, so eliminating ecDNA can induce cell differentiation and reverse the tumor phenotype. Studies have shown that COLO320-DM cells contain ecDNA micronuclei, and low concentrations of hydroxyurea can further induce their formation (105). Hydroxyurea induces DNA double-strand breaks, causing ecDNA to aggregate and lag behind mitotic chromosomes, eventually forming micronuclei (106). Subsequently, CRISPR Cas9 was employed to induce precise double-strand breaks in ecDNA. This approach confirmed that damaged ecDNA is prone to aggregation, and that aggregated ecDNA subsequently detaches from mitotic chromosomes, forming micronuclei (107). In addition, gemcitabine also promotes micronucleus formation of ecDNA in human ovarian cancer cells (108). The MDC1-TOPBP1-CIP2A complex mediates the tethering of chromosomal fragments, allowing them to be transmitted as a whole to daughter cells (109). Therefore, the MDC1-TOPBP1-CIP2A complex may also explain why broken ecDNA are more likely to aggregate, but this requires further research. Moreover, recent studies demonstrate that BRD4 plays a significant role in the nuclear segregation of ecDNA, and inhibition of BRD4 impairs ecDNA clustering during mitotic segregation, ultimately leading to micronucleation (101). Subsequently, ecDNA micronuclei are degraded in cells through autophagy or apoptosis-like processes, or extruded from cells through exocytosis-like mechanisms, becoming the main pathways for ecDNA elimination (105).