Human DNA is organized into 23 pairs of chromosomes, which provide the templates for the proteins that make up the body. Occasionally, chromosomes can break or mutate by mistake. This creates changes in molecular networks and can alter cell functions, driving abnormal cell growth in cancer. A dramatic type of genetic change in cancer is when extra copies of DNA are formed outside chromosomes. These extra DNA copies, or extrachromosomal DNA (ecDNA), were originally discovered in the 1960s in pediatric tumors (1) and called "double minutes" because they often appear as paired dots on chromosome spreads. In the following decades, they were found in many types of cancer and shown to carry oncogenes (2), thus dysregulating growth signals in cells. ecDNAs appear to have distinct properties from chromosomes. For example, they form circular rather than linear DNA structures (3). They also lack centromeres and therefore divide unevenly during cell division (4, 5). However, how ecDNAs affect oncogene activity and the evolutionary dynamics of cancer cells has been poorly understood.
Motivated to understand how oncogenes are dysregulated in cancer at the level of the genome, I started my doctoral thesis work by first taking on the challenge of isolating ecDNA molecules from cancer cells. At the time, DNA fluorescence in situ hybridization was used to visually identify extrachromosomal oncogenes by microscopy, and bulk DNA sequencing was used for demonstrating copy number-amplified ecDNA signals beyond the native cellular genome.
However, targeted sequence analyses of ecDNAs were difficult owing to the inability to separate them from the native chromosomes. We adapted a method called CRISPR-CATCH (CRISPR-Cas9-assisted targeting of chromosome segments) to physically isolate ecDNA molecules by exploiting the differential migration patterns of large circular and linear DNA in pulsed-field gel electrophoresis (6). Using this approach on ecDNAs linearized with CRISPR-Cas9, we successfully separated ecDNAs from the rest of the genome, enabling analysis of their genetic sequences and epigenetic features (6). We identified a number of mutations in ecDNA in patient-derived cancer samples, which displayed strong genetic divergence from native chromosomes and selection of activating mutations within oncogenes such as EGFR and NRAS. By phasing single-nucleotide variants, we inferred the chromosomal origin of the ecDNAs as well as an excision scar left behind on the corresponding chromosomal allele in these cancer cells, which supports a model of ecDNA formation in which DNA is excised from chromosomes. We further discovered that the EGFR oncogene promoter has reduced DNA methylation on ecDNA compared with chromosomes, suggesting a potential mechanism of oncogene up-regulation on ecDNAs. Finally, using a computational workflow for reconstruction of ecDNAs, we identified ecDNA structures with rearranged oncogene loci and genomic enhancer elements, suggesting oncogene dysregulation by means of rearrangement of regulatory DNA elements that modulate gene expression.
I next investigated the interactions between these oncogene loci and regulatory DNA elements to understand how oncogene expression may be dysregulated on ecDNAs beyond simple copy number amplifications. The three-dimensional context of genomic elements in the physical space of the cell nucleus heavily influences their interactions and activities, such as gene expression (7). ecDNA, by definition, exists outside of the normal chromosomal context, and this altered genomic context may change how regulatory elements interact with oncogenes. Even when accounting for increased copy numbers, ecDNAs enable disproportionately high levels of oncogene transcription compared with chromosomes. Using genomics and fluorescence imaging, we found that ecDNAs cluster with each other in the nucleus, a phenomenon we term ecDNA hubs (8). ecDNA hubs were observed in all cancer lines and patient tumors that we analyzed and involved well-known oncogenes such as MYC, FGFR2, EGFR, MDM2, MYCN, and CDK4. We discovered that ecDNA hubs promote intermolecular enhancer-oncogene interactions among many ecDNAs, driving oncogene expression. These interactions are mediated by chromatin regulators called bromodomain and extraterminal (BET) proteins and can be perturbed by chemical inhibition of BET proteins as well as targeted CRISPR interference, thereby reducing oncogene expression in cancer cells. These results demonstrate a distinctive mechanism by which ecDNA hubs drive cooperative, intermolecular oncogene activation (8, 9) and suggest a potential vulnerability of oncogene-addicted cells.
These cooperative interactions among many ecDNAs within cancer cells raised a key question: How are they distributed among a growing cancer cell population? Unlike chromosomes, ecDNA lacks centromeres and segregates unevenly during cell division (4, 5). Thus, different species of ecDNA were predicted to be randomly distributed among dividing cells. Surprisingly, we found that this prediction was not true. By measuring ecDNA copies in pairs of daughter cells after mitosis, we found that distinct ecDNA species are coinherited (10). That is, a cell with many copies of one type of ecDNA also tends to inherit many copies of another type of ecDNA (e.g., ecDNAs carrying different regulatory DNA elements or oncogene sequences). Using single-cell genomics and evolutionary modeling, we further showed that coinheritance establishes copy number correlation of distinct ecDNA species in a growing cell population. This phenomenon allows ecDNA species to continue to cooperate and interact after multiple cell generations and leads to coupled dynamics when cells are under selective pressures such as cancer drugs. Transcription inhibition reduced coinheritance, suggesting that transcriptional complexes within ecDNA clusters promote their cosegregation. Thus, collectives of ecDNAs are passed down together in dividing cancer cells, allowing winning combinations of genetic sequences to drive cancer activity.
As a whole, my thesis research has demonstrated how the emergence of circular ecDNA structures can lead to dysregulation of oncogene expression, with implications for how these alterations may affect cancer cell fitness and evolution. My work has also demonstrated that the cancer genome is highly dynamic. It rearranges in sequence to alter genes and their regulatory elements, it rearranges spatially to modulate gene expression, and it rearranges across cell generations to influence gene inheritance by dividing cells (see the figure). Looking into the future, a better understanding of how the cancer genome defies the normal constraints of chromosomes may help us update our thinking about how tumors evolve and ultimately help us create better treatment paradigms.