Gene regulation on extrachromosomal DNA

Abstract

Oncogene amplification on extrachromosomal DNA (ecDNA) is prevalent in human cancer and associated with poor patient outcomes. Clonal, megabase-sized circular ecDNAs in cancer are distinct from nonclonal, small sub-kilobase sized DNAs that may arise during normal tissue homeostasis. ecDNAs enable profound changes in gene regulation beyond copy number gains. An emerging principle of ecDNA regulation is the formation of ecDNA hubs, micron-sized nuclear structures of numerous copies of ecDNAs tethered by proteins in spatial proximity. ecDNA hubs enable cooperative and intermolecular sharing of DNA regulatory elements for potent and combinatorial gene activation. The 3D context of ecDNA shapes its gene expression potential, selection for clonal heterogeneity among ecDNAs, distribution through cell division, and reintegration into chromosomes. Technologies for studying gene regulation and structure of ecDNA are starting to answer long-held questions on the distinct rules governing cancer genes beyond chromsomes.

Introduction

The cancer genome undergoes extensive genetic alterations, often leading to dysregulation of gene expression. One important mechanism is amplification of oncogenes and drug resistance genes; this process increases the copies of genes that provide a selective advantage to cancer cells. While it has long been known that gene amplification can occur within or outside chromosomes, it is increasingly appreciated that the context and spatial architecture of gene amplification has an enormous impact on changes in gene expression which cannot solely be explained by gene copy number.

Major recent advances have been made regarding extrachromosomal DNA (ecDNA), a common mode of oncogene amplification. ecDNA was first observed in 1962 and described in detail by Cox et al. in 1965^1,2. Since then, ecDNA has been detected in nearly half of human cancer types, carrying oncogenes such as MYC, MYCN, EGFR, HER2 in tumor cell lines, patient-derived cell cultures and clinical tumor samples^3–13. ecDNA can also integrate into chromosomes and therefore may act as an intermediate step toward stable chromosomal gene amplification in a subset of cases^9,14–16. ecDNA is associated with poor patient outcomes even when compared to other forms of gene amplification⁵. These observations suggest that gene amplification in the context of ecDNA may have unique impacts on cellular programs that profoundly contribute to tumor pathogenesis and progression.

ecDNA refers to circular DNA molecules which are self-replicating, clonally selected and amplified, and range from 100 kilobases (kb) to several megabases (Mb) in size. They are selected in cancer cell populations as they provide a fitness advantage typically by carrying oncogenes and drug resistance genes. ecDNA was classically termed “double minutes”, which referred to paired, extrachromosomal chromatin bodies that can be microscopically observed on metaphase spreads. However, some extrachromosomal DNA molecules are submicroscopic and the majority of ecDNAs (~70–80%) appear as singletons on metaphase chromosome spreads rather than paired bodies^3,4,16–18. Therefore, the classic “double minute” structure describes only a fraction of extrachromosomal DNA, leading to a shift towards the more inclusive term “ecDNA”. These large clonal ecDNAs found in cancer cells should not be confused with another class of smaller, nonclonal, extrachromosomal elements termed eccDNAs, or extrachromosomal circular DNA elements. While eccDNA is sometimes used as a broad umbrella term for circular DNA, it typically refers to DNA circles which are found in normal tissues or as byproducts of programmed cell death^19,20, are not amplified or selected, are only up to ~10 kb in size, usually do not carry genic or regulatory sequences, and can span any part of the genome^20–22. Given the similarity in nomenclature despite stark differences in sequence, function, behavior, and cell types in which the DNA elements are observed, we have constructed a summary table here to clarify the main distinctions (Table 1). This Review focuses on large clonal ecDNAs observed across many cancer types.

Table 1.

Differences between ecDNA and eccDNA.

ecDNA	eccDNA
100 kb to several megabases	Several hundred to thousand bases19,22
Circular	Circular
Self-replicating	Not known to replicate
Clonally selected	Typically not amplified or selected
Contains oncogenes, drug resistance genes, or other genes that provide a selective advantage3–13,28,101	Sequences cover the entire genome with some hotspots; typically does not contain genic sequences19
Contains regulatory sequences such as oncogene enhancers24,46,48,53	Usually does not contain regulatory sequences
Found in cancer cells	Found in many cell lines and tissues, including healthy samples20,22,102–104
Can contain heavily recombined genomic sequences and specific mutations49,87	Not known to carry mutations

Genes encoded by ecDNA are expressed at much higher levels compared to the native chromosomal locus and linear gene amplicons, even when normalized for copy number^4,5,23. In addition to copy number amplification, ecDNA shows unique structural, genetic and epigenetic features that are conducive to gene activation, suggesting that there are fundamental differences in how amplified genes are regulated on ecDNA. Given the observation that ecDNA is prevalent in cancer, is a key form of oncogene amplification, and is linked to poor clinical outcomes, there is a need for a better understanding of how genes are regulated on ecDNA. In this Review, we highlight aspects of regulation of gene expression on ecDNA which differ from chromosomal DNA. These unique ecDNA features are linked to alterations in structure, sequence, chromatin composition, and contacts with DNA regulatory elements.

ecDNA enables high levels of oncogene transcription

Copy number amplification.

ecDNA is associated with increased oncogene expression compared to linear amplifications as well as the native chromosomal locus^4,5,23,24. This is partly driven by gene copy number amplification^4,5. As ecDNAs lack centromeres, they are distributed randomly among daughter nuclei during cell division^25–27 (Figure 1a). This random segregation results in copy number heterogeneity and selection of cells carrying ecDNAs that provide a fitness advantage (Figure 1b). This characteristic of extrachromosomal oncogene amplification can lead to up to several hundreds of ecDNAs in a single cell^4,15,24 and has been linked to rapid adaptation to selective pressures and development of therapeutic resistance^4,27–29. However, copy number alone does not fully explain the high level of oncogene expression observed in ecDNA+ cancer cells; patient tumors containing circular amplicons express amplified genes more highly than those containing linear or other types of rearranged amplicons even with copy number normalization⁵. This observation suggests that there are additional mechanisms that overexpress oncogenes on ecDNA in ways distinct from chromosomal regulation of gene expression.

Figure 1. Unique characteristics of ecDNA.

(a) Random segregation of ecDNA molecules during cell division allows multiple possible outcomes in the number of gene copies inherited. (b) Cancer cells containing ecDNAs, after undergoing multiple cell divisions with random ecDNA segregation, can have a wide range of copy numbers. ecDNAs that provide a fitness advantage to cancer cells can be selected. (c) ecDNAs have lower chromatin compaction and increased transcription.

Oncogene overexpression.

In addition to increased copy numbers, ecDNAs are associated with increased transcriptional activity and highly accessible chromatin^5,23,24 (Figure 1c). ecDNAs lack higher-order compaction via depletion of large nucleosome arrays, which may allow binding by transcription factors and access to gene loci by the transcriptional machinery. There appear to be multiple mechanisms causing this oncogene overexpression. First, ecDNA molecules physically cluster with one another in the nucleus and engage in intermolecular, combinatorial enhancer-promoter interactions. Second, the circular structure of ecDNA is a stable, covalently closed structure allowing increased chromatin cis-interactions compared to chromosomes. Third, extensive genome sequence rearrangements alter the regulatory context of gene loci. This is an area of active research and there are likely additional mechanisms driving oncogene overexpression. It also remains an open question whether the ecDNA characteristics which enable oncogene overexpression are consequences of selection for transcriptionally active molecules or inherent differences between circular extrachromosomal chromatin and native chromosomes. These mechanisms driving oncogene overexpression are described in detail in the following sections of this Review.

ecDNA hubs: a new nuclear structure for intermolecular gene activation

The three-dimensional structure of mammalian chromosomes is organized at various length scales: chromosome territories, compartments A and B, topologically associating domains (TADs), and long-range enhancer-promoter interactions spanning tens to hundreds of kilobases^30,31. On the finer scale, chromatin interactions, such as those occurring between enhancers and target genes, are most often found within TADs on the same DNA molecule (cis-interactions). While interchromosomal interactions have been documented, they represent exceptions rather than the norm^32,33 (reviewed by Maass et al.³⁴). In contrast, a cancer cell can have up to hundreds of ecDNA copies in the nucleus, raising the possibility that multiple ecDNA copies can interact with one another fostering new cooperative interactions.

Rather than being randomly scattered around, ecDNA molecules cluster with one another to form micron-sized hubs in the interphase nucleus^24,35. ecDNA hubs, or “extrasomes”, represent a counterpart to chromosomes as units of genetic information and organization (Table 2). Chromosomes have linear arrays of genes and permit gene activation by DNA regulatory elements on the same DNA molecule. In contrast, ecDNA hubs permit intermolecular gene activation of combinatorial enhancers and promoter elements in spatial proximity²⁴. Chromosomes are dispersed during interphase of the cell cycle and condense ~10,000-fold during mitosis. ecDNA hubs coalesce during interphase but disperse during mitosis²⁴. These fundamental differences distinguish DNA in the form of chromosomes versus ecDNA hubs in the same nucleus.

Table 2.

Organization of ecDNA hubs and chromosomes.

ecDNA hubs (“Extrasomes”)	Chromosomes
10–100s of ecDNAs in proximity and interact with one another; tethered by proteins	Endogenous gene loci covalently linked; one long piece of DNA per chromosome
Regulatory elements such as enhancers interact with target genes in cis and in trans	Regulatory elements primarily interact with genes in cis within the same TAD
ecDNA hubs coalesce during interphase but can dynamically break into smaller clusters of ecDNAs during mitosis	Chromosomes are dispersed during interphase and condense during mitosis

Previous studies also reported preferential localization of these ecDNA clusters at the nuclear periphery during G1 phase and M phase, although the significance and mechanism of this localization pattern is not well understood³⁶. As the nuclear periphery is a transcriptionally repressive environment while ecDNAs are highly transcriptionally active³⁷, their peripheral localization is counterintuitive and warrants further investigation into its generalizability and potential functional significance. ecDNA hubs are observed during mitosis with dynamic changes in size, suggesting that these clusters are not stable during DNA partitioning^24,25,38. Finally, double-strand DNA breaks in ecDNA have also been associated with aggregation of ecDNA molecules and formation of chromosomal tandem amplifications termed homogeneously staining regions (HSRs)³⁹, suggesting that ecDNA clustering may explain the formation of some chromosomal amplifications as well.

ecDNA hubs drive intermolecular oncogene activation.

Formation of nuclear ecDNA hubs is linked proportionally to the rate of oncogene transcription from each ecDNA molecule^24,35 (Figure 2a). ecDNA hubs bring 10–100 ecDNAs into proximity and enable intermolecular enhancer-promoter interactions, increasing the level of combinatorial enhancer input to oncogenes²⁴ (Figure 2b). Whereas genes on chromosomes are activated by enhancer and regulatory DNA elements on the same chromosome, ecDNA molecules can promiscuously engage enhancers on other ecDNAs within the spatial proximity of an ecDNA hub. Even two distinct ecDNAs derived orginally from two different chromosomes can cross activate each other via enhancer-promoter contacts²⁴. The ability to enact intermolecular gene activation appears to be a bright dividing line between normal cellular physiology and cancer cells harboring ecDNA.

Figure 2. ecDNA hubs drive oncogene expression and may shape cancer cell evolution.

(a) ecDNA clustering in hubs is associated with oncogene transcription. (b) Enhancer-promoter contacts in trans within ecDNA hubs enable combinatorial interactions. (c) Enhancers and oncogenes may be co-selected on two levels. First, enhancer-oncogene pairs on the same molecule which drive oncogene expression can be selected together. Second, distinct ecDNA molecules containing enhancers and oncogenes that interact intermolecularly may be co-selected. (d) Dynamic ecDNA hub transcriptional activity is linked to heterogeneous oncogene expression levels in a cell population. (e) ecDNA hubs may allow correlated integration of multiple amplicon copies into chromosomes to form HSRs. (f) ecDNA-chromosome contacts allow enhancers on ecDNAs to interact with chromosomal genes and activate transcription.

Potential mechanisms of ecDNA hub formation.

Small transient transcriptional hubs are necessary for gene transcription⁴⁰, and we speculate that ecDNA hubs are a kind of transcriptional hub mediated by protein-protein interactions which may be oncogene specific. This is supported by dispersal of MYC ecDNA hubs via inhibition of the bromodomain and extraterminal (BET) proteins in a colorectal cancer cell line²⁴. MYC ecDNA hubs are not disrupted by transcriptional inhibition with alpha-amanitin nor by 1,6-hexanediol²⁴, suggesting that ecDNA hubs do not depend on RNA polymerase II or specific interactions between intrinsically disordered regions (IDRs) which are sensitive to hexanediol such as Mediator 1⁴¹. As BET proteins can normally concentrate accessible DNA, exclude heterochromatin, and mediate long-range enhancer-promoter communication^42,43, it is possible that ecDNA hubs may coopt endogenous mechanisms of long-range gene looping within chromosomes to promote intermolecular chromatin interactions in ecDNA ensembles. Other than MYC which is regulated by BET proteins^44,45, this model predicts that other oncogenes amplified on ecDNA may also exploit their endogenous enhancer mechanisms to operate in ecDNA hubs. As functional enhancers are co-selected with EGFR on ecDNAs in glioblastoma⁴⁶, we speculate that proteins that mediate endogenous enhancer-EGFR interactions could be involved in ecDNA hub maintenance as well.

Dispersal of ecDNA hubs was associated with reduced oncogene expression in ecDNA-containing cells, suggesting that ecDNA hubs may be a vulnerability of these oncogene-addicted cells. This observation also implies that ecDNAs may depend on unique transcriptional regulators, warranting further investigation of distinct or common regulators of ecDNA hub formation and transcription across cancer cells with various ecDNA amplicons. As long noncoding RNAs (lncRNAs) are involved in the formation of interchromosomal interactions³⁴, additional studies may address whether lncRNAs play a role in ecDNA hub formation. Finally, it is still an open question whether ecDNA hubs inhabit specific territories in the nucleus in relation to other chromosomes. Chromosome territories are non-randomly distributed in the nucleus and can even be conserved across different species, suggesting functional importance of the radial organization of chromosomes in the nucleus^31,47. The radial, sub-nuclear arrangement of ecDNA hubs in relative to chromosome territories may provide novel insights into ecDNA functions.

Implications of ecDNA hubs for evolution of oncogene diversification, cooperation, and ecDNA reintegration

Given that ecDNA has been separated from the 3D genomic context of its chromosomal origin, it has been proposed that the co-selection of oncogenes and enhancers shapes ecDNA amplicon structures^46,48. With the observation of intermolecular interactions among ecDNAs carrying distinct enhancer elements, we propose a two-level model for oncogene-enhancer co-selection (Figure 2c). The first level of co-selection occurs to individual ecDNAs; molecules that possess functional enhancers can promote oncogene expression and provide better fitness to cancer cells compared to ones that do not. The second level of oncogene-enhancer co-selection occurs to the repertoire of ecDNAs in hubs. We predict that each ecDNA molecule does not need to contain the full set of enhancer elements for oncogene activation; rather, they exist as part of an ecDNA hub that facilitates chromatin interactions among a diverse repertoire of regulatory elements and promotes interactions between the target oncogene and functional enhancers, which may be located on distinct molecules. This model raises the intriguing concept that winning the clonal competition among cancer cells occurs through clonal cooperation among ecDNA molecules. Furthermore, the presence of functional regulatory elements in a cooperative ecDNA hub may increase tolerance of mutations on individual molecules. Others have previously reported ecDNA mutational diversity and rapid response to environmental changes⁴⁹, though further investigation is needed to measure mutational diversity in functional enhancers on ecDNAs.

ecDNA hubs are also associated with variable enhancer usage and cancer cell heterogeneity in oncogene activity²⁴ (Figure 2d). This may be attributed to differential enhancer-promoter interactions which occur in the context of ecDNA hubs. As dozens of ecDNA molecules can cluster together in many possible spatial configurations, this may provide an opportunity for ectopic enhancer-promoter interactions which do not normally occur on linear chromosomes. We speculate that these differential interactions contribute to the highly variable enhancer activities and enhancer rewiring on ecDNAs. Furthermore, we speculate that ecDNAs markedly extend the concept of cancer genetic heterogeneity, as tumor cells can contain ecDNAs with diverse sequences and different oncogenes and regulatory sequences, which can interact with each other and even potentially combine into larger, single circular elements. The potential for driving diversity and accelerated evolution is remarkable.

ecDNA-chromosome interactions.

The formation of ecDNA hubs may provide a palatable explanation for the well-known tendency of a subset of ecDNA+ cancer cells to develop homogeneously staining regions (HSRs, a type of tandem amplifications on chromosomes). Double-strand breaks in ecDNAs can trigger aggregation, micronucleus formation, and reintegration into chromosomal HSRs³⁹. ecDNAs that are spatially proximal in hubs could enable correlated DNA breaks⁵⁰ and concentrated DNA cargo, creating a potential set up for HSR formation (Figure 2e). Hub formation may also impact ecDNA segregation. Previous work has demonstrated that ecDNAs are transmitted into daughter cells in clusters during mitosis³⁸. Future studies may address whether an ecDNA hub serves as a unit of inheritance or merely as a transient congregation. Other cellular processes, such as DNA repair and replication, are also regulated by genome organization^51,52. For example, replication units, or replicons, form clusters in which replication origins fire synchronously⁵¹. Furthermore, genomic loci located in the nuclear interior contain early-replicating domains, while the nuclear periphery is associated with late-replicating domains⁵¹. Therefore, the observation of ecDNA hubs warrants further investigation into whether this unusual 3D organization of DNA molecules impacts these cellular processes.

In addition to regulatory interactions among ecDNAs, ecDNAs also interact with specific sites within chromosomes⁵³ (Figure 2f). These sites of ecDNA-chromosome interactions are associated with increased transcriptional activity and active histone H3K27 acetylation (H3K27ac) marks, suggesting that these may be functional regulatory interactions involving ecDNA enhancers transcriptionally active regions in chromosomes⁵³. Therefore, in addition to amplifying oncogenes, ecDNAs may also act as “mobile enhancer carriers” that modulate chromosomal gene expression. Further investigation may address how these ecDNA-chromosome interactions affect cancer cell fitness.

Genetic basis of regulatory rewiring on ecDNA

Genetic alterations and rearrangements on ecDNAs are frequently observed^{5,24,48,54–58}. As enhancer-promoter interactions are sensitive to physical distance between regulatory elements, sequence rearrangements can alter this physical distance and subsequently influence enhancer-promoter circuitry, leading to dysregulation of gene expression^59–62. As a platform for genomic rearrangements, ecDNAs frequently contains sequences originating from the same locus as the amplified-oncogene, as well as sequences originating from distal chromosomal regions or other chromosomes of origin^{24,48,54–58}. These rearrangements enable co-amplification of cognate as well as ectopic enhancers with oncogenes^46,48. Sequence rearrangements can also result in fusion gene transcripts, leading to dysregulation of gene expression via promoter hijacking^24,63. As ecDNA is associated with accelerated adaptation and selection, molecules with increased transcriptional activity due to regulatory rewiring can be selected by providing a fitness advantage to cancer cells harboring these molecules.

Enhancer hijacking.

ecDNAs, while containing sequences that originated from chromosomes, are uncoupled from their chromosomal loci of origin and the regulatory context thereof. Importantly, regulation of gene expression is tightly connected to non-coding regulatory elements, such as enhancers, which physically interact with target genes. Consistent with this idea, ecDNAs containing oncogenes typically co-amplify enhancer elements which upregulate oncogene expression^46,48 (Figure 2c). Unlike the native chromosomal locus, ecDNAs containing sequence rearrangements can create new regulatory circuitry between ectopic enhancers and oncogene promoters^46,48 (Figure 3a). These novel regulatory interactions include enhancers which are normally insulated from the target oncogene on the linear chromosome, or those which are normally located in distal chromosomal regions^46,48. This adoption of novel enhancers is termed enhancer hijacking^48,64–66. As ecDNA amplicon structures can vary greatly among tumor samples, including different amplified sequences and different structural rearrangements, there is likely a high level of diversity in enhancer-promoter circuitry among different tumors with ecDNA amplifications of a given oncogene.

Figure 3. Genetic and structural basis of the regulatory circuitry on ecDNA.

(a) Enhancer hijacking on ecDNA leads to ectopic enhancer-gene interactions by bringing distal enhancers into proximity via sequence rearrangement. (b) Promoter hijacking drives gene expression by structural rearrangement on ecDNA. (c) The circular structure of ecDNA brings regulatory elements into spatial proximity and increases their interactions, as compared to the more variable and dynamic loop structure of a chromosomal TAD.

Promoter hijacking.

Sequence rearrangements on ecDNA can also lead to gene fusions, resulting in upregulation of oncogenes via hijacking of highly active promoters (Figure 3b). An example of promoter hijacking was observed for the MYC oncogene on ecDNAs^24,57,67. MYC is located ~50 kb upstream of the long non-coding RNA (lncRNA) gene PVT1, which negatively regulates MYC expression via enhancer competition with the MYC promoter normally⁶⁸. Fusion of the PVT1 promoter with the coding sequence of MYC is proposed to overcome this negative regulatory feedback loop by direct linkage of the PVT1 promoter activation with MYC transcription⁶⁹. PVT1-MYC fusion has been reported in multiple cancer types including breast, colon, ovary, esophagus cancers and medulloblastomas^57,69–72. This rearrangement can be observed on ecDNAs^24,67 and is generally associated with focal copy number amplifications^57,69–72. PVT1-MYC fusion was observed in 60% of select MYC-amplified cases of Group 3 medulloblastomas⁶⁹, a cancer subtype that is 18% MYC ecDNA+⁷³. Therefore, PVT1-MYC fusion potentially accounts for a significant portion of medulloblastoma cases with ecDNA amplifications, although further systematic analyses are needed to provide more insight into how frequently promoter hijacking events are observed in the context of ecDNAs in various tumor types. A recent study showed viral-human hybrid ecDNAs containing sequences of human and human papillomaviral (HPV) origins in ~20% (6 out of 28) of HPV oropharyngeal cancer samples, including amplicons that enable high levels of viral-human fusion gene transcription driven by HPV promoters⁶³. These observations hint at promoter hijacking as a powerful mechanism for upregulating gene expression on ecDNAs.

Ectopic self-associating chromatin domain.

Most of the genome is organized into TADs, which are self-associating chromatin domains on the scale of 100 kilobases to over 1 Mb^74–78. DNA elements within a chromosomal TAD are in contact at much higher frequencies, enabling cis-interactions between regulatory elements including enhancer elements and gene promoters. These enhancer-promoter cis-interactions activate gene transcription and control cellular programs. Typically several hundred kilobases to several megabases, ecDNAs are on a similar length scale as chromosomal TADs. Nevertheless, ecDNAs differ from chromosomal TADs in two major ways. First, ecDNAs are circular as demonstrated by electron microscopy and pulsed field gel electrophoresis^23,79–82. Therefore, In contrast to chromosomal TADs, ecDNAs are covalently closed rather than stabilized by looping proteins such as cohesin and CTCF. Thus, while TADs are variable from cell to cell^83–85, the covalent circular structure of ecDNA is stable (Figure 3c). Consistent with this idea, gene loci and regulatory elements on a circular ecDNA interact with each other with much higher frequencies than corresponding loci on chromosomes^23,53. Second, while chromosomal TADs are relatively conserved across cell types^74,86, ecDNA amplicons can have variable boundaries and sequence arrangements in individual tumors, allowing non-interacting loci in the native chromosomal locus to interact with each other ectopically on the circle^23,46,48 (Figure 3a). Therefore, by bringing regulatory elements into proximity on the circular structure, ecDNAs effectively serve as a type of covalently closed, ectopic “TAD”. This circular structure also serves as the basis for enhancer hijacking via incorporation of distal enhancer elements into the same DNA circle as the target gene as described above.

Methods for studying gene regulation on ecDNA

ecDNAs containing oncogene amplifications arise from chromosomes originally and, therefore, comprise sequences that heavily overlap with chromosomal DNA. As such, bulk sequencing data typically do not provide specificity to ecDNAs over the native chromosomal loci containing the corresponding genes and must therefore be carefully interpreted. Imaging-based methods can provide good separation of chromosomal and ecDNA signals depending on the approach, but suffer from limited throughput and DNA sequence resolution. We have recently demonstrated a strategy for isolating ecDNAs and separating amplicons by size, providing empirical evidence for ecDNA structures and enabling targeted profiling of their genetic sequences as well as epigenomic landscapes⁸⁷.

Imaging-based methods.

ecDNAs were originally discovered by imaging of metaphase chromosome spreads from cancer cells¹. Cells arrested in metaphase are fixed and condensed mitotic chromosomes are physically spread out on a microscope slide, providing excellent separation of chromosomes and ecDNAs. Metaphase spreading followed by DNA fluorescence in-situ hybridization (FISH) further allows hybridization of a sequence-specific probe and detection of oncogenes on ecDNA (Figure 4a). This method typically requires cell culturing followed by metaphase arrest. DNA FISH can also be performed on cells in interphase without arrest (Figure 4a), allowing detection of oncogene amplifications in clinical tumor sections. Interphase DNA FISH provides information about the spatial distribution of ecDNAs in intact cancer cell nuclei and led to the discovery of ecDNA hubs²⁴. Interphase DNA FISH combined with nascent RNA FISH (using probes that target intronic sequences on pre-messenger RNA) also allows single-molecule assessment of transcriptional activity on ecDNAs²⁴. However, interphase FISH alone does not identify ecDNA presence definitively and, therefore, should be used in combination with other methods such as metaphase FISH. Finally, live cell imaging has been used to visualize ecDNA dynamics in live cancer cells during interphase and mitosis^24,27,35 (Figure 4a).

Figure 4. Technologies used to reveal ecDNA gene regulation and structure.

(a) Imaging of ecDNA in interphase cells or metaphase spreads. (b) Bulk and single-cell sequencing approaches to identify regulatory elements, map chromatin interactions and reveal cell heterogeneity in the context of ecDNA. (c) Isolation of ecDNA and targeted analyses. (d) Computational methods and tools for reconstruction of ecDNA sequence. (e) CRISPR interference of gene promoters and regulatory elements on ecDNA.

Bulk sequencing-based methods.

High-throughput sequencing has enabled detailed characterization of the cancer genome and epigenome and have provided novel insights into the spatial and structural bases of regulation of gene expression on ecDNA. Some examples are (1) the mapping of DNA regulatory elements using assay for transposase accessible chromatin with sequencing (ATAC-seq)⁸⁸ and chromatin immunoprecipitation sequencing (ChIP-seq) targeting markers of regulatory elements such as H3K27ac⁸⁹; (2) the identification of chromatin interactions associated with spatial organization involving regulatory elements using HiChIP^90,91, chromatin interaction analysis by paired-end tag sequencing (ChIA-PET)⁹², or chromatin interaction analysis with droplet sequencing (ChIA-Drop)⁹³; (3) the fine-scale assessment of variable regulatory element activities at the single-cell level using single-cell multiomics, allowing simultaneous measurements of RNA expression and chromatin accessibility^94–96 (Figure 4b). These various methods have been applied to studying gene regulation on ecDNAs^{23,24,46,48,53}. Sequencing signals represent all DNA material in the samples, including both chromosomal DNA and ecDNA. In cancer cells containing high copy numbers of ecDNA, the majority of sequencing signals is assumed to originate from ecDNA molecules. However, ecDNA-focused interpretations of bulk sequencing data are more challenging when ecDNA molecules are present at lower copy numbers or in a small subset of cells in a heterogeneous cell population. Furthermore, ecDNAs typically contain extensive structural rearrangements and can be a mixture of heterogeneous amplicons containing various sequence elements within a cancer cell population^{5,24,48,54–58,87}, thus altering the 2D distances between loci as compared to the chromosomal reference sequence. Therefore, interpretations of chromatin regulatory interactions in relation to gene loci can benefit from construction of custom rearranged ecDNA sequence maps in addition to alignment to the reference genome.

Isolation of ecDNA.

To better profile the genetic and epigenetic landscapes of ecDNAs, there is a need for new molecular methods to isolate ecDNAs from cancer cells for targeted profiling and comparisons between ecDNAs and chromosomes. A technique for unbiased isolation of DNA circles, termed Circle-seq, was previously developed for small eccDNA molecules^19–21 and recently applied to oncogenic ecDNAs⁹⁷ (Figure 4c). Circle-seq involves magnetic bead-based genomic DNA isolation, exonuclease digestion of linear DNA fragments followed by multiple displacement amplification (MDA) of remaining DNA. Application of this method on neuroblastoma samples enabled analysis of structural rearrangements on ecDNAs⁹⁷. However, as megabase-sized ecDNAs are extremely fragile in solution and prone to breakage, Circle-seq favors small ecDNA and eccDNA species. In-solution DNA isolation is not recommended for DNA molecules that are above 100 kb in size.

To isolate ecDNAs which are megabases in size as are commonly observed in cancer cells, a method termed CRISPR-CATCH (CRISPR-Cas9-Assisted Targeting of CHromosome segments) can be used^87,98 (Figure 4c). In this method, ultra high molecular weight genomic DNA is embedded in agarose to maintain DNA integrity^87,99. Following agarose entrapment of genomic DNA, CRISPR-Cas9 ribonucleoprotein is used to cleave ecDNA circles in vitro and the resulting DNA is separated in pulsed field gel electrophoresis (PFGE). As linearized ecDNAs represent specific amplicon sizes, they are separated by size in PFGE and extracted from the gel. CRISPR-CATCH allows targeted profiling of the ecDNA genetic sequence and epigenomic landscape. This method serves as an empirical approach for heterogeneous amplicon separation and sequence reconstruction. It also enables physical separation of chromosomal DNA and ecDNA from the same cell sample, allowing direct comparisons. However, this targeted approach can only enrich for ecDNAs containing the CRISPR-Cas9 target sequence (e.g. a known oncogene) and therefore cannot be used as a method for unbiased ecDNA detection.

Computational inference of ecDNA structure.

In parallel to molecular techniques for ecDNA isolation, computational tools can infer ecDNA amplicon structures from bulk sequencing data (Figure 4d). Whole-genome short-read sequencing data can be analyzed by AmpliconArchitect, which constructs breakpoint graphs based on structural rearrangements detected and infer amplicon structures⁵⁴. Sequence information from long DNA molecules can be used to provide more accurately phased structural arrangements. Such information can be collected using long-read sequencing or optical mapping (OM)¹⁰⁰, both of which have been applied to ecDNA amplicons to resolve complex structures^23,24,48. AmpliconReconstructor integrates short-read sequencing data and OM data for accurate reconstruction of ecDNA amplicons⁵⁵. Finally, ecDNA sequence maps can be orthogonally constructed using sequence data from CRISPR-CATCH⁸⁷ (Figure 4d). Analysis of ecDNA amplicon structures in the context of regulatory elements may provide a better picture of how regulatory circuitry can be rewired by structural rearrangements on ecDNA, including enhancer and promoter hijacking events described earlier.

Functional interrogation via CRISPR interference.

To interrogate the regulation of gene expression by perturbation, CRISPR interference (CRISPRi) has been used to both directly target gene promoters on ecDNA as well as target non-coding regulatory DNA elements like enhancers^23,24,46 (Figure 4e). Due to elevated ecDNA copy numbers, Cas9-mediated DNA cleavage can lead to many double-strand breaks and is likely to come with unintended effects caused by DNA damage response as well as chromosomal integration of ecDNAs⁵⁸. On the other hand, CRISPRi can effectively silence ecDNA promoters despite high copy numbers, though effects of enhancer targeting appear diminished potentially due to combinatorial enhancer-gene interactions and compensation by other enhancers within ecDNA hubs²⁴. Nonetheless, enhancer effects on cell survival and oncogene expression can be detected in more sensitive, pooled assays^24,46. Pooled perturbation of enhancers by CRISPRi has identified cognate as well as ectopic oncogene enhancers that upregulate oncogene expression on ecDNAs and improve ecDNA+ cancer cell survival. These studies demonstrated an enhancer hijacking mechanism as well as intermolecular cooperativity in ecDNA hubs, two driving forces of ecDNA evolution and oncogene upregulation^24,46.

Conclusions

While ecDNAs have been long known to be an important mechanism of oncogene amplification in cancer, their impacts on cancer development, progression, evolution, and drug resistance are only beginning to be appreciated. With recent advances in sequencing technologies, we are now able to obtain detailed information about alterations in the cancer genome including oncogene amplifications, as well as consequences in gene expression programs. While mapping sequencing data of cancer cells to the reference sequence shows amplified oncogene sequences and structural rearrangements, it obscures differences in the spatial distribution of these oncogene amplifications. With the renewed interest in extrachromosomal oncogene amplification, we are now equipped to study a wide range of cancer biology questions in the context of ecDNA, including how gene expression is regulated.

As collective attention shifts from a sequence-oriented view to interrogating molecular topology, architecture, and structural organization in cancer, ecDNA has emerged as a challenge at the interface of cancer genetics and epigenetics. ecDNA also poses exciting new challenges for our understanding of tumor evolution. As described, the concept of ecDNA hubs and intermolecular combinatorial interactions opens a new paradigm in understanding how physical architecture regulates gene regulation, as the fundamental unit of transcription shifts from the gene to the hub.

Future development.

The discoveries of ecDNA hubs and regulatory rewiring on ecDNAs have provided a strong basis for the identification of unique aspects of transcriptional regulation of oncogenes not previously appreciated. These unique regulatory mechanisms on ecDNAs may point to unique oncogene transcriptional dependencies of ecDNA+ cancer cells. Future studies dissecting these regulatory mechanisms may provide novel insights into potential vulnerabilities of ecDNA-driven and oncogene-addicted cancers. Genome-wide genetic screens as well as small-molecule screens with a focus on ecDNA+ cancer models may identify these potential vulnerabilities. The observation that ecDNA hubs drive extrachromosomal and chromosomal gene expression warrants further investigation of the their significance in in vivo models as well as models from various cancer types. These future investigations will address whether ecDNA hubs are universal, cancer type- or oncogene-specific. In addition, Large-scale studies tracking extrachromosomal oncogene contents within primary tumors as well as metastases in cancer patients during chemotherapy and immunotherapy may also address the long-standing question of whether ecDNAs provide an additional advantage to cancer cells undergoing selective pressure.

Ongoing efforts in method development allow ecDNA characterization with ever increasing resolution. As molecular and computational tools now enable isolation of ecDNAs as well as prediction and reconstruction from bulk cancer samples, these techniques may provide important insights into the prevalence and evolution of ecDNAs in clinical tumor samples. Finally, future development of empirical ecDNA detection methods in clinical samples is urgently needed. While oncogene copy number amplifications can be detected by interphase DNA FISH and whole-genome sequencing, it is still challenging to empirically and definitively attribute these amplifications to ecDNAs in clinical tumor samples. Development of such techniques will enable systematic examination of clinical outcomes associated with ecDNAs as well as sequence and structural features of ecDNAs in patient tumors. Together, these upcoming advances will more precisely pinpoint the role of ecDNA-based oncogene amplifications in cancer and identify new ways to target oncogene-addicted tumors therapeutically.