It was previously thought that genetic variation and V(D)J recombination were the main reasons for genome heterogeneity of different tissues from the same individual or different cells from the same tissue. However, recent studies have shown that extrachromosomal circular DNA (eccDNA) is an additional source of genomic heterogeneity. EccDNAs are a group of double-stranded circular DNA molecules that are derived and free from eukaryotic genome DNA. They could function in genome evolution and environmental adaptation, which depend on eccDNA sequence features. The high copy number and the significant transcriptional activity of eccDNA lead to the overexpression of the inhabiting genes (1). Additionally, eccDNAs could serve as mobile enhancers to trans-regulate chromosomal genes (2). Small eccDNAs are more widespread, but less is known about their function in cell biology. They are too small to contain protein-coding genes. MicroDNA can be released from normal and tumor tissues to plasma and serum, implying their roles in cell communication (3, 4). Transcription factor sponge is another speculated function of microDNA, where the accumulation of specific microDNA could titrate components of the replication or transcription machinery and lead to an inability to replicate or transcribe genomic DNA (5). Thus, establishing the associations of microDNA with economic phenotype or disease is an important direction for future exploration.

EccDNA was first detected and measured in pig sperm using electron microscopy technology, with a length of ~0.5–16.8 μm (1 μm ≈ 3,100 bp) (6, 7). The length of eccDNA in the Hela cell nucleus was estimated to be 0.2–19.8 μm, compared with 4.81 ± 0.24 μm (mean ± SD) of mitochondrial DNA, and the eccDNA count was 20% of the mitochondrial DNA count (8). When analyzing a set of tumor karyotypes, scientists found many small double chromatin bodies (double-minutes, DMs), sometimes in large numbers, in addition to the apparently structurally intact chromosome (9).

Small polydisperse circular DNA (spcDNA) was an obsolete concept to commonly characterize small eccDNAs with a length of ~0.05–2.00 μm (10). From the 1980s to the 1990s, the repetitive sequences of spcDNAs were widely observed, including short interspersed nuclear element (SINE), long interspersed nuclear element (LINE), tandem repeats, transposons, rDNA, and telomere DNA (11–18). Subsequently, spcDNAs with rDNA and telomere DNA were exclusively termed as extrachromosomal rDNA circles (ERCs) and extrachromosomal telomeric circles (t-circles), respectively. However, their lengths are larger than that of spcDNA, meaning that some ERCs and t-circles are not covered by spcDNA (19–21). With the help of high-throughput sequencing, microDNAs were identified. They have non-repetitive sequences with a length of about 200–400 bp and derive from 5′ UTRs, exons, and CpG islands (22). It is abundant, with several 100 to several 1,000 counts per cell (22, 23). Recently, the concept of extrachromosomal DNA (ecDNA) was developed to exclusively specify larger eccDNA in tumors, typically covering intact oncogenes, and 30% of ecDNAs exist as DMs (24). Further studies revealed that the deletion of large genomic fragments could be circularized into episomes and subsequently polymerized into DMs. Therefore, episomes are the precursors of DMs (25–30). In this review, we divide general eccDNA into two categories: narrow sense eccDNA with length < 100 kb and ecDNA covering DMs and episomes (1, 31, 32). Notably, eccDNA refers to general eccDNA in the following parts unless stated otherwise. Accordingly, we make a clear classification for eccDNAs in Figure 1.

Given the heterogeneity of eccDNAs in terms of sequence features, various molecular mechanisms may contribute to eccDNA biogenesis. Interestingly, all these mechanisms seem to be associated with DNA repair (33). We generalize these mechanisms into four categories: homologous recombination (HR), non-homologous end joining (NHEJ), DNA replication, and transcription (Figure 2). However, these potential models underlying the formation of different kinds of eccDNAs require further tests and verification.

DNA double-strand breaks could induce rDNA and tDNA to generate ERCs and t-circles, respectively, via loop structures mediated by HR (34). Among the sequencing reads mapped to ~100,000 eccDNAs in human muscle, only 8.1% are from genomic rDNA, 3.5% from LINE, 3.1% from SINE, 1.2% from satellite sequences, and 0.8% from tDNA. Most of the remaining reads are mapped to non-repetitive regions that could produce microDNA (23). The generation of microDNA is strongly associated with DNA mismatch repair. To comprehensively probe microDNA biogenesis, researchers knocked out the key proteins of NHEJ, HR, and mismatch repair (MMR) in a chicken DT40 cell line. They found that knocking out the MMR key protein, MSH3, could cause an 81% decrease of microDNA amount (35). Notably, the microdeletions by MMR were rare, occurring in one of ~400–4,000 alleles from the brain, and would be missed if genomic sequencing was not done at high depth. Nevertheless, the presence of microDNA from >100,000 sites in mouse, human, and chicken cells makes it unlikely that all of them are created by an excision event that leaves behind >100,000 somatically mosaic deletions on the chromosomes (22). Besides this, transcription in GC-rich regions and UTRs could generate triple-stranded DNA/RNA hybrids that function in DNA damage and repair processes and lead to microDNA formation (36). Origin-dependent inverted-repeat amplification may be involved in microDNA production as well. The nascent DNA strand could be circularized via short reverted repeats at both ends of the replication bubble (37). Inhibiting the expressions of BRCA1 and PRKDC, the key proteins for HR and NHEJ, respectively, lead to the reduction of ecDNA count in colon cancer cells, suggesting that HR and NHEJ activated by double-strand breaks and macrodeletions may be involved in ecDNA production (38, 39). The summarized mechanisms are listed in Table 1, and detailed information can be obtained from the corresponding references.

Large eccDNAs could be observed with a light microscope in karyotype analysis, but as for the small eccDNAs, electron microscopy is necessary, and their lengths can be estimated all at once (Figure 3) (8, 53). This estimation could be achieved by 2D electrophoresis as well, but its detection power ranges from 0.7 to 56.8 kb (19). Southern blotting enables 2D electrophoresis to reveal the sequence features of eccDNA (60). Interestingly, software, such as ECdetect, has been developed for moderate-through counting of eccDNAs in DAPI-stained cells (24).

Initiated with a large amount of cells, the CsCl-EB method is time consuming and labor intensive for eccDNA enrichment, where most nicked circles are missed. Therefore, this approach is being replaced by plasmid extraction kits (60–62). The enriched eccDNAs can be subjected to high-throughput sequencing to determine their locations and junction sites by Circulome-seq, mobilome-seq, Circle-seq, or CIDER-seq (Figure 3). Circulome-seq adopts Tn5 transposition-based fragmentation and a tagging system, which simplify the sequencing library construction. This method could detect eccDNAs in length ranging from several hundreds of base pairs to several hundreds of kilobase pairs (62). Mobilome-seq is highly suitable for the detection of retrotransposon eccDNAs. It removes linear DNA with exonuclease DNase and then performs rolling circle amplification (RCA) (63). Circle-seq combines endonuclease NotI with exonuclease DNase to fully eliminate linear DNA, potentially leading to an unexpected damage on eccDNAs; its detection power is ~1–38 kb (64). As for CIDER-seq, RCA is performed straightforwardly for the eccDNAs enriched by plasmid extraction kits without the removal of linear DNA. Given the undesired amplification of linear DNA, long-read single-molecule real-time sequencing is used to guarantee the amount of valid reads (i.e., split and discordant reads), which are essential for eccDNA calling (65). CIDER-seq has a similar power to Circulome-seq, but the former has a more accurate detection for <10 kb eccDNAs. Notably, it is advisable to introduce plasmids as a control or internal reference during eccDNA library construction.

Despite the above-mentioned methods, whole-genome sequencing (WGS) and assay for transposase-accessible chromatin using sequencing (ATAC-seq) can be used for eccDNA calling, given the presence of split and discordant reads in their sequencing data (24, 66, 67). It has been demonstrated that 100% of ecDNAs and 30% of eccDNAs (narrow sense) called from WGS were supported by Circle-seq (32). Prior enrichment of high-molecular-weight DNA is helpful in ecDNA identification (1). To date, various software packages have been developed to call eccDNA with sequencing data (Figure 3).

Telomeric arrays can be maintained through various mechanisms, such as telomerase activity or recombination. T-circles function in recombination-dependent maintenance pathways by serving as templates for the rolling circle synthesis of telomere DNA. This may be the same case for animal and plant t-circles (20, 69). It is estimated that 15% of human immortalized cell lines may maintain telomere length through alternative lengthening of telomeres (70). A total of ~100–1,000 copies of eukaryotic rDNA are tandemly organized in the genome to satisfy the need for ribosome synthesis. The production of ERC reduces the copy number of rDNA in the Drosophila genome that could recover in germ cells. Studies have found that ERC could reintegrate into the genome to maintain rDNA copy number by self-replication (40).

Asymmetric inheritance and self-replication lead to ERC accumulation in yeast mother cells. The number of ERCs per mother cell after 15 generations has been estimated at between 500 and 1,000. Mutations in SGS1 could result in ERC accumulation and age-associated phenotypes in addition to a shortened lifespan. Conversely, loss of the replication fork blocking protein Fob1 decreases the formation of ERCs and extends the lifespan of mother cells by 30–40% (71). Interestingly, genes present on low-copy plasmids confer fitness effects rarely and of much lower magnitude than those on multi-copy plasmids. Therefore, young yeast populations contain about 1,800 circular DNA species, and it is only with substantial accumulation of any given circular DNA that major phenotypic effects are likely to manifest (64, 71, 72). These facts demonstrate that ERC accumulation functions in yeast aging. It was proposed that yeast senescence may be the result of health sacrifice to environmental adaption by accumulating specific eccDNAs, which could titrate components of the replication or transcription machinery and lead to an inability to replicate or transcribe genomic DNA and, thus, growth arrest and eventual death (5). According to this assumption, CUP1 eccDNA enrichment in a CuSO₄ environment may contribute to yeast aging, but further research is still necessary (34).

DMs, typically bearing intact oncogenes, are tumor specific and have been detected in 182/200 kinds of tumors, and ~0.26–44% of cancer patients and 7–100% of cancer cells have DMs (50, 53, 73–79). ecDNA exclusively refers to larger eccDNA in tumors where 30% of ecDNAs existed as DMs (24). The survival rate of cancer patients with ecDNAs is significantly lower than those without ecDNAs, making ecDNA a potential prognostic marker (80, 81). As a carrier of oncogene amplification, ecDNAs are subjected to non-Mendelian inheritance, which enables tumors to achieve very high intratumoral genetic heterogeneity and evolve rapidly in response to changing conditions (24, 66, 82)—for example, EGRF, MET, or MYC ecDNAs can make tumor cells proliferate rapidly and further develop into tumor invasion and migration (51, 66). There is a substantial quantity of cell-free microDNAs in the plasma and serum from both healthy individuals and cancer patients (3, 4). Surprisingly, tumor-derived human microDNAs are detected in mouse circulation in a mouse xenograft model of human ovarian cancer, and tumor excision alters the length of these small molecules (3). Thus, circular DNA in the circulation is a previously unexplored pool of nucleic acids that could complement miRNA and linear DNA for diagnosis and intercellular communication (3).

It had been accepted that the high copy number of ecDNA leads to oncogene overexpression. However, the significant transcriptional activity of ecDNA does matter as well (1). The deficiency of chromatin high-order structure and suppressing histone modification make ecDNAs more accessible than their genome parallels, which instigates strong promoter–enhancer interactions. Furthermore, ecDNA enhancers can shake off the insulator shackles and lead to novel interactions with oncogenes, which causes additional expression (83, 84). More recently, researchers have reported the chromatin connectivity networks of ecDNA in cancer, revealing that ecDNA can function as mobile super-enhancers, which drive genome-wide transcriptional amplification, including that of oncogenes. These findings support an expanded role for ecDNA in trans-regulating chromosomal genes in promoting tumor growth (2).

EGFR VIII, an oncogenic variant, could accelerate glioblastoma growth, but it also makes cells more sensitive to the EGFR tyrosine kinase inhibitor (TKI) (85). After TKI treatment, the proportion of TKI-sensitive tumor cells with a high expression of EGFR VIII was significantly decreased, whereas cells with low EGFR VIII expression were increased (85). Studies have demonstrated that tumor TKI resistance is caused by the elimination of DMs containing EGFR VIII, which could reintegrate into the genome HSRs (86). However, after drug withdrawal, the reemergence of clonal EGFR mutations on ecDNA follows quickly (86). Through this mechanism, cancer cells can escape targeted oncogene therapy. Therefore, pulsatile intermittent treatment with much higher doses of TKI could potentially lead to better target inhibition and even possibly less toxicity relative to continuous dosing (24). Notably, the self-replication of EGFR DMs could also generate EGFR mutations, which would provide additional heterogeneity (86).

To our current knowledge, eccDNAs have been associated with animal phenotypes, including cattle color sideness (Cs) and pigeon muscle development. Cs is a dominantly inherited trait characterized by a white band along cattle spines. The dominance of the Cs allele is expected to reflect a gain of function resulting from the dysregulated expression of the translocated KIT gene. A 492-kb fragment containing the KIT gene on chromosome 6 produces a circular intermediate (now referred to as eccDNA) and integrates into chromosome 29 to form the Cs29 allele. Then, a 575-kb fragment containing the partial Cs29 allele is circularized and translocated to chromosome 6 as Cs6 allele (87). Regenberg et al. found that the number of eccDNA in king pigeon muscle is nine-fold higher than that of homing pigeons. Interestingly, eccDNAs bearing the AGRIN gene were identified (88). This gene encodes a membrane protein that is involved in the development of neuromuscular junctions, and its variations could lead to abnormal muscle development (88).

As for plants, Amaranthus palmeri can develop herbicide resistance to glyphosate by amplification of the EPSPS gene as eccDNA, the molecular target of glyphosate. These circular molecules can be transmitted to germ cells and drive rapid glyphosate resistance through genome plasticity and adaptive evolution (89, 90). Moreover, retrotransposons, such as EVD and Tos17, can produce eccDNAs and insert into the genome to improve the response to environmental stress through promoting DNA methylation and gene silencing at the transcriptional level (63).

The above-mentioned facts suggest that eccDNAs may be promising molecular markers in life science. However, the inability to provide biopsies of some tissues limits the use of large eccDNAs as biomarkers. MicroDNAs released to the circulation represent a previously unexplored pool of nucleic acids; although they are too small to contain protein-coding genes, they are sufficiently long to code for regulatory elements. Possible acting mechanisms of microDNA have been assumed based on indirect evidence, including cell communication, transcription factor sponges, and mobile enhancers to trans-regulate chromosomal genes (Figure 4). Thus, establishing the associations of microDNA with a particular phenotype or disease is an important direction for future exploration.

The occurrence of eccDNA is a ubiquitous, normal phenomenon in eukaryotic systems, including those of plants, yeasts, and animals. There are various types of eccDNA according to sequence feature and size. Several models have been proposed to explain the formation and proliferation of eccDNAs, but the underlying mechanisms and direct evidence for each model are still required. Regardless of the above-mentioned issues, the CRISPR-hapC system has been developed for genome haplotyping based on the generation of eccDNA in cells (91). This system can map haplotypes from a few 100 bases to over 200 Mb and will be important for genome research and haplotype-specific gene therapy.

ecDNA could drive oncogene amplification and has become a hotspot of research in tumor pathogenesis and evolution. Additionally, the survival rate of cancer patients with ecDNAs is significantly lower than those without ecDNAs, making ecDNA a potential prognostic marker (80, 81). However, an interesting question arises: Are tumor suppressor genes present in ecDNA? If they exist, what are their functions? In plants and animals, large eccDNAs bearing intact genes have been identified, and they play important roles in environmental stress response and phenotypic variations, respectively (87, 89, 90). These facts make eccDNA a particularly promising molecular marker for breeding. However, given the unavailability of some tissue biopsies, it may be difficult to use large eccDNAs as biomarkers for early diagnosis and breeding.

MicroDNA represents the majority of eccDNAs. It has been detected as abundant cell-free DNA in plasma and serum released both by normal and tumor tissues (3, 4). Thus, establishing the associations between microDNAs and economic phenotype or diseases is an important direction for future exploration. Fortunately, an eccDNA database (eccDNAdb, http://www.eccdnadb.net/) has been set up. It has recorded a total of 1,700,000 eccDNAs for humans, mice, and chickens, which will provide supporting data for association analyses. If there are positive results, then the mechanisms of their regulatory roles can be validated, including serving as sponges of transcription factors, carriers of regulatory RNAs in intercellular communication, or mobile enhancers to globally amplify chromosomal transcription. These mechanisms may also provide novel insights into the phenotypic effects of genome copy number variations (41, 45, 92, 93). All of these questions require a further in-depth exploration.

XC proposed the topic, retrieved literatures, provided outline, tables, and figures, and revised the manuscript. SW, LG, and WZ wrote the manuscript. JH and WS reviewed the final manuscript. All the authors read and approved the final manuscript.

This work was supported by the Key Research and Development Plan (modern agriculture) in Jiangsu Province (BE2018354), Jiangsu Agricultural Science and Technology Innovation Fund [XC(18)2003], major projects of Natural Science Research of colleges and universities in Jiangsu Province (17KJA230001), National Natural Science Foundation of China (31872333), major new varieties of agricultural projects in Jiangsu Province (PZCZ201739), National Natural Science Foundation of Cooperation on International Agricultural Research Organization (32061143036), and Open Project of Joint International Research Laboratory of Agriculture and Agri-Product Safety and the Ministry of Education of China (JILAR-KF202103).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.