Plasticity of extrachromosomal DNA segregation during drug adaptation

Abstract

Uneven segregation during mitosis is a striking feature of extrachromosomal DNA (ecDNA). Because ecDNA lacks a centromere, it is thought to segregate stochastically, generating intratumoral heterogeneity in genomic copy number. Drug treatment can readily change ecDNA copy number, enabling cells to acquire drug resistance, yet whether these changes reflect static selection of pre-existing clones or active reconfiguration under stress remains unresolved. To address this, we develop a high-throughput framework combining single-cell DNA sequencing with cellular barcoding for clonal tracking. Single-cell cloning reveals that not all clones exhibit identical segregation modes even under drug-free conditions. Under treatment, resistant populations do not simply arise from pre-existing clones with favorable ecDNA states; instead, some clones actively reconfigure their segregation behavior to generate resistant cells. Thus, although ecDNA generally segregates stochastically, it can undergo nonrandom, actively regulated segregation under drug stress, raising the possibility of therapeutically targeting ecDNA segregation mechanisms to counteract adaptive resistance.

Extrachromosomal DNA (ecDNA) carries cancer genes and is thought to segregate randomly, driving heterogeneity within tumors. Here, authors use single-cell tracking to show that some cancer cell clones can actively reconfigure their ecDNA segregation behavior under drug stress to gain resistance.

Introduction

Cancer cells continually reshape their genomes to adapt and survive under stress. Among the most dynamic forms of genome organization is extrachromosomal DNA (ecDNA)—a circular, acentromeric DNA element that frequently carries oncogenes^1,2. A hallmark of ecDNA is its uneven segregation during mitosis, which generates striking heterogeneity in oncogene copy number within tumors^3–7. This heterogeneity has been implicated in drug resistance and rapid clonal evolution^4,8–11.

Several studies have reported that, after drug exposure, resistant populations often show altered ecDNA copy numbers compared with the parental cells, suggesting that ecDNA dynamics contribute to adaptation^4,8–10. However, it is unclear whether these changes only reflect the selection of pre-existing clones with favorable ecDNA states or arise from active reconfiguration of ecDNA content during stress. This question lies at the heart of understanding whether ecDNA segregation is purely stochastic or can be actively modulated by the cell.

Evidence exists for both models. Treatment with doxorubicin decreases ecDNA copy number at the population level, with cells harboring higher copy numbers being more sensitive; this phenomenon is consistent with static selection¹². By contrast, BET inhibitors can disrupt ecDNA tethering to chromatin, leading to mis-segregation of ecDNA during mitosis and global changes in copy number, consistent with active reconfiguration¹³. These seemingly contradictory results imply that ecDNA segregation may not always follow a purely stochastic process but can be contextually altered through specific molecular interactions. Nonetheless, a coherent framework explaining when and how these modes coexist is still lacking. Because selection of pre-existing variants and active reconfiguration are likely to operate together in a context-dependent manner, their interplay complicates the interpretation of ecDNA-mediated adaptation. Furthermore, conventional approaches for tracking ecDNA dynamics have limited throughput and cannot evaluate diverse stress contexts or clonal responses simultaneously^12,14–17, making it difficult to achieve a comprehensive understanding of how ecDNA-mediated adaptation unfolds across heterogeneous populations.

To address these limitations, we establish a high-throughput single-cell system that quantifies ecDNA copy-number distributions across thousands of cells and their derived lineages under defined perturbations. Using this approach, we ask whether cells can actively alter ecDNA segregation behavior and whether such alterations underlie adaptive evolution under drug-induced stress.

Results

Establishment of a high-throughput framework for the quantitative analysis of ecDNA copy-number heterogeneity

We developed a high-throughput single-cell DNA sequencing (scDNA-seq) framework based on the Tapestri platform, optimized for precise quantification of copy-number distributions across user-defined genomic regions. This integrative framework combines droplet-based scDNA-seq, cellular barcoding, and clonal tracking for quantitative, high-resolution monitoring of ecDNA dynamics across ~11,000 individual cells and clonally related populations. We designed a fully customized primer panel that simultaneously amplifies target loci, control regions, and cellular barcode sequences, enabling both copy-number estimation and sample multiplexing. Multiplexing was achieved by combining two distinct cellular barcoding systems with a single nucleotide variant (SNV)-based indexing. To improve the accuracy of copy-number estimation, data from 40 evenly distributed control regions, obtained from both target cells and diploid RPE1 cells, were incorporated into a statistical normalization algorithm (Fig. 1a and Supplementary Figs. 1a–c and 2a–c). Throughout this study, the diploid cell line RPE1 served as the consistent normalization reference. Detailed methods are provided in the Supplementary Note 1.

Fig. 1. Establishment of a high-throughput framework for quantifying extrachromosomal DNA (ecDNA) copy-number distributions.

a Schematic of the framework for the quantitative analysis of ecDNA copy-number distribution using single-cell DNA sequencing (scDNA-seq) based on the Tapestri platform. Droplet-based targeted PCR was used to quantify copy numbers of ecDNA-associated genomic regions and analyze their distributions. Single-nucleotide variations (SNVs) and cellular barcodes facilitated sample multiplexing and independent analysis within a single experiment, enabling a high-throughput workflow. b Histograms showing single-cell copy-number distributions estimated by scDNA-seq across the indicated genomic regions and cell lines. c Representative metaphase FISH images showing the physical configuration of amplified loci in the indicated cell lines. MYC and ERBB2 are shown in red and green, respectively. d Gamma distribution fitting of copy-number distributions of each cell line shown in (b). e Heatmap of eight parameters for the cell lines shown in (b). Each column was normalized by Z-score scaling. f Scatter plot of the mean and standard deviation (SD) values showing simple and robust differences among cell lines.

We used this optimized system to analyze the genomic regions known or predicted to reside on ecDNA across cancer cell lines with focal amplification, with the diploid cell line RPE1 serving as the control. Among the loci examined, MYC displayed broad and heterogeneous single-cell copy-number distributions in COLO320DM, H2170, SNU-16, and COLO320HSR cells, whereas PC-3 cells showed only modest variability (Fig. 1b). Interphase fluorescence in situ hybridization (FISH)-based quantification supported these scDNA-seq measurements, with the mean and standard deviation (SD) of MYC signals showing broad concordance with the corresponding scDNA-seq estimates across cell lines (Supplementary Fig. 2d, e). As an additional check, copy numbers inferred from publicly available scATAC-seq data of COLO320DM and COLO320HSR were also consistent with the scDNA-seq estimates (Supplementary Fig. 2f). These results indicate that our scDNA-seq framework can capture cell-to-cell differences in amplified copy number with reasonable agreement to an independent image-based quantification method. We next examined the physical configuration of MYC amplification by metaphase FISH. COLO320DM, H2170, and SNU-16 cells showed extrachromosomal MYC signals consistent with ecDNA, whereas PC-3 and COLO320HSR cells showed chromosomally integrated amplification without detectable MYC ecDNA (Fig. 1c). Similarly, CDX2 in COLO320DM and ERBB2 in H2170—both of which have been reported to reside on ecDNA—exhibited broad copy-number distributions. Thus, this framework captures single-cell copy-number distributions across samples with distinct amplification structures and provides a quantitative basis for subsequent analysis of ecDNA-associated copy-number dynamics.

Next, we evaluated how best to describe and parameterize their shapes across samples to establish a quantitative basis for comparing ecDNA copy-number distributions. The details of the statistical assessment are provided in Supplementary Note 2 and Supplementary Fig. 2g. Based on these analyses, we defined three practical criteria for subsequent experiments as follows: (i) sampling at least 100 cells is sufficient to capture representative distributions; (ii) fitting single-cell histograms with a gamma distribution facilitates visualization and parameter extraction; and (iii) the combination of mean and SD provides a simple yet robust metric for comparative analyses. The copy-number distributions shown as histograms in Fig. 1b were fitted to gamma distributions, and eight candidate parameters were visualized as heatmaps together with a scatter plot of the mean and SD (Fig. 1d–f). This approach distinguishes cell lines not only by amplification status but also by the extent of heterogeneity within amplified regions, providing a quantitative framework for evaluating ecDNA copy-number variability at single-cell resolution.

Clonal diversity of ecDNA copy-number distributions

To test whether ecDNA segregation follows a purely stochastic model, we examined how copy-number distributions evolve when starting from a single cell. Under purely stochastic segregation across all cells, independently derived clones are expected to exhibit similar binomially shaped distributions, even if their mean copy numbers differ due to differences in initial copy number. Persistent interclonal differences, in contrast, would imply that segregation behavior is influenced by clone-specific mechanisms. To examine this, we combined single-cell cloning with cellular barcoding and scDNA-seq, allowing quantitative profiling of ecDNA copy-number distributions across multiple clones simultaneously (Fig. 2a).

Fig. 2. Clonal diversity of the copy-number distributions.

a Schematic of single-cell cloning with cellular barcode. After PseuMO-Tag lentiviral cellular barcodes were transduced into each cell line, single cells were isolated into 96-well plates. Each clone was cultured until it reached approximately 1–2 × 10⁵ cells and cryopreserved. Once a sufficient number of clones was established, pooled clones were analyzed using scDNA-seq. For COLO320DM, the same clonal pool was additionally analyzed with scRNA-seq, and each clone was cultured individually. After 18 additional cell divisions, scDNA-seq was performed. b Gamma distribution fitting of copy-number distributions of MYC for each clone derived from COLO320DM, H2170, and PC-3 cells. The dataset includes 5861 cells from 12 clones in COLO320DM; 7118 cells from 21 clones in H2170; and 14,796 cells from 25 clones in PC−3 cells. c Heatmap of eight parameters of MYC for each clone shown in (b), sorted in descending order of the mean value. Each column was normalized by Z-score scaling. d Scatter plot of the mean and SD values for MYC copy number showing differences among clones.

We applied this framework to three cancer cell lines—COLO320DM, H2170, and PC-3—representing ecDNA-positive and ecDNA-negative contexts. In the ecDNA-positive lines, loci such as MYC, CDX2, and ERBB2 exhibited broad single-cell copy-number distributions, consistent with the uneven segregation of ecDNA during mitosis (Fig. 2b–d and Supplementary Fig. 3a–c). By contrast, PC-3 cells displayed narrow distributions consistent with chromosome-integrated amplification (Fig. 2b). Notably, within each ecDNA-positive line, independently derived clones exhibited distinct copy-number distributions that deviated from the expectation of a uniform, random pattern. For example, in COLO320DM, some clones (e.g., clone 12) displayed broad distributions with large SD, whereas others (e.g., clone 2) maintained narrow distributions despite similarly high mean copy numbers. Clones 4 and 11, which had lower mean copy numbers, also exhibited consistently narrow distributions (Fig. 2c, d). Similar interclonal differences were also detected for other ecDNA-associated genes—CDX2 in COLO320DM and ERBB2 in H2170—whose copy numbers were correlated across clones (Supplementary Fig. 3d). These observations indicate that, even under identical conditions, ecDNA copy-number distributions do not converge to a single steady-state form across clones, suggesting that not all clones undergo purely random segregation of ecDNA.

Clone-specific MYC copy-number distributions and segregation patterns in COLO320DM

We next focused on 12 independently derived COLO320DM clones for more detailed characterization because this cell line showed the most pronounced interclonal variation in MYC copy-number distributions. Since COLO320DM cells have been reported to undergo chromosomal integration of MYC-amplified ecDNA, resulting in homogeneously staining regions (HSRs), we first examined whether the isolated clones retained MYC amplification as ecDNA or had either acquired or originally harbored chromosomally integrated MYC amplification. Metaphase FISH analysis revealed that clones 4, 6, and 7 predominantly showed HSR-like MYC signals, whereas the remaining clones retained extrachromosomal MYC signals consistent with ecDNA (Fig. 3a, b and Supplementary Fig. 4a). Thus, although HSR formation explained some of the interclonal differences, substantial variation in MYC copy-number distributions was still observed among ecDNA-retaining clones.

Fig. 3. Clonal diversity of the uneven segregation patterns and transcriptional profiles.

a Representative FISH images of COLO320DM clone 2, 4, 11, and 12. b Proportion of ecDNA-positive (DM) and ecDNA-negative (HSR) cells quantified from metaphase FISH images. The number of cells analyzed for each clone is indicated above each bar. c Gamma distribution fitting of MYC copy-number distributions in each COLO320DM clone. Orange curves represent the distributions at the same time point as in Fig. 2b–d. Purple curves represent the distributions after 18 additional cell divisions. d Left, Representative images showing ecDNA segregation to daughter cells, identified by Aurora B midbody staining (red). Right, Histograms showing segregation frequencies for each COLO320DM clone. Black dashed lines indicate simulated random segregation. P values derived from a two-sided Kolmogorov–Smirnov tests comparing observed and simulated distributions are shown. e The relationship between the two key parameters defining the beta-binomial distribution: the expected value of the segregation probability (μ_BB) and the intraclass correlation (ρ). Each point corresponds to the maximum likelihood estimate for a single clone. f Probability density functions of the folded segregation probability, denoted as p ®_BB = min⁡(p_BB,1-p_BB), derived from the beta-binomial model for each of the 12 clones. The curves represent the theoretical probability distributions of segregation to the daughter cell inheriting the minority fraction, bounded within the range of 0<p ®_BB ≤ 0.5. g Density plot of MYC expression levels for each clone. h Correlation between mean MYC copy number and mean MYC expression (left), and between SD of MYC copy number and MYC expression (right). P values were derived from two-sided Pearson correlation tests. i Correlation between mean MYC copy number and mean MYC target gene expression. P values were derived from a two-sided Pearson correlation test. Source data are provided as a Source Data file.

We next asked whether these copy-number distributional differences are transient or stable properties of individual clones. To this end, COLO320DM clones were continuously passaged for approximately 18 additional cell divisions and reanalyzed by scDNA-seq. The overall copy-number profiles of each clone were largely maintained over time, although the degree of dispersion and the relative abundance of cells with high or low ecDNA copy number showed some variation (Fig. 3c and Supplementary Fig. 4b). Interphase FISH performed in parallel supported these scDNA-seq measurements, with clone-specific differences in MYC signal distributions broadly consistent between the two methods (Supplementary Fig. 4c, d). These results indicate that clone-specific ecDNA copy-number states are not merely transient fluctuations immediately after cloning but can persist over multiple cell divisions.

We then investigated whether the observed interclonal differences in ecDNA copy-number distributions could be related to differences in ecDNA segregation behavior during cell division. To directly assess ecDNA inheritance between daughter cells, we combined Aurora Kinase B (AURKB) immunofluorescence, which marks the midbody region between recently divided daughter nuclei, with interphase MYC FISH. This approach enabled us to identify daughter-cell pairs and quantify the relative distribution of MYC ecDNA signals between paired nuclei. The degree of asymmetry in MYC signal partitioning differed among COLO320DM clones (Fig. 3d). Some clones showed relatively balanced distribution of MYC signals between daughter cells, whereas others exhibited more asymmetric partitioning. To further interpret these segregation patterns quantitatively, we performed mathematical modeling and simulation analyses based on the daughter-cell ecDNA count pairs, with details provided in Supplementary Note 3 and Supplementary Data 4. Using a beta-binomial framework, we decomposed the observed segregation patterns into allocation bias and overdispersion. This analysis suggested clone-dependent differences in the underlying segregation mode (Fig. 3e, f and Supplementary Data 5). These findings support the possibility that clone-specific ecDNA copy-number distributions are at least partly associated with differences in ecDNA segregation behavior, rather than reflecting only stochastic sampling of a common parental population.

Finally, we asked how clone-specific MYC copy-number distributions are reflected at the levels of chromatin accessibility and transcription. Taking advantage of our barcode-based framework, we performed scRNA-seq and scATAC-seq on the same pooled COLO320DM clone population analyzed by scDNA-seq, allowing clone-resolved comparison of MYC copy number, MYC expression, MYC-associated chromatin accessibility, and MYC-target gene expression (Figs. 2a and 3g–i and Supplementary Fig. 4e–i). Overall, the molecular outputs were broadly consistent with MYC copy-number states, with lower-copy clones tending to show lower MYC and MYC-target gene expression. However, this relationship was not uniform across clones; for example, clone 12 had high MYC copy number but did not show correspondingly high MYC expression, whereas clone 2 showed relatively high and uniform MYC expression (Fig. 3g–i). scATAC-seq–derived metrics—MYC gene score (Supplementary Fig. 4h) and inferred copy number (Supplementary Fig. 4i)—also captured clone-dependent differences and correlated significantly with the mean MYC copy number from scDNA-seq, but neither correlated with its SD, suggesting that these two modalities capture different layers of clonal variation. These results suggest that MYC amplification states are broadly reflected in chromatin and transcriptional outputs, while also indicating that MYC expression and downstream activity are not determined by DNA copy number alone.

Together, these observations indicate that COLO320DM-derived clones maintain distinct MYC copy-number distribution patterns over multiple generations. These differences were observed not only between ecDNA-type and HSR-type amplification, but also among ecDNA-retaining clones. Based on these results, we selected four representative clones for subsequent drug-response experiments, prioritizing clones whose MYC copy-number distribution patterns were relatively maintained after extended culture. Clone 12 was selected as a high-copy, highly heterogeneous ecDNA clone, clone 2 as a high-copy, moderately heterogeneous ecDNA clone, clone 11 as a low-copy, low-heterogeneity ecDNA clone, and clone 4 as an HSR-like clone. For clarity, we refer to these clones as EC12, EC2, EC11, and HSR4, respectively. This clone set allowed us to examine how distinct MYC amplification states and different degrees of ecDNA copy-number heterogeneity influence cellular responses to drug-induced stress.

Differential clonal responses to drug-induced ecDNA copy-number changes

We next asked whether these distinct ecDNA states translate into differential responses to environmental stress. To this end, we screened a panel of drugs and stress conditions to identify those in which ecDNA copy-number changes coincide with altered drug sensitivity. Among the conditions tested, the BET bromodomain inhibitor JQ1 and paclitaxel (PTX) were associated with both reductions in MYC copy number and the emergence of resistant populations specifically in the ecDNA-positive COLO320DM but not in the chromosomally amplified COLO320HSR (Fig. 4a, b and Supplementary Fig. 5a–c). Interphase FISH analysis showed similar changes in MYC signal counts, supporting the qPCR-based measurements (Fig. 4c). These results suggest that under JQ1 and PTX treatment, changes in ecDNA copy number are associated with drug resistance.

Fig. 4. Clonal diversity in responses to drug-induced changes in ecDNA copy-number.

a MYC copy number of COLO320DM and COLO320HSR cells after 14-day treatment with various stresses was measured by qPCR and normalized to the corresponding DMSO control. Data are mean ± SD (n = 3 biological replicates). P values were derived from a two-sided Welch’s t-test, without adjustment for multiple comparisons. PTX paclitaxel, HU hydroxyurea, DOX doxorubicin, OXA oxaliplatin, Rab rabusertib, Glucose glucose depletion, Glutamine glutamine depletion. b Cell viability assay of COLO320DM and COLO320HSR under the same conditions shown in (a). Data are mean ± SD (n = 3 biological replicates). c MYC copy number was estimated by counting ecDNA signals using interphase FISH after 14-day treatment and normalized to the corresponding DMSO control. The number of cells is indicated above each bar. P values were derived from a two-sided Wilcoxon rank-sum test, without adjustment for multiple comparisons. Data are representative of three independent experiments. Box plots show the median (center line), the interquartile range (box, 25th–75th percentiles), and whiskers extending to the most extreme data points within 1.5 × the interquartile range. Outliers are not shown. d Time-dependent changes in MYC copy number of EC2, EC11, EC12, and HSR4 treated with 500 nM JQ1, measured by qPCR and normalized to the corresponding DMSO control. The lines indicate the mean values (n = 3 biological replicates). e Cell viability assay of JQ1-treated EC2, EC11, EC12, and HSR4. Data are mean ± SD (n = 3 biological replicates). f Gamma distribution fitting of copy-number distributions of MYC in EC2, EC11, EC12, and HSR4 treated with DMSO or JQ1 treatment. g Scatter plot of the mean and SD values for MYC copy number showing differences among conditions. Source data are provided as a Source Data file.

To determine whether the effects of JQ1 and PTX differ among ecDNA-positive clones, we monitored the time-dependent changes in MYC copy number and cell viability in the four representative COLO320DM clones (EC2, EC11, EC12, and HSR4). JQ1 and PTX treatment gradually reduced MYC copy number in EC2, EC11, and EC12, but not in HSR4 (Fig. 4d and Supplementary Fig. 5d). Cell viability assays revealed that EC2, which showed the greatest reduction in MYC copy number, was the most resistant to JQ1. EC12, which exhibited a high and variable ecDNA copy number, was more sensitive than EC2. HSR4 displayed an intermediate level of sensitivity (Fig. 4e). These findings suggest that although JQ1 treatment reduces ecDNA copy number at the population level, the sensitivity to JQ1 cannot be explained solely by differences in copy-number magnitude and heterogeneity. In contrast, under PTX treatment, HSR4 was the most resistant (Supplementary Fig. 5e). Taken together, these results indicate that the JQ1 resistance of EC2 may be linked to alterations in ecDNA copy number.

To validate these observations at the single-cell level, we performed scDNA-seq on representative clones treated with JQ1. In EC2, the copy-number distributions of MYC and CDX2—both ecDNA-associated loci—shifted leftward, showing marked decreases in both the mean and SD. By contrast, non-ecDNA control regions, including ERBB2 and chr8q23.3 (a chromosomal region adjacent to MYC but not on the ecDNA), remained largely unchanged. In EC11, the mean copy numbers of both MYC and CDX2 decreased modestly, whereas their SDs were largely maintained. In EC12, MYC showed only a minor reduction, while CDX2 showed decreases in both mean and SD. In HSR4, MYC and CDX2 showed only minimal changes (Fig. 4f, g and Supplementary Fig. 5f, g). These clone-dependent patterns indicate that JQ1-associated copy-number reduction is most prominent in EC2 and is preferentially observed at ecDNA-associated loci. Consistent results were obtained in an independent experiment with shorter JQ1 exposure (7 days; Supplementary Fig. 6a–c), and FISH analysis after prolonged (56 days) treatment confirmed a reduction in ecDNA signals in EC2 (Supplementary Fig. 6d). Together, these data confirm that JQ1 reduces ecDNA copy number at the single-cell level in a clone-specific manner.

Clonal tracking reveals distinct patterns of ecDNA copy-number reduction across clones

Two models could explain the observed decrease in ecDNA copy number. The first is the static selection model, in which subclones with inherently low copy numbers preferentially survive JQ1 treatment. The second is the active reconfiguration model, in which subclones capable of generating low-copy cells, regardless of their original copy number, are the ones that survive (Fig. 5a). To distinguish between these models, we performed clonal tracking experiments. A lentiviral barcode library was introduced into each of the four representative COLO320DM clones, which were then evenly divided into two groups. One culture was analyzed by scDNA-seq to record baseline distributions, whereas the other culture was expanded, divided evenly, and exposed to various stresses for four weeks. Surviving clones under each condition were identified by barcode sequencing and linked to their corresponding scDNA-seq profiles, facilitating comparison of their pretreatment ecDNA states (Fig. 5b). If the static selection model is correct, JQ1-surviving clones should be those that already had low ecDNA copy numbers before treatment. In contrast, the active reconfiguration model predicts that surviving clones could originate from cells with either low or high pretreatment copy numbers. Thus, if a JQ1-surviving clone originally had a low copy number, either model could explain its behavior; however, if the clone originally had a high copy number, this would indicate that active reconfiguration had occurred (Fig. 5c).

Fig. 5. Clonal diversity in drug-resistant cell properties revealed by clonal tracking.

a Schematic of two hypothetical models for ecDNA behavior underlying survival under JQ1 treatment. b Schematic of the experimental design for high-throughput clonal tracking using scDNA-seq and amplicon-seq under multiple stresses. c Schematic of predicted experimental outcomes under the two hypotheses. d Gamma distribution fitting of MYC copy-number distributions in each clone under drug-free conditions. The DMSO-treated group is shown in blue and the JQ1-surviving subclones in red. e Scatter plot of the mean and SD values for MYC in each clone. The DMSO-treated group is shown in blue and the JQ1-surviving subclones in red.

Data from qPCR and scDNA-seq analyses showed that EC2 exhibited a marked reduction in copy number after JQ1 treatment. Interestingly, JQ1-surviving subclones derived from EC2 exhibited higher and more heterogeneous MYC and CDX2 copy-number distributions at baseline (i.e., before JQ1 treatment) than subclones surviving under DMSO. These findings suggest that JQ1 resistance in EC2 was not due to static selection of pre-existing low–copy-number cells, but rather involved active reconfiguration of ecDNA segregation. In contrast, PTX treatment also led to a reduction in copy number by qPCR, yet PTX did not show evidence for the type of active reconfiguration observed with JQ1 in our clonal tracking analysis. HSR4, which did not exhibit a copy-number change after JQ1 treatment, showed no association between MYC copy number and survival. EC11 and EC12, both of which showed JQ1-induced copy-number reduction by qPCR, displayed patterns of surviving subclones distinct from EC2. In EC11, JQ1-surviving subclones tended to have lower baseline copy numbers, opposite to the trend in EC2. In EC12, the baseline copy-number distributions of surviving subclones were nearly identical across all stress conditions, indicating that ecDNA copy number was largely unrelated to survival in this clone (Fig. 5d, e and Supplementary Fig. 7a, b).

Taken together, these findings indicate that the mechanisms underlying drug resistance vary among clones and drug treatments. In particular, at least in EC2, the reduction in ecDNA copy number upon JQ1 treatment is unlikely to result merely from static selection, but rather represents an active reconfiguration of ecDNA content.

JQ1 alters ecDNA copy-number dynamics and segregation in EC2

To directly test whether JQ1-induced changes in ecDNA copy number reflected active reconfiguration rather than static selection, we conducted another clonal tracking experiment. A lentiviral barcode library was introduced into EC2, which was then divided into two populations and treated with DMSO or JQ1 for 10 and 24 days, respectively, followed by scDNA-seq (Fig. 6a). If JQ1-surviving subclones also exhibit high ecDNA copy numbers under DMSO conditions, this suggests that active reconfiguration has occurred. The DMSO condition yielded 326 detectable subclones, whereas 119 subclones were recovered after JQ1 treatment (Supplementary Fig. 8a). Consistent with the previous results, MYC and CDX2 copy-number distributions shifted leftward after JQ1 exposure, whereas no changes were observed for the ERBB2 and chr8q23.3 regions (Supplementary Fig. 8b).

Fig. 6. Active reconfiguration of ecDNA under JQ1 treatment revealed by clonal tracking.

a Schematic of the experimental design for clonal tracking using single-cell DNA sequencing (scDNA-seq) analysis in EC2 and EC12. One group was treated with DMSO and the other with JQ1. b Gamma distribution fitting of MYC copy-number distributions under DMSO (left) and JQ1 (right) conditions in EC2 (top) and EC12 (bottom). Copy-number distributions of clones detected only in the DMSO-treated group are shown in green, those detected only in the JQ1-treated group are shown in blue, and detected in both groups are shown in orange. c Scatter plot of the mean and SD values for MYC copy number in the DMSO-treated group in EC2 (top) and EC12 (bottom). Clones detected only in the DMSO-treated group are shown in green, and those detected in both groups are shown in orange. d Comparison of the mean MYC copy number between the DMSO and JQ1 culture conditions in clones detected in both the DMSO- and JQ1-treated groups. e Histograms showing segregation frequencies in EC2 and EC12 after treatment with DMSO, JQ1 for 24 h, or JQ1 for 48 h. Black dashed lines indicate simulated random segregation. f Violin plots showing the distribution of the smaller fraction of ecDNA, defined as the proportion of ecDNA in the daughter cell inheriting fewer ecDNA copies, across four clones. P values were derived from a two-sided Wilcoxon rank-sum test, without adjustment for multiple comparisons. Data are representative of three independent experiments. Box plots show the median (center line), the interquartile range (box, 25th–75th percentiles), and whiskers extending to the most extreme data points within 1.5 × the interquartile range. Source data are provided as a Source Data file.

In the DMSO culture, approximately ten dominant subclones accounted for most of the population, indicating that clonal expansion was already non-uniform even under drug-free conditions. Notably, the 34 subclones that survived JQ1 were not among these dominant subclones, implying that survival was not driven by pre-existing proliferative strength (Supplementary Fig. 8c). Interestingly, under DMSO conditions, these JQ1-surviving subclones displayed higher and more heterogeneous MYC copy-number distributions, with greater mean and SD, compared with those that did not survive JQ1 treatment (Fig. 6b, c and Supplementary Fig. 9a, b). Paired analyses of the same subclones under DMSO and JQ1 conditions showed that MYC and CDX2 copy numbers decreased following treatment (Fig. 6d and Supplementary Fig. 9c), excluding the possibility that survival arose from pre-existing low-copy states. Together, these results provide direct evidence that JQ1 response in EC2 involves active reconfiguration of ecDNA content rather than static selection.

We performed the same clonal tracking analysis in EC12. In contrast to EC2, subclones that survived both DMSO and JQ1 conditions tended to exhibit slightly lower MYC and CDX2 copy numbers than those observed under DMSO alone, and paired comparisons showed minimal changes in copy number upon JQ1 treatment (Fig. 6b–d and Supplementary Fig. 9a–c). In addition, JQ1-surviving subclones in EC12 were also dominant under DMSO conditions (Supplementary Fig. 8c), suggesting that survival was associated with pre-existing fitness rather than dynamic changes in ecDNA content. These findings indicate that the mechanisms underlying JQ1 response differ between clones, with EC12 showing patterns more consistent with static selection than active reconfiguration.

To further examine whether JQ1 alters ecDNA segregation behavior, we performed combined AURKB immunofluorescence and FISH analysis across four representative clones. In EC2, JQ1 treatment was associated with a shift toward more unequal segregation patterns, whereas no apparent changes were observed in the other clones (Fig. 6e, f and Supplementary Fig. 9d). These observations raise the possibility that enhanced asymmetry in ecDNA segregation may contribute to the adaptive response to JQ1 in EC2.

Together, these results indicate that distinct clones can adopt different survival strategies under JQ1, with EC2 showing direct evidence of active ecDNA reconfiguration accompanied by altered segregation behavior.

Transcriptional profiling identifies EN2 as a candidate regulator of ecDNA reconfiguration in EC2

To investigate the molecular basis underlying the distinct behavior of EC2, we performed bulk RNA-seq on COLO320DM clones (EC2, EC11, EC12, and HSR4), their parental line, and COLO320HSR. Because transcriptional states associated with ecDNA are often linked to cell cycle–related genes, RNA was extracted under various culture and harvesting conditions to minimize potential confounding effects. Principal component analysis revealed that the first three principal components (PC1, PC2, and PC3) captured the major sources of biological variability in the dataset (Supplementary Fig. 10a). PC1 reflected transcriptional differences between DM clones and COLO320HSR; PC2 captured variability between COLO320HSR and HSR4; and PC3 represented variability between COLO320HSR/HSR4 and the other clones (Supplementary Fig. 10b–d). These findings indicate that EC2, EC11, and EC12 share similar transcriptional profiles, and that comparing EC2 with EC11 and EC12 may help identify gene expression programs that are specifically enriched in EC2. Differential expression analysis revealed 260 genes that were highly upregulated in EC2 (Fig. 7a and Supplementary Data 6). Notably, these activated genes were enriched for functions related to axon development and neuronal signal transduction (Supplementary Fig. 10e), suggesting that EC2 may exhibit features of a neuron-like transcriptional state. Such transcriptional features may be associated with differences in microtubule dynamics and intracellular transport processes, which could in turn influence chromosome-independent DNA partitioning during mitosis.

Fig. 7. Potential involvement of transcription factor EN2 in active reconfiguration in EC2.

a Volcano plot showing the differential gene expression between EC2, EC11, and EC12. Each dot represents a gene. Red dots indicate genes significantly upregulated in EC2. b, c Cell viability assays of each clone after 7-day treatment with indicated drugs. Data are mean ± SD (n = 3 biological replicates). P values were derived from a two-sided Welch’s t-test. d Growth curves showing JQ1 sensitivity after knockdown of genes upregulated in EC2. e Gamma distribution fitting of MYC copy-number distributions after EN2 knockdown under JQ1 treatment in EC2. f Scatter plot of the mean and SD values for MYC copy number after EN2 knockdown under JQ1 treatment in EC2. g Histograms showing segregation frequencies in EC2. h Violin plots showing the distribution of the smaller fraction of ecDNA. P values were derived from a two-sided Wilcoxon rank-sum test, without adjustment for multiple comparisons. Data are representative of three independent experiments. Box plots show the median (center line), the interquartile range (box, 25th–75th percentiles), and whiskers extending to the most extreme data points within 1.5 × the interquartile range. i The relationship between the expected value of the segregation probability (μ_BB) and the intraclass correlation (ρ) estimated by the beta-binomial model. Points represent the parameter estimates for each of the four experimental conditions. j Probability density functions of the folded segregation probability (p ®_BB) estimated using the beta-binomial model for the four distinct experimental conditions. The curves represent the theoretical probability distributions of segregation to the daughter cell inheriting the minority fraction, bounded within the range of 0 < p ®_BB ≤ 0.5. k Schematic of cellular behavior under JQ1 treatment in EC2 and EC12. Source data are provided as a Source Data file.

To explore this possibility, we assessed the effect of perturbing microtubule dynamics on JQ1 response. While low-dose nocodazole alone did not affect cell proliferation, its combination with JQ1 resulted in a marked increase in sensitivity in EC2 and EC11, but not in EC12 and HSR4 (Fig. 7b, c). These findings suggest that microtubule-dependent processes may contribute to the distinct response of EC2 to BET inhibition.

To identify specific molecular factors that may underlie this clone-specific behavior, we then turned to the candidate genes upregulated in EC2. Among these, we focused on Engrailed Homeobox 2 (EN2) as a potential regulator of its unique phenotype. siRNA-mediated knockdown of EN2 partially restored sensitivity to JQ1 in EC2 (Fig. 7d and Supplementary Fig. 11a). scDNA-seq analysis further showed that EN2 knockdown attenuated the JQ1-induced reduction in ecDNA copy number in EC2 (Fig. 7e, f and Supplementary Fig. 11b, c). Moreover, combined AURKB immunofluorescence and FISH analysis showed that EN2 knockdown abolished the JQ1-associated shift toward more unequal ecDNA segregation observed in EC2 (Fig. 7g, h). Mathematical modeling and simulation analyses of the same daughter-cell ecDNA count data further suggested that JQ1 alters the inferred segregation mode in EC2, including the allocation probability to each daughter cell and that EN2 knockdown attenuates this JQ1-associated change (Fig. 7i, j and Supplementary Data 8). This effect was not observed in the other three clones (Supplementary Fig. 11d) and was reproduced with an independent siRNA targeting EN2 (Supplementary Fig. 11e, f). These findings suggest that EN2 may contribute, at least in part, to the active reconfiguration of ecDNA in response to JQ1.

Together, these findings suggest that clone-specific transcriptional programs may shape distinct modes of ecDNA regulation under stress conditions, and identify EN2 as a candidate factor linking transcriptional state to ecDNA reconfiguration dynamics (Fig. 7k).

Discussion

In this study, we established a high-throughput and quantitative framework for measuring ecDNA copy-number distributions and their dynamics using scDNA-seq on the Tapestri platform. By integrating cellular barcoding, clone assignment, and clonal tracking, this approach enabled us to follow ecDNA copy-number states across multiple clones and stress conditions at single-cell resolution. Using this framework, we found that COLO320DM-derived clones maintained distinct MYC copy-number distributions and segregation patterns, even under drug-free conditions. Under JQ1 treatment, EC2 showed a reduction in ecDNA copy number that could not be explained simply by static selection of pre-existing low–copy-number cells. Instead, clonal tracking and segregation analyses suggested that EC2 undergoes active reconfiguration of ecDNA content and segregation behavior in response to BET inhibition. These findings indicate that ecDNA copy number is not only a static genomic feature but can also behave as a dynamic and context-dependent property during drug adaptation.

A major advantage of our approach is that it combines quantitative copy-number measurement with clone-resolved tracking. FISH remains a gold-standard method for detecting ecDNA and defining its physical configuration, and in this study, metaphase and interphase FISH were essential for validating both ecDNA structure and copy-number distributions. However, image-based approaches alone are difficult to scale across many clones, loci, and perturbation conditions, and they are not readily linked to lineage tracking. In contrast, our sequencing-based framework allows sample multiplexing and barcode-based clone assignment, making it possible to connect the pretreatment copy-number state of individual subclones with their subsequent survival under drug stress. Thus, the strength of this framework is not that it replaces FISH, but that it complements orthogonal FISH validation with high-throughput clonal tracking.

Our single-cell cloning experiments revealed that ecDNA maintenance and segregation behavior can differ stably among clonal lineages derived from a single parental population. Beyond differences in physical configuration (ecDNA-type vs HSR-like MYC amplification), we observed substantial diversity among ecDNA-retaining clones in their copy-number distributions and inferred segregation modes—diversity that persisted over multiple generations and that was supported by independent FISH-based and mathematical-modeling analyses. A central question raised by these observations is how a single parental population can sustain such distinct ecDNA-related phenotypes across clonal lineages. One possibility is that clone-specific transcriptional or chromatin states, themselves heritable through cell division, bias ecDNA segregation by tuning factors involved in ecDNA tethering, hub formation, or mitotic partitioning¹³. Under this view, what is often described as the stochastic nature of ecDNA inheritance can be modulated by inherited cellular contexts, providing a non-genetic layer of clonal variability that complements the random sampling component of ecDNA segregation. This framework predicts that the rules governing ecDNA inheritance may not be uniform across clones and may instead be tuned by the cellular state in which each clone resides.

An important implication of these findings is that drug-induced reductions in ecDNA copy number can arise through different mechanisms depending on the clone and the stress condition. In EC2, JQ1-surviving subclones did not originate from low-copy states; rather, they had relatively high and heterogeneous ecDNA copy-number distributions before treatment and then reduced their ecDNA copy number under JQ1. Together with the AURKB–FISH segregation analysis, the mathematical modeling of daughter-cell partitioning, the nocodazole sensitization, and the EN2 knockdown experiments, these findings provide multiple lines of evidence that JQ1 is associated with active reconfiguration of ecDNA segregation in EC2. This effect of JQ1 may relate to its direct biological actions: BET inhibition can perturb BRD4-dependent transcriptional activity, enhancer function, chromatin association, and mitotic inheritance of ecDNA¹³, all of which could directly affect ecDNA maintenance and segregation. By contrast, our clonal tracking analysis under PTX did not provide evidence for the same type of active reconfiguration, although this comparison is based on a more limited PTX-specific dataset. Because ecDNA lacks centromeres and is not subject to canonical spindle attachment, PTX—which stabilizes microtubules—may influence ecDNA dynamics indirectly rather than through direct perturbation of ecDNA-associated chromatin or segregation. Several questions remain open for future investigation, including the mechanistic basis of the PTX effect on ecDNA dynamics (which will require dedicated PTX-specific clonal tracking experiments), why specific ecDNA-associated loci are more susceptible to reconfiguration than others, and whether the active reconfiguration phenotype generalizes to ecDNA carrying other oncogenes or to other BET- or chromatin-associated mechanisms.

Our results also highlight that DNA copy number alone does not fully determine the molecular output of ecDNA-amplified loci. Across COLO320DM clones, MYC copy-number states were broadly reflected in MYC expression, MYC-target gene expression, and chromatin accessibility, but this relationship was not uniform. For example, EC12 showed high MYC copy number but did not exhibit proportionally high MYC expression, whereas EC2 showed relatively high and uniform MYC expression. In line with this uncoupling, scATAC-seq–derived metrics correlated with the mean MYC copy number across clones but did not recapitulate the within-clone variability captured by scDNA-seq. This may reflect that scATAC-seq integrates not only copy-number information but also enhancer accessibility and chromatin context, which can vary independently of DNA copy number across cells within a clone. Such uncoupling may also reflect the involvement of additional regulatory layers, including ecDNA-associated enhancer activity, chromatin accessibility, ecDNA hub formation, nuclear localization, and clone-specific trans-regulatory states^5,18–20. Therefore, ecDNA amplification should not be interpreted simply as a linear dosage effect. Rather, copy number, chromatin state, and transcriptional context together determine the functional output of ecDNA-containing loci.

The transcriptional features of EC2 may provide one possible link between cellular state and ecDNA reconfiguration. EC2 was enriched for genes associated with axon development and neuronal signal transduction—programs that involve cytoskeletal organization, intracellular transport, and microtubule-related processes, each of which could plausibly influence the behavior of acentric DNA elements during mitosis. Consistent with this, low-dose nocodazole sensitized EC2 specifically to JQ1, and EN2 knockdown attenuated both the JQ1-induced ecDNA copy-number reduction and the associated shift toward unequal segregation, with mathematical modeling indicating that this change in inferred segregation mode is itself counteracted by EN2 knockdown. Whether the cytoskeletal or microtubule-related features inferred from the transcriptional program of EC2 directly contribute to ecDNA partitioning, and whether comparable transcriptional contexts arise in other ecDNA-positive systems, will require systematic investigation across diverse ecDNA-bearing models. Together, these findings nominate EN2 as a candidate factor linking clone-specific transcriptional state to ecDNA reconfiguration dynamics, although the molecular relationship between EN2, microtubule-related processes, and ecDNA segregation remains to be defined.

Several limitations should be noted. First, the strongest evidence for active ecDNA reconfiguration in this study comes from EC2 under JQ1 treatment. Other clones and stress conditions showed different patterns, including behavior more consistent with static selection. Therefore, active reconfiguration should not yet be considered a universal property of ecDNA-containing cells. Rather, our data suggest that ecDNA-mediated adaptation can occur through multiple modes, including static selection and active reconfiguration, depending on the cellular context and stress condition. Second, although EN2 knockdown provides functional evidence implicating a candidate regulator, we have not fully dissected the molecular mechanism by which EN2 or the EC2 transcriptional state influences ecDNA segregation. Third, this study was performed using cancer cell lines, and extension to patient-derived models and clinical specimens will be essential to determine the general relevance of these findings. Future studies combining live-cell imaging, perturbation screens, direct analysis of ecDNA clustering or chromatin tethering, and patient-derived ecDNA-positive models will be needed to define the broader principles governing ecDNA copy-number adaptation.

Overall, our study provides a framework for linking ecDNA copy-number dynamics to clonal adaptation under stress. By combining scDNA-seq, cellular barcoding, FISH validation, clonal tracking, and mathematical modeling, we show that ecDNA copy number can behave as a dynamic and potentially regulatable property rather than a fixed genomic state. These findings expand the view of ecDNA from a static oncogene amplification unit to a flexible layer of genome plasticity that can contribute to drug adaptation. Understanding when cancer cells use static selection versus active ecDNA reconfiguration may help identify new strategies to interfere with ecDNA-mediated therapeutic resistance.

Methods

Cell culture

COLO320DM (JCRB0225) and PC-3 (JCRB9110) cells were purchased from Japanese Collection of Research Bioresources (Osaka, Japan). RPE1 (CRL-4000), H2170 (CRL-5928), SNU-16 (CRL-5974), and COLO320HSR (CCL-220.1) cells were purchased from the American Type Culture Collection (Manassas, VA). COLO320DM, COLO320HSR, and RPE1 cells were cultured in Dulbecco’s Modified Eagle’s Medium (DMEM) low glucose (Nacalai, Kyoto, Japan) supplemented with 10% FBS (Thermo Fisher Scientific, Waltham, MA). H2170 and SNU-16 cells were cultured in Roswell Park Memorial Institute (Nacalai) 1640 Medium supplemented with 10% FBS. PC-3 cells were cultured in Kaighn’s modification (F12K) of Ham’s F12 medium (Wako, Osaka, Japan) supplemented with 10% FBS. All cells were incubated at 37 °C, 20% O₂ and 5% CO₂.

Custom primer preparation for Single-cell DNA sequencing using the Tapestri Platform

Single-cell DNA analysis was conducted using the Tapestri Platform from Mission Bio (San Francisco, CA), which performs targeted PCR at the single-cell level. Custom primers were designed by Tapestri Designer (Mission Bio). We optimized the primers so that the SNV region or cellular barcode locus required for cell-line or barcode identification was located within 75 bp from the reverse primer. We purchased oligonucleotides from Thermo Fisher Scientific, in which a bead-annealing constant sequence was appended to the 5′ end of the designed forward primer and the Nextera Read 2 sequence was appended to the 5′ end of the designed reverse primer. The constant sequence and Nextera Read 2 sequence appended were GTACTCGCAGTAGTC and GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAG, respectively. For each run, the specific primers were mixed and diluted to 30 µM (forward primer mix) and 800 µM (reverse primer mix). To determine whether custom-designed primers could amplify target regions with efficiency comparable to that of commercially available primer panels, custom primers starting 1 bp upstream of the corresponding panel primers were mixed and evaluated in the same reaction. All custom primer sequences, excluding the common sequences added later, are listed in Supplementary Data 1 and 2. Supplementary Data 1 contains the primers used for the experiments, whereas Supplementary Data 2 includes the primers used for the validation of primer efficiency as described in Supplementary Note 1.

Single-cell DNA sequencing by Tapestri Platform

Single-cell DNA analysis was done using the Tapestri Platform with version 3 software and the reagents according to the version 3 user guide. Frozen cell stocks were thawed, washed once with phosphate-buffered saline (PBS), and passed through a 40-µm cell strainer. The cells were counted, mixed at various ratios of each analysis, and resuspended in Cell buffer at 3000 cells/µL. They were loaded onto Tapestri Platform, followed by encapsulation, lysis, barcoding, and targeted PCR according to the manufacturer’s instructions. Although the primers for the cancer-specific panels from Mission Bio were used as described in the user’s guide, we primarily used the custom primers described above. Following targeted PCR, the emulsion was disrupted, the reaction was cleaned enzymatically, and the product was purified with AMPure XP beads (Beckman Coulter, Brea, CA). To add indexes, a second PCR of 10 cycles was performed, followed by AMPure XP beads (Beckman Coulter), and eluted in 15 µL. Library quality was assessed by Quantus (Promega, Madison, WI) and TapeStation 4200 (Agilent Technologies, Santa Clara, CA) with D5000 Screen Tape. Next-generation sequencing (NGS) was performed in 2 × 75 bp paired-ends using NextSeq550 platform (Illumina, San Diego, CA, USA), which generates 50–60 million reads per library.

Amplicon sequencing from a single-cell DNA sequencing library for cell barcode enrichment

NGS of the Tapestri library yielded few reads, including the cellular barcode locus, which rendered barcode assignment to individual cells difficult. This likely occurs because PCR efficiency at this locus was low and barcode-containing amplicons were under-represented. Therefore, a targeted enrichment PCR of barcode-containing amplicons was performed to increase representation and enable barcode assignment. Two cellular barcode systems, PseuMO-Tag and ID vector, were used. For PseuMO-Tag, the barcode-containing amplicon was shorter compared with the other amplicons, which resulted in failure to bind to AMPure XP beads (Beckman Coulter). Therefore, after targeted PCR and cleanup, the supernatant obtained from the first AMPure XP beads (Beckman Coulter) purification was used as a template for nested PCR. For the first PCR, 1 µL of the one-third–diluted supernatant was used as the template in a total reaction volume of 25 µL, using the NEBNext High-Fidelity 2× PCR Master Mix (New England Biolabs, Ipswich, MA). The PCR conditions were as follows: 30 s denature step at 98 °C, followed by seven cycles of 98 °C for 10 s, 63 °C for 30 s, and 72 °C for 30 s, and finally 72 °C for 2 min. The primers used for the first PCR were as follows: Fw: 5′- CGTCGGCAGCGTCAGATG -3′, and Rv: 5′- GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGTCCGCTCGCTAGTTATTGCTCAACG -3′. To maintain barcode diversity, the reaction was initially run for seven cycles, then temporarily removed from the thermal cycler and placed on ice. The appropriate number of additional PCR cycles was determined by qPCR. The qPCR reaction mixture (total 15 µL) contained 2 µL of the first PCR reaction as the template, 0.75 µL each of 10 µM forward and reverse primers, 0.24 µL of SYBR Green, 0.3 µL of ROX reference dye, 3.46 µL of nuclease-free water, and 7.5 µL of NEBNext High-Fidelity 2× PCR Master Mix. The amplification program was identical to that of the first PCR. The number of cycles corresponding to one-third to one-fourth of the plateau phase of the qPCR amplification curve was then added to the original reaction (kept on ice) to complete the first PCR. The PCR products were purified using 1.8× AMPure beads (Beckman Coulter), and eluted in 17 µL. For the second PCR, 1 µL of the first PCR product template in total amount of 25 µL reaction volume was used with NEBNext High-Fidelity 2× PCR Master Mix (New England Biolabs). The PCR conditions were as follows: 30 s denature step at 98 °C, followed by 5 cycles of 98 °C for 10 s, 61 °C for 30 s, and 72 °C for 30 s, and finally 72 °C for 2 min. The primers used in the second PCR were as follows: Fw: 5′- AATGATACGGCGACCACCGAGATCTACACXXXXXXXXTCGTCGGCAGCGTCAGATG -3′, and Rv: 5′- CAAGCAGAAGACGGCATACGAGATXXXXXXXXGTCTCGTGGGCTCGGAGA -3′. X indicates the indices. The PCR products were purified using 0.8× → 1.3× SPRIselect (Beckman Coulter), and eluted in 17 µL. For the ID vector, PCR was carried out under the same conditions. The primers used for the first PCR were as follows: Fw: 5′- CGTCGGCAGCGTCAGATG -3′, and Rv: 5′- GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGCGAGGCAGGAAACAGTGACTAG -3′. After the initial seven cycles, the reaction was placed on ice, and qPCR was done to determine the appropriate number of additional cycles, based on the same procedure used for the PseuMO-Tag amplicon. After purification with 1.8× AMPure beads (Beckman Coulter), the second PCR was performed using the same primers and PCR conditions were similar to those used for the PseuMO-Tag amplicon. The PCR products were purified using 0.7× SPRIselect (Beckman Coulter), and eluted in 17 µL. Library quality was assessed by Quantus (Promega, Madison, WI) and TapeStation 4200 (Agilent Technologies) with D5000 Screen Tape. The expected amplicon size was 256 bp for the PseuMO-Tag and 374–382 bp for the ID vector. NGS was performed in 2 × 75 bp paired-ends using NextSeq550 platform (Illumina), which generates 5 million reads per library.

Single-cell DNA sequencing data processing pipeline and normalization

A custom computational pipeline was established to construct copy-number count matrices and to annotate cell identities according to SNV profiles (see “Code availability”). For each scDNA-seq dataset, paired-end FASTQ files were processed by extracting sequence reads and concatenating Read1 and Read2 using a tab delimiter, followed by sorting before inputting into the pipeline. Read1 consisted of the Tapestri barcode (TapeBC) and forward primer sequence, whereas Read2 contained the reverse primer and downstream genomic sequence. Based on the primer, Read2 sequences corresponded to copy number variant quantification, SNV detection, reference region, or cellular barcodes (PseuMO-Tag and ID vector). Within the pipeline, each read was scanned to extract the TapeBC. The reads sharing identical barcodes were aggregated, and per-target read counts were calculated to generate the copy-number matrix. The reads containing SNV information were then used to separately count wild-type and mutant alleles, thus producing an SNV count matrix from which allele frequencies (AF) were determined. Based on the AF thresholds, the cells were classified as wild-type (AF > 0.8), heterozygous (0.2 ≤ AF ≤ 0.8), or mutant/LOH (AF > 0.8 for mutant reads). These SNV patterns were compared with known mutation profiles for each cell line to annotate the corresponding cell identity, which was linked to the copy-number count matrix. Raw read counts were normalized based on reference regions with low variability (top 20 regions with the smallest coefficient of variation in RPE1 cells). For each cell, the mean count across the reference regions was used for depth normalization. Correction factors were then calculated for each target region by trimming extreme values and aligning the modal copy number of the RPE1 cells to two. Normalized and adjusted copy numbers were then calculated for all cells and exported for downstream analysis. RPE1 was used as the diploid normalization reference across all scDNA-seq experiments. Other cell lines, such as COLO320HSR and PC-3, were included as biological comparison samples representing distinct amplification states, and were not used as diploid normalization controls.

Statistical analysis and visualization of copy number

Density-scaled histograms were generated to visualize the distribution of copy numbers for each cell line. The cell lines harboring ecDNA frequently showed right-skewed distributions with extended tails. To characterize these distributions, the copy-number distributions were also visualized using the fitted gamma curves (see Supplementary Note 2). As quantitative indicators of the distribution shape, the mean, median, and mode (representing copy-number level), SD (representing dispersion), skewness (asymmetry), kurtosis (peakedness), and gamma distribution parameters, shape and rate were calculated. These parameters were visualized using heatmaps and scatter plots.

Single-cell DNA sequencing –PseuMO-Tag assignment

The clone-specific barcode sequences obtained from cDNA amplicon sequencing (amplicon-seq) were initially compiled as a whitelist (see “Preparation of amplicon-seq library from cDNA” Section). Based on the amplicon-seq FASTQ files enriched from the scDNA-seq library, the TapeBC and PseuMO-Tag sequence regions were extracted using cutadapt (Hamming distance ≤ 1). For each TapeBC, the read counts were aggregated across the clones, and the proportion of reads assigned to each clone was determined. The cells in which > 80% of the reads mapped to a single clone were designated as clone-assigned. The clone labels were merged with the normalized copy-number matrix for downstream analyses. Statistical analyses and data visualizations were similarly conducted on a per-clone basis, following the same procedures described above.

Single-cell DNA sequencing—ID assignment

ID-specific barcode sequences obtained from Sanger sequencing were first compiled as a whitelist (see “Sanger sequencing” section). The subsequent procedures for barcode extraction, read-count aggregation, clone assignment, and integration with the copy-number matrix were conducted in the same manner as described above.

Single-cell RNA sequencing analysis

Single-cell RNA sequencing and clone assignment were performed as previously reported²¹. To generate count matrix, we used 10× Genomics Cell Ranger v8.0 for 10× Chromium scRNA-seq. Downstream analysis was performed using R package Seurat (v5.1.0). The cells with over 25% mitochondrial RNA and fewer than 400 or more than 6000 features were filtered out. The gene expression counts were normalized by total cellular counts and scaled to 10,000 transcripts per cell. The normalized data were then log-transformed using Seurat’s NormalizeData() function, followed by scaling through the ScaleData() function. Principal component analysis (PCA) was performed with RunPCA() after identifying the top 2000 highly variable genes using FindVariableFeatures() with the vst method. Cell clustering was subsequently conducted using Seurat’s standard workflow, applying FindNeighbors() and FindClusters(). For visualization, Uniform Manifold Approximation and Projection (UMAP) was generated using the RunUMAP() function. After clone assignment, we calculated module scores for “MYC Targets V2” (MSigDB Hallmark 2020) using the addModuleScore function.

Single-cell ATAC sequencing analysis

Single-cell ATAC-seq experiments on our COLO320DM-derived clones and clone assignment were performed as previously reported²¹. All downstream analyses were performed using the ArchR package (v1.0.1). To compare scDNA-seq–based copy-number estimates with sequencing-based copy-number inference at the cell-line level, we downloaded publicly available scATAC-seq fragment files for COLO320DM and COLO320HSR cells (GSE160148). Amplicon copy numbers were inferred from background ATAC-seq signals as previously described²². Briefly, read counts in 3-Mb windows (1-Mb sliding steps) were calculated after filtering out ENCODE hg19 blacklist regions. Insertion rates per base pair were compared with those of 100 GC-content–matched neighboring intervals to compute the mean log₂ fold-change (FC). Copy numbers were estimated under a diploid model as CN = 2 × 2^FC, and the copy number for a given gene was defined as the mean across all overlapping genomic intervals. For clone-resolved comparison with scDNA-seq, the same copy-number inference procedure was applied to our scATAC-seq data from individual COLO320DM-derived clones to obtain clone-level MYC copy-number distributions. In addition, MYC gene scores were extracted using the ArchR GeneScoreMatrix function, and module scores for “MYC Targets V2” (MSigDB Hallmark 2020) were calculated using the addModuleScore function. All three metrics (inferred copy number, gene score, and module score) were aggregated by clone, and the clone-level mean and SD were compared with MYC copy numbers derived from scDNA-seq.

DNA FISH probe preparation

Bacterial artificial chromosome (BAC) clones in E. coli were purchased from Advanced GenoTechs (Ibaraki, Japan) and cultured in 200 mL of Lysogeny Broth (LB) medium containing 34 µg/mL chloramphenicol. The following day, BAC clones were extracted using QIAGEN Plasmid Maxi Kit (QIAGEN, Gelderland, Netherlands). Probes were enzymatically labeled at 15 °C for 12 h using Nick translation kit (Abbott, IL, USA) with either Fluorescein-12-dUTP (Roche, Switzerland) or ChromaTide Texas Red-12-dUTP (Thermo Fisher Scientific). The following day, labeled oligos were purified using MicroSpin G50 Columns (Roche, Switzerland) according to the manufacturer’s instructions. Briefly, columns were centrifuged at 740 × g for 1 min to remove storage buffer, samples were applied to the columns, and probes were recovered by centrifugation. The following BAC clones were used: RP11-1136L8(MYC) and RP11-94L15(ERBB2).

Metaphase chromosome spread

For COLO320DM and COLO320HSR, cells in the metaphase were collected following the treatment with KaryoMAX (Thermo Fisher Scientific) at 100 ng/mL for 6 h. For other cell lines, cells in the metaphase were collected following the treatment with Colcemid (AdipoGen, Switzerland) at 100 ng/mL for 6 h. All cell lines were washed once with PBS, and single-cell suspensions were incubated under hypotonic condition (3:7 PBS: deionized distilled water) for 5 min at room temperature. The samples were fixed in Carnoy’s fixative (3:1 methanol: acetic acid), and the cell pellet was collected by centrifugation. Samples were fixed additional three times with Carnoy’s fixative. Finally, they were resuspended in the fixative and dropped onto MAS-coated slide glass (Matsunami, MAS-01, Osaka, Japan).

Metaphase DNA FISH

Fixed cells on slide glasses were air-dried overnight. The hybridization mix was prepared by combining 50 µL of deionized formamide (Sigma-Aldrich, St. Louis, MO), 10 µL of 20× SSC (Sigma-Aldrich), 20 µL of 60% dextran sulfate (Wako), 10 µL of water, and 1 µL of Tween 20 (Wako), and thoroughly mixed. Next, the probe mix was prepared. For each slide, 10 µL of red-labeled probe, 10 µL of green-labeled probe, 2.5 µg of Human Cot-1 DNA (Thermo Fisher Scientific), 10 µg of sonicated salmon sperm DNA (BioDynamics Laboratory), 2.6 µL of 3 M Na₂CO₃ (Nacalai), and 65 µL of ice-cold 100% ethanol were mixed and centrifuged at 20,000 × g for 5 min at 4 °C. The supernatant was discarded, and the pellet was retained. Next, 10 µL of the hybridization mix was added to the pellet and mixed well, and the entire mixture was dropped onto a glass slide. A coverslip (Matsunami) was placed on top, sealed with paper bond (Kokuyo, Osaka, Japan), denatured at 80 °C for 1 min, and immediately incubated at 37 °C in a humidified light-protected chamber. The following day, the coverslip was removed, and the slides were washed twice with PBS. The samples were stained with 4′,6-diamidino-2-phenylindole (DAPI) (1:1000 in PBS) for 5 min and mounted with 10 µL of ProLong Gold with DAPI (Thermo Fisher Scientific). Images were captured using BZ-X700 microscope (Keyence, Osaka, Japan) with a 100× objective lens.

Interphase FISH

Cells were seeded on poly-L-lysine–coated (Sigma-Aldrich) coverslips. The following day, the cells were washed once with PBS and fixed with 2% paraformaldehyde (Electron Microscopy Sciences, Morgantown, PA) for 10 min at room temperature. The samples were then permeabilized with 0.2% Triton ×-100 (Wako) for 10 min at room temperature and washed once with PBS. Coverslips with attached cells were placed on a Coverslip Mini-Rack for 8 coverslips (Thermo Fisher Scientific), dehydrated in 100% ethanol for 2–3 min, and incubated in 70% formamide at 80 °C for 5 min for denaturation. Immediately afterward, the coverslips were transferred to ice-cold 70% ethanol for 3 min, followed by incubation in 100% ethanol for 2 min, and then air-dried thoroughly. The hybridization mix and probe mix were prepared as described above. For each sample, 10 µL of hybridization mix was added to the probe pellet to prepare the final hybridization solution. The mixture was denatured at 80 °C for 5 min and then annealed at 37 °C for 15 min. Ten microliters of the mixture was applied to a glass slide, and the coverslip was placed on top and sealed with paper bond. The samples were incubated at 37 °C in a humidified light-protected chamber overnight. The following day, the coverslips were washed twice with PBS. The samples were stained with 4′,6-diamidino-2-phenylindole (DAPI) (1:1000 in PBS) for 5 min and mounted with 10 µL of ProLong Gold with DAPI (Thermo Fisher Scientific). Images were captured using a CellVoyager CQ1 imaging system (Yokogawa, Tokyo, Japan) with a 60× objective lens.

Dual immunofluorescence–FISH

Cells were fixed and permeabilized with 2% paraformaldehyde (Sigma-Aldrich) as described above. After fixation, the samples were blocked with 3% bovine serum albumin (Wako) for 5 min at room temperature. The samples were then incubated with primary antibody diluted in blocking buffer (1:500) for 2 h at room temperature, followed by two washes with PBS. Subsequently, the samples were incubated with secondary antibody diluted in blocking buffer (1:1000) for 1 h at room temperature and washed twice with PBS. The samples were then refixed with 2% paraformaldehyde (Sigma-Aldrich) and incubated in 70% formamide at 80 °C for 13 min for denaturation. Subsequent procedures were performed as described above. The following antibodies were used: Aurora B antibody (Active Motif, Carlsbad, CA, #39261) as the primary antibody and goat anti-rabbit IgG (H + L) highly cross-adsorbed secondary antibody conjugated with Alexa Fluor 568 (Thermo Fisher Scientific, A-11036).

Quantification of FISH foci and pixel intensity

Quantification of FISH foci and pixel intensity was performed using the Find Maxima function and pixel intensity measurements in ImageJ. For FISH foci analysis, only nuclei with an area between 0.5 and 1.0 were included, whereas for pixel intensity measurements, nuclei with an area between 0.3 and 1.5 were analyzed to exclude abnormally small or large nuclei. To analyze pairs of daughter cells using Aurora B immunofluorescence in dual immunofluorescence–FISH experiments, the positions of Aurora B signals and nuclei were converted into coordinate information to identify daughter cell pairs computationally. See “Code availability”.

Preparation of PseuMO-Tag lentiviruses

PseuMO-Tag lentiviruses were prepared as previously reported²¹. Briefly, the construct was derived from the pLenti-CMV-GFP-Puro vector (Addgene #17448), with a strong CMV promoter immediately upstream of the barcode sequence. Barcode sequence was inserted by Golden Gate assembly. Lentiviruses were produced using FuGENE HD Transfection Reagent (Promega). Newly generated plasmids are available from the corresponding author upon reasonable request.

ID vector construction

The construct was modified based on the pLenti-CMV-GFP-Puro vector (Addgene #17448), with a strong CMV promoter immediately upstream of the barcode sequence. The original vector was digested with EcoRI and KpnI to remove the PGK promoter and puromycin resistance gene, and replaced with the EF-1α core promoter and an EGFP-P2A-puromycin resistance cassette. The vector was then digested with ClaI and SalI to remove the original CMV promoter and EGFP, and a new CMV promoter, followed by a stuffer sequence for the cellular barcode, was inserted. The resulting plasmid served as the backbone for the cellular barcode vector. Next, 10 µg of the plasmid was digested with BsmBI-v2, purified using the QIAquick Gel Extraction Kit (QIAGEN), and eluted in 20 µL of elution buffer. For insertion, 2 µL of 100 µM barcode oligonucleotide and 1 µL of 100 µM reverse primer (5′-TGCAGCATGCGTCTCACAACG-3′) were mixed with 25 µL of NEBNext High-Fidelity 2× PCR Master Mix (New England Biolabs) and 22 µL of water to produce double-stranded barcode DNA (Thermo Fisher Scientific). PCR was carried out using the following conditions: 98 °C for 2 min; 10 cycles of 65 °C for 30 s and 72 °C for 10 s; and a final extension at 72 °C for 2 min. The PCR product was purified using a QIAquick PCR purification kit (QIAGEN) and eluted in 20 µL. The backbone plasmid (1.5 µg) and barcode insert (33 ng) were ligated by Golden Gate assembly. The reaction mixture consisted of 2 µL of r3.1 buffer (New England Biolabs), 1 µL of BsmBI-v2 (New England Biolabs), 1 µL of T4 DNA ligase (New England Biolabs), 2 µL of 10× T4 DNA ligase buffer (New England Biolabs), and nuclease-free water in a total volume of 20 µL. The reaction was incubated in a thermal cycler under the following conditions: 100 cycles of 42 °C for 2 min and 16 °C for 5 min, followed by incubation at 50 °C for 10 min and 80 °C for 20 min. The reaction product was purified using DNA Clean & Concentrator Kit (Zymo Research, Irvine, CA) and eluted in 10 µL. The purified plasmid was transformed into NEB Stable Competent E. coli (New England Biolabs), which were cultured in 500 mL LB medium containing 100 µg/mL ampicillin at 30 °C overnight with shaking. Plasmid DNA was then extracted using QIAGEN Plasmid Plus Maxi Kit (QIAGEN). To isolate individual barcode plasmids from the mixed-barcode plasmid pool, the mixture was transformed into NEB Stable Competent E. coli (New England Biolabs). The following day, 20 colonies were randomly selected and cultured in 3 mL of LB medium at 37 °C with shaking overnight. Plasmid DNA was purified using the Monarch Plasmid Miniprep Kit (New England Biolabs). Newly generated plasmids are available from the corresponding author upon reasonable request.

Sanger sequencing

The barcode sequence for each plasmid was validated by Sanger sequencing. The reaction mixture (10 µL total volume) consisted of 150–300 ng of plasmid DNA, 1 µL of 3.2 µM primer, 2 µL of BigDye Terminator v3.1 (Thermo Fisher Scientific), 1 µL of 5× sequencing buffer, and nuclease-free water. The reactions were incubated as follows: 96 °C for 1 min, followed by 25 cycles of 96 °C for 10 s, 50 °C for 3 s, and finally 60 °C for 75 s. The sequencing primer used was 5′-TAAGCAGAGCTATGGTGAGC-3′. The products were purified using a Gel Filtration Cartridge (Edge BioSystems, San Jose, CA) and transferred to a MicroAmp Optical 96-Well Reaction Plate (Thermo Fisher Scientific), denatured at 94 °C for 2 min, and subjected to capillary electrophoresis on a 3500 × L Genetic Analyzer (Thermo Fisher Scientific).

ID vector transfection

For each well of a 96-well plate, 5 µL of Opti-MEM (Thermo Fisher Scientific) and 0.32 µL of FuGENE HD Transfection Reagent (Promega) were mixed and incubated at room temperature for 5 min. The mixture was then added to 80 ng of ID vector that was pre-aliquoted into another tube, followed by incubation at room temperature for 15 min. Next, 45 µL of the culture medium was added, and the mixture was gently pipetted twice. Finally, 50 µL of medium was removed from each well, and the prepared transfection mixture was added.

Reagents and special medium

JQ1, PTX, oxaliplatin, doxorubicin, and hydroxyurea were purchased from Selleck Biotech (Kanagawa, Japan). Glutamine depleted DMEM and glucose depleted DMEM were purchased from Nacalai. Medium containing drugs was replaced every 3–4 days.

RNA extraction and reverse transcription

Total RNA was extracted using RNeasy Mini/Micro Kit (QIAGEN) according to the manufacturer’s instructions. Reverse transcription was performed with 10 ng of RNAs using PrimeScript RT Master Mix (TAKARA, Shiga, Japan).

Preparation of amplicon-seq library from cDNA

To prepare a list of PseuMO-Tag sequences for each clone, amplicon-seq for the barcode locus was performed. The library was prepared using nested PCR as previously reported²¹. For the first PCR, 1 µL of the cDNA template in a total volume of 25 µL was used with NEBNext High-Fidelity 2× PCR Master Mix (New England Biolabs). The PCR conditions were as follows: 30 s denature step at 98 °C, followed by 35 cycles of 98 °C for 10 s, 67 °C for 30 s, and 72 °C for 30 s, and finally 72 °C for 2 min. The primers used for the first PCR were as follows: Fw: 5′- CACCATCGTGGAACAGTACG -3′, and Rv: 5′- GTCCGCTCGCTAGTTATTGC -3′. The PCR products were purified using 0.6× SPRIselect beads (Beckman Coulter), and eluted in 12 µL. For the second PCR, 1 µL of the first PCR product template in a total volume of 50 µL was used with NEBNext High-Fidelity 2× PCR Master Mix (New England Biolabs). The PCR conditions were as follows: 30 s denature step at 98 °C, followed by 8 cycles of 98 °C for 10 s, 67 °C for 30 s, and 72 °C for 30 s, and finally 72 °C for 2 min. The primers used for the second PCR were as follows: Fw: 5′- AATGATACGGCGACCACCGAGATCTACACXXXXXXXXACACTCTTTCCCTACACGACGCTCTTCCGATCTGCAGGAAACAGCTATGACTATGC -3′, and Rv: 5′- CAAGCAGAAGACGGCATACGAGATXXXXXXXXGTGACTGGAGTTCAGACGTGTGCTCTTCCGATCTGTCCGCTCGCTAGTTATTGC -3′. X indicates the indices. The PCR products were purified using 0.8× → 1.2× SPRIselect (Beckman Coulter), and eluted in 12 µL. Library quality was assessed by Quantus (Promega) and TapeStation 4200 (Agilent Technologies) with D1000 Screen Tape. NGS was performed in 2 × 38 bp paired-ends using NextSeq550 platform (Illumina), which generates 0.5–1 million reads per library.

DNA extraction

Quantitative PCR (qPCR) was used to evaluate changes in ecDNA copy number following drug treatment. DNA was extracted from cells following lysis. The cell pellets were collected and resuspended in 36 µL of 50 mM NaOH, followed by incubation at 95 °C for 10 min. After cooling to room temperature, 4 µL of 1 M Tris-HCl (pH 8.0) was added to neutralize the solution. The mixture was then diluted with 120 µL of nuclease-free water. NaOH solution was freshly prepared by appropriately diluting from a 1 M stock. For other experimental purposes, DNA was extracted using QIAamp UCP DNA Micro Kit (QIAGEN) or DNeasy Blood & Tissue Kit (QIAGEN) according to the manufacturer’s instructions.

Quantitative PCR

Quantitative PCR was done using TB Green® Premix Ex Taq™ II (Tli RNaseH Plus) (TAKARA) and StepOnePlus (Thermo Fischer Scientific), according to the manufacturer’s instructions. To determine the internal control, as shown in Supplementary Fig. 5a, 0.6 ng of DNA was used as a template. To evaluate copy number changes following drug treatment, 2 µL of the lysate-extracted DNA described above was used as the template per well. The internal control was chr5q23.1, which showed no difference in amplification compared with RPE1 (Supplementary Fig. 5a). Amplification levels were normalized relative to chr5q23.1 levels, using the ∆∆CT method. The primers were as follows: Chr4-Fw 5′- CATGCTCTATCTACACTTGTAAGCCA -3′, Chr4-Rv 5′- GCCCTGCAGCCTAGGAATG -3′, Chr5q23.1-Fw 5′- TTTGTTCTACTACAAAGACTCATGCAC -3′, Chr5q23.1-Rv 5′- CTGGGTTGGTTCCAGGTCTT -3′, Chr6-Fw 5′- AGAAGGGATTACACTATCTGTCACTC -3′, Chr6-Rv 5′- AACCACCATACCCAGCCATG -3′, Chr17-Fw 5′- GCCAGACCTGTGAATATTTTATGTTAC -3′, Chr17-Rv 5′- TTCTTCTCCCTCCCTCGTCC -3′, Chr22-Fw 5′- ACCCTCCTATTAAAATGCCTCTGA -3′, Chr22-Rv 5′- ACGGCGCATGAACAGAAAAC -3′, MYC-Fw 5′- GGACGACGAGACCTTCATCA -3′, MYC-Rv 5′- GAGGCCAGCTTCTCTGAGAC -3′, and ERBB2-Fw 5′- GCTGGTACTTTGAGCCTTCACA -3′, ERBB2-Rv 5′- GAAGGCGGGAGACATATGGG -3′.

Preparation of amplicon-seq library from DNA

To preserve barcode diversity, 50 ng of DNA was used as the template. To maintain barcode diversity, the first PCR was performed for seven cycles under the same temperature conditions as described above. Subsequent purification and the second PCR were performed as described above. NGS was performed in 2 × 38 bp paired-ends using NextSeq550 platform (Illumina), which generates 2–5 million reads per library.

Cell viability assay

COLO320DM and COLO320HSR were seeded into 96-well plates at 2000 cells/well with three replicates, and the drugs were added the following day. After 7 days, 10 µL of CCK-8 reagent from Cell Counting Kit-8 (Dojindo, Kumamoto, Japan) was added to each well, and then the absorbance at 450 nm was measured using BioTek Gen5 (Agilent Technologies). The absorbance was normalized to that of DMSO-treated wells.

Cell proliferation assay

Cell proliferation was measured using the IncuCyte S3 Live-Cell Analysis System (Sartorius, Germany). Similar to the cell viability assay, cells were seeded into 96-well plates at 2000 cells per well in triplicate, and drugs were added the following day. Confluency values were measured at 0 and 144 h, and growth rates were calculated and compared across conditions.

Bulk RNA sequencing library preparation

Total RNA was extracted with RNeasy Plus Mini Kit (QIAGEN). Each library was constructed from 1 µg of RNA using the SMARTer Stranded Total RNA Sample Prep Kit—HI Mammalian (TAKARA), according to the manufacturer’s instructions. Library quality was assessed by Quantus (Promega) and TapeStation 4200 (Agilent Technologies) with D5000 Screen Tape. Next generation sequencing was performed in 2 × 38 bp paired-ends using NextSeq550 platform (Illumina), which generates 20 million reads per library.

Bulk RNA sequencing data analysis

RNA sequencing data were analyzed as previously reported²³. Briefly, the raw reads were trimmed by Skewer (v0.2.2), followed by mapping to GRCh38 genome by STAR (v2.7.8a), and then counted by featureCounts (v2.0.10). Differential gene expression analysis was performed with edgeR’s glmQLFTest (v3.32.1). Gene Ontology enrichment analysis was performed by ClusterProfiler’s enrichGO() function.

Transfection of siRNAs

Silencer Select siRNAs were purchased from Thermo Fisher Scientific. Cells were seeded, and the following day siRNAs (10 nM final concentration) were transfected using Lipofectamine RNAiMAX (Thermo Fisher Scientific) according to the manufacturer’s instructions. The medium containing siRNA was replaced every 3–4 days. Total RNA was extracted, and knockdown efficiency was confirmed by qPCR. Primer sequences are listed in Supplementary Data 7. The following siRNAs were used: 4390843 (siCtrl_1), 4390846 (siCtrl_2), s2388 (CAPN6), s15638 (AP3B2), s4674 (EN2_1), s22934 (NMNAT2), s5398 (FUCA2), and s4676 (EN2_2).

Statistics and reproducibility

Statistical analyses were performed using R (v4.5.2). Data are presented as mean ± SD unless otherwise indicated. Two-sample comparisons were performed using two-sided Welch’s t-tests. Comparisons of ecDNA segregation ratios were performed using two-sided Wilcoxon rank-sum tests. Distribution comparisons were performed using two-sided Kolmogorov–Smirnov tests. Correlation analyses were performed using two-sided Pearson correlation tests. For image-based analyses, values above the 95th percentile within each experimental group were excluded. Nuclei outside the predefined area ranges were also excluded to remove abnormally small or large nuclei: 0.3–1.5 for foci analysis and 0.3–1.2 for pixel-intensity analysis. The experiments were not randomized. The investigators were not blinded to allocation during experiments and outcome assessment. Unless otherwise indicated, experiments were performed independently three times. For all analyses, statistical significance was set at p < 0.05.

Clonal tracking by scDNA-seq and amplicon-seq (Fig. 5 and Supplementary Fig. 7)

COLO320DM-derived clones EC2, HSR4, EC11, and EC12 cells were seeded into 96-well plates at 10,000 cells per well. The following day, cells were transduced with PseuMO-Tag lentiviruses (MOI = 3) supplemented with 10 µg/mL polybrene. One day after transduction, cells were trypsinized, dissociated, and resuspended in 20 mL of DMEM. The suspension was distributed into a new 96-well plate at 150 µL per well (Split 1). When the colonies reached 4–8 cells per clone (4 days after plating for all clones), the number of clones containing ≥ 4 cells per clone was counted. The cells from each well were then trypsinized again to obtain single-cell suspensions. Wells were pooled so that each mixture contained approximately 100–120 clones, and the mixed suspension was equally divided into two wells of a new 96-well plate (Split 2). The following day, one well of each pair was treated with 1:1000 DMSO, and the other was maintained without any treatment. For the DMSO-treated group, the cells were cultured until the colonies reached 100–150 cells per colony (EC2, 9 days; HSR4, 9 days; EC11, 14 days; EC12, 8 days). Next, the ID vector was transfected as described above, and the cells were collected 18–20 h later. To obtain a total of approximately 600 clones, cells from six corresponding well pairs were pooled, generating six mixed samples for storage. The clones remained spatially separated throughout the culture period, without any physical contact or competition. After EC11 collection, all four clones were pooled and subjected to scDNA-seq. For the untreated paired wells, six corresponding wells were harvested on the same day as the DMSO group, pooled, and split into 18 wells of a 96-well plate (Split 3). The following day, the cultures were treated with 1:1000 DMSO, 500 nM JQ1, 100 nM PTX, 50 µM hydroxyurea, 300 nM doxorubicin, 10 µM oxaliplatin, and 500 nM rabusertib, glucose depletion, or glutamine depletion. The medium was replaced every 3–4 days. After four weeks, the cells were collected for DNA extraction and amplicon-seq library preparation (see “DNA extraction” Section and “Preparation of amplicon-seq from DNA” Section). Based on the amplicon-seq results, the clones with barcodes detected by two or more sequencing reads were defined as surviving clones under each stress condition. These surviving clones were associated with the scDNA-seq data. For each clone, the mean copy number across all targeted regions was determined and used for comparative statistical analyses of copy-number distributions.

Clonal tracking by scDNA-seq (Fig. 6 and Supplementary Figs. 8 and 9)

COLO320DM EC2 cells were seeded into 96-well plates at 10,000 cells per well. The following day, cells were transduced with PseuMO-Tag lentiviruses (MOI = 3) supplemented with 10 µg/mL polybrene. The day after transduction, the cells were trypsinized, dissociated, and resuspended in 10 mL of DMEM. The suspension was distributed into a new 96-well plate as follows: (i) 100 µL per well for 48 wells, and (ii) 60 µL per well for additional 48 wells (Split 1). Each well represented an independent clone that could be tracked using a unique combination of barcodes. When the colonies reached 4–8 cells per clone (4 days after plating), the number of clones containing ≥ 4 cells was counted. Cells were again trypsinized to obtain single-cell suspensions. To prepare mixed populations containing approximately 50–60 clones, one well from group (i) and one well from group (ii) were combined, and equally divided into two new wells of a 96-well plate. This procedure was repeated for all pairs, resulting in 48 paired wells, each containing 50–60 mixed clones (Split 2). The following day, one well for each pair was treated with 500 nM JQ1, whereas the other was treated with DMSO as a control. Cells were cultured until colonies reached 100–150 cells per colony (10 days for the DMSO group and 24 days for the JQ1 group). At that point, the ID vector was transfected to label each condition, and cells were collected 18–20 h after transfection. As the goal was to analyze approximately 300 clones, cells from six corresponding well pairs were pooled, generating six mixed wells for storage. The transfection procedure followed the protocol described below (see “ID vector transfection” section). Throughout the culture period, individual clones remained spatially separated, with no physical contact or competition between the clones. After collection, samples from the JQ1 and DMSO groups were pooled for scDNA-seq. During the analysis, clones detected in the DMSO and JQ1 groups were extracted separately. Of these, we focused on two subsets: clones detected only in the DMSO group, but not in the JQ1 group, and those detected in both the DMSO and JQ1 groups. The differences in their copy-number distributions were then analyzed and compared.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

C.S. and R.M. conceived and designed the study, developed methodologies, and wrote the manuscript. C.S. performed most of the experiments and data analyses. K.M. developed and provided methodologies and technical advice. K.K. performed part of the data analyses. Y.M. contributed to the data analysis and interpretation, as well as to the study design. L.Y. and M.N. performed part of the experiments. R.-S.N. and T.H. provided technical guidance and oversight for the FISH experiments and analyses. M.A., S. Iwami, and S. Iwanami performed the mathematical modeling and analyses. R.M. supervised the study and secured funding. All authors discussed the results and contributed to manuscript editing.

Peer review

Peer review information

Nature Communications thanks Anton Henssen and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported in part by JSPS KAKENHI (grant numbers JP26K18692, JP24K18385, and JP23KJ2162 to C.S.; JP25K22584, JP23K18247, and JP23K27446 to K.M.; JP25H01036, JP24K02312, and JP24H00037 to R.M.); the Japan Agency for Medical Research and Development (AMED) (grant number JP24ama221606 and JP26ama221228 to R.M.); the Japanese Cancer Association-Kobayashi Foundation for Cancer Research (to C.S.); the Kato Memorial Bioscience Foundation (to K.M.); the Uehara Memorial Foundation (to K.M.); the Takeda Science Foundation (to K.M.); the Mochida Memorial Foundation for Medical and Pharmaceutical Research (to K.M.); and the Princess Takamatsu Cancer Research Fund (to R.M.).

Data availability

Data generated in this study have been deposited in the BioStudies database under accession code E-MTAB-16119. This study also used previously published single-cell ATAC-seq data under accession code GSE160148. Source Data are provided with this paper. Source data are provided with this paper.

Code availability

Processed single-cell objects and custom PseuMO-Tag Decoder to reproduce analyses and figures is available at [https://github.com/Chikako-Shibata/ecDNA_single_cell_DNA]. Code for quantification of ecDNA in FISH images is available at [https://github.com/Chikako-Shibata/Quantification-of-ecDNA-in-FISH-images]. All code is available under the MIT License.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-76566-5.