ecDNA replication is disorganised and vulnerable to replication stress

31 32 Extrachromosomal DNA (ecDNA) is a critical driver of cancer progression, contributing to 33 tumour growth, evolution, and therapeutic resistance through oncogene amplification. Despite 34 its significance, the replication of ecDNA remains poorly understood. In this study, we 35 investigated the replication dynamics of ecDNA using high -resolution replication timing 36 analysis (Repli-seq) and DNAscent, a method for measuring origin firing and replication fork 37 movement based on ultra-long read Oxford Nanopore Sequencing, that we applied to both bulk 38 DNA and to ecDNA isolated with FACS -based Isolation of Native ecDNA (FINE), a new 39

for isolating intact, chromatinised ecDNA without DNA or protein digestion. We 40 demonstrate that ecDNA in the COLO 320DM colorectal cancer cell line exhibits largely 41 asynchronous replication throughout the S phase, contrasting with the conserved replication 42 timing of the corresponding normal linear chromosomal DNA in RPE -1 cells and the 43 chromosomally reintegrated ecDNA in COLO 320HSR, which forms a homogen eously 44 staining region. Replication origins on ecDNA are redistributed, and replication forks exhibit 45 reduced velocity and increased stalling, particularly near the c -MYC oncogene. Under 46 replication stress induced by hydroxyurea treatment, ecDNA replication is further 47 compromised, leading to altered origin activation, reduced fork velocity and eventual ecDNA 48 depletion from cells. Our findings reveal fundamental differences in the replication dynamics 49 of ecDNA, providing insights that could inform the development of therapies targeting ecDNA-50 associated oncogene amplification in cancer. 51 52 53

54 Oncogene amplification is a critical driver of cancer progression, contributing to tumour 55 growth, evolution and resistance to therapy. Traditionally, oncogene amplification has been 56 associated with copy number increase on linear chromosomes, observed as homogeneously 57 staining regions (HSRs) in chromosome G -banding (1). However, recent studies have 58 highlighted an alternative and highly potent mechanism of oncogene amplification through 59 extrachromosomal DNA (ecDNA) (2, 3). 60 61 ecDNA molecules comprise large, double stranded DNA fragments averaging between 1-3 Mb 62 in size (2) that have been detected in approximately 17 % of all cancers (4). These circular 63 DNA elements frequently harbour oncogenes or immunomodulatory genes (3, 5, 6) , driving 64 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 3 the aggressive behaviour of tumours. ecDNA -based oncogenes are highly transcribed , even 65 when adjusted for their high copy number, which often exceeds 100 copies per cell (7). 66 67 Unlike canonical chromosomes, ecDNA lacks centromeres, leading to its uneven segregation 68 during cell division (8, 9). This unique flexibility of ecDNA contributes to the rapid evolution 69 of tumours by promoting intratumour genetic heterogeneity (9) allowing cancer cells to rapidly 70 adapt to selective pressures, including therapeutic interventions (9‑13). Consequently, the 71 presence of ecDNA is associated with poorer outcomes, including reduced survival rates 72 compared to tumours without such amplifications (4, 5). 73 74 Despite the recognised importance of ecDNA in cancer, much remains unknown about its DNA 75 replication dynamics. While ecDNA clearly replicates (14, 15), and likely only once per cell 76 cycle during S phase (16), details such as the precise timing of replication, the distribution of 77 replication origins, replication fork velocity and the frequency of replication stalling remain 78 poorly understood. For instance, it is unclear whether the replication origins and replication 79 timing found in a chromosomal segment are preserved when present in ecDNA. Moreover, 80 ecDNA is characterised by elevated replication stress (17), and its loss appears to be linked 81 primarily to a further increase in this stress followed by micronuclei -mediated ecDNA 82 elimination, particularly following treatment with hydroxyurea (HU) (18‑21). A more complete 83 understanding of ecDNA replication may enable the development of strategies to control 84 ecDNA copy number and, consequently, oncogenic drive in certain cancers. 85 86 To address these questions, we combine assessment of the replication timing programme using 87 Repli-seq (22, 23) with DNAscent (24), a method for determining the velocity and direction of 88 the nascent DNA synthesis in single DNA molecules, applying Oxford Nanopore Technologies 89 (ONT) long-read sequencing of COLO 320DM cells and matched ecDNA -negative controls. 90 Further, to directly detect replicating ecDNA, we introduce FINE (FACS -based Isolation of 91 Native ecDNA), a method for isolating largely intact, chromatinised ecDNA without requiring 92 DNA or protein digestion. We show that replication timing programme in ecDNA is 93 significantly disrupted. Further, replication in ecDNA appears more sensitive to stress induced 94 by depletion of ribonucleotide pools with HU than chromosomal DNA, which leads to 95 replication stress-induced ecDNA loss. 96 97 98 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 4

& Methods 99 100 Cell Culture 101 Human colorectal adenocarcinoma cell lines COLO 320DM and COLO 320HSR were 102 obtained from the American Type Culture Collection (ATCC) and maintained in RPMI -1640 103 medium (Gibco) supplemented with 10% (vol/vol) fetal bovine serum (FBS; Sigma -Aldrich). 104 hTERT-immortalized retinal pigment epithelial cells (RPE -1) were also sourced from ATCC 105 and cultured in a 1:1 mixture of Dulbecco’s Modified Eagle’s Medium and Ham’s F12 106 (DMEM, Gibco) supplemented with 10% (vol/vol) FBS. All cell lines were incubated in a 107 humidified atmosphere of 5% CO2 at 37°C. Cell lines were regularly tested for mycoplasma 108 contamination. 109 110 Metaphase Chromosome Spreads 111 Cells were arrested in metaphase by treatment with either 0.1 µg/ml KaryoMAX Colcemid 112 solution in PBS (Gibco) for RPE -1 cells, or 0.4 µM nocodazole (Sigma -Aldrich) for COLO 113 320DM and COLO 320HSR cells, for 3 h at 37°C. After arrest, cells were washed with PBS 114 and gently resuspended in a pre-warmed 75 mM potassium chloride (KCl) hypotonic solution, 115 followed by a 20 min incubation at 37°C to swell the cells. 116 117 Cells were pre -fixed by adding 10% volume of freshly prepared Carnoy’s fixative (3:1 118 methanol:acetic acid, vol/vol), then washed twice with ice -cold Carnoy’s fixative and stored 119 overnight at −20°C. The following day, the cell suspension was dropped from a height of 120 approximately 50 cm onto ice -cold glass slides held at a 45° angle. Slides were matured by 121 incubation at 65°C for 1 h, followed by washing in 2× saline -sodium citrate (SSC) buffer. 122 Chromosomes were stained with 1 µg/ml Hoechst 33258 (Sigma-Aldrich) for visualisation. 123 124 Fluorescence In Situ Hybridization (FISH) 125 Fixed metaphase spreads were washed in 2× saline -sodium citrate (SSC) buffer and 126 sequentially dehydrated ethanol series (70%, 85%, 100%) for 2 min each. Fluorescence in situ 127 hybridization (FISH) probes (Empire Genomics; green CHR08-10-GR for Chromosome 8 and 128 orange MYC-20-OR for c-Myc) were applied to the slides, which were then sealed with rubber 129 cement. Denaturation was performed on a hot plate at 75°C for 7 min, followed by 130 hybridization for 16 h in a humidified chamber at 37°C. 131 132 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 5 After hybridization, coverslips were removed, and slides were washed in 0.3% Igepal CA-630 133 (Sigma-Aldrich) in 0.4× SSC at 73°C for 2 min, followed by a wash in 0.1% Igepal CA-630 in 134 2× SSC at room temperature for 2 min. The slides were stained with 1 µg/ml Hoechst 33258 135 (Sigma-Aldrich) and washed again in 2× SSC before being mounted with ProLong Diamond 136 Antifade Mountant (Thermo Fisher Scientific) and covered with a coverslip. 137 138 FISH images were captured using a Zeiss LSM780 confocal microscope with a 63× objective. 139 Image analysis was performed using Fiji (ImageJ-based software) (25). 140 141 Chromosome Flow Cytometry and Sorting 142 Actively dividing COLO 320DM and COLO 320HSR cells were synchronised in mitosis by 143 treating with 0.4 µM nocodazole (Sigma -Aldrich) for 13 h. For RPE -1 cells, 100 µl of 144 Colcemid (Gibco KaryoMAX Colcemid Solution in PBS) was added to 30 ml of culture 145 medium. Following mitotic arrest, cells were harvested by mitotic shake -off and collected by 146 centrifugation at 400g for 5 min at room temperature. Cells were resuspended in the medium 147 containing nocodazole and incubated with 0.1 µg/ml Latrunculin B (Cambridge Bioscience 148 CAY10010631) at 37°C for 1 h. 149 150 Cells were pelleted by centrifugation (400g, 5 min) and resuspended in a hypotonic solution 151 containing 75 mM KCl, 10 mM MgSO 4, 0.2 mM spermine, and 0.5 mM spermidine, pH 8.0. 152 The cell suspension was incubated at room temperature for 20 min, followed by centrifugation 153 at 300g for 5 min. Cells were then resuspended in ice-cold PAB buffer (150 mM Tris-HCl, pH 154 8.0, 0.25% Triton X -100, 20 mM KCl, 0.5 mM spermine, 1 mM EDTA, 5 mM EGTA). 155 Chromosomes were released by gently flicking the tube, and the quality of chromosome release 156 was checked by mixing 10 µl of sample with 1 µl of propidium iodide (PI) and visualising 157 under a microscope. 158 159 The isolated chromosomes were stained with a combination of 5 µg/ml DAPI, 50 µg/ml 160 Chromomycin A3, and 10 mM MgSO4. The staining reaction was carried out in a low volume 161 (300 µl) for 1 h at room temperature, followed by dilution with 2–3 ml of PAB buffer. Samples 162 were then incubated for an additional 30 min. 163 164 Chromosomes were analysed and sorted using a BD FACSAria Fusion Flow Cytometer with 165 a 70-µm or 100-µm nozzle, set at a flow rate of < 25,000 events/sec. The 585 nm filter was 166 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 6 used to detect Chromomycin A3 and 450 nm filter was used to detect DAPI. Sorted 167 chromosomes were maintained in PBS containing 1 mM EDTA for further analysis. 168 169 Whole-Genome Sequencing (WGS) and ecDNA Assembly 170 Genomic DNA was extracted from cultured cells using the Monarch Genomic DNA 171 Purification Kit (New England Biolabs) according to the manufacturer’s protocol. DNA 172 libraries were prepared using the NEBNext® Ultra™ II FS DNA Library Prep Kit for Illumina 173 (New England Biolabs), following the manufacturer’s instructions. Library quantification and 174 dilution were performed using the Qubit dsDNA High Sensitivity Quantification Assay 175 (Thermo Fisher Scientific). Sequencing was carried out on an Illumina NextSeq 2000 platform. 176 177 The resulting FASTQ files were aligned to the human reference genome (hg38) using BWA -178 MEM. AmpliconArchitect (AA; https://github.com/virajbdeshpande/AmpliconArchitect; (26)) 179 was employed to detect genomic amplifications. The output from AA was integrated with 180 optical genome mapping data to facilitate the assembly of extrachromosomal DNA (ecDNA) 181 structures using the Amplicon Classifier tool. Optical genome mapping was performed by the 182 Wellcome Sanger Institute using the Saphyr® instrument (Bionano Genomics). Cell pellets 183 were prepared following standard Bionano protocols. DNA was labelled by DLE-1 and de novo 184 assembled using the Bionano Access software. 185 186 Repli-seq analysis of DNA replication timing 187 Repli-seq was performed as previously described with modifications (23, 27). Briefly, actively 188 proliferating cells were cultured in T175 flasks under standard conditions and pulse -labelled 189 with 100 µM BrdU (Sigma -Aldrich) for 30 min at 37°C. After labelling, cells were washed 190 twice with ice-cold PBS and fixed by adding 75% (vol/vol) ice -cold ethanol dropwise while 191 vortexing gently. Cells were stored at −20°C for at least 16 h. 192 For FACS, fixed cells were washed with 1% (vol/vol) FBS in PBS and stained with a solution 193 of propidium iodide (50 μg/ml, Sigma -Aldrich) and RNase A (20 μg/ml, Sigma -Aldrich) in 194 PBS/1% FBS. After 30 min of incubation at room temperature in the dark, cells were filtered 195 through a 37 -μm nylon mesh and sorted by flow cytometry (BD FACSAria II) into five 196 fractions representing different stages of S phase based on DNA content. 197 198 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 7 Genomic DNA was extracted from sorted cells using the Zymo Quick -DNA Microprep Kit 199 (Zymo Research) following the manufacturer’s instructions. DNA was fragmented to an 200 average size of 200 bp using a Covaris M220 ultrasonicator set to 75 W peak incident power, 201 10% duty cycle, 200 cycles per burst, for 260 seconds. Fragment size was verified using an 202 Agilent Bioanalyzer 2100 with a high-sensitivity DNA chip. 203 204 Next, libraries were prepared using the NEBNext Ultra DNA Library Prep Kit for Illumina 205 (NEB), according to the manufacturer’s instructions. BrdU -labeled DNA was 206 immunoprecipitated by first denaturing the DNA at 95°C for 5 min, followed by incubation 207 with an anti-BrdU antibody (BD Biosciences). Immunoprecipitated complexes were recovered 208 by incubation with rabbit anti-mouse IgG and centrifugation. After washing, the samples were 209 digested overnight with proteinase K (0.25 mg/ml) at 56°C. DNA was purified using the DNA 210 Clean & Concentrator-5 kit (Zymo Research). The purified DNA was amplified, indexed, and 211 prepared for sequencing using NEBNext Ultra II DNA Library Prep Kit for Illumina. 212 Sequencing was performed on an Illumina NextSeq 2000 platform. 213 Sequencing reads were aligned to the human reference genome (GRCh38) using BWA-MEM. 214 The aligned data were indexed, and PCR duplicates and reads with multiple alignments were 215 removed using SAMtools (28). The resulting BAM files were normalized to reads per kilobase 216 per million (RPKM) using bamCoverage. The normalized data were visualised in the 217 Integrative Genomics Viewer (IGV) (29). 218 219 The statistical analysis of the synchronicity of the replication was achieved using Rao’s 220 quadratic entropy estimate. We consider bins of length 10kB, to which reads are mapped and 221 normalised between five fractions encompassing the S phase, giving a categorical distribution 222 over five classes. As a measure of disorder, we use Rao's quadratic entropy (30, 31) with a 223 circular metric, which is more suitable for modelling cycling cells than alternatives not 224 employing the temporal order of the categories, such as Shannon's entropy (32). The amplified 225 fragment of length 1.6 Mb hence contains 160 bins, each associated with a single Rao's 226 quadratic entropy estimate. We average this value over all bins, to associate a single estimate 227 σ to each 1.6 Mb genomic fragment. 228 229 To assess the degree of extremity of the observed value, we calculate the empirical distribution 230 of Rao's quadratic entropy estimate over fragments of matched length chosen uniformly along 231 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 8 the whole genome, with the exception of fragments of length 5 Mb from each end to avoid 232 sequencing artifacts. To avoid correlations between closely-located fragments we ensure 10Mb 233 spaces between subsequent fragments. This yields N = 273 fragments distributed along 22 234 chromosomes. After removing fragments containing bins without reads, we use N = 223 235 fragments of 1.6 Mb to estimate the empirical distribution. We compared the empirical 236 distributions of RQE between three cell lines using two-sample Kolmogorov-Smirnov test. All 237 three comparisons returned highly significant p-values (< 10-19 for HSR – DM; < 10-8 for RPE-238 1 – HSR; 0.003 for RPE-1 – DM), suggesting minor experiment-specific differences. However, 239 qualitatively the distributions look similar (Figure 2B). To assess the extremity of the RQE 240 value associated with the MYC fragment, we calculated the tail probabilities p = P(X >= X c-241 MYC). We obtain p_DM < 0.01 and p_HSR < 0.01, with p_RPE-1 = 0.23. Our results therefore 242 suggest that the replication of the amplified fragment in COLO 320DM and COLO 320HSR 243 proceeds in a less ordered fashion than for the large majority of the matching fragments in these 244 cell lines. 245 246 RT–qPCR 247 Quantitative PCR (qPCR) was employed to estimate copy number variations of ecDNA by 248 targeting the c-MYC gene (TaqMan™ Copy Number Assay, Assay ID: Hs00292858_cn). The 249 human RNase P gene (TaqMan ™ Copy Number Reference Assay, RNase P; Applied 250 Biosystems, Cat. No. 4403328) was used as the reference. 251 252 Reactions were prepared using TaqPath ™ ProAmp™ Master Mix (Thermo Fisher Scientific, 253 Cat. No. A30865) according to the manufacturer's instructions. Each 20 µL reaction consisted 254 of 10 µL of 2× TaqPath ™ ProAmp™ Master Mix, 1 µL of 20× TaqMan ™ Copy Number 255 Assay, 1 µL of 20× TaqMan ™ Copy Number Reference Assay (RNase P), and 2 µL of 256 genomic DNA (1.6 ng). Amplification was conducted on an Viia7 (ThermoFisher) cycler under 257 the following thermal cycling conditions: initial denaturation at 95°C for 10 minutes, followed 258 by 40 cycles of 95°C for 15 seconds and 60°C for 60 seconds. 259 260 Relative copy number was calculated using the comparative Ct (ΔΔCt) method, normalising 261 target Ct values to the reference RNase P gene and comparing them to a TK -6 sample with a 262 known copy number. All reactions were performed in technical quadruplicate across three 263 biological replicates. 264 265 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 9 Replication Dynamics and Origin Analysis with DNAscent 266 Replication velocity and origins were analysed using DNAscent. Actively dividing cells in 267 COLO 320DM-HU sample were treated with 50 µM hydroxyurea for 24 h to induce replication 268 stress. After the optional treatment, cells were pulse -labelled with 50 µM EdU for 6 min, 269 washed twice with warm PBS, and then treated with 50 µM BrdU for 6 min. Cells were washed 270 again with warm PBS and incubated with 100 µM thymidine for 1 h. After a final PBS wash, 271 fresh complete medium was added. 272 273 For PromethION (ONT) sequencing, cells were harvested 1 h after thymidine treatment, 274 washed with ice -cold PBS containing 1% BSA, and flash -frozen in liquid nitrogen. 275 Approximately 6 million cells were used for each experiment. DNA libraries were prepared for 276 ultra-long sequencing using the Ultra -Long DNA Sequencing Kit (SQK -ULK1, ONT) and 277 sequenced on a Nanopore PromethION platform. For one of the untreated DM samples 278 (DM_untreated_rep2) adaptive sampling targeting chromosome 8 was performed 279 (Supplementary Table 1). An initial attempt at enriching circles with pulse field gel 280 electrophoresis resulted in a low N 50 and no enrichment (Supplementary Table 1). N50 is 281 calculated for reads passing the DNAscent quality criteria (minimum alignment length of 20 282 kb and minimum mapping quality of 20). Then, all reads are sorted by length in descending 283 order, and the cumulative sum over all read lengths is calculated. To determine N50, the read 284 where the cumulative sum up to that read is equal or greater than half of the overall cumulative 285 sum, is determined, and this read’s length is the N50. An overview of all datasets is provided 286 in Supplementary Table 1. 287 288 289 Alternatively, 30 min after cell labelling with BrdU and EdU, cells were arrested in mitosis by 290 overnight treatment with 0.4 µM nocodazole. Following ecDNA release and FACS sorting (as 291 described above), ecDNA was concentrated by centrifugation at 24,000g for 1 h at 4°C. DNA 292 libraries were prepared using the Ligation Sequencing Kit (SQK -LSK110) and sequenced on 293 a ONT MinION platform, following the manufacturer’s instructions. 294 295 Base calling of the raw sequencing files from ONT PromethION or MinION platform (R9.4.1 296 Nanopore, fast5 files available on ENA under accession PRJEB83636) was performed using 297 GUPPY (version 5.0.16) with the configuration for R9 -sequenced DNA 298 (dna_r9.4.1_450bps_fast.cfg) and the output reads (fastq format) that passed and failed the base 299 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 10 calling quality metrics were used for subsequent processing. Alignment of the reads to the hg38 300

genome (release GRCh38.p14) or our custom ecDNA reference map (see above) was 301 performed with minimap2 (version 2.24). The output from minimap2 (sam format) was then 302 converted to bam format and sorted and indexed using samtools (version 1.14). Next, 303 DNAscent (version 3.1.2) subprogrammes index, detect, forkSense and bedgraph 304 (visualisation) were run and replication forks and origins were visualised using a genome 305 browser (Integrative Genomics Viewer, version 2.15.1). 306 307 Calculation of replication fork speed and outlier removal 308 The DNAscent forkSense output (bed file format, available on ENA under accession 309 PRJEB83636) was used for custom downstream processing using Python (version 3.8.2), 310 starting with calculation of fork speed. Replication fork speed and stall scores from DNAscent 311 were obtained as previously described (33). We only include replication forks where both the 312 fork speed and stall score could be calculated for downstream analyses i.e., fork tracks were 313 not near the end of the read. Reads that mapped to chromosome Y were removed as the cell 314 line originates from the tumour of a female patient (34). For all fork speed figures in the main 315 body, we apply outlier filtering based on the interquartile range (IQR). Outliers were removed 316 based on the IQR. The 1st (Q1, 25%) and 3rd quartiles (Q3, 75%) were calculated, and the IQR 317 was defined as the difference between Q3 and Q1. Outliers were identified as data points lying 318 below the lower bound (Q1 - 1.5 * IQR) or above the upper bound (Q3 + 1.5 * IQR) and were 319 excluded from further analysis. For the supplementary histograms of fork speed, no IQR 320 filtering was applied to present the raw data (Supplementary Figure 5). A replicate comparison 321 for the PromethION runs is provided in Supplementary Figure 6. Data cleaning, outlier 322 removal, statistics and visualisations were performed using custom Python (version 3.11.5) 323 scripts, except for circular plots which were created in R (see below). 324 325 FACS enrichment analysis 326 To assess the enrichment factor achieved through FACS sorting, we compared the ratio of reads 327 mapped to ecDNA and to the human reference genome, focusing only on reads that passed the 328 DNAscent minimum quality criteria (minimum alignment length of 20 kb and minimum 329 mapping quality of 20). 330 331 Visualisation of ecDNA maps with origin densities, fork speeds and stall scores 332 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 11 Circular ecDNA visualisations were generated using the circlize package (version 0.4.16) (35) 333 in R (version 4.3.3) using the IQR filtered replication forks as input for fork speed and stall 334 scores and DNAscent forkSense origins for origin of replication analyses. In brief, a custom 335 backbone was created using the AmpliconArchitect (26) generated ecDNA reference map. For 336 origin of replication visualisations, origins from all biological replicates of COLO 320DM 337 (‘DM’, 4 replicates), COLO 320HSR (‘HSR’, 4 replicates) or DM HU treated (‘DM -HU’, 3 338 replicates) were pooled. Origins were grouped into 50 kb segments and normalised by the 339 number of reads in each segment to account for varying read depth within and across samples. 340 The number of reads in each segment was obtained from the aligned bam file and only reads 341 that passed the DNAscent quality criteria (minimum alignment length of 20 kb and minimum 342 mapping quality of 20) were included for the read depth assessment. Aligned read length and 343 mapping quality were extracted from the bam file using the pysam package 344 (https://github.com/pysam-developers/pysam) in Python. For replication fork speed and stall 345 scores, replication forks for each group (DM, HSR, DM -HU) from all replicates were pooled 346 and grouped into 20 kb segments. Fork speed and stall score averages across all forks in each 347 segment were calculated for visualisation. G-quadruplex scores for the ecDNA reference map 348 were obtained using G4Hunter (36) with a window size of 25 and a threshold of 1.5. 349 350 351

352 353 Molecular Characterisation and Isolation of ecDNA 354 355 To investigate ecDNA replication, we used the colorectal cancer cell line COLO 320DM (34), 356 which carries an amplification of the c -MYC locus on multiple ecDNAs. A cell line (COLO 357 320HSR) derived from the same patient tumour provides an interesting comparison in that it 358 harbours linear amplifications of a region of chromosome 8 containing c -MYC on another 359 chromosome (34). We also used an unrelated, immortalised, but untransformed cell line RPE-360 1 as a control (37) (Figure 1A). We characterised the ecDNA structure in our isolate of COLO 361 320DM, using AmpliconArchitect (26) and AmpliconReconstructor (38) algorithms that 362 integrate optical genome mapping with whole -genome short -read sequencing (Figure 1B). 363 COLO 320DM contained a heterogeneous population of ecDNA containing the fragment of 364 chromosome 8 surrounding c-MYC oncogene, and for our analysis we selected the predominant 365 1.5Mb species obtained using AmpliconArchitect (Figure 1B). 366 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 12 367 DAPI Chr 8 c-Myc RPE-1 COLO 320HSR COLO 320DM A E Cell preparation a. M phase synchronisation b. Inhibition of actin polymerisation c. Hypotonic swelling d. DNA release in PAB buffer Cell lysis Sorting e. Staining f. FACS c-Myc PCAT1 PVT1 Chr 8: Chr ?: Chr 8: c-Myc COLO 320DM 1,580,856 bp Sequence G4 score GC content DAPI Chr 8 c-Myc DAPI Chr 8 c-Myc c-Myc HSR amplification c-Myc ecDNA amplification CB D Chromomycin A3 DAPI 21 19 18 9-12 13 8 4 RPE-1 COLO 320HSR COLO 320DM n = 4n = 3 No enrichment FACS enriched Mapped ecDNA / hg38 ratio 0.0 0.1 0.2 0.3 0.4 0.5 0.6 11.2x p = 0.034 0kb 100kb 200kb 300kb 1300kb 1400kb 400kb 500kb 600kb 700kb 800kb 900kb 1000kb 1100kb 1200kb 1500kb .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 13 Figure 1. Characterisation of ecDNA in COLO 320DM and its isolation by FACS. A. 368 Representative DNA FISH image of c -MYC localisation in RPE -1, COLO 320HSR and COLO 369 320DM. ecDNA or the HSR were labelled by FISH with a c-MYC probe. Slides were stained with 370 DAPI, 5-fluorescein (centromeric region of Chromosome 8) and 5-TAMRA (c- MYC). Scale bar, 10 371 μm. The cartoons provide a graphic representation of the MYC amplification in COLO 320HSR 372 and DM. B. Composite graph depicting the ecDNA structure in COLO 320DM generated using 373 short reads and optical genome mapping with Amplicon Reconstructor. G4 calls from G4Hunter 374 (36) above threshold 1.5 (purple dots); grey dotted lines are shown at G4Hunter scores 2.0 and 375 3.0); GC content (blue line); grey dotted line shown at 50%). C. Experimental workflow of the 376 FACS-based protocol for ecDNA isolation involving cell preparation, DNA release and FACS 377 sorting. D. Flow cytometry plots of chromosomes (top) and the region containing ecDNA and 378 debris (bottom) from RPE-1 (left), COLO 320HSR (centre) and COLO 320DM (right) cell lines. The 379 scales represent linear mean fluorescence intensity (MFI) but do not reflect detector voltage gains. 380 Chromosome detection was performed using a detector gain of 622V (DAPI) and 651V 381 (Chromomycin A3) for COLO 320DM; 622V (DAPI) and 604V (Chromomycin A3) for RPE -1; and 382 668V (DAPI) and 713V (Chromomycin A3) for COLO 320HSR. ecDNA detection was set using a 383 gain of 750V (DAPI) and 750V (Chromomycin A3) for COLO 320DM; 750V (DAPI) and 750V 384 (Chromomycin A3) for RPE-1; and 801V (DAPI) and 875V (Chromomycin A3) for COLO 320HSR. 385 E. Ratio of ONT long read counts of sequencing reads aligned to the ecDNA region. Left: whole 386 genome results excluding adaptive sampling sequencing, right: following FACS -based ecDNA 387 purification. n indicates number of biological replicates. Error bars represent standard deviation. 388 389 Existing methods for ecDNA isolation pose significant challenges, requiring ecDNA cleavage 390 or chromatin digestion, resulting in the extraction of limited data with high levels of noise (6, 391 39). To overcome these limitations, we refined established protocols for mitotic chromosome 392 karyotyping by employing fluorescence -activated cell sorting (FACS) at the limits of its 393 capabilities to detect and isolate ecDNA. In addition to previously tested approaches, we 394 introduced Latrunculin B treatment to reduce the internal structural integrity of the cells by 395 disrupting the actin cytoskeleton, which allowed us to omit the vortexing step. As a result, 396 mitotic DNA, including both chromosomes and ecDNA, was released with minimal shearing, 397 generating less debris and improving resolution, allowing for clearer visualisation of ecDNA 398 that would otherwise be obscured by larger, chromosome -derived, DNA fragments (Figure 399 1C). This method is efficient in visualising and sorting intact chromosomes in chromosomally 400 stable cell lines such as RPE-1 (Figure 1D), as well as in aneuploid cancer cell lines like COLO 401 320DM and COLO 320HSR (Figure 1D). 402 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 14 As anticipated, the substantial chromosomal heterogeneity observed in metaphase spreads of 403 COLO 320DM led to less distinct chromosome profiles (Figure 1D). However, by increasing 404 the voltage during sorting, we were able to detect low -molecular-weight DNA populations 405 exclusive to ecDNA-positive cell lines, which were absent in ecDNA-negative COLO 320HSR 406 and RPE-1 (Figure 1D), suggesting that these populations represent ecDNA rather than simple 407 debris. Subsequent sequencing and copy number analysis revealed a 11 to 40-fold enrichment 408 of ecDNA following FACS sorting compared to sequencing without prior isolation, further 409 confirming that the isolated DNA molecules were ecDNA (Figure 1E). 410 411 Replication Timing of ecDNA 412 To examine the timing of ecDNA replication, we employed the Repli -Seq technique, 413 commonly used to assess replication timing in chromosomal DNA (22). We used five FACS 414 gates across the cell cycle which provided the temporal resolution necessary not only to 415 distinguish between early - and late- replicating regions, but also to assess whether ecDNA 416 replication is synchronous within the population and whether it reflects the normal timing of 417 the region when in a chromosomal context. 418 We performed Repli -Seq in COLO 320DM, COLO 320HSR and RPE -1. Each cell line was 419 labelled with BrdU for 30 minutes, fixed in Carnoy’s fixative, stained with propidium iodide, 420 and subjected to sorting into 5 separate fractions before deep sequencing of each fraction 421 (Figure 2A). 422 To depict replication timing of ecDNA, we plotted the fraction of reads in windows of 10 kb 423 across the genome for each S-phase bin (Figure 2B). Examination of a representative genomic 424 region on chromosome 20 in the three cells lines suggested that they exhibit qualitatively 425 similar replication timing profiles, with minor experiment-specific differences (Figure 2B; see 426 Methods). To test this hypothesis across the genome, we measured replication synchronicity 427 using the normalised mean Rao’s quadratic entropy (RQE) (30, 31), which is an information 428 theory-based measure of distribution diversity, that considers the temporal ordering of the S -429 phase fractions (see Methods). In this analysis, a fragment replicating fully in a single fraction 430 of S phase has a RQE of zero, while a fragment replicating uniformly across all five subphases 431 achieves RQE of exactly 1. We plotted RQE of 1.6 Mb fragments (chosen to match the size of 432 the ecDNA in COLO320DM) across the whole genome (Figure 2C). While differences in the 433 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 15 distribution of synchronicity of replication timing are observed across the genome in the three 434 lines, as would be expected for transformed cells (40, 41), there is significant overlap in the 435 distributions. Notably, none of the 1.5Mb fragments sampled was found to replicate as 436 asynchronously as ecDNA (Figure 2B). 437 438 However, replication of the ecDNA region in COLO 320DM appears considerably less 439 synchronous than in the RPE-1 control with ongoing replication observed throughout the circle 440 in the entire S phase (Figure 2D). This observation is supported by an RQE of 0.96, close to 441 the maximum possible value of 1 and represents an outlier in the distribution of replication 442 timing synchronicities in all three cell lines (Figure 2C). The same region in COLO 320HSR, 443 in which the circle DNA is chromosomally integrated as an array, exhibits broadly late 444 replication timing throughout, in contrast to the wave of replication that broadly passes left to 445 right through the region in RPE -1 cells as S phase progresses (Figure 2D), but the overall 446 synchronicity of the region is similar to that seen in RPE-1. Thus, replication on the ecDNA in 447 COLO 320DM is significantly less ordered in terms of timing than the genome as a whole. 448 We next investigated replication fork velocity and origin distribution in the COLO320DM 449 ecDNA. 450 Replication origin distribution on ecDNA 451 Given the less synchronized replication observed in ecDNA (Figure 2), we explored whether 452 ecDNA employs a different number and/or position of origins compared with the 453 corresponding genomic sequence in a non -ecDNA context. We employed DNAscent (24) to 454 identify replication forks and origins of replication in the sequences containing c -MYC 455 amplified either on ecDNA or in the HSR. In brief, cells were incubated with two nucleoside 456 analogues EdU and BrdU for 6 minutes each, which were incorporated into newly synthesised 457 DNA during replication. The isolated DNA was then subjected to ultra-long ONT sequencing 458 (Figure 3A). The characteristic disruptions in electrical current caused by EdU and BrdU as 459 the labelled DNA passes through the sequencing pores allows the probability of their presence 460 to be calculated by the DNAscent model on top of the canonical DNA sequence. The 461 distribution of EdU and BrdU along the individual DNA molecules is indicated as a probability 462 (Figure 3B). Applying density -based segmentation to these probabilities creates replication 463 tracts that identify fork direction, velocity, stalling and replication origins (24). 464 465 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 16 466 467 Figure 2. ecDNA replicates throughout S phase in COLO 320DM. A. Graphical representation 468 of Repli -seq workflow. B. Replication timing analysis of a 10 Mb region on chromosome 20 469 (40,000,000 – 50,000,000) in RPE -1, COLO 320DM and COLO 320HSR. Coverage shown as 470 reads per kilobase per million mapped reads (RPKM), group normalised to the highest peak in the 471 visualised region between samples from the same cell line. Lower panels: visualisation of the 472 proportion of reads in each S phase bin as a function of genome location. C. Rao entropy of 473 replication timing. Replication timing was analysed using 1.6  Mb genomic bins (matching the 474 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 17 ecDNA size in COLO 320DM), spaced 10 Mb apart. The Rao entropy for each bin was calculated 475 for all three cell lines. Dashed lines indicate the Rao entropy of c -MYC gene in each respective 476 cell line. D. Replication timing analysis of the region around ecDNA locus on chromosome 8 in 477 RPE-1, COLO 320DM (DM) and COLO 320HSR (HSR). Coverage shown as RPKM, group 478 normalised to the highest peak in the visualised region between samples from the same cell line. 479 Lower panels: visualisation of the proportion of reads in each S phase bin as a function of genome 480 location. ecDNA region highlighted with red box. c -MYC labelled in red. Note that the ecDNA 481 regions in COLO 320DM and COLO 320HS R are shown aligned contiguously with adjacent 482 sequence on chromosome 8 although the integration of the sequence in COLO 320HSR is not on 483 chromosome 8 (Figure 1A). 484 485 We initially employed DNAscent on FACS-isolated ecDNA. While this allowed identification 486 of replication forks and replication origins in ecDNA (Supplementary Figure 1), the throughput 487 was not satisfactory and the need to concentrate the circle DNA from the relatively large 488 volumes generated by FACS sorting led to DNA breakage and relatively short reads in the 489 ONT sequencing runs (Supplementary Table 1). We therefore moved to using whole genome 490 sequencing using the ONT PromethION instrument. 491 492 Using PromethION sequencing on unselected DNA we were able to achieve an average N50 493 of 91.1 +/ - 16.7 kb (Supplementary Table 1). From this data, we identified 1312 reads with 494 origin of replication calls in COLO 320DM and 1115 reads with origin calls in COLO 320HSR 495 DNA. Of these, 81 and 43 respectively mapped to the 1.6 Mb interval covered by the ecDNA 496 (Supplementary Table 1). The sites to which origins were mapped in the ecDNA interval 497 exhibited no significant increase in GC content or G4 prevalence over randomly selected 498 regions from the same interval (Supplementary Figure 2 & 3). We examined whether the 499 distribution of origins is conserved between these amplification types, given differences in 500 replication timing (Figure 2D). There is not a significantly higher origin density in the ecDNA 501 of COLO 320DM compared with the equivalent interval in COLO 320HSR, consistent with 502 the genome-wide analysis (Supplementary Figure 4). Further specific analysis of the region 503 around c-MYC, which contains known sites of replication initiation (42), revealed origins near 504 the c-MYC and PVT-1 loci in both cell lines. However, these did not align with the origins 505 within c -MYC gene observed in other systems (Figure 3C & D), highlighting potential 506 differences in the replication patterns of the amplified c-MYC locus in COLO320 cell lines (43, 507 44) and consistent with the disorganised replication timing of this region in the COLO320 DM 508 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 18 ecDNA (Figure 2). Origin distribution across the ecDNA differed significantly from a uniform 509 distribution based on origin counts in 50kb segments for all 3 conditions as determined by a 510 one-sample Kolmogorov-Smirnov test (COLO 320DM: KS statistic = 0.91, p-value = 1.9e-34; 511 COLO 320HSR KS statistic = 0.76, p-value = 7.6e-9; COLO320 DM HU-treated KS statistic 512 = 0.92, p-value = 5.9e-36). 513 514 DNA replication is slower on ecDNA compared to chromosomal DNA 515 To further investigate the differences in replication dynamics between ecDNA and the 516 chromosomes, we measured replication fork velocity and stalling in ecDNA -positive and 517 ecDNA-negative cell lines. Using ONT sequencing and DNAscent, we detected individual 518 nascent DNA fragments labelled with EdU and BrdU (Figure 4A). By dividing the length of 519 continuous stretches of labelled DNA by the pulse durations we calculated replication fork 520 velocity - longer tracts indicate faster overall fork progression. Additionally, we examined the 521 pattern of signal decay to assess fork stalling: an abrupt loss of signal corresponds to replication 522 fork stalling or termination, while a gradual decline reflects normal fork progression as BrdU 523 levels diminish over time due to the thymidine chase (Figure 4A). This approach enabled us to 524 directly measure and compare replication fork dynamics on ecDNA and chromosomal DNA, 525 revealing statistically significant differences in fork velocity and stalling patterns (24). 526 527 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 19 528 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 20 Figure 3: Location of origins of replication. A. Schematic of base analogue pulsing protocol 529 followed by either FACS sorting or whole genome DNA extraction for subsequent ultra -long 530 Nanopore sequencing and DNAscent analysis. B. Representative raw base analogue 531 incorporation probabilities and DNAscent segmentation of origins of replication each represented 532 by four tracks (Upper tracks: Raw BrdU and EdU probabilities (prob.) at thymidine positions; 533 Lower tracks BrdU and EdU segmentation (seg.) derived from the raw probabilities). Both origins 534 map to ecDNA in the untreated COLO 320DM cell line. (Coordinates on ecDNA map to 535 chromosome chr8_126425747-127997820 bp. The top origin track maps within the reconstructed 536 ecDNA between 793,008 to 801,870 bp; the lower example to 470,941 to 470,941 bp). Raw 537 DNAscent probabilities for EdU or BrdU are shown on a scale from 0 to 1, segmentation is binary. 538 C. & D. Distribution of origins in C. COLO 320DM and D. Colo 320HSR on the ecDNA reference 539 map (Low density = light green shading to high density = dark green shading). Orange: SNS-seq 540 origin locations (43); yellow: Ini-seq 2 origins (44); green: DNAscent origin locations; Blue line: 541 GC content. Zoom in shows the region around c-MYC (1250 kb to 1350 kb). 542 543 Global fork velocity was reduced genome-wide in ecDNA-positive COLO 320DM compared 544 to ecDNA -null COLO 320HSR (Figure 4B). Moreover, ecDNA -positive cells displayed a 545 higher frequency of fork stalling (Figure 4C), though this increase in stalling did not account 546 for the reduced fork velocity since forks selected for matched levels of stalling still exhibited 547 reduced fork velocity in COLO 320DM (Supplementary Figure 5). Consistent with these 548 findings, replication forks on each chromosome in COLO 320HSR were progressing more 549 rapidly than on its counterpart in COLO 320DM cells (Figure 4D). 550 To further characterise ecDNA replication, we compared the replication dynamics of ecDNA 551 with its corresponding amplified sequence on the HSR in COLO 320HSR cells. Replication of 552 the ecDNA was approximately 6.5% slower (1.27 kb/min vs 1.18 kb/min), although no 553 significant difference in fork stalling rates was observed (Figure 4E & F). Similarly, ecDNA 554 replication in COLO 320DM was slower than chromosomal DNA replication (1.18 kb/min vs 555 1.25 kb/min; Figure 4G & H), with a marginally higher stalling rate (0.41 vs 0.38). 556 We next examined the distribution of replication fork velocities across COLO 320DM ecDNA 557 and the corresponding sequence in COLO 320HSR. Mean fork velocity, visualised in 20 kb 558 bins, ranged from 0.7 to 1.8 kb/min, with most regions showing concordant replication 559 velocities between ecDNA and HSR (Figure 4I & J). However, regions such as the c -MYC 560 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 21 locus exhibited stark differences in replication dynamics. While the c-MYC region was one of 561 the fastest -replicating in the HSR, it was the slowest -replicating on ecDNA, indicating 562 substantial differences in replication dynamics between ecDNA and chromosomal DNA 563 (Figure 4I & J). A similar pattern was observed for the stall rate (Figure 4I & J), with ecDNA 564 displaying elevated stalling in regions where HSR replication was most efficient. These 565 findings suggest that ecDNA replication diverges significantly from chromosomal HSR 566 replication, even in the regions with the same underlying sequence and, overall, consistent with 567 replication on ecDNA being more intrinsically stressed (17). 568 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 22 569 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 23 570 Figure 4. Replication dynamics on ecDNA. A. Representative DNAscent tracks for leftward -571 moving replication forks (BrdU probability (prob.) and segmentation (seg): red; EdU: blue. The top 572 example shows a stalled fork, characterised by a sharp drop-off in BrdU incorporation probability 573 at the fork tip (upper tracks; stall score of 0.9991, fork speed of 1.79 kb/min). The lower example 574 tracks show a replication fork with no fork stalling (stall score of 0.2657, fork speed 1.36 kb/min) 575 represented by a smooth decrease in BrdU probabilities at the fork tip. Both examples are from 576 the Colo 320HSR cell line within the ecDNA region from chromosome 8. Raw DNAscent 577 probabilities for EdU or BrdU are shown on a scale from 0 to 1, segmentation is binary. B. & C. 578 Comparison of genome-wide fork speeds (B) and stall scores (C) in COLO 320DM (DM; red) and 579 Colo 320HSR (HSR; blue). D. Median fork speed per chromosome in COLO 320DM (DM) and 580 COLO 320HSR (HSR) cell lines. For the COLO 320DM cell line, the median fork speed on the 581 ecDNA is shown as a red dot with black outline. E. & F. Comparison of fork speed (E) and stall 582 scores (F) of forks mapped to the ecDNA interval in COLO 320DM (circular, light red) and COLO 583 320HSR (chromosomally reintegrated, light blue). G. & H. Comparison of fork speed (G) and stall 584 scores (H) within COLO 320DM cell line between forks mapped to the ecDNA interval (dark 585 orange) and forks mapped to chromosomes (excluding chromosome 8 from which the ecDNA 586 originates, light orange). I. & J. Visualisation of replication fork speeds (I) and stall scores (J) 587 averaged across 20kb segments of the ecDNA interval in COLO 320DM and J. COLO 320HSR 588 cell line. Outer track, fork speeds; second track, stall score; third track, GC content (blue line); 589 Inner track, genes. Fork speeds and stall scores representation uses the same range in I and J 590 with the most extreme values across both cell lines determining the minimum and maximum 591 colour shades. All p-values for boxplots with a fork speed are obtained from a two-sided Welch’s 592 t-test with no assumption of equal variances and all p -values in boxplots for stall scores are 593 obtained from a two-sided non-parametric Wilcoxon Rank Sum test. Statistical significance: ns = 594 not significant (p ≥ 0.05), * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001. 595 596 Hydroxyurea disrupts ecDNA replication, leading to its depletion from cells 597 Consistent with previous studies (18‑21), hydroxyurea treatment led to a marked reduction in 598 ecDNA levels in COLO 320DM cells (Figure 5A). This was reversed on withdrawing HU 599 (Figure 5A), suggesting strong selective pressure to maintain ecDNA in these cells. To further 600 explore this phenomenon, we analysed ecDNA replication dynamics under HU treatment using 601 DNAscent. 602 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 24 First, we investigated changes in origin distribution under HU treatment, as reduced fork 603 velocity can trigger activation of dormant origins to compensate for slower replication (45, 46). 604 Upon HU treatment, we observed increased origin activation both genome-wide and on ecDNA 605 (Supplementary Figure 4), particularly around the c-MYC locus (Supplementary Figure 2). This 606 shows that dormant origins are activated both genome -wide and on ecDNA under replication 607 stress. 608 Hydroxyurea treatment also resulted in a genome -wide reduction in fork velocity of 20.7% 609 (Figure 5B; 0.99 vs 1.25 kb/min), with every chromosome exhibiting decreased replication 610 speed (Figure 5D). Notably, DNAscent-detected fork stalling also decreased in HU treated cells 611 (Figure 5C). This suggests that the reduced velocity of DNA synthesis resulting from 612 nucleotide depletion leads to fewer sharply cut -off tracts of DNA synthesis, which likely 613 represent stalled forks (24), is perhaps counterintuitive. However, such resolution of stalls has 614 previous not been accessible to techniques like DNA combing due to their inability to 615 distinguish between gradual decline of the signal, and abrupt signal loss suggesting stalling 616 (47). The observed reduction in the stall score may reflect either the slower forks in HU less 617 frequently encountering synthesis impediments, or the slower replication allows more 618 opportunity to deal with problems on the template ‘on-the-fly’ allowing synthesis to continue. 619 This reduction in fork velocity, observed both on ecDNA and genome -wide, may also lead to 620 underreplication, which might be mitigated by the increased activation of dormant origins in 621 response to HU-induced replication stress. 622 When comparing specifically ecDNA replication under HU treatment, the replication slowing 623 effect was even more pronounced. Fork velocity on ecDNA dropped significantly by 22.9% 624 (0.91 vs 1.18 kb/min), making it the slowest replicating fraction relative to all chromosomes 625 (Figure 5D & E). The stalling rate also decreased from 0.41 to 0.31 (Figure 5F), indicating that 626 stalled forks do not account for the overall reduction in replication speed. Even under HU 627 treatment, ecDNA replication remained slower than chromosomal DNA (Figure 5G; 0.91 vs 1 628 kb/min), with an even higher relative decrease in replication velocity (5.6% vs 9% reduction) 629 (Supplementary Figure 7). Notably, the stalling rate across both fractions dropped to nearly 630 identical levels (0.31 vs 0.30) (Figure 5H). These effects were observed in both ecDNA and 631 chromosomal DNA but to different extents. Replication fork velocity decreased proportionally 632 more on ecDNA, while the stalling rate was similarly reduced in both fractions (Supplementary 633 Figure 7), which may contribute to the loss of ecDNA from cells under HU treatment. 634 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 25 Further analysis of 20 kb fragments across ecDNA confirmed a significant reduction in both 635 replication velocity and stalling (Figure 5I & J). However, the distribution of fast - and slow-636 replicating regions shifted markedly. Regions surrounding c -MYC and PVT -1, previously 637 among the slowest replicating areas, became the fastest replicating upon HU treatment, even 638 surpassing their pre -treatment speeds. This suggests that HU has its greatest impact on the 639 fastest replicating regions, indicating a complex, region -specific influence on ecDNA 640 replication. In contrast, the stalling rate across the other regions of ecDNA decreased more 641 uniformly, suggesting a differential impact of HU on replication dynamics that may be driven 642 by the variations in underlying sequence or local chromatin environment (Figure 5I & J). 643 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 26 644 Fork speed 0.8 1.2 1.5 Stall score 0.2 0.6 0kb 100kb 200kb 300kb 1300kb 1400kb 400kb 500kb 600kb 700kb 800kb 900kb 1000kb 1100kb 1200kb 1500kb COLO 320DM HU 0kb 100kb 200kb 300kb 1300kb 1400kb 400kb 500kb 600kb 700kb 800kb 900kb 1000kb 1100kb 1200kb 1500kb COLO 320DM ** ** ** **** ** * * 100 80 60 40 20 0 0 5 10 15 20 25 Time (days) Normalised ecDNA copy number HU 50µM HU washoff (day 10) control Fork speed (kb.min-1) 0.0 0.5 1.0 1.5 2.0 2.5 3.0 Stall score 0.2 0.4 0.6 0.8 1.0 0.0 **** **** 1.25 0.99 n=12078 n=20409 n=20409n=12078 DM DM +HU 0.39 0.30 DM DM +HU Fork speed (kb.min-1) 0.0 0.5 1.0 1.5 2.0 2.5 3.0 ******** genome (all chr) genome (all chr) ecDNA mapped ecDNA mapped Stall score 0.2 0.4 0.6 0.8 1.0 0.0 1.18 0.91 0.41 0.31 n = 907 n = 887 DM DM +HU n = 907 n = 887 DM DM +HU Fork speed (kb.min-1) 0.0 0.5 1.0 1.5 2.0 2.5 3.0 Stall score 0.2 0.4 0.6 0.8 1.0 0.0 n = 887 n = 18677 **** n = 887 n = 18677 0.91 1.00 0.300.31 DM +HU ecDNA DM +HU genome DM +HU ecDNA DM +HU genome DM +HU ecDNA vs genome DM +HU ecDNA vs genome * ecDNA chr1 chr2 chr3 chr4 chr5 chr6 chr7 chr8 chr9 chr10 chr11 chr12 chr13 chr14 chr15 chr16 chr17 chr18 chr19 chr20 chr21 chr22 chr23 mean median 1.30 1.25 1.20 1.15 1.10 1.05 1.00 Fork speed median (kb.min-1) 0.95 0.90 DM DM +HU Sequence GC content c−MYC PVT1 PCAT1 A B C D E F G H I J .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 27 Figure 5. Hydroxyurea treatment slows down DNA replication on ecDNA. A. Copy number 645 variation, estimated by qPCR, in COLO 320DM with HU treatment (green), upon HU removal (blue) 646 and without treatment (red). Pair-wise comparisons are assessed with a two-sided Wilcoxon Rank 647 Sum test. B. & C. Fork speeds (C) and stall scores (D) in untreated (red) and HU -treated (green) 648 COLO 320DM cell lines. D. Median fork speed per chromosome in untreated (DM) and HU treated 649 (DM_HU) COLO 320DM cell lines. The median fork speed on ecDNA is shown as a red dot with a 650 black edge. E. & F. Comparison of fork speed (F) and stall scores (G) stall on ecDNA region 651 mapped forks in untreated (medium red) and HU-treated (medium green) COLO 320DM cell line. 652 G. & H. Comparison of fork speed (H) and stall scores (I) in forks mapped to ecDNA (medium 653 green) and forks mapped to chromosomes (excluding chromosome 8), light green from HU -654 treated COLO 320DM cells. I. & J. Visualisation of replication fork speeds and stall scores 655 averaged across 20kb segments in the ecDNA interval of untreated (J) and HU-treated (K) COLO 656 320DM cells. Fork speeds lower than the average fork speed (light grey) are shaded in blue, higher 657 in red and stall score averages are all low to moderate (light to medium grey). GC content is shown 658 as a blue line and position of c -MYC and PVT1 genes are highlighted by pink and blue blocks, 659 respectively. Colouring of fork speeds and stall scores is the same in I and J with the most extreme 660 values across both cell lines setting the minimum and maximum colour shades. All p values for 661 panel A are obtained from a Wilcoxon Rank Sum test. All p-values for boxplots with a fork speed 662 are obtained from a two -sided Welch’s t-test with no assumption of equal variances and all p -663 values in boxplots for stall scores are obtained from a two-sided non-parametric Wilcoxon Rank 664 Sum test. Statistical significance: ns = not significant (p ≥ 0.05), * p < 0.05, ** p < 0.01, *** p < 665 0.001, **** p < 0.0001. 666 667

668 This study reveals that ecDNA replicates asynchronously throughout the S phase, contrasting 669 with the well -defined replication timing of linear chromosomal DNA. This asynchrony 670 suggests that ecDNA lacks the strict temporal regulation of replication initiation found in 671 chromosomal DNA, possibly due to its previously reported generally accessible euchromatic 672 nature (7, 48). 673 Given this largely asynchronous pattern of replication on ecDNA, we explored whether ecDNA 674 employs a different number or arrangement of replication origins, compared with the 675 corresponding chromosomal sequence. This property of ecDNA replication could influence 676 both replication dynamics and potential replication-transcription conflicts. The analysis of the 677 distribution of replication origins revealed an increased number of origins on ecDNA upon HU 678 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 28 treatment. The activation of additional origins on ecDNA under replication stress, suggests that 679 the well-documented compensatory mechanism by which cryptic origins are fired to ensure 680 complete replication despite slower fork progression (45, 46) is active on ecDNA. This 681 increased origin firing may also reflect the less strict replication pattern of ecDNA, act as an 682 adaptation to increased replication stress and may also contribute to replication –transcription 683 conflicts, given the high transcriptional activity of oncogenes on ecDNA (2, 7, 17). 684 The mildly reduced fork velocity and elevated stalling rates on ecDNA under normal conditions 685 suggest that replication forks on ecDNA are constitutively subject to some degree of replication 686 stress. Hydroxyurea (HU) treatment results in an expected significant decrease in fork speed, 687 but surprisingly also causes a reduction in stalling, suggesting that slower replication forks are 688 less prone to complete stalling of DNA synthesis. This finding contrasts with previous reports 689 that HU treatment is linked to increased fork stalling (49). One explanation is linked to the fact 690 that previously used DNA fibre combing analyses could not distinguish between slower 691 replication and increased stalling. In contrast, our observations reveal that the reduced fork 692 velocity reflects genuine slowing of the replication fork rather than increased stalling, as 693 DNAscent enables the differentiation of a gradual decay of the signal - suggesting slower 694 progression - and a sudden stop, suggesting fork stalling. This distinction cannot be achieved 695 using traditional DNA fibre methods. 696 We also hypothesise that the depletion of ecDNA under HU treatment is likely due to its 697 vulnerability to replication stress, as suggested by more marked slowing of replication on 698 ecDNA compared with chromosomal DNA in the presence of HU. This supports the potential 699 therapeutic strategy of inducing replication stress to selectively target ecDNA-bearing tumour 700 cells (11, 17, 19). 701 Applying DNAscent to ecDNA allowed us to map replication origins with high precision on 702 single DNA molecules without requiring amplification, reducing noise and artefacts associated 703 with other high-throughput methods. Our study offers new insights into the distinct replication 704 dynamics of ecDNA. In addition, we developed a new method for ecDNA isolation, FINE, 705 which minimises ecDNA processing and offers high specificity, potentially enabling the study 706 of ecDNA properties such as chromatin composition and three -dimensional structure through 707 proteomics and imaging. While FINE is limited by relatively low yields of ecDNA in large 708 volumes and incompatibility with some replication assays like Repli -seq, it provides a 709 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 29 potentially valuable tool for ecDNA detection and characterisation, e.g. through proteomics or 710 structural analysis. 711 While our study sheds light on ecDNA replication dynamics, it raises questions about the 712 molecular mechanisms governing replication initiation and regulation on ecDNA. The roles of 713 specific proteins or epigenetic marks in the altered replication timing and origin activation 714 patterns remain to be elucidated. Notably, ecDNA exhibits a more open chromatin 715 conformation than chromosomal DNA (7), which may facilitate replication by improving the 716 accessibility to the replication machinery, or conversely, impede it by generating high levels 717 of replication-transcription conflicts (17). Furthermore, the mechanisms governing replication 718 licensing on ecDNA remains unresolved. If, consistent with the previous reports (16), ecDNA 719 is replicated only once per cell cycle, its copy number can increase only through random 720 segregation and subsequent selection, rather than through multiple rounds of replication (50). 721 In any case, understanding ecDNA licensing remains a fundamental issue that requires further 722 investigation. Additionally, the interplay between replication and transcription machinery on 723 ecDNA warrants further investigation, particularly given the potential for replication –724 transcription conflicts to contribute to genomic instability (17, 51) . We hypothesise that 725 increased transcription (7, 17) together with slower ecDNA replication (particularly under HU 726 treatment), and the euchromatic state of ecDNA may collectively contribute to the dysregulated 727 origin landscape. 728 Finally, ecDNA biology exhibits radical heterogeneity between cell lines and even among 729 individual ecDNA molecules. Therefore, future studies should examine the applicability of 730 these findings across cancer types and ecDNA compositions. Understanding whether the 731 vulnerabilities of ecDNA replication can be generalized will be crucial for developing targeted 732 therapeutic interventions. The application of advanced replication analysis techniques and 733 improved ecDNA isolation methods opens new avenues for research into diverse aspects of 734 ecDNA biology and its role in cancer progression. Investigating how cancer cells compensate 735 for ecDNA loss and whether they can develop resistance to replication-targeting therapies will 736 determine future treatment strategies. 737 738 739 740 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 30 741 Data Availability 742 743 Sequencing data can be accessed at the Gene Expression Omnibus archive with the accession 744 number GSE186675. The code for the ini -seq 2 origin caller can be found at 745 https://github.com/Sale-lab. The ONT sequencing data can be found on ENA under the 746 accession number PRJEB83636. The code for the DNAscent -associated analyses is publicly 747 available under https://github.com/Pfuderer/ecDNA_replication_dynamics. 748 749

750 751 We would like to thank the FACS facility at the MRC Laboratory of Molecular Biology, 752 especially P. A. Penttilä, M. Daly, Y. Li, F. Zhang, and D. Nolan, for support. We also thank 753 Alastair Crisp for support with AmpliconArchitect ecDNA structure analysis, Welcome Sanger 754 Sequencing R&D facility, in particular Iraad Bronner for Nanopore and OGM sequencing, and 755 members of the Sale and Chin labs for helpful discussions. 756 757 Contributions 758 759 J.J.J. and J.E.S. conceptualised the project. J.J.J., G.P., P.L.P., J.E.S., and M.A.B. designed the 760 experiments. J.J.J. performed wet lab experiments and conducted bioinformatic Repli -seq 761 analyses. P.L.P. carried out bioinformatic analyses of DNA replication fork dynamics. P.C. 762 performed replication timing statistical analyses and visualisation. J.J.J. and P.L.P. wrote the 763 manuscript with input from all authors. J.E.S. and M.A.B. supervised the project. 764 765 Funding 766 767 This work was supported by the MRC grant (MC_U105178808; J.J.J., J.S.). J.J.J. was 768 supported by Boehringer Ingelheim Fonds PhD Fellowship. P.L.P. was funded by the Cancer 769 Research UK Cambridge Centre (C9685/A25117) via a Cancer Research UK Cambridge 770 Centre PhD Studentship to PLP. P.C. was supported through the ETH AI Centre fellowship 771 programme. This work was performed using resources provided by the Cambridge Service for 772 Data Driven Discovery (CSD3) operated by the University of Cambridge Research Computing 773 Service (www.csd3.cam.ac.uk), provided by Dell EMC and Intel using Tier -2 funding from 774 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 31 the Engineering and Physical Sciences Research Council (capital grant EP/T022159/1), and 775 DiRAC funding from the Science and Technology Facilities Council (www.dirac.ac.uk). G.P. 776 was funded by a Marie Skłodowska -Curie European Postdoctoral Fellowship (897663). The 777 funders had no role in study design, data collection and analysis, decision to publish or 778 preparation of the manuscript. 779 780 Conflict of interest statement. JES is Senior Executive Editor of NAR. PLP received funding 781 from Oxford Nanopore Technologies to present parts of this work at the UK DNA Replication 782 Meeting. GP is co -founder of, and holds stock in, Alia Therapeutics, a genome editing 783 company. 784 785 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 32

786 787 1. Storlazzi,C.T., Lonoce,A., Guastadisegni,M.C., Trombetta,D., D’Addabbo,P., 788 Daniele,G., L’Abbate,A., Macchia,G., Surace,C., Kok,K., Ullmann,R., 789 Purgato,S., Palumbo,O., Carella,M., Ambros,P.F. and Rocchi,M. (2010) Gene 790 amplification as double minutes or homogeneously staining regions in solid 791 tumors: origin and structure. Genome Res 20, 1198-1206. 792 2. Turner,K.M., Deshpande,V., Beyter,D., Koga,T., Rusert,J., Lee,C., Li,B., 793 Arden,K., Ren,B., Nathanson,D.A., Kornblum,H.I., Taylor,M.D., Kaushal,S., 794 Cavenee,W.K., Wechsler-Reya,R., Furnari,F.B., Vandenberg,S.R., Rao,P.N., 795 Wahl,G.M., Bafna,V. and Mischel,P.S. (2017) Extrachromosomal oncogene 796 amplification drives tumour evolution and genetic heterogeneity. Nature 543, 797 122-125. 798 3. Verhaak,R.G.W., Bafna,V. and Mischel,P.S. (2019) Extrachromosomal 799 oncogene amplification in tumour pathogenesis and evolution. Nat Rev Cancer 800 19, 283-288. 801 4. Bailey,C., Pich,O., Thol,K., Watkins,T.B.K., Luebeck,J., Rowan,A., Stavrou,G., 802 Weiser,N.E., Dameracharla,B., Bentham,R., Lu,W.-T., Kittel,J., Yang,S.Y.C., 803 Howitt,B.E., Sharma,N., Litovchenko,M., Salgado,R., Hung,K.L., Cornish,A.J., 804 Moore,D.A., Houlston,R.S., Bafna,V., Chang,H.Y., Nik-Zainal,S., Kanu,N., 805 McGranahan,N., Genomics England Consortium, Flanagan,A.M., Mischel,P.S., 806 Jamal-Hanjani,M. and Swanton,C. (2024) Origins and impact of 807 extrachromosomal DNA. Nature 635, 193-200. 808 5. Kim,H., Nguyen,N.P., Turner,K., Wu,S., Gujar,A.D., Luebeck,J., Liu,J., 809 Deshpande,V., Rajkumar,U., Namburi,S., Amin,S.B., Yi,E., Menghi,F., 810 Schulte,J.H., Henssen,A.G., Chang,H.Y., Beck,C.R., Mischel,P.S., Bafna,V. and 811 Verhaak,R.G.W. (2020) Extrachromosomal DNA is associated with oncogene 812 amplification and poor outcome across multiple cancers. Nat Genet 52, 891-897. 813 6. Koche,R.P., Rodriguez-Fos,E., Helmsauer,K., Burkert,M., MacArthur,I.C., 814 Maag,J., Chamorro,R., Munoz-Perez,N., Puiggròs,M., Dorado Garcia,H., Bei,Y., 815 Röefzaad,C., Bardinet,V., Szymansky,A., Winkler,A., Thole,T., Timme,N., 816 Kasack,K., Fuchs,S., Klironomos,F., Thiessen,N., Blanc,E., Schmelz,K., 817 Künkele,A., Hundsdörfer,P., Rosswog,C., Theissen,J., Beule,D., Deubzer,H., 818 Sauer,S., Toedling,J., Fischer,M., Hertwig,F., Schwarz,R.F., Eggert,A., 819 Torrents,D., Schulte,J.H. and Henssen,A.G. (2020) Extrachromosomal circular 820 DNA drives oncogenic genome remodeling in neuroblastoma. Nat Genet 52, 29-821 34. 822 7. Wu,S., Turner,K.M., Nguyen,N., Raviram,R., Erb,M., Santini,J., Luebeck,J., 823 Rajkumar,U., Diao,Y., Li,B., Zhang,W., Jameson,N., Corces,M.R., Granja,J.M., 824 Chen,X., Coruh,C., Abnousi,A., Houston,J., Ye,Z., Hu,R., Yu,M., Kim,H., 825 Law,J.A., Verhaak,R.G.W., Hu,M., Furnari,F.B., Chang,H.Y., Ren,B., Bafna,V. 826 and Mischel,P.S. (2019) Circular ecDNA promotes accessible chromatin and 827 high oncogene expression. Nature 575, 699-703. 828 8. Hung,K.L., Jones,M.G., Wong,I.T.-L., Curtis,E.J., Lange,J.T., He,B.J., 829 Luebeck,J., Schmargon,R., Scanu,E., Brückner,L., Yan,X., Li,R., Gnanasekar,A., 830 Chamorro González,R., Belk,J.A., Liu,Z., Melillo,B., Bafna,V., Dörr,J.R., 831 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 33 Werner,B., Huang,W., Cravatt,B.F., Henssen,A.G., Mischel,P.S. and Chang,H.Y. 832 (2024) Coordinated inheritance of extrachromosomal DNAs in cancer cells. 833 Nature 635, 201-209. 834 9. Lange,J.T., Rose,J.C., Chen,C.Y., Pichugin,Y., Xie,L., Tang,J., Hung,K.L., 835 Yost,K.E., Shi,Q., Erb,M.L., Rajkumar,U., Wu,S., Taschner-Mandl,S., 836 Bernkopf,M., Swanton,C., Liu,Z., Huang,W., Chang,H.Y., Bafna,V., 837 Henssen,A.G., Werner,B. and Mischel,P.S. (2022) The evolutionary dynamics of 838 extrachromosomal DNA in human cancers. Nat Genet 54, 1527-1533. 839 10. Shoshani,O., Brunner,S.F., Yaeger,R., Ly,P., Nechemia-Arbely,Y., Kim,D.H., 840 Fang,R., Castillon,G.A., Yu,M., Li,J.S.Z., Sun,Y., Ellisman,M.H., Ren,B., 841 Campbell,P.J. and Cleveland,D.W. (2021) Chromothripsis drives the evolution 842 of gene amplification in cancer. Nature 591, 137-141. 843 11. Nathanson,D.A., Gini,B., Mottahedeh,J., Visnyei,K., Koga,T., Gomez,G., 844 Eskin,A., Hwang,K., Wang,J., Masui,K., Paucar,A., Yang,H., Ohashi,M., Zhu,S., 845 Wykosky,J., Reed,R., Nelson,S.F., Cloughesy,T.F., James,C.D., Rao,P.N., 846 Kornblum,H.I., Heath,J.R., Cavenee,W.K., Furnari,F.B. and Mischel,P.S. (2014) 847 Targeted therapy resistance mediated by dynamic regulation of 848 extrachromosomal mutant EGFR DNA. Science 343, 72-76. 849 12. Deng,X., Zhang,L., Zhang,Y., Yan,Y., Xu,Z., Dong,S. and Fu,S. (2006) Double 850 minute chromosomes in mouse methotrexate-resistant cells studied by atomic 851 force microscopy. Biochem Biophys Res Commun 346, 1228-1233. 852 13. Hahn,P.J., Nevaldine,B. and Longo,J.A. (1992) Molecular structure and 853 evolution of double-minute chromosomes in methotrexate-resistant cultured 854 mouse cells. Mol Cell Biol 12, 2911-2918. 855 14. Kaufman,R.J., Brown,P.C. and Schimke,R.T. (1979) Amplified dihydrofolate 856 reductase genes in unstably methotrexate-resistant cells are associated with 857 double minute chromosomes. Proceedings of the National Academy of Sciences 858 76, 5669-5673. 859 15. Levan,A. and Levan,G. (1978) Have double minutes functioning centromeres? 860 Hereditas 88, 81-92. 861 16. Barker,P.E., Drwinga,H.L., Hittelman,W.N. and Maddox,A.M. (1980) Double 862 minutes replicate once during S phase of the cell cycle. Exp Cell Res 130, 353-863 360. 864 17. Tang,J., Weiser,N.E., Wang,G., Chowdhry,S., Curtis,E.J., Zhao,Y., Wong,I.T.-865 L., Marinov,G.K., Li,R., Hanoian,P., Tse,E., Mojica,S.G., Hansen,R., Plum,J., 866 Steffy,A., Milutinovic,S., Meyer,S.T., Luebeck,J., Wang,Y., Zhang,S., 867 Altemose,N., Curtis,C., Greenleaf,W.J., Bafna,V., Benkovic,S.J., Pinkerton,A.B., 868 Kasibhatla,S., Hassig,C.A., Mischel,P.S. and Chang,H.Y. (2024) Enhancing 869 transcription-replication conflict targets ecDNA-positive cancers. Nature 635, 870 210-218. 871 18. Shimizu,N., Nakamura,H., Kadota,T., Kitajima,K., Oda,T., Hirano,T. and 872 Utiyama,H. (1994) Loss of amplified c-myc genes in the spontaneously 873 differentiated HL-60 cells Cancer Research 54, 3561-3567. 874 19. Von Hoff,D.D., McGill,J.R., Forseth,B.J., Davidson,K.K., 875 Bradley,T.P.R.V.D.D., Van Devanter,D.R. and Wahl,G.M. (1992) Elimination of 876 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 34 extrachromosomally amplified MYC genes from human tumor cells reduces their 877 tumorigenicity. Proceedings of the National Academy of Sciences 89, 8165-8169. 878 20. Von Hoff,D.D., Waddelow,T., Forseth,B., Davidson,K., Scott,J. and Wahl,G. 879 (1991) Hydroxyurea accelerates loss of extrachromosomally amplified genes 880 from tumor cells Cancer research 51, 6273-6279. 881 21. Eckhardt,S.G., Dai,A., Davidson,K.K., Forseth,B.J., Wahl,G.M. and Von 882 Hoff,D.D. (1994) Induction of differentiation in HL60 cells by the reduction of 883 extrachromosomally amplified c-myc. Proceedings of the National Academy of 884 Sciences 91, 6674-6678. 885 22. Zhao,P.A., Sasaki,T. and Gilbert,D.M. (2020) High-resolution Repli-Seq defines 886 the temporal choreography of initiation, elongation and termination of replication 887 in mammalian cells. Genome Biol 21, 76. 888 23. Marchal,C., Sasaki,T., Vera,D., Wilson,K., Sima,J., Rivera-Mulia,J.C., Trevilla-889 García,C., Nogues,C., Nafie,E. and Gilbert,D.M. (2018) Genome-wide analysis 890 of replication timing by next-generation sequencing with E/L Repli-seq. Nat 891 Protoc 13, 819-839. 892 24. Boemo,M.A. (2021) DNAscent v2: detecting replication forks in nanopore 893 sequencing data with deep learning. BMC Genomics 22, 430. 894 25. Schindelin,J., Arganda-Carreras,I., Frise,E., Kaynig,V., Longair,M., Pietzsch,T., 895 Preibisch,S., Rueden,C., Saalfeld,S., Schmid,B., Tinevez,J.-Y., White,D.J., 896 Hartenstein,V., Eliceiri,K., Tomancak,P. and Cardona,A. (2012) Fiji: an open-897 source platform for biological-image analysis Nature Methods 9, 676-682. 898 26. Deshpande,V., Luebeck,J., Nguyen,N.D., Bakhtiari,M., Turner,K.M., Schwab,R., 899 Carter,H., Mischel,P.S. and Bafna,V. (2019) Exploring the landscape of focal 900 amplifications in cancer using AmpliconArchitect. Nat Commun 10, 392. 901 27. Ryba,T., Battaglia,D., Pope,B.D., Hiratani,I. and Gilbert,D.M. (2011) Genome-902 scale analysis of replication timing: from bench to bioinformatics. Nat Protoc 6, 903 870-895. 904 28. Li,H., Handsaker,B., Wysoker,A., Fennell,T., Ruan,J., Homer,N., Marth,G., 905 Abecasis,G., Durbin,R. and 1000,G.P.D.P.S. (2009) The Sequence 906 Alignment/Map format and SAMtools. Bioinformatics 25, 2078-2079. 907 29. Robinson,J.T., Thorvaldsdóttir,H., Winckler,W., Guttman,M., Lander,E.S., 908 Getz,G. and Mesirov,J.P. (2011) Integrative genomics viewer. Nat Biotechnol 909 29, 24-26. 910 30. Botta‐Dukát,Z. (2005) Rao’s quadratic entropy as a measure of functional 911 diversity based on multiple traits Journal of Vegetation Science 16, 533-540. 912 31. Rao,C.R. (1980) Diversity and dissimilarity coefficients: A unified approach 913 Theoretical Population Biology 21, 24-43. 914 32. Shannon,C.E. (1948) A mathematical theory of communication The Bell System 915 Technical Journal 27, 379-423. 916 33. Jones,M.J.K., Rai,S.K., Pfuderer,P.L., Bonfim-Melo,A., Pagan,J.K., Clarke,P.R., 917 McClelland,S.E. and Boemo,M.A. (2022) A high-resolution, nanopore-based 918 artificial intelligence assay for DNA replication stress in human cancer cells 919 bioRxiv 4. 920 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 35 34. Quinn,L.A., Moore,G.E., Morgan,R.T. and Woods,L.K. (1979) Cell lines from 921 human colon carcinoma with unusual cell products, double minutes, and 922 homogeneously staining regions Cancer Research 39, 4914-4924. 923 35. Gu,Z., Gu,L., Eils,R., Schlesner,M. and Brors,B. (2014) circlize Implements and 924 enhances circular visualization in R. Bioinformatics 30, 2811-2812. 925 36. Bedrat,A., Lacroix,L. and Mergny,J.L. (2016) Re-evaluation of G-quadruplex 926 propensity with G4Hunter. Nucleic Acids Res 44, 1746-1759. 927 37. Bodnar,A.G., Ouellette,M., Frolkis,M., Holt,S.E., Chiu,C.-P., Morin,G.B., 928 Harley,C.B., Shay,J.W., Lichtsteiner,S. and Wright,W.E. (1998) Extension of 929 life-span by introduction of telomerase into normal human cells Science 279, 930 349-352. 931 38. Luebeck,J., Coruh,C., Dehkordi,S.R., Lange,J.T., Turner,K.M., Deshpande,V., 932 Pai,D.A., Zhang,C., Rajkumar,U., Law,J.A., Mischel,P.S. and Bafna,V. (2020) 933 AmpliconReconstructor integrates NGS and optical mapping to resolve the 934 complex structures of focal amplifications. Nat Commun 11, 4374. 935 39. Hung,K.L., Luebeck,J., Dehkordi,S.R., Colón,C.I., Li,R., Wong,I.T., Coruh,C., 936 Dharanipragada,P., Lomeli,S.H., Weiser,N.E., Moriceau,G., Zhang,X., Bailey,C., 937 Houlahan,K.E., Yang,W., González,R.C., Swanton,C., Curtis,C., Jamal-938 Hanjani,M., Henssen,A.G., Law,J.A., Greenleaf,W.J., Lo,R.S., Mischel,P.S., 939 Bafna,V. and Chang,H.Y. (2022) Targeted profiling of human extrachromosomal 940 DNA by CRISPR-CATCH. Nat Genet 54, 1746-1754. 941 40. Dileep,V. and Gilbert,D.M. (2018) Single-cell replication profiling to measure 942 stochastic variation in mammalian replication timing. Nat Commun 9, 427. 943 41. Rivera-Mulia,J.C. and Gilbert,D.M. (2016) Replication timing and 944 transcriptional control: beyond cause and effect-part III. Curr Opin Cell Biol 40, 945 168-178. 946 42. Vassilev,L. and Johnson,E.M. (1990) An initiation zone of chromosomal DNA 947 replication located upstream of the c-myc gene in proliferating HeLa cells. Mol 948 Cell Biol 10, 4899-4904. 949 43. Akerman,I., Kasaai,B., Bazarova,A., Sang,P.B., Peiffer,I., Artufel,M., Derelle,R., 950 Smith,G., Rodriguez-Martinez,M., Romano,M., Kinet,S., Tino,P., Theillet,C., 951 Taylor,N., Ballester,B. and Méchali,M. (2020) A predictable conserved DNA 952 base composition signature defines human core DNA replication origins. Nat 953 Commun 11, 4826. 954 44. Guilbaud,G., Murat,P., Wilkes,H.S., Lerner,L.K., Sale,J.E. and Krude,T. (2022) 955 Determination of human DNA replication origin position and efficiency reveals 956 principles of initiation zone organisation. Nucleic Acids Res 50, 7436-7450. 957 45. Ge,X.Q., Jackson,D.A. and Blow,J.J. (2007) Dormant origins licensed by excess 958 Mcm2-7 are required for human cells to survive replicative stress. Genes Dev 21, 959 3331-3341. 960 46. Ibarra,A., Schwob,E. and Méndez,J. (2008) Excess MCM proteins protect human 961 cells from replicative stress by licensing backup origins of replication 962 Proceedings of the National Academy of Sciences 105, 8956-8961. 963 47. Hyrien,O., Guilbaud,G. and Krude,T. (2025) The double life of mammalian 964 DNA replication origins. Genes Dev 39, 304-324. 965 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint 36 48. Pope,B.D., Ryba,T., Dileep,V., Yue,F., Wu,W., Denas,O., Vera,D.L., Wang,Y., 966 Hansen,R.S., Canfield,T.K., Thurman,R.E., Cheng,Y., Gülsoy,G., Dennis,J.H., 967 Snyder,M.P., Stamatoyannopoulos,J.A., Taylor,J., Hardison,R.C., Kahveci,T., 968 Ren,B. and Gilbert,D.M. (2014) Topologically associating domains are stable 969 units of replication-timing regulation. Nature 515, 402-405. 970 49. Petermann,E., Orta,M.L., Issaeva,N., Schultz,N. and Helleday,T. (2010) 971 Hydroxyurea-stalled replication forks become progressively inactivated and 972 require two different RAD51-mediated pathways for restart and repair. Mol Cell 973 37, 492-502. 974 50. Ilić,M., Zaalberg,I.C., Raaijmakers,J.A. and Medema,R.H. (2022) Life of double 975 minutes: generation, maintenance, and elimination. Chromosoma 131, 107-125. 976 51. Helmrich,A., Ballarino,M. and Tora,L. (2011) Collisions between replication and 977 transcription complexes cause common fragile site instability at the longest 978 human genes. Mol Cell 44, 966-977. 979 980 981 982 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 23, 2025. ; https://doi.org/10.1101/2025.03.22.644567doi: bioRxiv preprint