Targeted profiling of human extrachromosomal DNA by CRISPR-CATCH

Abstract

Extrachromosomal DNA (ecDNA) is a common mode of oncogene amplification but is challenging to analyze. Here, we adapt CRISPR-CATCH, in vitro CRISPR-Cas9 treatment and pulsed field gel electrophoresis of agarose-entrapped genomic DNA, previously developed for bacterial chromosome segments, to isolate megabase-sized human ecDNAs. We demonstrate strong enrichment of ecDNA molecules containing EGFR, FGFR2 and MYC from human cancer cells and NRAS ecDNA from human metastatic melanoma with acquired therapeutic resistance. Targeted enrichment of ecDNA versus chromosomal DNA enabled phasing of genetic variants, identified the presence of an EGFRvIII mutation exclusively on ecDNAs and supported an excision model of ecDNA genesis in a glioblastoma model. CRISPR-CATCH followed by nanopore sequencing enabled single-molecule ecDNA methylation profiling and revealed hypomethylation of the EGFR promoter on ecDNAs. We distinguished heterogeneous ecDNA species within the same sample by size and sequence with base-pair resolution and discovered functionally specialized ecDNAs that amplify select enhancers or oncogene-coding sequences.

Fig. 1. Isolation of megabase-sized ecDNA and its native chromosomal locus from the same cancer cell sample by CRISPR-CATCH.

a, Experimental workflow for enrichment of ecDNA and its corresponding chromosomal locus from the same cell sample. b, A representative DNA FISH image on a metaphase spread from a GBM39 glioblastoma cell showing extrachromosomal EGFR signals and multiple chromosome 7 (chr7) signals (n = 65 cells). Quantification of copy numbers is shown in Extended Data Fig. 2a. DAPI, 4,6-diamidino-2-phenylindole. c, Design of CRISPR sgRNAs for linearizing ecDNA circles or extracting the native chromosomal locus. d, PFGE images showing linearized ecDNA molecules and the chromosomal locus after treatment with indicated guides (Methods; guide sequences in Supplementary Table 1). Boxed regions indicate parts of the gel that were extracted for DNA isolation. GBM39 ecDNA cutting and fractionation by PFGE were reproduced in three independent experiments. e, Normalized short-read sequencing coverage of the expected ecDNA locus in unenriched WGS or after CRISPR-CATCH (guide A). f, Fraction of total sequencing reads aligning to the expected ecDNA locus in unenriched WGS or after CRISPR-CATCH (guide A). g, Sequencing tracks showing coverages for enriched ecDNA and its chromosomal locus at the zoomed-in locations compared to WGS. Orange arrows indicate locations of sgRNA targets. Source data

Fig. 2. Isolation of ecDNA from a flash-frozen metastatic melanoma tumor.

a, Melanoma patient treatment timeline. NRAS ecDNAs were detected in cutaneous metastasis by AmpliconArchitect. Human figure was created with BioRender.com. b, A schematic for the tumor processing and electrodepletion protocol for preparing tumor DNA for CRISPR-CATCH. c, Normalized short-read sequencing coverage of the expected NRAS ecDNA in melanoma patient tumor (Pt9) after CRISPR-CATCH (guide 194; guide sequence in Supplementary Table 1). Amplicon size from sequencing (890.9 kb) was in agreement with molecule size shown by PFGE (750–945 kb). d, Sequencing coverage of an NRAS G12R mutation identified on ecDNA (top). Sequencing reads supporting single-nucleotide variant (SNV) identification (bottom). VAF, variant allele frequency.

Fig. 3. Phasing of SVs and SNVs for ecDNA and its native chromosomal locus identified the chromosomal origin of ecDNA.

a, Isolation of ecDNA and the corresponding chromosomal locus from GBM39 neurospheres by CRISPR-CATCH followed by mutation analysis using short-read sequencing. b, Barplot showing relative sequencing coverage of ecDNA (guide A) and chromosomal DNA (guide E + F) (left) and variant allele frequencies (VAFs) of the EGFRvIII mutant on ecDNA and chromosomal DNA (middle). Sequencing coverage and junction reads supporting the EGFRvIII mutation and wild-type (WT; right). c, Bimodal distribution of VAFs of SNVs identified within the ecDNA-amplified region in bulk WGS (top). VAFs of SNVs classified by CRISPR-CATCH as either ecDNA-specific or chromosome-specific (bottom). d, Schematic of chromosomes with or without deletion of the ecDNA-amplified region (top). Density plots showing VAFs of non-homozygous SNVs (VAF < 0.99) in WGS 20 Mb upstream or downstream of the region corresponding to the ecDNA amplicon on chromosomes (bottom left). VAF of the SV resulting from deletion of the ecDNA amplicon region and reference sequence without deletion in WGS (bottom right). e, Sequence of genomic events leading to ecDNA amplification and chromosome 7 copy gain in GBM39 cells (top). Visualization of all allele-specific genetic variants on ecDNA and chromosomal DNA and their parental alleles of origin identified by CRISPR-CATCH (bottom). Source data

Fig. 4. Comparison of CpG methylation statuses of ecDNA and its native chromosomal locus showed hypomethylation of gene promoters on ecDNA.

a, Isolation of ecDNA (guide A) and the corresponding chromosomal locus (guides E + F) from GBM39 neurospheres by CRISPR-CATCH followed by detection of 5mC-CpG methylation by nanopore sequencing. b, Negative correlation between mean methylation probabilities of ATAC-seq peaks and their ATAC-seq signals (Pearson’s R, two-sided test; error bands represent 95% confidence intervals). c, Aggregated levels of relative CpG methylation of ecDNA compared to the chromosomal locus at top 50 ATAC-seq peaks in the ecDNA-amplified region. Mean methylation frequencies were calculated in 100-bp windows sliding every 10 bp. Relative frequencies were quantified from standardized residuals for a linear regression model for mean frequencies on ecDNA vs. chromosomal DNA (Methods). d, Bulk ATAC-seq track with differentially methylated regions annotated (Methods; two-sided z-test, P values were Benjamini-Hochberg adjusted; regions with P < 0.005 were considered significant). e, Relative CpG methylation of ecDNA compared to the chromosomal locus in differential regions and concordance with accessibility by ATAC-seq and nucleosome positioning by MNase-seq. Mean methylation frequencies were calculated in 100-bp windows sliding every 10 bp. Relative frequencies were quantified from standardized residuals for a linear regression model for mean frequencies on ecDNA vs. chromosomal DNA (Methods). f, From top to bottom: Loess-smoothed methylation probability around the EGFR promoter (error band represents 95% confidence intervals); nanopore sequencing reads showing CpG methylation calls (gray denotes regions with no CpG sites); heatmap showing co-occurrence probabilities of unmethylated CpG sites on the same molecules; heatmap showing co-occurrence probabilities of methylated CpG sites on the same molecules (Methods). Reads were collected using a MinION sequencer (Oxford Nanopore Technologies).

Fig. 5. Identification of diverse ecDNA species revealed heterogeneous structural rearrangements and an altered enhancer landscape.

a, Analysis of ecDNA structure using CRISPR-CATCH. ecDNA species are separated by size in PFGE and sequenced. AmpliconArchitect generates CN-aware breakpoint graphs, which are used in combination with molecule sizes from PFGE to find paths and identify candidate ecDNA structures. b, PFGE image for SNU16 after treatment with independent sgRNAs targeting either the FGFR2 or MYC locus (guide sequences in Supplementary Table 1). PFGE result is representative of two independent experiments. Bands passing all quality filters for reconstruction are shown in blue. c, From top to bottom: WGS, ATAC-seq, BRD4 and H3K27ac chromatin immunoprecipitation with sequencing (ChIP-seq); heatmap showing enrichment of multiple structurally distinct ecDNA species by CRISPR-CATCH. ecDNA species were isolated from bands shown in PFGE gels in panel b and Extended Data Fig. 8c. Orange arrows on the top mark all sgRNA target sites. d–f, ecDNA reconstructions using CRISPR-CATCH data (outer rings indicate coordinates along reference genome and gene bodies, and thin gray bands mark connections between sequence segments). OM patterns (orange rings) and assembled contigs (blue rings, contig IDs indicated) validated CRISPR-CATCH reconstructions. Orange arrows mark sgRNA target sites. An FGFR2 ecDNA structure containing the full amplicon locus was reconstructed from band ‘i’ as shown in panel d. A MYC ecDNA reconstructed from band ‘d’ containing sequences from chromosomes 8 and 11 is shown in panel e. A short-form FGFR2 ecDNA reconstructed from band ‘m’ is shown in panel f. g, Sequencing coverage of ecDNAs (bands 15 and 30 correspond to bands extracted from gel in Extended Data Fig. 8c). Region highlighted in green denotes enhancer amplification. ATAC-seq, BRD4 and H3K27ac ChIP-seq show locations of enhancers. Orange arrows mark sgRNA target sites. h, Schematic showing diverse ecDNA structures and an altered enhancer landscape revealed by CRISPR-CATCH. FC, fold change. Source data

Extended Data Fig. 1. Agarose entrapment of genomic DNA preserves intact ecDNA.

(a) A PFGE image showing DNA fragmentation after in-solution HMW DNA isolation as compared with intact agarose-embedded DNA trapped in the loading well. DNA fragmentation was reproduced in two independent experiments. (b) A PFGE image showing complete digestion of fragmented in-solution HMW DNA after a 5-day exonuclease treatment. One independent experiment was performed. (c) Analysis of ecDNA amplicon sizes predicted by AmpliconArchitect in TCGA tumor samples. (d) A PFGE image showing size ladders and GBM39 ultrahigh-molecular weight (UHMW) genomic DNA without in vitro CRISPR-Cas9 linearization (representative of three independent experiments). UHMW DNA was trapped in the loading well and the upper compression zone. Source data

Extended Data Fig. 2. Enrichment of circular ecDNA by CRISPR-CATCH.

(a) Left: quantification of EGFR and chromosome 7 copy numbers in GBM39 cells using DNA FISH on metaphase spreads (n = 65 cells; box center line, median; box limits, upper and lower quartiles; box whiskers, 1.5× interquartile range). Right: number of GBM39 cells with 4, 5, 6, 7, or 8 copies of chromosome 7. An example FISH image is shown in Fig. 1b. (b) Full sequencing tracks showing coverage for isolated ecDNA and its chromosomal locus at the EGFR amplified region compared to WGS. Zoomed-in tracks are shown in Fig. 1f. Orange arrows indicate locations of sgRNA targets. (c) Chromosomal overhangs from chromosome-targeting guides (guides C-H) outside of the ecDNA-amplified region were used for calculating sequencing coverage of the chromosomal allele. The mean coverage of the 5’ and 3’ chromosomal overhangs was calculated. The coverage of ecDNA alleles was calculated by subtracting chromosomal coverage from total coverage in the ecDNA-amplified region. (d) Relative sequencing coverage of chromosomal DNA and ecDNA alleles in WGS or CRISPR-CATCH samples.

Extended Data Fig. 3. Tumor processing and ecDNA enrichment from patient tumor samples using CRISPR-CATCH.

(a) A PFGE image showing presence of DNA bands from S. cerevisiae and H. wingei DNA size markers with or without electrodepletion. One independent experiment was performed. (b) A PFGE image showing linearized ecDNA molecules from SNU16 cells containing FGFR2 ecDNAs after electrodepletion and treatment with an FGFR2 guide (guide 17; guide sequence in Supplementary Table 1). One independent experiment was performed. (c) AmpliconArchitect breakpoint graph from bulk WGS of melanoma patient tumor Pt9 showing amplification of NRAS. (d) A PFGE image from melanoma patient sample Pt9 after electrodepletion and CRISPR-CATCH using NRAS-targeting guide 194 (guide sequence in Supplementary Table 1). Brackets on the right correspond to gel-extracted regions shown in Fig. 2c. One independent experiment was performed. (e) Top: raw Hi-C contact heatmap for the SK-MEL-5 melanoma cell line (40-kb resolution). Bottom: sequencing track showing CRISPR-CATCH-enrichment of the NRAS ecDNA from melanoma patient tumor Pt9. (f) Layered H3K27ac ChIP-seq tracks from 7 cell lines (GM12878, H1-hESC, HSMM, HUVEC, K562, NHEK, NHLF) in ENCODE using the UCSC Genome Browser. Brown arc marks ecDNA breakpoints. Shaded brown region marks the NRAS ecDNA amplicon detected in patient sample Pt9. Source data

Extended Data Fig. 4. Phasing of SNVs for ecDNA and its native chromosomal locus.

(a) VAFs of SNVs identified in the ecDNA-amplified region and its native chromosomal locus in various CRISPR-CATCH treatments. Letters denote sgRNAs used (A-H). (b) Left: VAFs of two ecDNA- and chromosome-specific SNVs in WGS, isolated ecDNA or chromosomal molecules using CRISPR-CATCH. Right: sequencing reads supporting SNV identification. (c) VAFs of SNVs classified as ecDNA- or chromosome-specific SNVs in various CRISPR-CATCH treatments. Black lines connect identical SNVs detected in WGS and indicated CRISPR-CATCH treatments. Low-frequency allele-specific SNVs are defined as SNVs with VAFs < 0.5 and are marked in red. Horizontal lines in lilac in the chromosome SNV plot represent levels of chromosomal enrichment corresponding to Extended Data Fig. 2d. (d) VAFs of a low-frequency somatic ecDNA SNV in WGS, isolated ecDNA or chromosomal molecules using CRISPR-CATCH.

Extended Data Fig. 5. Quantification of 5mC-CpG methylation probability of ecDNA and the native chromosomal locus.

(a) Aggregated CpG methylation probability of ecDNA and chromosomal DNA at the top 50 ATAC-seq peaks with highest coverage in the amplified region. Mean methylation frequencies were calculated in 100-bp windows sliding every 10 bp. (b) Linear regression model of mean methylation probabilities of ecDNA vs chromosomal DNA. Mean methylation probabilities were calculated in 100-bp windows sliding every 10 bp in the ecDNA-amplified region. Each point represents a window mean. Brown arrow demonstrates the standardized residual of a data point from the regression line. (c) Relative CpG methylation of ecDNA compared to the chromosomal locus in differential regions shown as absolute differences in methylation frequencies. Regions shown correspond to differentially methylated regions in Fig. 4d,e. Mean methylation frequencies were calculated in 100-bp windows sliding every 10 bp. Normalized sequencing coverage tracks are shown on the bottom of each plot. (d) Relative CpG methylation of ecDNA compared to the chromosomal locus and nucleosome positioning by MNase-seq, zooming into indicated gene promoters. Regions shown correspond to differentially methylated regions in Fig. 4d,e. Mean methylation frequencies and MNase-seq coverage were calculated in 100-bp windows sliding every 10 bp. Relative frequencies were quantified from standardized residuals for a linear regression model for mean frequencies on ecDNA vs chromosomal DNA (Methods).

Extended Data Fig. 6. Reconstruction of a 1.258 Mb ecDNA from GBM39 neurospheres.

(a) AmpliconArchitect breakpoint graphs for CRISPR-CATCH-isolated ecDNAs using guides A and/or B as in Fig. 1 (guide sequences in Supplementary Table 1). (b) Reconstructed ecDNA circles from CRISPR-CATCH data using independent sgRNAs showing equivalent ecDNA structures (outer rings; thin gray bands mark connections between sequence segments). Sequencing coverage is shown along the reconstructed circle (inner rings). Orange arrows mark sgRNA target sites. Coordinate tick marks are printed in 10-kb units. AmpliconArchitect segment IDs and orientations are annotated.

Extended Data Fig. 7. CRISPR-CATCH enables disambiguation of heterogeneous structural rearrangements on individual ecDNA species.

(a) AmpliconArchitect breakpoint graph from bulk WGS of stomach cancer SNU16 cells showing significantly amplified sequences from chromosomes 8, 10, and 11. (b) An example of an AmpliconArchitect breakpoint graph for a CRISPR-CATCH-separated ecDNA species (band ‘d’) from SNU16 cells showing greatly simplified breakpoints connecting only sequences from chromosomes 8 and 11. Gray vertical lines represent genomic coverage from WGS data and black horizontal lines indicate the estimated copy number of the region. Colored arcs represent breakpoint junctions, and the orientation of those junctions is specified by the color. Red and brown arcs preserve the orientation of the genome, with red reflecting breakpoints supported by reads in the proper orientation and brown reflecting breakpoints supported by reads in the everted orientation. Teal and magenta arcs indicate breakpoints leading to a change in genome orientation before and after the breakpoint where teal breakpoints are supported by both paired-end reads mapping to the forward strand and magenta breakpoints are supported by both paired-end reads mapping to the reverse strand.

Extended Data Fig. 8. Enrichment of multiple ecDNA species from the SNU16 stomach cancer cell line.

(a) Sequencing coverage of multiple ecDNA species from SNU16 cells after CRISPR-CATCH isolation at the FGFR2, MYC and CD44 loci. Bands a-w correspond to extracted bands shown in Fig. 5b. Bands corresponding to unresolved DNA content in the compression zone are labeled CZ. (b) ecDNA reconstruction using CRISPR-CATCH data (outer rings; thin gray bands mark connections between sequence segments). Optical mapping patterns (orange rings) and assembled contigs (blue rings, contig IDs indicated) validated CRISPR-CATCH reconstructions. Orange arrow marks sgRNA target site. Shown is an FGFR2 ecDNA structure reconstructed from band ‘p’, equivalent to that reconstructed from band ‘i’ (Fig. 5d) using an independent sgRNA. (c) PFGE image for SNU16 after treatment with independent sgRNAs targeting either the FGFR2 or MYC gene bodies or enhancers (FGFR2 gene body: guide 17; MYC gene body: guide 5; guide sequences in Supplementary Table 1). One independent experiment was performed. (d) Short-read sequencing coverage tracks of multiple ecDNA species from SNU16 cells after CRISPR-CATCH isolation at the FGFR2, MYC and CD44 loci. Bands 1–44 correspond to extracted bands shown in c. Orange arrows mark sgRNA target sites. Source data

Extended Data Fig. 9. Validation of ecDNA species in SNU16 cells mapped by CRISPR-CATCH.

(a) Sequencing coverage of ecDNAs isolated from SNU16 cells (bands extracted from gel in Extended Data Figure 8c). Region highlighted in purple is connected to MYC or FGFR2. ATAC-seq, BRD4 and H3K27ac ChIP-seq show that an enhancer is located in the rearranged region. Orange arrows mark sgRNA target sites. (b) Left: ecDNA species targeted by dual-color FISH. Right: representative two-color DNA FISH image on a metaphase spread showing instances of specialized ecDNAs containing either FGFR2 (green) or the enhancer region (red, identified in Fig. 5g, chr10:122988480–123026871), as well as ecDNA species with colocalized oncogene and enhancers (n = 69 cells). (c) From top to bottom: WGS coverage of ecDNA-amplified regions; connected DNA segments on ecDNAs identified by CRISPR-CATCH (boxes 1–4 mark coordinates targeted by pairs of FISH probes in panel e and f); unnormalized background signals from chromatin conformation capture using H3K27ac HiChIP; connected DNA segments predicted from WGS data using AmpliconArchitect. (d) Levels of unnormalized HiChIP interactions between inter-chromosomal DNA segments and their co-occurrence on ecDNA as identified by CRISPR-CATCH compared to WGS. Connected ecDNA segments identified by CRISPR-CATCH were strongly supported by HiChIP signals. (e) Top: FISH probes targeting either the chromosome 8 or 10 segment located on ecDNAs in SNU16 cells. Bottom: representative two-color DNA FISH images on metaphase spreads for quantifying colocalization of the chromosome 8 and 10 ecDNA segments marked in the CRISPR-CATCH heatmap in c (regions 1–4). Red DNA FISH probe targets MYC. Green DNA FISH probes target the following: region 1, chr10:122309127–122477445 (n = 11 cells); region 2, chr10:122635712–122782544 (n = 11 cells); region 3, chr10:122973293–123129601 (n = 12 cells); region 4, chr10:123300005–123474433 (n = 11 cells). (f) Frequencies of red-green colocalized FISH signals (probe pairs 1–4 correspond to regions targeted in e). The number of colocalized over total signals and the number of cells assessed are shown above each bar.

Extended Data Fig. 10. Recommended usage of CRISPR-CATCH.

A recommended workflow for using CRISPR-CATCH in complement to WGS, DNA FISH and optical mapping for analysis of ecDNAs in cancer samples.