A recipe for chaos: Extrachromosomal DNA and the hallmarks of cancer.

Introduction
The publication of the Hallmarks of Cancer in Cell twenty-six years ago was a watershed moment. By providing an organizing principle and shared language used by layman and experts alike, the Hallmarks concept has shaped cancer research from fundamental mechanism to diagnostics and drug discovery. Since that time, the hallmarks have continued to evolve, to parse cancer into a series of clear, understandable, and measurable features1–4. Recent technological advances and the ability to deploy them to study patients have also yielded some surprises, from rethinking the deterministic role of mutations in cancer, to recognizing the differential impacts of mutations, structural variation, and chromosomal alterations, to observing the critical interplay between genes, environment, and the immune system in cancer formation and progression. It may be time to again revisit the Hallmarks of Cancer in light of these new developments, including the increasingly recognized role for biological processes whose role in cancer was not previously appreciated. Extrachromosomal DNA (ecDNA) is one of them.
The past decade has revealed a surprisingly important role for ecDNA and its non-Mendelian inheritance in cancer. The application of modern molecular techniques to an observation made over 60 years ago, has opened a new window and provided a potentially unifying explanation for long standing questions in cancer biology such as – why some cancers are so heterogeneous, how they achieve such high levels of oncogene amplification but can change DNA copy number so quickly to resist treatment, and why patients whose tumors have these ecDNAs, have significantly shorter overall survival5. Beyond existing as a form of gene amplification, ecDNA is also an epigenetic problem that changes how genes are organized, regulated and expressed. It also appears to exert influence on the tumor microenvironment, playing a role in suppressing innate and adaptive immunity. Thus, the integration of ecDNA into the Hallmarks of Cancer framework is now warranted. This review will highlight the impact of ecDNA on the Hallmarks of Cancer, especially genome instability, a crucial enabling feature, in a way that affects many of the Hallmarks of Cancer. We will discuss what has been learned about ecDNA biology to date, its role in human cancers, and its enabling relationship to the Hallmarks of Cancer.

Extrachromosomal DNA (ecDNA)
EcDNAs are megabase-sized DNA circles that can encode a combination of oncogenes, immune escape genes, and DNA regulatory elements6,7, driving high copy number amplifications that are almost exclusively observed in cancerous but not in normal human genome. Being unbound by chromosomes, ecDNAs lack centromeres and telomeres, enabling their rapid and uneven segregation into daughter cells through chromosomal hitchhiking at every successive rounds of mitotic cell divisions, stirring a level of remarkably heterogeneous genetic landscape8–10. Aside from driving heterogeneity at the copy number level, distinct ecDNA species which encode different genetic elements can exist within a single cancer cell, further exacerbating genetic diversity at the genetic content level11. Unlike circular mitochondrial DNA that lacks histone proteins and nucleosome structures, ecDNAs are packaged into highly accessible chromatin with a preponderance of active histone marks7. Despite the normal arrangement of the DNA into nucleosomes and an intact domain structure, it became clear that ecDNAs lack the ability to undergo higher-order compaction to exclude the transcriptional machinery. Thus, ecDNAs are powerful units that rewire genetic circuits to maximize oncogene expression, well-proven by the fact that ecDNA-encoded oncogenes are among the top 1% of genes expressed in the cancer genome even when normalized for gene copy number7.

Early cytogenetic observations of ecDNAs
Prior to the molecular era, the most common way to interrogate cancer cells was to look at them under a microscope and examine the morphology of their chromosomes during cell division. In 1965, Arthur Spriggs and colleagues did just that and reported, “the presence of very small double chromatin bodies” in fresh tumor sample metaphase preparations12, which he called double minutes because they were often, although not always, doublets. It turns out that only about 30% of these are doublets6, hence the switch to the descriptive nomenclature of ecDNA. At the time, the significance of such extrachromosomal DNA particles was not clear. Of note, human genomes are laced with small circular DNA elements, including rDNAs, microDNAs13, and even small extrachromosomal circular DNAs called eccDNAs14, they are rarely amplified and most of which do not contain genes. Although eccDNAs may play a role in cancer, they will not be the focus of this perspective.
Between 1979 and 1981, Robert Schimke and colleagues, including Fred Alt, Rodney Kellems, and Daniel Haber, identified an “unstable” form of the amplified dihydrofolate reductase gene on double minute chromosomes in mouse cancer cells and mouse fibroblasts, that was associated with methotrexate resistance15–17. Following up on the idea that this could be relevant to human cancer, Garrett Brodeur and colleagues detected double minute chromosomes in human neuroblastoma samples, and Kari Alitalo, Manfred Schwab, Chien-Chi Lin, Harold Varmus, and Michael Bishop identified c-MYC amplification on double minutes in a human colon cancer cell line18, while Alt and colleagues detected MYCN amplification on double minutes in a neuroepithelioma cell line19. Continued progress was made through the studies of Geoffrey Wahl, Elena Guilioto, Michelle Debatisse and Bernard Malfoy, George Stark and others20,21. Wahl, along with Dan Van Hoff and others continued to find evidence to suggest importance in human cancer22. However, despite key work from a small number of researchers including Noriaki Shimizu, Songbin Fu, and Clelia Tiziana Storlazzi, the relative importance of ecDNA remained unclear23–27.
As technologies advanced, the field moved from observing cells in metaphase, to gene microarrays and then mapping of the entire human genome along with next generation sequencing, cancer biologists unsurprisingly shifted to reading and interpreting DNA sequences. In this process, the localization of cancer gene alterations became inferential, as short sequencing reads were mapped back to a normal reference genome. Extrachromosomal DNA largely disappeared from the literature, being considered a rare (1.4% of cancers) event of unclear significance28.

Chromosomal assumptions – a blind spot for ecDNA
Chromosomes are the terrain on which genes are written, providing the physical unit for genetic inheritance in all three kingdoms of life. Consequently, chromosomal inheritance underlies the most basic assumptions about how genes are passed during cell division. From Boveri’s chromosomal theory of cancer29 to the Philadelphia chromosome30–32, chromosomes play an outsized role in our basic understanding of cancer, serving as a reference point for interpreting DNA sequencing studies to decipher the mechanisms of tumor formation, progression, and treatment resistance. Advanced sequencing technologies rely on the implicit assumption that genes are on chromosomes, thereby following the rules of Mendelian inheritance. Advanced DNA sequencing technologies enabled the generation of remarkably detailed cancer genome maps of many different tumor types. To construct these maps, DNA sequence reads are mapped back to the reference human genome coming from normal cells, potentially mistakenly assigning ecDNA-based amplifications to the chromosomes from which the genes arose. Similarly, while techniques to study the clonal evolution of tumors using phylogenetic approaches remain effective for resolving the subclonal architecture based on chromosomal single nucleotide variants (SNVs) and structural variants (SVs), they implicitly assume chromosomal inheritance for all genomic elements. This assumption can result in misleading conclusions when inferring the genomic location and rapid evolutionary dynamics of genes amplified on ecDNA, which are key drivers of tumor fitness.

Revisiting the problem with new tools in hand
In 2014, a set of paradoxical observations about glioblastoma, a highly lethal brain cancer, brought the problem of ecDNA into stark new focus. Glioblastomas frequently harbor amplified EGFRvIII, a gain-of-function deletion mutant of the EGFR oncogene. First, the extent of intratumoral heterogeneity of EGFRvIII protein expression from cell to cell was too great to be easily explained by classical human genetics. It was postulated that this heterogeneity could contribute to treatment resistance, but it was unclear how. Second, when tumor cells that expressed either very high or very low levels of EGFRvIII protein were sorted and put into culture, or into a mouse brain, they gave rise to tumor populations that recapitulated the extent of intratumoral heterogeneity found in the original tumor, which again was difficult to understand. Third, when EGFRvIII amplified glioblastoma cells were treated with an EGFR tyrosine kinase inhibitor in a dish, in a mouse, or in patients, they quickly became resistant to treatment by dramatically lowering EGFRvIII levels, which rapidly returned once drug was removed33. Looking at the chromosomes, rather than reading the sequences, unraveled the mystery. EGFRvIII was not encoded on chromosomes, but rather on ecDNA elements. Treatment with an EGFR tyrosine kinase inhibitor resulted in relatively rapid resistance, concomitant with a large decline in EGFRvIII oncogene DNA and protein levels that returned to baseline upon discontinuation of drug treatment33. These results indicated that dynamic regulation of oncogene levels on ecDNA could be a major contributor to intratumoral genetic heterogeneity and targeted therapy resistance. Subsequent studies demonstrated that increased copy number of oncogenes amplified on ecDNA can also contribute to targeted therapy resistance, as has been demonstrated by BRAF or NRAS amplification on ecDNA in tumor cells from melanoma patients and patient-derived xenografts that developed resistance to MAPK inhibitors34,35.
At this point, the critical gaps in the collective understanding of oncogene amplification became apparent. How often are driver oncogenes amplified on ecDNA in cancer, and does the location of amplification matter? These important questions have now been addressed by leveraging powerful new DNA sequencing, computational, epigenetic, and imaging analysis tools.

EcDNA in human cancers
Through analysis of whole genome sequencing, structural modeling, and cytogenetic analysis of 17 different cancer types on metaphase spreads, the prevalence of ecDNAs in human cancers was clarified. The most prevalently amplified driver oncogenes in cancer, including EGFR, MYC, CCND1, MYCN, amongst others, are frequently found on ecDNA, and such ecDNA was not detected in the normal cells studied6. Importantly, ecDNA particles lacked centromeres. Mathematical modeling suggested that ecDNA would drive intratumoral genetic heterogeneity, which was empirically verified6,8–10, raising the possibility that it could play an important role in accelerated tumor evolution.
The development of AmpliconArchitect36 and other computational tools made it possible to identify ecDNA from whole genome sequencing data. With their application to public datasets, the prevalence of ecDNA across human cancer is now being recognized (Figure 1). In a study of sequencing data from nearly 15,000 adult patients of 39 different tumor types, ecDNA was detected in 17.1% of samples, with particularly high frequencies in glioblastoma, sarcomas, breast, lung, upper GI, and genitourinary cancers, amongst others, and was significantly associated with advanced metastatic disease and shorter overall survival5, observations that were also identified in other datasets37,38. EcDNA is not confined to adult tumors, being similarly found in 18% of medulloblastoma in children, where it is also linked with significantly shorter survival39.

EcDNA in some pre-cancerous lesions
Recent work demonstrates that ecDNA can also be found in high-grade dysplasia before the onset of frank cancer40 or in dysplastic regions adjacent to cancer42. In patients with Barrett’s esophagus, in two clinical cohorts, ecDNA was detected in high-grade dysplasia, and its presence was tightly linked to the development of esophageal adenocarcinoma40. Similarly, the presence of ecDNA in regions of high-grade dysplasia adjacent to urothelial cancers, suggests that it is involved in malignant transformation42. Houlahan et al. demonstrated that in high-risk ER+ and HER2+ breast cancers, cyclic ecDNA amplifications were already present in pre-invasive ductal carcinoma in situ (DCIS) lesions43. The study further revealed a unique mechanistic link where estrogen receptor (ER) signaling induces the formation of R-loops, which subsequently serve as substrates for APOBEC3B-mediated editing and double-stranded breaks. These early structural scars suggest that ecDNA may be established well before invasive progression, acting as a primary engine for replication stress and immune evasion from the disease’s inception. In light of these data, and the role in transformation demonstrated in mouse genetic models, there is a growing recognition that ecDNA might be more broadly involved in cancer formation, although current knowledge is limited. There is a need to investigate its presence across other pre-invasive cancer types to determine its viability as a screening marker.

EcDNA and the Hallmarks of Cancer

ecDNA and the unstable genome
The cell has long been considered the unit for studying genome instability and variation. However, for cancers with ecDNA, the unit driving variation can be the ecDNA particle and its subsequent inheritance, as well as the epigenomic, regulatory, and functional consequences. New techniques including single molecule methods, and live cell imaging to track ecDNAs over time have begun to provide unprecedented insight into how ecDNA directly and indirectly impacts several established hallmarks of cancers, highlighting ecDNA’s contribution to fostering genome instability and mutation, sustaining proliferative signaling, evading growth suppressors, and avoiding immune destruction. This section on the life cycle of ecDNA (Figure 2) will explore the fundamental contributions of ecDNAs to the hallmarks of cancer.

ecDNA formation
The concept of genome instability encompasses a variety of different flavors of genomic alterations of DNA in cancer cells, including changes in sequence, copy number, structure, or topology. Remarkably, ecDNA is a potent contributor to each of these components, fundamentally contributing to genome instability and mutation. Genome instability and mutation are not only foundational hallmarks of cancer, but they are also enabling features that lead to the acquisition of other cancer hallmarks2. When Schimke and colleagues suggested that ecDNA was an “unstable” form of gene amplification15, they were presciently foreshadowing the pivotal role for ecDNA in promoting genome instability.
The extent of genome instability in ecDNA-containing tumors is vast, because it can be both a cause and a consequence of genome instability. Compelling correlative data suggests that features known to contribute to genome instability may also contribute to the initial formation of ecDNA. In their 2011 update, Hanahan and Weinberg pointed to the importance of a surveillance system that monitor genome integrity, including through the ‘guardian of the genome’ TP5345; and via the DNA maintenance machinery through ‘caretakers of the genome’ with roles in 1) detecting DNA damage, 2) repairing DNA damage, and 3) suppressing mutagenic molecules48. Both the ‘guardian’ and ‘caretakers’ are impaired in tumors that have ecDNA. In addition, beyond the guardian’s role, TP53 is a canonical tumor suppressor that negatively regulate cell proliferation2, thus it serves as a key mechanism to evading growth suppressors. Corroborating ecDNA’s contribution to this cancer hallmark, TP53 loss is significantly associated with ecDNA, particularly in endometrial, renal, and luminal estrogen receptor positive breast cancer5,46, and also in patients with Barrett’s esophagus who develop esophageal adenocarcinoma40. In some cancer types, TP53 loss was not associated with ecDNA, such as glioblastoma and sarcomas, in which other tumor suppressor losses such CDKN2A and NF1 were detected5.
In addition to obvious tumor suppressor losses, the regulation of total genome copy number is often aberrant, with whole genome doubling being strongly associated with ecDNA formation5, underscoring the underlying genome instability of ecDNA-containing cells. Interestingly, ecDNAs are less likely to be observed in tumors with underlying mismatch repair deficiency and DNA polymerase δ 1 or DNA polymerase ε deficiency (POLD1/POLEd)5. The hypermutated genomic landscape and copy number alteration status seemingly exist in mutual exclusivity in certain cancers such as colorectal cancers, where some hypermutated colorectal cancers are relatively chromosomally stable with far less somatic copy number alterations49,50. Such mutual exclusivity suggests hypermutation and copy number alteration could be alternative mechanisms to confer proliferative advantage, which warrants future mechanistic studies to dissect whether they follow distinct evolutionary trajectories that result in functional redundancy.
EcDNA formation arising from paired double strand breaks usually occurs in the context of a tumor suppressor loss, including TP535,40, resulting in selection for ecDNAs bearing a gain of function element, such as an oncogene as illustrated in one type of DHFR/methotrexate model47. The causative role for paired double strand breaks in ecDNA formation has been further experimentally established, as demonstrated using CRISPR-C51, and more recently, with Cre-Lox to generate ecDNA-driven cancer in autochthonous mouse models52. EcDNA has also been shown to arise from chromothripsis, in which shattered pieces of DNAs resulting from lagging chromosomes that transit into micronuclei are subject to nuclease attack, after which DNA fragments can ligate into circular ecDNA structures47. Genetically engineering the Y chromosome and forcing it to mis-segregate in studies conducted by Peter Ly and Don Cleveland’s labs, confirmed that chromothripsis can give rise to ecDNA. Through these ecDNA biogenesis models, specific DNA damage and repair proteins associated with ecDNA formation have been implicated, including those involved in non-homologous end joining34,37,47, the Fanconi anemia pathway44, and nucleases including the recently described role for N4PB2 that appears to be important for ecDNA formation from lagging chromosomes in micronuclei53. We anticipate that the mechanism of ecDNA formation may vary by tumor tissue type5.

ecDNA remodeling
Time-inferred mutational signature analyses strongly suggest that ecDNAs continue to be remodeled after they are formed. This ongoing remodeling further enhances genome instability and mutation on ecDNA itself, which, through selection, can enhance tumor cell fitness. For instance, smoking and APOBEC-mediated cytosine deamination are frequently found in ecDNA-containing tumors, and signatures of homologous recombination deficiency and APOBEC activity become prominent once ecDNAs have formed5. In agreement with these findings, by mapping clustered mutations across 2,583 genomes, Bergstrom et al. found prominent clustered APOBEC3-mediated mutagenesis (kataegis) a common feature on existing ecDNA particles54, resulting in strand-coordinated mutational bursts. In urothelial cancer, APOBEC3-induced mutations act as early clonal drivers, while platinum-based chemotherapy triggers late mutational bursts55. Crucially, they found that ecDNA-forming SVs, such as those amplifying CCND1, not only persist through systemic therapy but also increase in complexity and copy number. Additional layer of regulatory plasticity has been observed in a study involving paired high-grade serous ovarian cancer (HGSOC) samples whereby ecDNAs in metastatic sites undergo active DNA demethylation, driving the expression of transposable elements such as LINE-1 and pathways associated with aggressive progression56.
EcDNAs are also a potent source for structural variation to enhance genome instability. Chromosomal rearrangements and translocations are a well-recognized source of gene fusions such as BCR-ABL, activating the biochemical activity of a proto-oncogene (i.e the ABL tyrosine kinase activity). Until recently, the link between ecDNA and gene fusions was not known. Recent data demonstrate that genomic rearrangements resulting in amplified fusion transcripts are prominent on ecDNA. Integrated analysis of whole genome and transcriptome sequencing data from cancer samples and cell lines of many different tumor types, demonstrated that ecDNAs have the highest rate of oncogene fusion events of any copy number alterations, establishing ecDNAs as the main source of gene fusions in solid tumors57.
EcDNAs can also generate genome instability by integrating into chromosomes as homogenously staining regions (HSRs)58 that create new cis-regulatory interactions. The frequency of ecDNA chromosomal integration, its preferred sites, as well as its causes, are currently not well understood. Importantly, ecDNAs are not the only source for HSR formation. Other forms of amplification, including breakage-fusion-bridges can also give rise to HSRs59.
Amplification of transposable elements on ecDNA generates yet another form of genome instability. 3D mapping of the architecture of MYC-amplified ecDNA in colorectal cancer cells demonstrated that transposable elements can be frequently amplified and reactivated on ecDNA, and that they have functional enhancer activity that drives transcription60. Taken together, it is clear that there are multiple ways that contribute to the ongoing remodeling of ecDNAs in cancer. The rapid selection for ecDNAs that enhance tumor cell fitness, is consistent with the finding that ecDNAs are more likely to be retained than chromosomal amplifications in longitudinal samples38.

ecDNA expression
As described above, oncogenes borne on ecDNAs, such as EGFR, MYC, FGFR2, KRAS, and CDK4, are among the top 1% of genes expressed in the cancer genome. The genetic elements encoded on ecDNA often regulate key signaling nodes that drive cancer cells toward ceaseless proliferation, fundamentally contributing to sustained proliferative signaling. EcDNAs are known to increase the transcriptional output of their oncogene cargo by mechanisms distinct from chromosomal DNA. First, ecDNAs have highly accessible chromatin, and such open chromatin structure is accompanied by the lack of higher-order chromatin compaction. Enhanced ultra-long-range internal regulatory interactions on ecDNA are observed, as the circular topology relocate distal enhancers close to oncogenes in cis (also referred to as enhancer hijacking)7. Second, multiple copies of ecDNAs often come into physical proximity to share regulatory elements61. Such cooperative intermolecular interactions of enhancers and gene promoters, which are encoded on ecDNA, facilitate the formation of micron-sized collections of 10 to 100 copies or more of ecDNA molecules in cell nucleus, which is termed ecDNA hubs61. Compared to singleton ecDNAs, ecDNAs in hubs are more transcriptionally active as revealed by FISH-based imaging. EcDNA hubs provide cancer cells with a massive transcriptional advantage, because ecDNAs that originate from different chromosomes can congregate together and share gene regulatory elements in the same ecDNA hubs, thereby allowing transcriptional rewiring that is otherwise not possible for genes located on different chromosomes61. Importantly, targeting proteins that maintain ecDNA hubs structure such as BRD4 through JQ1 inhibition, greatly attenuates their transcriptional advantage. Third, since gene fusions are frequently found on ecDNAs57, fusion transcripts arising from these aberrant gene rearrangements may endow cancer cells with growth advantage by sustaining proliferative signaling. A salient example is the frequent gene fusions linking the long non-coding RNA PVT1 on the 5’ end to diverse oncogenes at the 3’ end on ecDNAs (e.g. PVT1-MYC, PVT1-CASC11). PVT1 5’ fusion was found to increase the messenger RNA stability of any linked short-lived transcript, revealing a new gain-of-function mechanism to activate MYC and other oncogenes57.
Altogether, this form of altered cis-regulation as well as unique forms of trans regulation described above, have profound effects on ecDNA-mediated gene transcription and cancer genome expression. This heightened transcription, in turn, creates increased opportunity for DNA damage, likely, at least in part, through pervasive collisions between replication-transcription machineries on ecDNA that promote replication stress and damage to the ecDNAs themselves62, further exacerbating genome instability and mutation.

ecDNA inheritance
One of the most important aspects of ecDNA is what they lack – centromeres. In chromosomal inheritance, spindle fibers attach to centromeres through kinetochores to separate sister chromatids during cell division63, ensuring equal segregation of genetic information from mother to daughter cells. Without centromeres to equally distribute genetic information to daughter cells, ecDNAs are randomly segregated, resulting in intratumoral genetic heterogeneity9 and fostering rapid genome evolution, as tumor cells that contain an ecDNA with elements that enhance fitness are rapidly selected, including resistance to treatment9. One can speculate that the random segregation of ecDNA in cancer cells bears some similarities to how genetic information can be asymmetrically inherited in bacteria through circular chromosomes and plasmids. Daughter cells do not always inherit the same DNA as the mother cell during cell division. Consequently, cell populations rapidly evolve, including to drive antibiotic resistance. The analogy is imperfect, in that horizontal transfer of ecDNA has yet to be established, but the concept of rapid evolution driven by unequal inheritance of circular DNA elements provides an illuminating analogy.
As cancer cells with ecDNA divide, the net effect of the numerical and structural evolution on ecDNA is dynamically and unevenly reset simply by how ecDNAs are inherited through non-Mendelian principles9, stimulating ongoing genome instability in terms of sequence, copy number, and structure. Of note, individual ecDNAs can contain single or multiple oncogenes, and multiple ecDNA species encoding combinations of oncogenes, immunomodulatory genes, and/or regulatory elements can be found within a single cancer cell5,11. Yet paradoxically, if ecDNAs gain transcriptional advantage by congregating together in ecDNA hubs, random segregation would often break up advantageous combinations of ecDNA species in daughter cells, based on the expectation of chromosomal behavior of independent assortment, known as “Mendel’s third law” of genetics. Instead, Hung et al. discovered that ecDNA hubs can be coordinately inherited by daughter cells after cell division, a process termed ecDNA cosegregation, thereby ensuring the continuity of fitness advantages across somatic cell generations11. During mitosis, chromosomes are dramatically condensed and cease transcription. In contrast, ecDNAs retain their open chromatin architecture, and the continuation of transcription initiation is required to sustain enhancer-promoter contacts between ecDNA species in ecDNA hubs that allow for their coordinated inheritance. En masse, the overall distribution of ecDNA particles remains random, but interacting ecDNA particles co-segregate by staying together during mitosis, and being parceled randomly to daughter cells as a unit. This process allows winning combinations of genetic and epigenetic states to be transmitted across cell generations, thereby stabilizing oncogene cooperation that sustains proliferative signaling, fueling rapid tumor evolution under therapeutic pressure11. Such unique behavior of ecDNA inheritance may represent a vulnerability for targeted therapeutic strategies.
The absence of centromeres implies that ecDNAs are at risk of being left behind in the cytoplasm with every cell division. This challenge is analogous to the problem faced by DNA viruses, which have evolved specific DNA binding proteins and viral DNA elements to tether viral episomes onto human chromosomes, a process termed mitotic hitchhiking. Sankar et al. discovered that human ecDNAs possess a family of DNA elements, termed retention elements, that promote the tethering of episomes to mitotic chromosomes64. Adding retention elements to heterologous episomes promotes their retention, and inhibiting the activity of retention elements in ecDNAs led to increased untethering and increased ecDNA loss. Notably, retention elements have a natural function: they are gene promoters that uniquely retain transcription factor binding during mitosis, which are termed mitotic bookmarks. When chromosomes condense during mitosis, chromosomal transcription is shut off, but the presence of DNA-binding factors on a subset of genes promote the re-activation of these genes in the G1 phase of the next cell cycle, thereby propagating gene expression memory. Retention elements in ecDNAs contact enhancer elements on chromosomes that are also mitotically bookmarked; in effect, ecDNA retention is a co-option of the enhancer-promoter contacts involved in mitotic bookmarking64. Mitotic retention is critical for ecDNA immortality and thus represents a distinguishing feature of ecDNA biology that may be amenable for therapeutic targeting.
In addition, micronuclei are frequently found in ecDNA-containing cells, consistent with the notion that the presence of micronuclei is a feature of genome instability. Importantly, ecDNA amplicons are more frequently found inside micronuclei structures than chromosomal amplicons65. Such ecDNA-enriched micronuclei are often asymmetrically inherited into one of the daughter cells during mitosis, further driving the uneven inheritance of ecDNAs. Altogether, these interactions further enable ecDNAs to break Mendel’s law of independent assortment, adding to the surprising tricks that ecDNA use to generate genetic variation. Non-Mendelian inheritance is a central feature of ecDNA biology, explaining the surprising biology of some of the most aggressive forms of cancer – intratumoral genetic heterogeneity, massive oncogene copy number, and rapid genome evolution and treatment resistance.

ecDNA selection
The vastly heterogenous genetic landscape facilitated by ecDNAs serves as a key principle allowing for selection of cells with the best fitness corresponding to its tumor microenvironment. Analysis of ecDNA in human tumors over time40 reveals that ecDNAs are constantly evolved to further stir genome instability and mutation. Once an oncogene is present in multiple copies of ecDNAs, each copy is subject to mutation and selection for gain-of-function (GOF) variants. This process does not only give cancer cells more shots on goal to obtain an advantageous mutation, but also allow them to keep the elements that have most GOF mutations, which are usually oncogenes, thus sustaining proliferative signaling. Well-known GOF mutations frequently found in cancer cells, such as EGFRvIII or KRAS(G12C), lock their downstream signaling proteins into a constitutively ON state. While these GOF mutations are also known to be present in ecDNA-containing cells, it was unclear whether the GOF mutations are located on ecDNAs, chromosomes, or both. Recent development of methods to biochemically separate ecDNAs from chromosomal DNAs followed by sequencing, confirmed the GOF mutation arise from ecDNA instead of the chromosomal copy66. Thus, the elevated copy number and the mutagenic nature of ecDNAs make ecDNA a powerful platform for rapid propagation of GOF mutants that confer cancer cell fitness66,67.
The dynamic nature of ecDNAs endows cancer cells with the ability to rapidly respond to stress, or resist to treatment. EcDNAs allow for rapid, almost real-time, dosage control, shedding the amplified oncogene when under drug pressure and re-amplifying when the drug is removed to regain proliferative signaling33. The reversible, dynamic copy number change driven by ecDNA stands in stark contrast to stable chromosomal amplifications, suggesting that ecDNA could contribute to treatment resistance11,33 and shorter survival in patients, as has been recently demonstrated5. Initially, many researchers suspected that ecDNA was a resistance mechanism that arises once tumors were established, along the lines of how DHFR amplification occurs in response to methotrexate treatment or NRAS and BRAF amplification on ecDNA becomes apparent in melanomas that develop resistance to MAPK inhibitors34,35, or, for example, in prostate cancers that became resistant to anti-androgen therapy, in which AR or MYC amplification on ecDNA can be detected68. Through patient-derived (PD) model systems such as organoids (PDOs) and xenografts (PDX), the role ecDNA plays in driving plasticity and treatment resistance has been demonstrated. In pancreatic ductal adenocarcinoma (PDAC), ecDNA is a primary driver of phenotypic plasticity through MYC-heterogeneity69. By utilizing patient-derived organoids, the study demonstrated that MYC-bearing ecDNAs allow cancer cells to rapidly adapt to environmental stressors, such as the withdrawal of essential stromal niche factors like WNT. Similarly, in small-cell lung cancer (SCLC), PDX models have linked ecDNAs carrying MYC-family genes to acquired resistance against platinum–etoposide and olaparib–temozolomide. Notably, while non-MYC ecDNAs appear across all disease stages, ecDNAs harboring MYC/L/N amplifications are found almost exclusively in relapsed patients, suggesting they are a specific mechanism for surviving intensive DNA-damaging therapies70.
Taken together, these ecDNA-unique features create a powerful recipe for making tumors increasingly diverse through productive forms of genome instability and mutations, allowing ecDNA-bearing tumors to sustain proliferative signaling while evading growth suppressors’ control (Figure 2).

The Emerging Nexus of ecDNA and Immune Evasion
EcDNA can be a powerful architect of the immune evasive tumor microenvironment (TME), functioning to avoid immune destruction. Initial pan-cancer genomic and transcriptomic analyses provided the first glimpse into this relationship. By integrating whole genome sequencing with multi-platform immune infiltration data obtained from over 1,600 TCGA samples, we learned that ecDNAs can amplify immune regulatory genes that are associated with CD8 depletion, and showed reduced levels of cytotoxic CD8+ T cells and attenuated cytolytic activity in ecDNA-containing tumors compared to their ecDNA-negative counterparts71. In addition, these tumors often exhibit a systematic downregulation of the MHC class I and II antigen-presenting machinery, suggesting a fundamental defect in immune recognition71. A subsequent study employing more stringent ecDNA classification and robust gene feature-selection algorithms has reinforced these findings46. These analyses identified a core gene set predictive of ecDNA-associated immune evasion, characterized by the global suppression of lymphocyte activation and T cell-mediated signaling, suggesting a link between ecDNA and immune escape46.
While bulk sequencing provided a macroscopic correlation, it often obscured the intricate cellular heterogeneity of the TME. Advancements in single-cell RNA sequencing (scRNA-seq) and spatial profiling technologies have enabled higher-resolution investigations of ecDNA-driven immune evasion. In urothelial carcinoma, single-cell analysis has shown that ecDNA-positive tumors are specifically enriched for immunosuppressive regulatory T cells. Furthermore, these tumors exhibit frequent loss of heterozygosity at HLA class I loci and downregulated expression of B2M, significantly impairing their antigen-presenting capacity42. Similar spatial and single-cell patterns have been observed in small cell lung cancer, where imaging mass cytometry revealed the physical exclusion of T cells and NK cells from ecDNA-rich tumor regions72. Collectively, these studies raise an important hypothesis that ecDNA may act as a TME remodeler.
The mechanisms by which ecDNA facilitates immune evasion likely stem from its high copy number, hyper-accessible chromatin, and unique genetic cargo. As carriers of oncogenes, such as MYC, KRAS, and ERBB2, as well as immunomodulatory genes, ecDNAs leverage their highly accessible chromatin and high copy number to reach elevated expression levels5,7. These oncogenes are well-known regulators of the immune response73. For example, MYC has been shown to regulate immune evasion, capable of inducing PD-L1 expression, suppressing type I interferon signaling, and altering the cancer cell secretome to favor immunosuppression74.
As discussed above, ecDNAs may carry immunomodulatory genes. This was first highlighted in human papillomavirus (HPV)-associated head and neck cancers, where the CD274 (PD-L1) gene was found co-amplified on human-viral hybrid ecDNAs41. More recent large-scale efforts, such as the Genomics England study, indicate that approximately 34% of ecDNA-containing tumors harbor immunomodulatory genes5. Intriguingly, 41.5% of these immunomodulatory ecDNAs lack canonical oncogenes. Tumors harboring immunomodulatory ecDNAs exhibit reduced T cell fraction compared to those containing oncogene-only ecDNAs5. This suggests a profound selective pressure to maintain ecDNA species that prioritize extrinsic immune evasion as well as cell-intrinsic proliferative signaling.
While a growing body of evidence supports a correlation between ecDNA and an immune evasive phenotype, a fundamental question remains unanswered: Why do ecDNA-driven cancers display more profound immunosuppression than those harboring other forms of genetic amplification? Other genomic alterations, such as enhancer hijacking due to chromosome translocation, chromosomal amplifications, or even epigenetic derepression, can similarly drive massive oncogene expression and immunomodulatory gene activity. What is the ecDNA-specific factor that endows this unique immune evasive capacity?
Addressing this question faces several difficulties. First, traditional clinical data derived from bulk-cell sequencing lacks the resolution to accurately deconvolute the cellular complexity of the TME. Second, while emerging single-cell and spatial profiling technologies offer high-resolution insights, establishing definitive causality in human tissues remains a critical challenge due to a persistent lack of isogenic controls. This makes it difficult to disentangle the specific biological effects of ecDNA from other forms of chromosomal amplification. Finally, although syngeneic mouse models carrying ecDNA-positive tumors are becoming available52, we have yet to generate the isogenic controls required for a direct comparison between ecDNA and chromosomal amplification.
A unique opportunity to bridge this causality gap has emerged from genetically engineered mouse models, such as the classic KPfC (KrasLSL-G12D/+; Trp53fl/fl; Pdx1Cre/+) pancreatic ductal adenocarcinoma (PDAC) model, in which the mutant Kras oncogene may spontaneously amplify on ecDNAs or HSRs within the same tumor75. After isolating single-cell clones, a recent preprint study75 demonstrates that although these isogenic cancer cell clones proliferate similarly in vitro, ecDNA-containing tumors are significantly more aggressive and establish an immune evasive TME more rapidly in vivo. This divergent behavior is attributed to the random segregation of ecDNAs, which generates a subset of Kras super-expressor cells characterized by very high Kras copy numbers. These super-expressors reprogram the tumor microenvironment by secreting amphiregulin, an EGF-like growth factor that drives the expansion of myofibroblastic cancer-associated fibroblasts and subsequently blocks T cell infiltration75. By characterizing ecDNA-driven super-expressors as the primary architects of the TME, these findings offer a framework for studying the causal relationship between ecDNA and immune evasion.
The benefits of ecDNA-driven super-expression come at a metabolic and cellular cost, including elevated replication stress75, which is linked to genome instability76. To survive this oncogene overdose-induced stress77, ecDNA-driven cancers must systematically suppress the cell-intrinsic pathways that would otherwise detect such genomic instability. A key emerging observation is that the cGAS-STING cytosolic DNA-sensing pathway is frequently silenced in these cancers to prevent the recognition of ecDNA as a pathological danger signal. Unlike stably inherited chromosomes, ecDNAs may occasionally be expelled into the cytosol, likely due to defective segregation during mitosis, a phenomenon recently described in a preprint study78. Consequently, another preprint study demonstrates that when cGAS expression is ectopically restored, it selectively recognizes these cytosolic ecDNAs, triggering a robust innate immune response to the ecDNA-positive cancer cells79. This selective vulnerability reveals a unique therapeutic opportunity. Leveraging mRNA lipid nanoparticle technology to deliver cGAS can effectively turn the unique genetic signature of ecDNA into a precision-targeted destruction signal, inhibiting ecDNA-containing tumors by utilizing their own genomic instability against them79. Interestingly, it was found that activation of the cGAS-STING pathway impedes the formation of ecDNA79. Although the molecular mechanism remains to be elucidated, this finding suggests that the cGAS-STING pathway may serve as a barrier to ecDNA-driven tumorigenesis.
These recent studies have begun to reveal the mechanism of ecDNA-mediated immune evasion; however, significant questions remain. Importantly, it remains to be determined whether the expression heterogeneity inherent to different ecDNA-encoded genetic elements universally drives the immune evasive phenotype observed across various malignancies. Furthermore, it is not yet clear how cancer cells harboring traditional oncogenic ecDNAs might collaborate with those carrying specialized immunomodulatory and inflammatory genes to orchestrate TME remodeling. Beyond the impact of ecDNA-driven cancer on its surroundings, a provocative and unresolved question is whether a pre-existing immunosuppressive microenvironment facilitates the de novo formation and clonal expansion of ecDNAs. Addressing these complexities is a non-trivial task, as it will require high-fidelity modeling and precise isogenic controls, using both syngeneic mouse models and patient-derived organoids, to definitively disentangle causality from correlation.

EcDNA and other cancer hallmarks
Although the links between ecDNA and some of the other hallmarks of cancer have yet to be fully established, emerging data suggest that ecDNA is associated with most, if not all of the Hallmarks and enabling features (Figure 3). EcDNA represents a radical departure from traditional models of tumor promotion and mutation, driving a hyper-responsiveness to the microenvironment and facilitating rapid adaptation to external stimuli.
EcDNA-containing cells have shown traits indicative of deregulated cellular energetics. For examples, glioblastoma cells with EGFRvIII amplification on ecDNA are relatively less sensitive to glucose withdrawal-mediated cell death than are their isogenic counterparts in which the amplicon has integrated into chromosomal HSRs, likely due to the ability of ecDNA glioblastoma cells to rapidly modulate oncogene copy number and proliferation rate9. Transcriptomic analysis on the TCGA cohort provided much insight on additional deregulated cellular processes of ecDNA-containing cells80, revealing their reduced pro-apoptotic gene expression (resisting cell death), enhanced pro-angiogenic gene expression (inducing angiogenesis), elevated proinflammatory gene expression (tumor-promoting inflammation), and upregulated EMT gene expression (activating invasion and metastasis). Although the mechanisms remain to be elucidated, ecDNA is more frequently found in metastatic disease and in tumors that are pre-treated compared to newly diagnosed cancers5,38. The observation of ecDNA correlating with late-stage and treated disease has led to work focused on the mutational events that underlie this observation. Moreover, the recent discovery of ecDNA-borne TERT encoding telomerase implies that ecDNA can contribute to the indefinite replication of the host cell81, supporting their role in enabling replicative immortality. Ongoing and future research is now poised to further dissect the molecular mechanistic underpinnings on how ecDNA precisely drives these cancer hallmarks.

EcDNA as future cancer diagnostics
Driven by the observation that ecDNA is not a static genomic feature but a highly dynamic and resilient engine of somatic evolution, the clinical impact of ecDNA is warranted. Its increased frequency in late-stage and metastatic disease may be a direct reflection of selective advantages, which allows tumors to both persist through the bottleneck of intensive therapy and colonize distant microenvironments. Developing tools to detect ecDNA in blood will be critical for gaining more insight into when ecDNA arises and how it contributes to cancer progression.
Early research82 suggested that the fragmentation patterns might also contain clues to epigenetic activity in the cell of origin, and play a role in detection of diseased cells, including tumor cells. The ideas have led to the development of methods for identifying circulating tumor DNA (ctDNA) detection using cfDNA sampled from blood. The methods rely on differences in ‘fragmentomic features,’ including fragmentation length patterns, location relative to nucleosomes, methylation, GC content to enrich and or detect ctDNA83,84. Other methods have utilized mutational patterns, including SNPs, copy number variation and aneuploidy signatures85,86. CtDNA analysis is increasingly utilized for early detection and minimal residual disease monitoring87, in a variety of solid tumors including gliomas, breast cancers, metastatic colorectal, metastatic castration resistant prostate cancer86, small cell lung cancer88, non-small cell lung cancer89–91 and other cancer subtypes. Together, these developments raise the potential of ecDNA detection using cfDNA from patients92. In a recent study, cfChiP-seq detected highly amplified oncogenes MYC and MYCL in SCLC patients, consistent with ecDNA observed in matched tumor tissue using whole genome sequencing and AmpliconSuite88. EcDNA are typically high in copy number, contain many structural variants not found in normal genome, and have highly accessible chromatin likely leading to unique fragmentation. Furthermore, plasma cfDNA sequencing using tumor-informed or hybrid-capture approaches can identify focal high-copy amplifications and structural features consistent with ecDNA92. Together, these methodologies establish a framework for integrating ecDNA analyses across tissue and blood, with significant implications for prognosis, relapse and metastases prediction, and longitudinal disease monitoring.

The deeper lesson – ecDNA is a Hallmark of some cancers
The Hallmarks of Cancer concept, from its initial presentation to its updates, has provided a shared lexicon for describing distinct measurable features and enabling characteristics of cancer. How we think and communicate about cancer has changed in response, including how we diagnose tumors, what we measure, and even the processes that we aim to target. Therein lies the power of the Hallmarks concept and why it so deeply resonates. In this review, we have described research developments and discoveries that have prompted a re-evaluation of some of our core ideas about the molecular basis of cancer and, through doing so, helped to explain some of the most puzzling aspects of its biology. The non-Mendelian inheritance achieved by ecDNA is a powerful concept with profound implications — accelerated evolution — that helps demystify apparently discordant features of cancer. Isn’t it odd that a tumor can have high oncogene copy number while maintaining still high levels of cell-to-cell variation in the expression of that same oncogene? How is it possible that you treat a tumor with a tyrosine kinase inhibitor and the oncogene copy number rapidly drops or rises, and when you remove it, it rapidly comes back to baseline again? Why do people whose cancers have ecDNA do so much worse than most other cancer patients? By framing new discoveries that have addressed these questions within the context of Hallmarks of Cancer, we begin to see the importance of ecDNA.
Powerful new technologies, from single cell to single molecule approaches applied to tissue and blood, will surely change our ability to detect ecDNA, including early in the course of disease development or treatment resistance. As the biology unravels, new targets will be uncovered that will likely require the most modern chemistry approaches, from activity-based protein profiling to molecular glues and degraders, for translation into effective treatments for some of the sickest of all cancer patients whose tumor types often fall into the untreatable category93. At present, a deeper lesson is clear. Cancers that have ecDNA-generated amplification, close to 20% of all people with cancer and likely more with advanced disease, are different, and the biochemical mechanisms that support ecDNA, which is distinct from other types of genome instability, are at the heart of its genomic chaos. Dare we say, for that subset of tumors, is ecDNA a new Hallmark of Cancer?